The primary objective of the European Space Agency’s 7th Earth Explorer mission, BIOMASS, is to determine the worldwide distribution of forest above-ground biomass (AGB) in order to reduce the major uncertainties in calculations of carbon stocks and fluxes associated with the terrestrial biosphere, including carbon fluxes associated with Land Use Change, forest degradation and forest regrowth. It also has important secondary objectives, viz. sub-surface mapping in arid zones, icesheet motion, production of a “bare earth” Digital Terrain Model, and mapping of ionospheric structure along its dawn-dusk orbit. The mission is expected to launch in mid-2024, and will consist of three mission phases: (1) the initial 6-month Commissioning Phase; (2) a Tomographic Phase (TomoSAR) of ~14 months, which will give a single global tomographic coverage; and (3) the Interferometric Phase (PolInSAR), which occupies the rest of the 5-year lifetime of the mission, and will produce 4-5 global coverages with dual-baseline polarimetric interferometry, each requiring ~7 months. Since the BIOMASS radar is not permitted to operate within line-of-sight of the US DoD Space Object Tracking Radars (SOTR), “global” as used here excludes N America, Europe and the Arctic, but includes the whole tropical belt, the whole of Eurasia east of the Urals, Asia, the Middle East and the whole Southern hemisphere. The SOTR exclusion zone thus has little effect on the primary objectives of the mission. A major part of the Commissioning Phase concerns characterisation of the instrument, in particular radiometric calibration, measurements of its cross-talk and channel imbalance properties, and estimation of the 2-D antenna pattern. This relies on the use of a single transponder in New Norcia, Western Australia (sites in Europe or N America are not possible because of SOTR restrictions). Algorithms to estimate the polarimetric properties of the radar sensor from the transponder data have been developed, but are complex because they must account for the significant Faraday rotation (up to several tens of degrees) likely at this location, which lies at magnetic mid-latitudes. Because of the special orbits needed for TomoSAR and PolinSAR, the transponder will only be revisited at times separated by several months during the mission, unless special calibration orbits are scheduled. Routine radiometric and polarimetric monitoring of instrument performance therefore requires the use of natural targets or targets of opportunity. Another task in the Commissioning Phase is therefore to investigate the properties of candidate natural targets, including Dome-C in Antarctica and desert sites (e.g. in N Sudan) for radiometric monitoring, and extended natural targets to estimate polarimetric performance and stability. The algorithms to use extended targets for monitoring the polarimetric performance of the sensor are still being refined. Although the algorithms to produce the three main BIOMASS products (above-ground biomass [AGB], forest height and severe forest disturbance) are currently well-defined, they have differing areas of weakness and error. AGB estimation relies on cancellation of the ground and double-bounce return in the data (Mariotti d'Alessandro et al., 2020). This isolates the canopy return, for which the intensity of the HH, VV and HV polarizations is described by a three-parameter scattering model whose parameters need to be inferred from the data. Calibration is then required using reference data, which currently is planned to be provided by GEDI data. This requires estimates of the mean AGB from GEDI in the 200 m grid cells of BIOMASS, together with their estimated Standard Error (SE). If this SE is too large then the GEDI data cannot sufficiently constrain the inversion. As a result, the GEDI reference data are sparse and methods to propagate into regions without such data form a fundamental part of the AGB algorithm, as described in Soja et al. (2023). The forest height product is currently the most robust of the three main products, though its error properties still need to be fully characterised. Forest disturbance detection is based on testing for significant change in the polarimetric covariance matrix (Conradsen et al. (2016). Although this is statistically well-founded, attribution of detected change is problematic since environmental changes (e.g. moisture changes in the canopy or soil) can trigger a detection. Work has been performed to identify if the nature of the scattering has changed (e.g. changes in the ratio of volume and surface scattering) but a consolidated algorithm using this information is not yet in place. Three other areas that need better definition are the validation plans for the three products, the science exploitation plan, and the role of BIOMASS in an overall global observing system. Major advances have been made in establishing in situ and airborne lidar datasets that can be used for validation (e.g. see https://geo-trees.org/), but a formal validation plan has not yet been articulated. Research into how BIOMASS time series can improve estimates of carbon fluxes has been carried out, but in the absence of actual time series data this is still speculative. Nonetheless, with the mission being only a year away it is becoming urgent to set up the studies and structures allowing immediate exploitation of BIOMASS data. Finally, the observational environment in which BIOMASS will sit is completely different from when the mission was proposed, with many other relevant sensors and numerous published biomass products from space-based data. It is clear that our quantification of biomass and its changes, and hence meeting the BIOMASS primary objective, relies on intelligently combining these other data with the unique capabilities of BIOMASS. Although this can only be accomplished once the BIOMASS data are available, available campaign and satellite datasets already allow us to begin exploring methods to take this more complete view of the role of BIOMASS. Conradsen K, Nielsen AA and Skriver H (2016). Determining the Points of Change in Time Series of Polarimetric SAR Data, IEEE Trans Geosci. Remote Sensing, 54, No. 5, 3007-3024 Mariotti d'Alessandro, M et al. (2020). Interferometric ground cancellation for above ground biomass estimation, IEEE Trans. Geosci. Remote Sensing, doi:10.1109/TGRS.2020.2976854 Soja M J et al (2023). The global forest biomass estimation algorithm for ESA’s BIOMASS mission, PolinSAR-BIOMASS 2023

The development phase of the Biomass mission is nearing completion. The satellite is assembled and has entered its system environmental test campaign. Three major tests - mechanical, thermal vacuum and electromagnetic compatibility - are executed in this campaign to verify that the satellite can sustain the launch and space environments and will function properly therein. The mechanical tests which simulate the launch loads have already been performed successfully. The satellite needs to be opened one more time to make some late exchanges in the radar instrument. These changes do not invalidate the mechanical test campaign but need to be performed before the thermal vacuum test. In parallel to these activities, the work on the various elements of the ground segment is peaking. Specific to Biomass is the ground calibration transponder, a 5 m antenna placed in New Norcia, Australia which will be assembled there later this year to primarily support the commissioning phase and the performance monitoring throughout the lifetime of the mission. Substantial attention is also given to the finalization of data processors and the consolidation of the mission performance. The presentation will give the latest development status and provide the outlook towards launch which is planned for Q3 2024.

Biomass is a fully-polarimetric interferometric P-Band SAR mission with a centre frequency of 435 MHz and a 6 MHz bandwidth. The primary objective of Biomass is to determine the worldwide distribution of forest above-ground biomass in order to reduce the major uncertainties in calculations of carbon stocks and fluxes associated with the terrestrial biosphere. The Biomass satellite is planned to be launched in June 2024. This talk will present an overview of the Biomass key performance parameters and performance drivers and the in-Orbit calibration and performance verification approach. The Biomass P-band SAR will operate in a quad-Pol mode where the V- polarisation and H-polarisation pulses are transmitted alternatively while backscattered signals are received simultaneously in both polarisations. The critical polarimetric performance parameters are the combined Tx and Rx Channel Imbalance between V and H and the Cross Talk. Another key performance parameter directly linked to the mission’s primary objective of Biomass retrieval is the radiometric stability which is required to be better than 0.5 dB at 1 sigma. The table below summarizes the main mission performance figures for the Biomass mission. Performance parameter Requirement Geolocation accuracy 25 m Noise Equivalent Sigma Nought (NESN) -27.0 dB Total Ambiguity Ratio (TAR) -18.0 dB Spatial Resolution 60 m (cross-track) 50 m (along-track) Radiometric Bias 0.3 dB Radiometric Stability 0.5 dB Cross-talk -30 dB Channel Imbalance -25 dB (Tx and Rx combined) Peak SideLobe Ratio (PSLR) -16.0 dB (along-track), -16.0 dB (cross-track) Integrated SideLobe Ratio (ISLR) -9.0 dB The in-orbit commissioning and performance verification activities are currently under definition. The in-orbit commissioning will rely on a dedicated fully polarimetric active transponder which will be installed at the ESA New Norcia Site, Western Australia. In addition, Targets of Opportunity will be imaged. Three commissioning parts with different activities are currently foreseen. Following the platform and instrument in-orbit tests, the in-flight antenna pattern characterisation using the transponder begins. This activity focuses on acquiring 21 azimuth antenna pattern cuts at different elevation angles as the Biomass satellite transits over the transponder in a drifting orbit with a revisit every 3 days. In this phase, the transponder will be operated in the Signal Acquisition mode. The Antenna pattern characterisation consists of the measurement of the patterns generated by each single patch of the Biomass satellite Feed Array (doublet). Thus for each overpass, 4 different antenna pattern azimuth cuts are collected. The 21 overpasses provide doublet patterns measurements at different elevations. For each doublet, the 2D antenna pattern (only main lobe) is determined by means of a polynomial fitting. The resulting antenna patterns are then averaged with their corresponding reference patterns obtained from on-ground characterisation (RASP) and the final H and V patterns are obtained combining the doublet characterised patterns. Following the antenna pattern characterisation, the satellite will be put in a zero-drift orbit which allows to perform repeated acquisitions over selected target areas with the same geometry. The focus is to characterise the radiometric stability and other performance parameters by means of the Persistent Scatterer calibration technique. Also, information on the temporal de-correlation of natural targets (as acquired during the operational phase) will be obtained. Following this, the satellite will again be placed in a drifting orbit with the transponder being revisited every 3 days at different elevation angles and in each one of the 3 swaths. In this phase, the calibration and verification of the key performance parameters will take place primarily based on transponder measurements, supplemented by Targets of Opportunities. The transponder will be operated in the echo generation mode. This mode enables the re-transmission of four echoes for each received pulse from Biomass corresponding to the four possible polarimetric signatures. Finally, the satellite will be put into the tomographic orbit to start the operational phase.

Biomass is a fully-polarimetric interferometric P-Band SAR mission with a centre frequency of 435 MHz and a 6 MHz bandwidth. The primary objective of Biomass is to determine the worldwide distribution of forest aboveground biomass in order to reduce the major uncertainties in calculations of carbon stocks and fluxes associated with the terrestrial biosphere. The Biomass satellite is planned to be launched in June 2024. The Biomass mission will last at least 5 years after commissioning, and consists of two phases, i.e. a Tomographic (TOM) and an Interferometric phase (INT). The INT phase will provide interferometric polarimetric observations to achieve the mission objectives. SAR tomography allows the investigation of forested areas by resolving the vertical structure of the forest. To optimize the performance of Polarimetric Interferometric (PolInSAR) retrievals over the full range of forest heights, a dual baseline concept has been implemented. This concept results in three baseline pairs, one long and two short baseline pairs. During the TOM phase, six spatial baselines (i.e. seven SAR measurements) will be acquired. A key driver in the orbit selection are the revisit time requirements for repeat-pass interferometry and tomography. Temporal decorrelation is mainly due to environmental and scatterers response changes occurring between the acquisitions. Since it cannot be controlled or compensated, it is the most critical performance parameter in a repeat-pass PolInSAR. Forest height inversion using PolInSAR requires the repeat interval to be short enough to maintain high temporal coherence between SAR acquisitions. Large baselines between acquisitions allow to compensate for errors introduced by temporal and other decorrelation sources. For Biomass a baseline of 30% and of 15% of the critical baseline, during INT and TOM phases respectively have been selected as trade-off. These baselines are achieved by means of drifting orbits. The Biomass mission will operate in a near 3 day repeat cycle orbit in both TOM and INT phases in order to minimize temporal de-correlation. To minimise the ionospheric effects, the local time of the orbital node at the equator has been chosen around 06:00 AM/PM sun synchronous orbit. To minimise temporal de-correlation, the Biomass mission will operate in an orbit with a repeat cycle of 3 days and a cycle length of 44 orbits, which corresponds to 14+2/3 revolutions per day. To reach global coverage, a three swaths imaging approach is implemented, achieved by using spacecraft attitude roll and combined with satellite repositioning manoeuvres (SRM). The SRMs are performed by placing the spacecraft in a higher orbit to drift relative to the nominal ground track. After the drift phase, the spacecraft is lowered back to the nominal orbit in order to access the next ground region to be imaged.

This abstract has been accepted at the Polinsar&Biomass workshop in the Session: Biomass Mission - Overview

This presentation has been accepted in the session Biomass Mission - Products and Algorithms

Scheduled for launch around mid 2024, ESA BIOMASS is designed to acquire Synthetic Aperture Radar (SAR) multiple passes to support the scientific community in gaining a better understanding of global forest status [1]. BIOMASS will be the first spaceborne P-band SAR, featuring full polarimetry, repeat pass interferometry and tomography. The P-band allows sensitivity to the woody elements of trees, polarimetry and multiple baselines allow polarimetric interferometry and tomography for the retrieval of forest structure and changes over time. The orbits are designed to acquire globally subject to Space Object Tracking Radars (SOTR) restrictions, which exclude North America and Europe from coverage. The acquisition timeline is studied to have a few months Commissioning Phase, followed by about 14 months of Tomographic (TOM) phase and about five years of operational Interferometric (INT) Phase. During TOM seven passes, ascending and descending, will be acquired for each location (with a repeat period of four days), enabling tomography, whereas during INT a dual-baseline configuration is specifically conceived for PolInSAR. P-band also allows unprecedented science opportunities, like under-vegetation topography retrieval and sub-surface desert and ice penetration. In order to generate forest products it is paramount to first calibrate the acquisitions and remove all residual disturbances that prevent from correct retrieval. Level-1 (L1) processor [2] already foresees polarimetric-radiometric calibration and a partial compensation of Ionosphere, nonetheless a finer level of calibration is needed for interferometric processing. In this presentation the design of the BIOMASS inteferometric calibration processor is outlined. The processor is composed of five main modules, briefly described in the following. The first module performs coregistration of the stacks in two steps, first from an external DEM, then refined coregistration from data to correct for residual orbital inaccuracies and Ionosphere-induced shifts. A model inversion of the estimated shifts from incoherent speckle tracking prevents from generating artifacts [3]. The background Ionosphere module applies split-spectrum processing [4] to estimate and remove low-pass differential ionospheric phase screens. Due to BIOMASS narrow bandwidth, only the linear component of the screens is retrieved and averaging over a large area is required to meet a satisfactory accuracy. The baseline correction module implements model inversion of dual-squint interferometric phase [3] to recover residual differential orbit inaccuracies. The model assumes dual-squint interferometric phase is mostly insensitive to topography, troposphere and scene displacements. The number of BIOMASS independent looks allows estimating only some components of the baseline errors, with preference on retrieving the constant along-track error which cannot be retrieved from coregistration. The fast Ionosphere module relies on partial L1 correction [2] and applies tomographic reconstruction techniques to the multi-squint interferometric phase [5] to estimate fast Ionosphere. Partial refocusing is applied within compensation to improve coherence. The correction is applied at high latitudes, where fast Ionosphere variations are more likely to occur. The last module performs additional phase calibration based on SKP decomposition [6] and generates phase screens containing both terrain topography and residual low-pass disturbances. The application of these screens is optional and aimed to produce stacks ready for Level 2 processing (i.e., retrieval of forest products) and SAR tomography [6]. Currently the processor is being developed and validated by unit testing each module with simulations and campaign acquisitions. Airborne acquisitions are reprocessed to mimic BIOMASS parameters, meaning the signal is filtered to 6MHz and BIOMASS azimuth bandwidth. The processor is currently being developed for operational use, and is planned to be validated against end-to-end simulations. References - --------- [1] Francesco Banda, Davide Giudici, Thuy Le Toan, Mauro Mariotti d’Alessandro, Kostas Papathanassiou, Shaun Quegan, Guido Riembauer, Klaus Scipal, Maciej Soja, Stefano Tebaldini, Lars Ulander, and Ludovic Villard, “The BIOMASS Level 2 Prototype Processor: Design and Experimental Results of Above-Ground Biomass Estimation,” Remote Sensing, vol. 12, no. 6, 2020. [2] P. Prats-Iraola, K. Papathanassiou, J. S. Kim, M. Rodriguez-Cassola, D. D’Aria, R. Piantanida, A. Valentino, D. Giudici, M. Jaeger, R. Scheiber, M. Pinheiro, N. Yague-Martinez, M. Nannini, V. Gracheva, T. Guardabrazo, and A. Moreira, “The BIOMASS Ground Processor Prototype: An Overview,” in EUSAR 2018; 12th European Conference on Synthetic Aperture Radar, 2018. [3] Simone Mancon, Andrea Monti Guarnieri, Davide Giudici, and Stefano Tebaldini, “On the Phase Calibration by Multi- squint Analysis in TOPSAR and Stripmap Interferometry,” IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 1, pp. 134–147, 2017. [4] Giorgio Gomba, Alessandro Parizzi, Francesco De Zan, Michael Eineder, and Richard Bamler, “Toward Operational Compensation of Ionospheric Effects in SAR Interferograms: The Split-Spectrum Method,” IEEE Transactions on Geoscience and Remote Sensing, vol. 54, no. 3, pp. 1446–1461, 2016. [5] Stefano Tebaldini, Mauro Mariotti d’Alessandro, Jun Su Kim, and Kostas Papathanassiou, “Ionosphere vertical profiling from biomass multi-squint InSAR,” in 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2017, pp. 5866–5869. [6] Francesco Banda, Mauro Mariotti d’Alessandro, and Stefano Tebaldini, “Ground and Volume Decomposition as a Proxy for AGB from P-Band SAR Data,” Remote Sensing, vol. 12, no. 2, 2020.

ESA’s 7th Earth Explorer satellite BIOMASS [1] is scheduled for launch in late 2024. It will carry on-board the first spaceborne P-band (435 MHz) synthetic aperture radar (SAR) sensor. Its main objective will be to acquire annual, near-global maps of above-ground forest biomass density (AGBD), forest height, and forest disturbance. In this paper, we give an overview of the Global Forest Biomass Estimation Algorithm for BIOMASS (BIO4BIO), to be implemented in the ground segment. P-band SAR backscatter intensity has a well-known sensitivity to AGBD, although the strong influences of soil moisture and topography often have a detrimental effect on the AGBD estimation performance. To this end, BIO4BIO uses a novel technique called ground cancellation (GC) [2], which is able to suppress ground-level scattering and provide a canopy backscatter (σ_cnp^0 ), which is a better proxy of AGBD. To describe the dependence of canopy backscatter on AGBD, BIO4BIO will use a three-parameter power law model [3]: σ_cnp^0=LB^a cos^n⁡ϑ (1) where B is AGBD in ton/ha, ϑ is the local incidence angle, and L, a, and n are unknown model parameters. Experimental results have shown a large variability of parameters L, a, and n in space and time, as well as across forest types and polarisations [3]. While much of it can be attributed to changes in soil and vegetation moisture, forest structure, and topography, the exact nature of these variabilities remains unknown, partly due to the limited extent of experimental P-band SAR data. Hence, these parameters will need to be estimated using local reference AGBD data, and aspects such as forest type and temporal variability will need to be considered. To this end, BIO4BIO will rely on the following principles: AGBD mapping will be conducted in non-overlapping 3x3 deg2 blocks, with model parameters estimated using data sampled from a 5x5 deg2 neighbourhood. This will allow model parameters to vary in space to reflect large-scale moisture, altitude, and climatological changes. Thanks to the larger sampling area, adjacent blocks will share some training data, thus improving the continuity of AGBD across block boundaries. The parameter estimation and AGBD mapping processes will, however, remain independent in adjacent blocks, thus allowing significant parallelisation of the algorithm. Each of the parameters will have a pre-defined variability across polarisation, forest type, and stacks (independent acquisitions). While the variability patterns will be modifiable, the basic scenario will allow all parameters to be polarisation-dependent. Moreover, L will attain different values for different stacks (to compensate for some moisture variability, which often affects the absolute value of σ_cnp^0), while a will attain different values for different forest types (to compensate for different forest structures). This approach reduces the number of parameters that need to be estimated, while allowing the model to fit to vastly different conditions and enforcing some physical interpretation onto the parameters. Within each 5x5 deg2 neighbourhood, the model parameters will be estimated from external reference data on AGBD. Many different sources have been considered for this task, with the final choice being a combination of NASA GEDI Level 4A product [4], [5] and the pan-boreal AGBD map generated mainly from ICESat-2 data . Both of these datasets rely on spaceborne laser scanning data, which are known to have a good sensitivity to the vertical forest structure (and thus AGBD), but poor spatiotemporal coverage. Hence, they are well-suited for training data. Both products will have to be pre-processed to match the resolution of BIOMASS. For GEDI, the Level 4A product consists of 25 m diameter footprints, the ICESat-2 product consists of a 30 m-resolution map. Both of these will be aggregated to the resolution of BIOMASS, which is 200 m×200 m and re-sampled onto the GLO-90 grid used for the Copernicus DEM (3 arcsec latitudinal spacing, 3-6 arcsec longitudinal spacing, depending on the latitude ). The combination of the GEDI and ICESat-2 datasets is expected to provide high-quality AGBD reference data for most parts of the world, especially in the overlap zone, where both datasets will be available (see Figure 1d). However, a few open issues remain. 1) Firstly, it is possible that some areas will have insufficient training data, particularly in the tropics, where GEDI footprints are sparse due to cloud cover and large distance between satellite orbits (Figure 1e). 2) Secondly, some land areas in the south will remain uncovered by either datasets, e.g., the southernmost part of Patagonia (Figure 1f). 3) Thirdly, it is possible that in some areas, AGBD and canopy backscatter will have a very low correlation, for example in areas of significant topography, high average AGBD, and low AGBD interval. To address these issues, the algorithm will be evaluated in two iterations. The first iteration will use the GEDI and ICESat-2 data as reference, while the second iteration will use the output AGBD maps from the previous iteration as reference. This will allow the algorithm to fill in possible gaps occurring due to any of the aforementioned three issues, by estimating the parameters using nearby AGBD estimates as training data and thus enforcing AGBD continuity in space. Another potential issue is the propagation of biases from the training data to the output product. Although bias can be present in any type of reference data, in laser scanning there are some well-known effects that need to be considered (e.g., large trees obscuring the underlying vegetation, topographic effects). In the case of BIO4BIO, this issue will be addressed by a careful screening of footprints used for training data and a close contact with scientists involved in the generation of GEDI and ICESat-2 products. The latter will allow us to implement the latest findings in our work.

BIOMASS is ESA's seventh Earth Explorer, launching in 2024. Its primary aim is to support the scientific community with Synthetic Aperture Radar (SAR) acquisitions over forested areas to gain a better understanding of global forest status and dynamics [1]. BIOMASS features a 435 MHz signal with 6 MHz bandwidth, full polarimetry and a repeat cycle designed specifically to acquire in tomographic and interferometric modes. P-band was chosen due to its high sensitivity to the woody elements of trees, its resistance to temporal decorrelation and the fact that at this frequency ionospheric effects are not so strong as to be unmanageable. BIOMASS coverage will be global except for Europe and North America due to international frequency regulations. The three-day repeat cycle helps to preserve coherence for interferometry and tomography [2]. The first acquisition phase is Commissioning, which is dedicated to instrument calibration and preliminary testing. This is followed by the Tomographic Phase (TOM), during which seven passes are collected over each location, in ascending and descending geometry, covering the entire globe in about 14 months. This allows tomographic processing, i.e. 3D forest structure mapping, as well as the possibility of retrieving sub-canopy terrain topography. The last phase, continuing to the end of the mission, is Interferometric (INT), during which PolInSAR [3, 4] techniques are applied under a dual-baseline configuration and for which each global coverage takes about seven months. This presentation focuses on Level-2 (L2) processing, i.e. the generation of forest products from BIOMASS acquisitions. The three products foreseen for routine generation are Forest Disturbance (FD), Forest height (FH) and Above Ground Biomass (AGB). The starting point for the generation of forest products is Level-1 (L1) coregistered and calibrated stacks produced by the BIOMASS interferometric processor, which is the topic of a companion talk at this symposium [5]. L1 stacks are phase calibrated to remove terrain topography and residual disturbances, after which retrieval of forest structure and changes is possible. All the L2 processing algorithms exploit ground/volume separation or ground rejection techniques, as ground scattering requires complex modelling and hinders forest parameter retrieval. In particular, the FD and AGB algorithms implement ground cancellation [4] as a first step. This recent technique coherently subtracts L1 interferometric calibrated and terrain-flattened acquisitions to cancel/reduce terrain and enhance the canopy signal. The FD algorithm then implements a statistical change detection approach [6] to detect deforestation within a time series of acquisitions. The AGB algorithm inverts a semi-empirical scattering model relating the canopy signal to AGB density (AGBD) [7] and is trained on external reference AGBD [8]. This algorithm is designed to perform global estimation through a block processing scheme which constrains the spatial variability of ancillary model parameters. The FH algorithm is based on PolInSAR inversion supported by tomographic information [3], which helps to reduce the number of parameters to be estimated. Dual-baseline interferometry allows temporal decorrelation to be accounted for, but single baseline inversion is also possible in contingency cases. Preliminarily testing of the algorithms has been successful on airborne campaign acquisitions reprocessed to BIOMASS resolution. Current work is dedicated to optimising these techniques with the aim of integrating them into the operational processor and official ESA BIOMASS software ecosystem [9]. References - --------- [1] Francesco Banda, Davide Giudici, Thuy Le Toan, Mauro Mariotti d’Alessandro, Kostas Papathanassiou, Shaun Quegan, Guido Riembauer, Klaus Scipal, Maciej Soja, Stefano Tebaldini, Lars Ulander, and Ludovic Villard, “The BIOMASS Level 2 Prototype Processor: Design and Experimental Results of Above-Ground Biomass Estimation,” Remote Sensing, vol. 12, no. 6, 2020. [2] Reigber, Andreas, and Alberto Moreira. "First demonstration of airborne SAR tomography using multibaseline L-band data." IEEE Transactions on Geoscience and Remote Sensing 38.5 (2000): 2142-2152. [3] Guliaev, Roman, et al. "Forest height estimation by means of TanDEM-X InSAR and waveform lidar data." IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 14 (2021): 3084-3094. [4] M. Mariotti d’Alessandro, S. Tebaldini, S. Quegan, M. J. Soja, L. M. H. Ulander and K. Scipal, "Interferometric Ground Cancellation for Above Ground Biomass Estimation," in IEEE Transactions on Geoscience and Remote Sensing, vol. 58, no. 9, pp. 6410-6419, Sept. 2020 [5] BIOMASS INTERFEROMETRIC PROCESSOR: CURRENT STATUS, POLINSAR 2023 [6] Conradsen, Knut, Allan Aasbjerg Nielsen, and Henning Skriver. "Determining the points of change in time series of polarimetric SAR data." IEEE Transactions on Geoscience and Remote Sensing 54.5 (2016): 3007-3024. [7] Soja, M., Quegan, S., Mariotti d’Alessandro, M. Banda, F. Scipal, K., Tebaldini, S. Ulander, L.M.H. “Mapping above-ground biomass in tropical forests with ground-cancelled P-band SAR and limited reference data”, Remote Sens. Environ. Volume 253, February 2021, 112153 [8] Dubayah, Ralph, et al. "The Global Ecosystem Dynamics Investigation: High-resolution laser ranging of the Earth’s forests and topography." Science of remote sensing 1 (2020): 100002. [9] www.biopal.org/docs/ecosystem/

Selected as European Space Agency’s seventh Earth Explorer in May 2013, the BIOMASS mission will provide crucial information about the state of our forests and how they are changing [Le Toan et al., 2011]. This mission is being designed to provide, for the first time from space, P-band Synthetic Aperture Radar measurements to determine the amount of biomass and carbon stored in forests [Le Toan et al., 1992]. The data will be used to further our knowledge of the role forests play in the carbon cycle. Earth Observation platforms play an essential role in the new European Space Agency Earth Observation strategy implementing a significant evolution of the ground segment. Earth Observation Platforms are gradually appearing in Europe, at national level or at European scale, and outside Europe. For example, the Thematic Exploitation Platforms currently developed at European Space Agency are a category of exploitation platforms dedicated to geophysical themes. The Thematic Exploitation Platforms under development are instrumental as pilots to formulate future technical and economical approaches for networking exploitation platforms (European ecosystem of Earth Observation Platforms), which is a central idea of the Earth Observation ground segment evolution. In this context of an innovative sensor and a changing ground segment, the concept of Mission Algorithm and Analysis Platform dedicated to the BIOMASS, to the NASA-ISRO SAR (NISAR) mission [Rosen et al., 2015] and to the NASA Global Ecosystem Dynamics Investigation (GEDI) mission [GEDI] mission is proposed. Developed in a collaborative way between ESA and NASA, this Mission Algorithm and Analysis Platform will implement, as part of the payload data ground segment, a virtual open and collaborative environment. The goal is to bring together data centre (Earth Observation and non- Earth Observation data), computing resources and hosted processing, collaborative tools (processing tools, data mining tools, user tools, …), concurrent design and test bench functions, accounting tools to manage resource utilisation, communication tools (social network) and documentation. This platform will give the opportunity, for the first time, to manage the community of users of the BIOMASS mission thanks to this innovative concept. To best ensure that users can collaborate across the platform and to access needed resources, the MAAP requires all data, algorithms, and software to conform to open access and open-source policies. As an example of best collaborative and open-source practices, most of the BIOMASS Processing Suite (BPS) will be made openly available within the MAAP. This Processing Suite contains all elements to generate the BIOMASS upper-level data products and is currently in development under the umbrella of the open-source project called BioPAL [BioPAL]. BioPAL is developed in a coherent manner, putting a modular architecture and reproducible software design in place. BioPAL aims to factorize the development and testing of common elements across different BIOMASS processors. The architecture of this scientific software makes lower-level bricks and functionalities available through a well-documented Application Programming Interface (API) to foster the reuse and continuous development of processing algorithms from the BIOMASS user community. [Le Toan et al., 2011] T. Le Toan, S. Quegan, M. Davidson, H. Balzter, P. Paillou, K. Papathanassiou, S. Plummer, F. Rocca, S. Saatchi, H. Shugart and L. Ulander, “The BIOMASS Mission: Mapping global forest biomass to better understand the terrestrial carbon cycle”, Remote Sensing of Environment, Vol. 115, No. 11, pp. 2850-2860, June 2011. [Le Toan et al., 1992] T. Le Toan, A. Beaudoin, et al., “Relating forest biomass to SAR data”, IEEE Transactions on Geoscience and Remote Sensing, Vol. 30, No. 2, pp. 403-411, March 1992. [Rosen et al., 2015] P.A. Rosen, S. Hensley, S. Shaffer, L. Veilleux, M. Chakraborty, T. Misra, R. Bhan, V. Raju Sagi and R. Satish, "The NASA-ISRO SAR mission - An international space partnership for science and societal benefit", IEEE Radar Conference (RadarCon), pp. 1610-1613, 10-15 May 2015. [GEDI] https://science.nasa.gov/missions/gedi [BioPAL] BioPAL project site: http://www.biopal.org

Tomographic Synthetic Aperture Radar (SAR), TomoSAR, is an electromagnetic imaging technique able to map the 3D reflectivity of complex environments. It is generally implemented using radar signals acquired over a 2D aperture, i.e. using a set of coherent 2D SAR images measured from slightly shifted trajectories. Forested areas can be efficiently characterized using TomoSAR acquisitions operated at lower frequency bands, such as P or L bands, as larger wavelength values ensure both a deep penetration of dense forest covers, as well as a better preservation of signal coherence over time. Polarimetric TomoSAR (PolTomoSAR) uses polarization diversity in order to further discriminate signals originating from the canopy or from the underlying ground of a forest. Several applications of PolTomoSAR advanced signal processing techniques on data acquired during airborne campaigns, have demonstrated that this type of configuration could be exploited in order to estimate some key parameters of forest covers, such as tree height, ground topography, biomass, structure... The upcoming launch of the European Space Agency (ESA) BIOMASS spaceborne mission, that includes a polarimetric SAR device operating at P band, will provide a unique opportunity to accurately study densely forested regions, and in particular tropical regions. The estimation of the performance bounds, i.e. the best achievable precision for a given observed medium and given acquisition conditions, reveals a fundamental importance for both the evaluation of the different existing estimation approaches, but also for the validation of the mission products. This paper proposes to first set up a methodological background for the computation of PolTomoSAR performance bounds that remains valid for single-polarization observations, multi-polarized ones, in PolinSAR configuration, or using more than two baselines. The performance of different modes is evaluated using Bayesian inference [1], and statistical processing classically used for the resolution of inverse problems [2]. Among all the descriptors of a forest cover, three main parameters are considered: the Digital Terrain Model (DTM), i.e. the topography of the underlying ground, the Canopy Height Model (CHM) or tree height and the Above Ground Biomass (AGB). As AGB is not directly estimated by PolTomoSAR imaging, various proxies are investigated [3]. The second part of this work concerns the influence of spatial resolutions, i.e. in range and azimuth directions, on the performance of PolTomoSAR. The particular case of BIOMASS with resolution of 12.5 m in azimuth and 25 m in slant range is investigated and compared with airborne measurements with a resolution of approximately 1m. Deteriorated resolution properties are accounted for using two approaches. The first one is based on an low-resolution adaptation of forest structure description with and unchanged theoretical derivation of the performance limits. The second technique extends the modeling of the forest reflectivity to include an additional axis in the inference process. Interestingly, temporal decorrelation of the SAR echos can be very easily integrated in the performance study in order to directly assess the influence of this effect of BIOMASS products. The validity and accuracy of the proposed methods are illustrated using the PolTomoSAR data set acquired at P band over the tropical forest test site of Paracou, in French Guiana, during the TropiSAR campaign in the summer 2009, using the ONERA’s SETHI system. It consists of six polarimetric and interferometric images having a spatial resolution of 1.5m in azimuth and 1.2m in range [4]. From this airborne data set, images having the spatial resolution of spaceborne BIOMASS acquisitions, 12.5m in azimuth and 25m in range [5, 6], are generated. Different temporal decorrelation scenarios are investigated using realistic decorrelation patterns derived from [7]. References [1] George Casella and Roger Berger, Statistical Inference, Duxbury Resource Center, June 2001. [2] Albert Tarantola, Inverse Problem Theory and Methods for Model Parameter Estimation, Society for Industrial and Applied Mathematics, USA, 2004. [3] D. Ho Tong Minh, T. L. Toan, F. Rocca, S. Tebaldini, M. M. d’Alessandro, and L. Villard, “Relating p-band synthetic aperture radar tomography to tropical forest biomass,” IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 2, pp. 967–979, 2014. [4] Pascale Dubois-Fernandez, Hélène Oriot, Colette Coulombeix, Hubert Cantalloube, Olivier Plessis, Thuy Le Toan, Sandrine Daniel, Jerome Chave, Lilian Blanc, and Malcolm Davidson, “Tropisar, a sar data acquisition campaign in french guiana,” 07 2010, pp. 1 – 4. [5] Maciej J. Soja, Shaun Quegan, Mauro M. d’Alessandro, Francesco Banda, Klaus Scipal, Stefano Tebaldini, and Lars M.H. Ulander, “Mapping above-ground biomass in tropical forests with ground-cancelled p-band sar and limited reference data,” Remote Sensing of Environment, vol. 253, pp. 112153, 2021. [6] L. Ferro-Famil, Y. Huang, L. Villard, T. Le Toan, and T. Koleck, “Comparison of biomass acquisition modes for the characterization of forests,” in 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, 2021, pp. 1545–1548. [7] S. El Idrissi Essebtey, L. Villard, P. Borderies, T. Koleck, B. Burban and T. Le Toan, "Long-Term Trends of P-Band Temporal Decorrelation Over a Tropical Dense Forest-Experimental Results for the BIOMASS Mission," in IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-15, 2022

Context: Tomography is a useful technique to reconstruct the reflectivity of volumetric environment using a stack of coherent 2D SAR images. This approach is particularly well suited to monitor and characterize forest environments. The upcoming ESA BIOMASS mission, whose main objective is to map the above ground biomass of forested areas at global scale and particularly over tropical regions, will use this observation mode in its initial phase (first 18 months). The main feature one may derive from 3D cube of reflectivity are the forest tree reflectivity, the tree height and the underlying ground topography. Several techniques have been proposed for such a characterization and this paper proposes a new one particularly simple to apply to practical scenarios and provide highly accurate estimates. Parametric tomography: In SAR tomography, the reflectivity distribution in the vertical dimension is usually estimated using non-parametric spectral estimation techniques such as Beamforming or Capon [1,2]. Such techniques have a limited resolution and their performances are affected by the presence of spurious peaks and sidelobes. Recently, some adaptive estimation approaches have been proposed that directly model the vertical reflectivity distribution as a combination of function basis. An adaptive example can be found in [3] in which wavelength are used as function basis. As explained in [4], this technique can be assimilated to deconvolution to access the high-resolution profile that have been averaged by the vertical impulse response of the tomographic system. This estimation approach is particularly demanding and relies on an optimization of a complex criteria which requires a large amount of computation. One of the conclusions of the work done in [4] is that the vertical reflectivity profile can be modeled with only two function bases which agrees with the original work of [5,6]. They considered independent ground and volume contributions. In this paper, we proposed to model the vertical reflectivity profile of tropical forest at P-band using a fixed number of function bases and not an adaptive function bases in order to reduce the complexity. This basis with fixed functions represents the contribution from the ground and the forest canopy. Two examples of function bases are retained. The well-focused ground response is assimilated to a Dirac reflectivity profile whereas two families of functions are proposed for modeling the volume response: the exponential profile proposed by [5] which corresponds to a constant extinction within the homogenous layer or the gaussian profile proposed in [7] which account for a linear trend in the extinction. It is very important to note that this work is done in single polarization. The choice of a single polarization represents a semi-efficient gain of computation time and complexity with respect to fully polarimetric techniques. Profile model performance comparison: The proposed two components estimation technique is applied to the polarimetric SAR tomographic dataset acquired by ONERA over French Guiana within the frame of the TropiSAR campaign [8]. Measures show that both models are able to retrieve ground topography and tree height with good accuracy. For the exponential model, a good estimation of both ground and tree height requires to use a range of value of extinction coefficient that is significantly different from the range of values specified in the literature. This is identified as an ambiguity of this model linked to its asymmetric structure, whereas the gaussian model is very fast and does not meet such limitation. Comparison with results derived from the polarimetric TomoSAR SKP decomposition: The parameters are compared to those obtained using the SKP, Sum of Kronecker Products, decomposition approach [6]. Results indicate that the very simple approach proposed in this paper achieves the same level of quality than the SKP technique concerning the ground topography and volume center elevation of the forest. Both single polarization waveforms and SKP technique can be adapted to tree height using a thresholding of the tomographic intensity. These comparable levels of performance show that our technique is robust and give excellent results using a single polarization channel. This conclusion shall be assessed over other test sites. References: [1] F. Gini, F. Lombardini, “Multibaseline cross-track SAR interferometry : a signal processing perspective”, IEEE Aerospace and Electronic Systems magazine, Volume 20, Issue 8, pp71 – 93, August 2005 [2] H. Aghababaei, G. Ferraioli, L. Ferro-Famil, Y. Huang, M. Mariotti D’Alessandro, V. Pascazio, G. Schirinzi, S. Tebaldini, “Forest SAR Tomography : Principles and Applications”, IEEE Geoscience and Remote Sensing Magazine, Volume 8, Issue 2, pp30 – 45, June 2020 [3] E.Aguilera, M. Nannini and A.Reigber, “Wavelet-based compressed sensing for SAR Tomography of forested areas”, IEEE Transactions on Geoscience and Remote Sensing, Volume 51, Issue 12, pp5283 – 5295, December 2013 [4] L. Ferro-Famil, Y. Huang and N. Ge, “Estimation of the vertical structure of a tropical forest using basis functions and parametric SAR Tomography”, in IGARSS 2022 – 2022 IEEE International Geoscience and Remote Sensing Symposium, July 2022, pp.599-602 [5] S.R. Cloude and K.P. Papathanassiou, “Three-stage inversion process for polarimetric SAR interferometry”, IEE Procedings – Radar, Sonar and Navigation, Volume 150, Issue 3, pp125 – 134, June 2003 [6] S. Tebaldini, “Single and multipolarimetric SAR tomography of Forested Areas : A parametric approach”, IEEE Transactions on Geoscience and Remote sensing, Volume 48, Issue 5, pp2375 – 2387, May 2010 [7] F. Garestier and T. Le Toan, “Estimation of the Backscatter Vertical profile of a Pine Forest Using Single Baseline P-band (Pol-)InSAR Data”, IEEE Transactions on Geoscience and Remote Sensing, Volume 48, Issue 9, pp3340-3348, September 2010 [8] P. Dubois-Fernandez, T. Le Toan, S. Daniel, H. Oriot, J. Chave, L. Blanc, L. Villard, « The TropiSAR Airborne campaign in French Guiana: objectives, description, and observed temporal behavior of the backscatter signal”, IEEE Transactions on Geoscience and Remote Sensing, Volume 50, Issue 8, pp3228 – 3241, August 2012

1. INTRODUCTION SAR tomography (TomoSAR) represents a unique way to characterize forested areas from their 3-D density of reflectivity. It has been shown in numerous [1, 2, 3, 4, 5, 6] studies that coherent 2-D SAR images with an intermediate horizontal resolution may be processed by spectral analysis techniques, in order to focus 3-D cubes of reflectivity of the observed media. This type of electromagnetic description of a forest is generally sufficient to estimate its main features, such as underlying ground topog- raphy, tree height as well as above-ground biomass. The European Space Agency (ESA)’s BIOMASS spaceborne mission, that includes a SAR device operating over multiple polarizations at P-band. One of the original features of this mission concerns its ability to provide 3-D SAR reconstruction via the use of polarimetric and tomographic (PolTomoSAR) processing. In this paper, a characterization of tropical forests is conducted by applying tomographic processing to both high-resolution airborne SAR data and simulated BIOMASS data, obtained by applying specific restriction to the airborne data set. Main forest descriptors, such as underlying topography and tree top height, are estimated and compared with airborne lidar-derived Digital Terrain Model (DTM) and Digital Surface Model (DSM). Moreover, the impact of resolution on the methodological approach for tropical forest characterization is also discussed in the paper. 2. STUDY DATA SET AND CONFIGURATIONS The methods proposed in this paper are applied to a PolTomoSAR data set acquired at P band over the tropical forest test site of Paracou, in French Guiana, during the TropiSAR campaign in the summer 2009, using the ONERA’s SETHI system. It consists of six polarimetric and interferometric images having a spatial resolution of 1.5m in azimuth and 1.2m in range [7]. The time interval between the acqusitions is small enough so that temporal decorrelation effects may not be considered as significant over the whole acquisition duration. From this airborne data set, images having the spatial resolution of spaceborne BIOMASS acquisitions, 12.5m in azimuth and 25m in range [8, 9], are generated. One may note that this transformation cannot be simply summarized as a low-pass filtering of the high-resolution images, but requires to bring back focused images into the radar signal domain before applying spectral restrictions. The resulting images has a much lower resolution than the original ones and modified interferometric correlation properties, due to range decorrelation, which affects InSAR features in the case of small signal bandwidth [10]. DTM and DSM estimates derived from airborne lidar acquisitions and a ground sampling of 1mx1m are used as references for the evaluation of tomographic products. 3. FOREST CHARACTERIZATION BY SAR TOMOGRAPHY In this paper, tomographic processing for forest characterization is led in both single-polarization and fully polarimetric TomoSAR configurations. 3.1. Single-polarization tomographic approach A single-polarization tomographic approach is proposed, based on a fitting between a parametric model of forest vertical structure and the measured coherence values from all the interferometric pairs of acquired SLC images. 3.2. Fully polarimetric tomographic estimator Using PolTomoSAR data, the SKP method [3] uses a sum of Kronecker products to model the polarimetric radar backscattering from the ground and canopy layers in forested areas. This method provides an undeniable performance for ground and volume separation in elevation as shown in [3]. In this paper, this polarimetric tomographic approach is applied to estimate the DTM and DSM of forested areas, whose results are compared with those obtained by the proposed single-polarization approach. 4. ESTIMATED FOREST PARAMETERS The proposed tomographic approaches are applied to high-resolution TropiSAR tomographic data set. The underlying ground elevation and tree top height of a densely-forested area in Paracou are estimated, and the derived DTM and DSM are illustrated in Fig 1. Using now the simulated BIOMASS data set, the estimated DTM and DSM are illustrated in Fig. 2. More analysis and comparison will be provided in the final paper.

Accurately estimating tropical forest structure is challenging but essential for understanding carbon cycling, biodiversity, and ecosystem functioning. Active sensors such as P-band Synthetic Aperture Radar (SAR) and Light Detection and Ranging (LiDAR) can penetrate thick vegetation layers and provide three-dimensional information on forest structure. NASA's Global Ecosystem Dynamics Investigation (GEDI) mission and the upcoming ESA's BIOMASS mission are designed to provide spaceborne active data. Meanwhile, SAR tomography (TomoSAR) can generate three-dimensional images by acquiring multiple P-band SAR data over the same areas. This study evaluates the potential of GEDI and TomoSAR acquisitions for accurately estimating tropical forest vertical structures. We analyze airborne P-band TomoSAR, airborne LiDAR, and spaceborne GEDI LiDAR data at tropical forest sites in South America and Africa. Our results show that GEDI and P-band TomoSAR can directly measure surface, vegetation heights, and vertical profiles with high resolution and precision [1]. Airborne TomoSAR has higher quality than GEDI due to better penetration properties and accuracy. However, GEDI has a vegetation height root-mean-square error of less than 5 m for an average forest height value of around 30 m, similar to the expected performance of the future spaceborne BIOMASS mission [2]. We recommend extracting GEDI data from all continents to maintain consistent global performance. GEDI's Relative Height (RH) metric is the most accurate, filtered by full power shots with a sensitivity greater than 98%. These results suggest that GEDI measurements with a sensitivity greater than 98% can provide a good reference for calibrating the BIOMASS mission algorithms [3]. Furthermore, we show that the location of the phase center in TomoSAR is consistently lower than in GEDI, with a range of 2-4m, depending on the height and polarization of the forest layers. This finding demonstrates that P-band SAR can penetrate the ground layer better than GEDI signals [4]. GEDI and TomoSAR data can accurately capture vertical information in the canopy levels (between 10m-40m), with a strong correlation between GEDI and TomoSAR reflectivity in the canopy layers. The highest correlation was observed around 30m above the ground, which aligns with previous research in developing algorithms for the BIOMASS mission in aboveground biomass retrieval [5]. Together, TomoSAR and GEDI are robust and comparable in studying tropical forests and support the BIOMASS mission for global biomass mapping. Reference: [1] Y. -N. Ngo, Y. Huang, D. Ho Tong Minh, L. Ferro-Famil, I. Fayad and N. Baghdadi, "Tropical Forest Vertical Structure Characterization: From GEDI to P-Band SAR Tomography," in IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1-5, 2022, Art no. 7004705, doi: 10.1109/LGRS.2022.3208744. [2] D. Ho Tong Minh, S. Tebaldini, F. Rocca, T. Le Toan, L. Villard and P. C. Dubois-Fernandez, "Capabilities of BIOMASS Tomography for Investigating Tropical Forests," in IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 2, pp. 965-975, Feb. 2015, doi: 10.1109/TGRS.2014.2331142. [3] Ngo, Y.-N.; Ho Tong Minh, D.; Baghdadi, N.; Fayad, I. Tropical Forest Top Height by GEDI: From Sparse Coverage to Continuous Data. Remote Sens. 2023, 15, 975. https://doi.org/10.3390/rs15040975 [4] Ngo, Y.-N.; Ho Tong Minh, D.; Baghdadi, N.; Fayad, I.; L. Ferro-Famil; Y. Huang. Phenomenology in tropical forests through GEDI and 3D SAR tomography. In preparation [5] D. Ho Tong Minh, T. Le Toan, F. Rocca, S. Tebaldini, L. Villard, M. Réjou-Méchain, O. L. Phillips, T. R. Feldpausch, P. Dubois-Fernandez, K. Scipal, J. Chave, "SAR tomography for the retrieval of forest biomass and height: Cross-validation at two tropical forest sites in French Guiana", Remote Sensing of Environment, vol.175, pp.138, 2016

Tropical forests play a critical role in regulating the Earth´s climate through providing diverse and valuable ecosystem services. For example, they are an important carbon storage and a main cradle for planetary biodiversity. Thus, an accurate monitoring of tropical forest biomass and structure over time is of crucial importance. Also, understanding the impact of human activities on these valuable ecosystems is essential. While the estimation of a single tree biomass can already be complex – as it is a function of wood density, trunk and branch volume – the biomass estimation of a whole stand is a far more challenging task. Even more so at highly complex and diverse tropical forest sites. Current remote sensing based approaches evolve around the use of allometric relationships at different spatial and temporal scales. Three main input parameters are used for the biomass estimation: the tree height at the respective resolution cell size, the allometric factor and the allometric exponent. The allometric factor accounts for density variations in the forest stand. The allometric exponent on the other hand represents species composition and variations in the growing conditions, thus involves a more profound knowledge of the forests and the implementation of forest field inventory data. Actual implementations of such an approach by means of TanDEM-X / GEDI measurements rely on forest height estimates obtained from the combination of interferometric TanDEM-X and GEDI waveform measurements, as well as allometric parameters. These parameters are derived from GEDI measurements and are spatially adapted to the horizontal forest structure index derived from interferometric TanDEM-X measurements. This way, a forest height map at 25 m spatial resolution is generated matching the GEDI footprint of equally 25 m radius, while the horizontal structure index is obtained at 1 Ha scale. With the new ESA BIOMASS mission on the horizon, an exiting new era for forest remote sensing will start in the near future. Given the expected spatial resolution of the resulting forest biomass products in the range of 200 m, a considerable resolution scale difference emerges between the current approach and the possible future implementation to the ESA BIOMASS mission. The question is how the allometric relation changes when the different parameters in it are obtained from different missions and / or at different scales. To gain knowledge on this issue, a preliminary study was carried out using the NASA LVIS datasets acquired in 2016 (Gabon, Lopé national park) and 2019 (Costa Rica, La Selva/Braulio Carillo national park). Both the LVIS forest height and biomass products were aggregated to different spatial scales to analyze the changes in the subsequently derived allometric relationships. Spatial resolutions of 25 m to 200 m were addressed. While the allometric parameters remain nearly constant for the coarser resolution scales, a considerable change is visible between the 25 m scale the 50 m resolution scale. Thus, the changes in forest structure must happen within this range. Also, all allometric functions appear insensitive to lower tree canopy heights, especially in the 10 m to 50 m range. The derived allometric functions were subsequently re-applied to the existing LVIS forest height map. As an example, considerable changes in forest biomass are visible in the Lopé reserve. Applying the derived 25 m resolution scale allometry to both 25 m and 100 m forests renders a decrease in biomass of about 10% that emerges primarily from the changing canopy height with scale (e.g. spatial resolution). Comparing the 25 m and 100 m allometries on a 100 m canopy height map results in a more drastic decrease of about 15%. A detailed discussion of the scale effect on the allometric biomass estimation performance will be further discussed by means of simulated and real experimental data.

Model-based (PM) forest height inversion from Polarimetric Interferometric Synthetic Aperture Radar (Pol-InSAR) measurements is today an established application demonstrated and validated at large scales for a wide variety of boreal and tropical forest sites at different frequencies (from X- down to P- band) [1], [2]. Although the estimation performance obtained may depend on the individual observation spaces in each case, it is generally (very) convincing. However, as with any model-based inversion approach, there are inherent limitations that can restrict expected performance depending on the individual case [3]. The parameterization (modeling) of the vertical reflectivity function (e.g., the vertical distribution of scatters in the forest) must be abstract and general in order to describe as many forest conditions as possible with the smallest possible set of parameters (for example in the form of a two-layer model, as in the case of the popular RVoG model) [4]. Hence, it cannot account for the high natural spatial variability of forest and terrain conditions. This mismatch between the model and data results in reduced estimation accuracy, expressed in most cases by an increased variance [5]. When the dimension of the observation space becomes small compared to the number of parameters used to describe the model, the simplified assumptions that are then required for an inversion affect the performance even more. In addition to estimation variance, biases are also introduced [5]. On the other side, machine learning (ML) approaches show impressive capabilities when it comes to resolving inversion problems and have been used for Forest height inversion in several test cases [6]. Yet, the interpretability and robustness of ML approaches for solving under-determined inverse problems are questionable without embedding prior knowledge and domain expertise. The fusion of physics-based models and DL models into so-called hybrid models can be essential to develop new models with optimum top performance and robustness. With this contribution, we propose a novel hybrid modeling approach that integrates machine learning (ML) and physical modeling (PM). Indeed, a hybrid modeling (ML + PM) approach is used to describe the spatial variability of the vertical reflectivity function which is used for the PM-based inversion of forest height from interferometric coherence measurements [7]–[9]. The vertical reflectivity function is defined by a set of coefficients by means of a Legendre series decomposition [7] and is derived using a MultiLayer Perceptron (MLP). We incorporate multi-model data for the forest height estimation from single baseline single-polarimetric TanDEM-X interferometric coherence measurements. We discuss the challenges that need to be addressed and how to integrate a multi-modal (LiDAR, multi-spectral images, Polarimetric SAR, etc.) in a general ML framework constrained with the PM. Emphasis will be given in discussing the use of quad-polarimetric multi-frequency datasets as inputs to the PM via the MLP, such as ALOS-2/PALSAR-2 and/or (simulated) BIOMASS data sets. [1] V. Carcarra-Bes, M. Pardini, C. Choi, R. Guliaev, and K. P. Papathanassiou, “Tandem-X and Gedi Data Fusion for a Continuous Forest Height Mapping at Large Scales,” pp. 796–799, 2021, doi: 10.1109/igarss47720.2021.9554655. [2] S. K. Lee et al., “Spaceborne GEDI and TanDEM-X data fusion for enhanced forest height and biomass,” AGUFM, vol. 2018, pp. B44E-14, 2018, Accessed: May 03, 2021. [Online]. Available: https://ui.adsabs.harvard.edu/abs/2018AGUFM.B44E..14L/abstract [3] A. T. Caicoya, F. Kugler, K. Papathanassiou, P. Biber, and H. Pretzsch, “Biomass estimation as a function of vertical forest structure and forest height - Potential and limitations for Radar Remote Sensing,” in 8th European Conference on Synthetic Aperture Radar, 2010, pp. 1–4. [4] K. P. Papathanassiou and S. R. Cloude, “Single-Baseline Polarimetric SAR Interferometry,” 2001. [5] K. Papathanassiou et al., “Forest Parameter Estimation by Means of Multi-Baseline Pol-Insar Techniques: State-of-the-Art and Future Challenges,” International Geoscience and Remote Sensing Symposium (IGARSS), vol. 2022-July, pp. 1123–1124, 2022, doi: 10.1109/IGARSS46834.2022.9884936. [6] M. Pourshamsi, M. Garcia, M. Lavalle, and H. Balzter, “A Machine-Learning Approach to PolInSAR and LiDAR Data Fusion for Improved Tropical Forest Canopy Height Estimation Using NASA AfriSAR Campaign Data,” IEEE J Sel Top Appl Earth Obs Remote Sens, vol. 11, no. 10, pp. 3453–3463, Oct. 2018, doi: 10.1109/JSTARS.2018.2868119. [7] S. R. Cloude, “Polarization coherence tomography,” Radio Sci, vol. 41, no. 4, Jul. 2006, doi: 10.1029/2005RS003436. [8] S. R. Cloude, “Polarimetric sar interferometry,” IEEE Transactions on Geoscience and Remote Sensing, vol. 36, no. 5 PART 1, pp. 1551–1565, 1998, doi: 10.1109/36.718859. [9] K. P. Papathanassiou and S. R. Cloude, “Single-baseline polarimetric SAR interferometry,” IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 11, pp. 2352–2363, Nov. 2001, doi: 10.1109/36.964971.

A method is implemented to detect forest height, structure and species using L- and P-band data over hemi-boreal swedish forest. Firstly, it is presented that L-band is less sensitive to the forest structure. a priori information on forest structure allows to estimate the forest height accurately. Furthermore it is indicated that the underlying ground topography is estimated using polarimetric sub-space decomposition at P-band to minimize the overestimation (bias). Then the Gaussian backscatter model is employed to extract the forest vertical structure at P-band. At low frequency, the inverted profiles differ significantly from one polarization to another and it indicates P-band sensitivity to the heterogeneity of the forest canopy. According to these measurements, it is shown that tree species, as pines and spruces, were presenting different signatures related to their vertical structure.

This work presents the first results of canopy height mapping with L-Band SAOCOM date by means of Polarimetric SAR Interferometry (Pol-InSAR). SAOCOM stands for Satélite Argentino de Observación con Microondas and it is operated by CONAE, the argentinean space agency. For this study case, a colocated temporal series of SAOCOM full polarimetric images covering the years 2021 and 2022 was acquired, with a temporal baseline of 8 to 16 days between each acquisition. The study area corresponds to Corrientes, Argentina, which is one of the main forest production regions in the country, and where field measurements provided by local producers were available in order to validate the obtained results, along with canopy height measurements from the Global Ecosystem Dynamics Investigation (GEDI) mission. The main planted species in Corrientes are Pinus and Eucalyptus, with heights ranging from 10 to 35 meters at the age of maturity. In the first step of this research, we carried out an exploratory assessment of the different decorrelation sources affecting the interferometric pairs. On one side, the temporal decorrelation term was analyzed with support of the different temporal spans between the existing images. On the other hand, the volumetric decorrelation was assessed by analyzing the sensitivity of the interferometric coherence to forest canopy height coming from field inventories and GEDI measurements. The results showed that, provided the shortest temporal span was guaranteed (that is, 8 days), a relationship between canopy height and interferometric coherence could be observed and estimated by models such as the Random Volume over Ground (RVoG) one. In a second step, a multi-baseline Pol-InSAR approach was adopted, following the methodology proposed by Denbina and Simard (2018) and Pourshamsi et al (2018) with Land Vegetation and Ice Sensor (LVIS) and UAVSAR data. Field data measurements and GEDI transects acquired over the study periods were used to select the RVoG-inverted maps that best represented the different forest heights present in the study areas. The results show that this multi-baseline selection approach is more appropriate than a single-baseline one, given the high variability that SAOCOM presents in the spatial baseline between acquisitions, and the lack of orbital control over this parameter, unlike other missions such as Sentinel-1 or UAVSAR. This study presents, for the first time, canopy height maps of 8-day temporal baselines, L-band, orbital interferograms, along a temporal window of almost two years. The conclusions yielded by this preliminary research are of great importance for the understanding of the Pol-InSAR capabilities of SAOCOM in the field of forest height mapping. The generalization of these results to other types of forests, such as boreal, tropical, or Mediterranean forests, is a foremost task, especially regarding the future L-band SAR missions like ROSE-L and NISAR. It is also important to assess the suitability of the RVoG model to characterize the structure of these forests with L-band data retrieved by the argentinean constellation.

Several SAR missions are designed to acquire global near zero-baseline interferometric time-series over vegetated land. The availability of a dense data set of systematic SAR acquisitions offers opportunities to map forest properties from new radar observables. Previous works have modeled the link between forest characteristics and the radar backscatter intensity [1]–[3] and interferometric coherence [4]-[7]. Considering the capabilities of the present and future SAR missions such as BIOMASS (ESA), NISAR (NASA/ISRO) and ROSE-L (ESA), this work addresses the role that interferometric time-series of temporal coherence plays in mapping forest properties with a model-based approach. We adopt the extended random-motion-over-ground (RMoG+) model [8] to capture the temporal decorrelation effects caused by the motion of the scatterers within the canopy layer and the changes in the ground dielectric properties. In [9]-[10], we devised a model inversion strategy to map forest height from time-series of C-band interferometric coherence and backscatter supported by sparse lidar data. Here, we extend the analysis to L-band to assess the interferometric coherence sensitivity to forest properties as well as to evaluate the inversion algorithm performance at lower frequencies. This work compares the L-band temporal coherence over two vegetated areas, one in Northern Spain and one in Laos. For both study sites, the L-band data set is a time-series of 9 ALOS-2 SLCs acquired every 14 days. The Spain data set covers the 2020 summer season (from June to September) whereas the Laos data set was acquired during the 2020 fall season (from September to December). The performed data analysis allows to assess the influence of the differences in the sites’s characteristics such as topography, vegetation type, and meteorological conditions, on both the L-band observables and the inversion algorithm performances. Along with the SAR data sets, LiDAR GEDI L2-L3 metrics, the ESA World Cover Map, and the ESA’s CCI Biomass map are introduced to constrain the analysis. As a first step, we simulate time-series of L-band InSAR coherence to study the RMoG+ coherence for the specific case of the L-band ALOS-2 interferometer. The RMoG+ model sensitivity analysis enables to address the role of each model parameter affecting the temporal coherence as well as to define a validity range for parameters estimation. Then, the ALOS-2 SLCs are processed using the JPL’s ISCE3 (InSAR Scientific Computing Environment) software to generate the interferometric data set. For each study site, we process a total of 36 coherence maps, with time intervals ranging from 14 up to 112 days. The long-time intervals are used to evaluate the long-term coherence maps which are required to include the sites’ topography in the forest parameters retrieval, as shown in [10]. To optimize the fitting of the RMoG+ model, various metrics such as the mean, median or maximum coherence are explored when dealing with multiple interferometric pairs having the same temporal gap. Prior to forest parameters retrieval, we compare model predictions with observations to assess the correlations between the radar observables and both system properties and sites’ characteristics such as GEDI tree height, above-ground biomass and topography using the COPDEM 3Sec digital elevation model. Finally, tree height and other parameters are estimated by fitting the RMoG+ model to the coherence time-series according to the tree-stages procedure proposed in [10]. The model parameters estimation algorithm involves the use of sparse LiDAR GEDI L2 tree height which is introduced to constrain the data model inversion. Along with the coherence time-series, the co- and cross-polarization ALOS-2 backscatter maps are introduced to reduce the number of unknown RMoG+ model parameters. Namely, the Water Cloud Model is fit to the dual-polarization backscatter data binned according to GEDI L2 canopy height sampled at 1 m intervals to estimate the unknown canopy extinction coefficient needed in the RMoG+. For each study site, the generated output is the 90-m resolution canopy height map derived from ALOS-2 coherence time-series. Validation of these outcomes is conducted using independent samples of LiDAR GEDI L2 tree height and the GEDI L3 interpolated data set. [1] Santoro, M., Cartus, O., and Fransson, J. E. “Integration of allometric equations in the water cloud model towards an improved retrieval of forest stem volume with L-band SAR data in Sweden”. Remote Sensing of Environment, vol. 253, pp. 112235, 2021 [2] Santoro, M. and Cartus, O. “Research pathways of forest above-ground biomass estimation based on SAR backscatter and interferometric SAR observations”. Remote Sensing, vol. 10, pp.608, 2018 [3] Cartus, O., Santoro, M., Wegmüller, U., and Rommen, B. “Bench-marking the retrieval of biomass in boreal forests using P-band SAR backscatter with multi-temporal C- and L-band observations”. Remote Sensing, vol. 11, pp.1695, (2019) [4] O. Cartus, M. Santoro, U. Wegmüller, N. Labrière, and J. Chave, “Sentinel-1 coherence for mapping above-ground biomass in semiarid forest areas,” IEEE Geosc. and Rem. Sens. Lett., vol. 19, pp. 1–5, 2021. [5] F. Sica, A. Pulella, M. Nannini, M. Pinheiro, and P. Rizzoli, “Repeat- pass SAR interferometry for land cover classification: A methodology using Sentinel-1 Short-Time-Series,” Remote Sensing of Environment, vol. 232, pp. 111277, 2019. [6] Seppi, S.A., C. López-Martinez, M. J. Joseau. "Assessment of L-Band SAOCOM InSAR Coherence and Its Comparison with C-Band: A Case Study over Managed Forests in Argentina." Remote Sensing, vol. 14.22, pp. 5652, 2022 [7] M. Lavalle, U. Khati, G. Shiroma, and B. Chapman, “Assessment of InSAR and PolSAR time-series from the 2019 NASA AM-PM campaign for biomass estimation,” in 2020 IGARSS, September 2020. [8] M. Lavalle, M. Simard, and S. Hensley, “A temporal decorrelation model for polarimetric radar interferometers,” Geosc. and Rem. Sens., IEEE Trans. on, vol. 50, no. 7, pp. 2880–2888, July 2012. [9] Lavalle, M., et al. "Global Sentinel-1 InSAR Coherence: Opportunities for Model-Based Estimation of Land Parameters." IGARSS 2022-2022 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2022. [10] Lavalle, M. et. al. “Model-based Retrieval of Forest Parameters from Sentinel-1 Coherence and Backscatter Time-series.”, Geosc. and Rem. Sens. Letters, (2022) accepted.

Spaceborne SAR Interferometry (InSAR) is not only well-suited for wall-to-wall mapping but also sensitive to vertical structures of natural scenarios (e.g., forest and glacier etc.,). Featured by good penetration capabilities, radar waves operated at lower frequencies (e.g., P-, L-) are able to capture information from underlying-ground, tree-trunks, multi-bounce scattering under forest canopy, while the higher frequency radar waves (i.e., C-, X-) are more likely to detect the information mainly from forest crown. In fact, a small amount of ground scattering information is still present in the radar observations at higher frequency as long as high-resolution signals are used[1]. Similar to LiDAR measurements, high-resolution radar signals are possible to penetrate gaps in the midst of clustered “hard” targets (typically for dense tropical forest), allowing us to develop a dedicated approach for determining underlying ground positions from high-resolution InSAR information. The original few-look (i.e., 2, 3 or 4) interferometric ground-finding approaches estimate underlying ground out of a phase histogram over a small area achievable either by manual interpretation [1] or via a statistical relationship decided by coherent electromagnetic scattering simulation with limited field inventory measurements[2]. The latter approach made one step further towards an automated processing for wall-to-wall mapping. However, such an effort had to assume a constant statistical model to estimate underlying topography over an entire InSAR scene due to a limited priori knowledge of ground positions. Actually, the statistical relationships for ground determination varies with respect to forest species, climate conditions, radar frequencies, spatial resolution, and surface slopes, etc. Many of the above factors (like forest species, topographical conditions) inherently manifest themselves in a spatially varying pattern over spaceborne InSAR scenes at large scale. A new perspective is offered by taking advantage of extensively but sparsely distributed spaceborne GEDI data to capture the spatially varying patterns, and resolve a spatially-varying statistical relationship by directly correlating the statistical moments of the observed few-look InSAR phase-center height with ground elevation (DTM) of each GEDI sample. The derived statistical coefficients in discrete distribution could help us to establish a statistical function of different factors, and thus invert underlying topography for the entire scene. As validated against NASA’s LVIS LiDAR data over a Howland research forest area in the U.S state of Maine, the underlying ground is estimated to an accuracy of RMSE of 3.48m at a spatial resolution of 50 m (See Figure 1) using a single pair of TanDEM-X InSAR data with overlapping GEDI data. Accurate information of underlying ground is able to facilitate subsequent inversion of top canopy height with InSAR derived phase-center products and RVoG model [3, 4]. This few-look InSAR histogram approach exhibits great potentials to capture structure parameters of forest scenarios (especially for those isolated scatterers acting as typical permanent scatterers) when only single baseline and single polarization InSAR data are available. Still, as shown in Figure 1 (c), some residuals remain and drive us to further analyse and refine this approach using the coherent electromagnetic simulation and real experiments over a forest scenario characterized by new ESA TomoSense campaign[5-7].

Mangrove forests are vitally important coastal ecosystems [1]. Despite their small coverage area, they are extremely productive, carbon rich ecosystems which are inherently vulnerable to sea level rise as well as human activities. Mangrove forests grow in many areas where field inventories are sparse, making remote sensing a valuable tool for monitoring and understanding these areas, and for reducing the uncertainty in mangrove biomass estimates [1,2]. We have experimented with estimating mangrove vertical structure parameters such as canopy height using data fusion of remote sensing data collected during the AfriSAR campaign [3]. We use UAVSAR PolInSAR and TomoSAR data fused with airborne lidar data from LVIS to improve the accuracy of PolInSAR and TomoSAR estimates of forest height and biomass. Data fusion is helpful for this approach as mangrove forests pose particular challenges for vertical structure estimation due to their unique vertical structure and above-ground root systems, which are different than other forest species. Model-based PolInSAR and TomoSAR forest height estimation approaches that were developed for non-mangrove forests are not necessarily applicable to the unique structure of mangroves. In addition, mangroves are prone to flooding which can impact the SAR coherence, and change the observed polarimetric scattering mechanisms. We have expanded on previous work in SAR/lidar data fusion [4] to train a deep neural network (DNN) regressor to estimate forest canopy height and biomass (using allometric relationships) from UAVSAR data, with LVIS lidar data used to train the network. Our previous work used a support vector machine (SVM) or DNN classifier to choose between baselines, and for each baseline, a model-based inversion approach was used to estimate forest canopy height. In this study, the DNN is trained to estimate the mangrove canopy height directly from the input PolInSAR features. The study areas are Mondah Forest and Pongara National Park, in Gabon, where UAVSAR and LVIS data were acquired in 2016 as part of the AfriSAR campaign [3]. The DNN is trained specifically on mangrove forest only, excluding other forest types. We will compare the results to a more generalized network trained on a wider range of forest types, including inland and boreal forest, to show the difference in performance between the specialized and generalized networks. The results are validated by comparison with lidar data not used in training the network, and in situ field plots in Pongara National Park, Gabon. References [1] Simard, M., Fatoyinbo, L., Smetanka, C. et al. Mangrove canopy height globally related to precipitation, temperature and cyclone frequency. Nature Geosci 12, 40–45 (2019). https://doi.org/10.1038/s41561-018-0279-1 [2] Simard, M., K. Zhang, V.H. Rivera-Monroy, M.S. Ross, P.L. Ruiz, E. Castaneda-Moya, R.R. Twilley, E. Rodriguez. 2006. Mapping Height and Biomass of Mangrove Forests in Everglades National Park with SRTM Elevation Data. Photogrammetric Engineering and Remote Sensing 72(3): 299-311. [3] Fatoyinbo, L., Armston, J., Simard, M., Saatchi, S., Denbina, M., Lavalle, M., Hofton, M., Tang, H., Marselis, S. Pinto, N., Hancock, S., Hawkins, B., Duncanson, L., Blair, B., Hansen, C., Lou, Y., Dubayah, R., Hensley, S., Silva, C., Hibbard, K. (2021). The NASA AfriSAR Campaign: Airborne SAR and Lidar Measurements of Tropical Forest Structure and Biomass in Support of Future Space Missions. 10.1002/essoar.10506317.1. [4] Denbina, M., Simard, M., and Hawkins, B. Forest Height Estimation Using Multibaseline PolInSAR and Sparse Lidar Data Fusion. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 10, pp. 3415-3433, Oct. 2018, doi: 10.1109/JSTARS.2018.2841388.

Forest structure refers to the 3D size, location and arrangement of trees, trunks and branches in a forest [1][2] and reflects therefore the actual forest state and evolution [3]-[5]. Accordingly, forest structure is an important parameter for assessing forest succession, productivity [6], biomass and biodiversity [7]-[9]. While numerous structure indices have been proposed and used in different studies [10] [11], defined primarily from field but also from remote sensing data, all of them fail in representing forest structure in a unique and comprehensive way, in the sense of allowing a univocal characterization of the possible tree distributions within a stand [12]-[15]. Nevertheless, despite the difficulties in defining an index appropriate for a wide span of applications and forest types, there is a common understanding that in order to express forest structure two complementary aspects of forest structure needs to be considered, namely the structural heterogeneity in the horizontal and in the vertical dimension. SAR Interferometry and SAR Tomography have been proven to be able to characterize physical forest structure [16]-[19]. The basic idea is to quantify the horizontal and vertical structural heterogeneity through the variability of tomographically reconstructed 3D reflectivity. In the case of SAR interferometry, where usually only one single-pass interferometric acquisition is available (with an appropriate vertical wavenumber), the reconstruction of a vertical reflectivity profile is not possible; at least not in a conventional tomographic way. However, the histogram of the interferometric phases (or alternatively of the converted phase center height) over a large enough area provides in many cases an approximation of the vertical reflectivity profiles. Such profiles are referred as canopy height profiles (CHP). In the case of TanDEM-X the high attenuation at X-band combined with the high spatial resolution of the interferograms support the correlation of phase center height variation to (top) canopy height variation, allowing the use of the derived CHP to extract relevant horizontal structure information. In this paper we investigate the phase center height variation - expressed at different spatial scales. For this a multi-scale wavelet analysis of the TanDEM-X phase center heights is proposed by means of a stationary discrete wavelet transform (SWT) implemented with an á-trous decomposition scheme [20]. For each level of the decomposition, following a dyadic scale, a set of wavelet coefficients was derived. The wavelet spectrum is computed by taking an averaged sum over the wavelet coefficients within the estimation window and maps the textural differences of the observed forest patches between the different scales. First, the results obtained from forest stands simulations obtained by using FORMIND, an individual and process-based forest gap model designed especially for tropical and temperate forests, are discussed. Forest gap models are well-established tools to investigate forest dynamics, able to account for processes such as growth, mortality and regeneration on the tree level and thus to simulate a wide range of different forest structure types [21][22]. A detailed description of the model can be found in [23]. In this work three different scenarios of forest succession over 500 years have been simulated: an undisturbed reference scenario and two disturbed scenarios one for logging and one for fire events. From the FORMIND dataset TanDEM-X phase center heights have been derived and used for the wavelet decomposition. Every scenario has its own distinctive wavelet signature, and the signatures of logging and fire events are clearly distinguishable from the ones of the natural undisturbed forest scenario. In a second step, the wavelet analysis is applied to real TanDEM-X data over different tropical forest sites and the obtained results are discussed and validated against reference data/ maps.

As a part of the TomoSense ESA campaign, Tomographic Synthetic Aperture Radar (TomoSAR) was applied to image the 3D reflectivity of a temperate forest located in the Kermeter area of Eifel National Park in North Rhine-Westphalia, Germany. The test site consisted of dominant beech and spruce forest, which provided a unique opportunity to study these two species separately, in addition to temperate forest in general. The topography is pronounced, with many ground slopes in the range of 20-40 ◦, allowing for the analysis of ground slope effects. Data points each covering a 0.5 ha area are studied, 531 at P-band and 663 at L-band (both monostatic and bistatic modes), for HH, HV and VV polarizations. An Airborne Lidar Scan (ALS) derived Above-Ground Biomass (AGB) estimation map is used as a reference for the sensitivity assessment. The ALS estimation show an R2 of 0.95 and an RMSE of 27 t/ha compared to in-situ forest plot AGB estimates. Vertical Reflectivity Profile (VRP), i.e. backscatter, observations show a clear AGB dependence, with characteristics depending on forest type. A high AGB is connected to a large portion of the intensity being located at a height about 10 m to 30 m above ground, with the height of maximum intensity in this interval increasing with AGB. This behavior is expected since a large forest biomass would imply many, tall and thick trees, often with large branches, contributing to the backscattered intensity. The main difference between the species is that spruce VRPs grow strongly in intensity and slightly in height with AGB, while beech VRPs grow strongly in height but not in intensity. This behaviour is speculated to originate from the forest structure, where a large spruce tree may contain a larger number of branches distributed along its height, thereby increasing its backscatter intensity, while a large beech tree may have a fairly constant number and size of branches located at the very top as it grows in height, hence not increasing its backscatter intensity. This has an impact on AGB retrieval, since vertical resolution is found necessary to estimate beech forest AGB. TomoSAR AGB retrieval is evaluated using three methods: total intensity integral (M1), representing performance without vertical resolution (such as 2D SAR), canopy intensity integral (M2), using the intensity of the 20 m to 30 m height layer, and canopy intensity fraction (M3), which is a normalized metric corresponding to M2 divided by M1. An exponential model is trained on 50 % of the data points. Three categories are studied: temperate forest (a combination of all forest types), spruce forest and beech forest. M1 show a sensitivity to spruce AGB at both P- and L-band, especially for HV. No frequency band show any significant sensitivity for beech. In turn, M2 greatly improves the sensitivity. In general, temperate forest at P-/L-band HV show an R2 of 0.53/0.43 and an RMSE of 38/36 t/ha (15/14 %). Correspondingly, M3 show an R2 of 0.48/0.45 and an RMSE of 40/35 t/ha (16/13 %). Compared with M2, spruce AGB sensitivity is reduced at both bands while that of beech is increased, especially at L-band. This emphasizes a dependence on height and low correlation with absolute intensity of beech forest AGB, while spruce forest AGB retrieval perform worse when the absolute intensity information is removed. The ground slope is shown to be a severe nuisance factor for P-band retrieval performance. When limiting the ground slope angle to below 10 ◦, the R2 measure for M2 over temperate forest is improved from 0.53 to 0.77, with the RMSE changing from 38 t/ha (15 %) to 29 t/ha (11 %). The most significant improvement is seen for spruce P-band HV, where R2 changes from 0.47 to 0.86. Temperate forest using M3 show a similar but slightly weaker improvement as M2.

By coherently combining interferometric SAR acquisitions at different polarization states, Polarimetric SAR Interferometry (Pol-InSAR) [1] allows to locate in height different scattering mechanisms present in the resolution cell. Together with the use of scattering models of the vegetation, as the well-known two-layer Random Volume over Ground (RVoG) model [2], Pol-InSAR has demonstrated its potential in forest monitoring applications. While the potential and limitations of Pol-InSAR measurements for reconstructing forest structure (parameters) are today well understood, the question of how to qualitatively and quantitatively characterize forest (structure) change using Pol-InSAR measurements is – besides first attempts [3, 4] – far from answered. This is primarily because of our incomplete understanding of how to parameterize forest change(s) and on how to resolve ambiguities arising from the superposition of structural and weather / seasonal changes. Both terms can be attributed to the lack of appropriate experimental “change” data sets as well as the absence of an established change processing methodology. In this sense, this paper constitutes a first step into the characterization of forest change by exploring and analyzing new Pol-InSAR data sets. Experimental SAR data acquired at low frequency bands, i.e. L- and P-band, capable of penetrating through the forest canopy until the underlying ground, are evaluated. In particular, data acquired by the DLR’s airborne F-SAR sensor in the frame of the TempoSAR22 campaign are available. The campaign is carried out over the temperate forest of Traunstein, in the South-East of Germany. The data set includes fully polarimetric multi-baseline data acquired at two different dates. In each date, different flights, each with different tracks (spatial / temporal baselines), were acquired. Therefore, this data set allows the study of dielectric changes throughout the daily cycle of the forest. Possible changes in the daily cycle of the forest are evaluated by exploiting the geometrical representation of Pol-InSAR data on the complex plane, the so-called coherence region [5]. According to the RVoG model, the coherence region follows a line segment that models the vertical profile of the forest canopy at pixel level. Under this assumption, a Pol-InSAR technique that is able to separate the scattering response from the underlying ground and the volume of the forest canopy [6] is applied. The obtained ground and volume coherences are the extremes that define the RVoG model line segment. A first evaluation is carried out over pixels corresponding to forest areas of data acquired at zero baseline, thus avoiding possible residual geometrical decorrelation effects. Under such conditions, a change in the ground and volume contributions translates into a change in the coherence region, and thus, into a change in the RVoG model line. The changes are as well analyzed in terms of the different observations of the master and slave covariance matrices of the interferometric pair. This provides a link between possible interferometric and polarimetric [7] changes. The analysis of such changes will be extended to non-zero spatial baselines. Further results of this study aim to help devise new strategies for the exploitation of future SAR missions, as the upcoming ESA BIOMASS. References [1] K. P. Papathanassiou and S. R. Cloude, “Single-baseline polarimetric SAR interferometry,” in IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 11, pp. 2352-2363, Nov. 2001, DOI: 10.1109/36.964971. [2] S. R. Cloude and K. P. Papathanassiou, “Three-stage inversion process for polarimetric SAR interferometry”, in IEE Proceedings-Radar, Sonar and Navigation, vol. 150, no. 3, pp. 125-134, June 2003, DOI: 10.1049/ip-rsn:20030449. [3] M. Tello, V. Cazcarra-Bes, M. Pardini and K. Papathanassiou, “Forest Structure Characterization From SAR Tomography at L-Band,” in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 10, pp. 3402-3414, Oct. 2018, DOI: 10.1109/JSTARS.2018.2859050. [4] V. Cazcarra-Bes, M. Pardini and K. Papathanassiou, “Definition of Tomographic SAR Configurations for Forest Structure Applications at L-Band,” in IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1-5, 2022, Art no. 4002605, DOI: 10.1109/LGRS.2020.3027439. [5] T. Flynn, M. Tabb and R. Carande, “Coherence region shape extraction for vegetation parameter estimation in polarimetric SAR interferometry,” IEEE International Geoscience and Remote Sensing Symposium, Toronto, ON, Canada, 2002, pp. 2596-2598 vol.5, DOI: 10.1109/IGARSS.2002.1026712. [6] A. Alonso-Gonzalez and K. P. Papathanassiou, “Multibaseline Two Layer Model PolInSAR Ground and Volume Separation,” in EUSAR 2018; 12th European Conference on Synthetic Aperture Radar, pp. 1-5. VDE, June 2018. [7] A. Alonso-González, C. López-Martínez, K. P. Papathanassiou and I. Hajnsek, “Polarimetric SAR Time Series Change Analysis Over Agricultural Areas,” in IEEE Transactions on Geoscience and Remote Sensing, vol. 58, no. 10, pp. 7317-7330, Oct. 2020, DOI: 10.1109/TGRS.2020.2981929.

One of the important applications of Synthetic Aperture Radar (SAR) time-series data is the detection of deforestation, especially in tropical and sub-tropical forests. The launch of C-band Sentinel-1 has increased the capability for operational monitoring of forests globally. Furthermore, the release of Level-2 L-band ALOS-2/PALSAR-2 ScanSAR data last year over many of the regions of interest has provided us more data for understanding the abilities and limitations of the SAR data to detect deforestation activities. The upcoming launch of L- and S-band NASA ISRO SAR (NISAR) mission and P-band ESA BIOMASS mission would provide us with unprecedented amount of time-series SAR data across the tropics which would potentially improve our understanding of interaction of microwaves with forest structures. SAR-based change detection techniques can be incoherent (use only amplitude), coherent (use phase or interferometric coherence) or use polarimetric information. Using polarimetric SAR data, changes has been estimated using polarimetric coherence matrix, target polarimetric signatures, feature vectors, dual-pol alpha angle, Markov random fields for multichannel SAR data, generalized maximum likelihood test for covariance matrices and using Bayesian statistics for change detection in a time-series. In this work, we use three major deforestation algorithms – the cumulative sums of change (CUMSUM) (Manogaran & Lopez, 2018), Omnibus test for change detection (OD) (Conradsen et al., 2016) and Wishart change detection (WD) (Conradsen et al., 2003). The three methods are applied to C-band Sentinel-1 SLC and L-band ALOS-2/PALSAR-2 dual-polarimetric SAR SLC data collected over three distinct forest sites – Haldwani forest (India), Central Kalimantan Forest (Indonesia) and Akanda National Park (Gabon). The forests in Haldwani are sub-tropical deciduous forests with a distinct senescence; the forests in Kalimantan are prone to under-canopy flooding and are tropical peat-swamp forests; the forests in Akanda National Park are dense tropical forests along with mangroves. All the three regions have seen planned and unplanned deforestation events due to commercial logging, illegal logging as well as expansion of agriculture lands In CUMSUM, the backscatter in HH/HV for ALOS-2 and VV/VH for Sentinel-1 along with dual-pol alpha angle and backscatter ratios are used to estimate deforestation. This is a sequential analysis technique and highlights one change according to the user-defined threshold in the forward direction for time-series data. In case of OD and WD, the polarimetric information content in the C2 matrix is used in addition to the backscatter. This helps in understanding the impact of polarimetric information for change detection. OD test is an extended version of bi-temporal Wishart test which operates on a time-series data rather than a pair of datatsets following complex Wishart distribution. Apart from the advantage from the added information in the form of polarimetric parameters of the time-series data, OD provides multiple change points in the time-series which proves beneficial if the temporal range under consideration is more or the study area is rapidly changing. The accuracy of the above mentioned algorithms is being studied for each of the three different kind of forest ecosystems thereby aiming towards finding an optimum way of change detection on a global scale. A few important observations from current analysis are: • Changes in under-canopy conditions (such as flooding) can lead to sudden increase in backscatter and false positives • Observational gaps in Sentinel-1 can result in loss of deforestation information due to lower saturation of C-band backscatter • Further integration of ancillary information in the form of soil moisture and precipitation are important to reduce uncertainties in deforestation alerts References: Conradsen, K., Nielsen, A. A., Schou, J., & Skriver, H. (2003). A test statistic in the complex Wishart distribution and its application to change detection in polarimetric SAR data. IEEE Transactions on Geoscience and Remote Sensing, 41(1), 4–19. https://doi.org/10.1109/TGRS.2002.808066 Conradsen, K., Nielsen, A. A., & Skriver, H. (2016). Determining the Points of Change in Time Series of Polarimetric SAR Data. IEEE Transactions on Geoscience and Remote Sensing, 54(5), 3007–3024. https://doi.org/10.1109/TGRS.2015.2510160 Manogaran, G., & Lopez, D. (2018). Spatial cumulative sum algorithm with big data analytics for climate change detection. Computers & Electrical Engineering, 65, 207–221. https://doi.org/10.1016/j.compeleceng.2017.04.006

Monitoring crop harvest and tillage using Sentinel-1 coherence change detection and polarimetric parameters Xianfeng Jiaoa, Heather McNarina, Marco van der Kooijb, Amanda Halsteada, Andrew Davidsona, Pamela Joossec aAgriculture and Agri-food Canada, Agri-Environment Division, Ottawa, Ontario, Canada bVanderkooij Consult, Ottawa, Ontario, Canada cAgriculture and Agri-food Canada, Harrow Research & Development Centre, Harrow, Ontario, Canada During the growing season, crop canopies protect the soil surface. However, if fields are tilled after crop harvest, soils lose the benefit of the protection of post-harvest residue. This loss of residue cover elevates the risk of wind and water erosion, degrades soil structure, and reduces the ability of soil to sequester carbon. Reducing or eliminating tillage improves soil health, and the adoption of this beneficial management practice (BMP) is encouraged. Agriculture and Agri-Food Canada (AAFC) researchers are developing methods to monitor the uptake of this BMP using satellite technologies. Synthetic Aperture Radars (SARs) are of particular interest given their near-all-weather data collection capability, an important advantage in monitoring harvest and tillage during the fall and spring. Both SAR intensity and phase are impacted by changes in surface roughness and, as such, could provided information relating to tillage activities. Our research investigated whether Coherence Change Detection (CCD) and polarimetric information could determine the timing of field were harvesting and tilling. A time series of Sentinel-1 Interferometric Wide (IW) dual pol (VV,VH) mode Single Look Complex (SLC) data were acquired for the area of Russell-Embrun in eastern Canada during the fall of 2022. This is an area of corn and soybean production and August to November is an active period of harvest and tillage. Two adjacent Sentinel-1A swaths overlap this site, which provided 17 images for the study site. A time series of 12-day repeat-pass interferometric coherence was created using the VV polarization. Sentinel-1 SLC data were converted to a 2 × 2 covariance matrix (C2). Based on the C2 matrix, four Stokes parameters (S0, S1, S2, and S3) and various polarization parameters were derived. These parameters included degree of polarization (DOP), degree of linear polarization (DOLP), linear polarization ratio (LPR), the two eigenvalues (l1, l2), a dual-polarized version of the entropy/alpha decomposition, and the Shannon entropy. Observations of field conditions were gathered on the same day as the Sentinel-1 overpasses, using roadside windshield surveys. Thirty-five corn and forty-five soybean fields were visited 12 times from August 26 to November 23, 2022. Field observations included whether the field had been harvested, tilled or if other management practices (such as manure applications) had taken place. Soil moisture status (dry, moist, wet) was also visually assessed. Observations of harvest and tillage events were compared to CCD products and temporal changes in polarimetric parameters. CCD values fell below 0.2 when crops were harvested or soils tilled. The drop in coherence at the time of harvest/tillage was preceded and followed by higher coherence indicating periods when field conditions were stable. Determining whether changes identified in CCD products were due to harvest, tillage or other management practices required additional information. Several parameters, including VV and VH backscatter, S0, S1, l1, and l2 are sensitive to the target structure and temporal trends revealed that these parameters were tracking “crop ripening-harvest-tillage” patterns. These parameters significantly decreased as crops senesced but stayed stable or increased slightly after harvesting and tillage. A simple logic model was created to determine if the change identified by the CCD products was a result of harvest or tillage.

There is global interest in monitoring both the area and species of agricultural crops across massive land areas for food security, trade, marketing and planning purposes. Routine and efficient satellite-based crop monitoring and management in tropical-, or other cloud-affected areas of the globe has been a major challenge for the agriculture sector. As one of largest grain producers, Australian dryland crops are planted across nearly 26 million hectares annually, with most of the production occurring in the wheatbelts of WA, NSW, Victoria, SA and Queensland. The Australian cropping scene is variable where farmers grow multiple crop species to diversify their income base and manage biological risks from diseases. In any given year the area planted to a particular species may change, depending on the climatic conditions and prevailing commodity prices. Thus, predicting and monitoring the area planted to a particular crop species is complex and dynamic. Optical imagery can be used to monitor plant growth through time, but frequent rainfalls and cloud during the growing season of rainfed crops in Australia often restraint the usage of optical satellite imagery. Therefore, development of new science capability and the next generation of crop monitoring information products that discriminate between crop species, predict crop area and yields becomes significantly important for agribusiness. The advent of a new generation radar satellites, especially Sentinel-1 mission of the European Copernicus Program offering the routinely global C-band SAR data coverage and free data access policy, opens up a major opportunity to enhance current, mostly-optical satellite crop mapping approaches. Remote identification and monitoring of agricultural crop growth and early estimates of crop types and/or potential yield have become possible. They can help improve on the more commonly used time-intensive field-based estimates. In this paper we will introduce how we process the large volume of Sentinel-1 IW dual-pol data acquired over the whole state of Victoria into analysis ready data including normalised radar backscatters of Gamma0 and polarimetric decomposition parameters for the 2016 growing season. Especially before polarimetric decomposition, a thermal noise removal method is implemented for Sentinel-1 dual-pol SLC data processing. Together with ground sampling data, Random Forest classifier, which focuses on model prediction based on known properties learned from the training data, has been implemented to S1 time series with different length of dates. Discriminant analysis by means of canonical variate analysis (CVA) is carried out for the pre-processed S1 data and training datasets. CVA finds the linear combinations of the original variables or image bands that maximise the differences between reference classes relative to the variation within the classes. Separation scores of crop discrimination are derived by various combinations of image bands. It shows the discrimination contribution of each band. Also similar analysis is implemented to different lengths of Sentinel-1 dates. That demonstrates how polarimetric analysis and S1 times will improve the crop type classification accuracies. There were 5 adjacent Sentinel-1 orbits to cover the state of Victoria, but ground sampling data were mainly collected within an intensive agriculture zone fallen in the footprint of the second orbit. To relief the temporal and spatial variations, a classifier propagation method is introduced to create the training datasets for classifying the data of neighbouring orbits. It is based on the dominant probabilities of classification layers derived by Random Forest. After repeating the process of classification propagation, crop maps for all 5 orbits are obtained and then mosaicked into a crop map for the whole state of Victoria. The prospective study demonstrated the improved crop type classification accuracies were achieved with polarimetric analysis of Sentinel-1 time series and showed the very promising application potentials for broad-scale crop mapping, especially during the growing season while often affected by the rainfall and clouds. This SAR approach combining Landsat imagery for dryland crop mapping has rolled out in Australia to form the core part of annually national-scale crop mapping since 2018.

Rapid and accurate crop type mapping is of great significance for agricultural management and sustainable development. Time-series multi-polarization synthetic aperture radar (SAR) data can reflect the dynamical variations of crop canopy and soil during various phenological stages, which is suitable for obtaining the large-scale distribution of crop types and continuously monitoring crops. Particularly, the dual-polarization Sentinel-1 satellite with its short revisit time (12 or 6 days) and large swaths (260 km in range) has strongly promoted the development of this application process. According to different utilization forms and understandings of time-series SAR datasets, existing classification methods can be mainly divided into two categories. 1) The first type of methods considers time-series datasets as the simple stack of multiple features, such as the joint polarimetric covariance matrixes or the joint feature vectors. Such stacks of features have been applied to machine learning algorithms to separate various crop types. The common machine learning algorithms include the maximum likelihood classification (MLC), decision tree (DT), random forest (RF), support vector machine (SVM), etc. This type of methods has been developed rapidly with the advancement of machine learning algorithms. However, these methods have not revealed the dynamical characteristics of SAR datasets in the temporal dimension. 2) The second type of time-series classification method uses phenological knowledge or time-varying characteristic to describe temporal context information between various acquisitions and improve crop discrimination. It applies phenological knowledge to improve feature selection or referring sequence pattern. Besides, some statistical models as the Hidden Markov Model (HMM) or the Conditional Random Fields (CRF) are applied into the crop classification, for explaining the temporal dependencies between single images of time-series datasets caused by crop phenology. Particularly, the exploration of the phenological sequence concept and temporal context is important for identifying crop types in the regions with random agroclimatic conditions or agricultural practices, where a given crop type might be recently sown in one parcel and ready to harvest in the neighbor parcel. Such uncertainty of crop phenological evolutions is presented as the time-series unalignment of different parcels from a given crop type, and it introduces obvious classification errors. Based on this, the classification method based on the time-series alignment of time-varying feature curves has been widely used, which can take the uncertainty of the phenological cycles of crops into account. The most classical method is the nearest neighbor (NN) classifier based on the dynamic time warping (DTW) alignment. While the DTW alignment does not consider the local shape of the curves and temporal ranges, and the NN classifier is inadequate in generalization, which restricts the accuracy of crop type mapping. In this report, a pairwise proximity function support vector machine (ppfSVM) classification method with the time-weighted shapeDTW (TWshapeDTW) alignment will be proposed. Firstly, the novel time-series alignment method simultaneously considers the local shape of the curve and the temporal range of crops. Besides, the novel ppfSVM classifier with the time-series alignment kernel is established. Such kernel matrix considers dual-similarity metrics of multiple features, and it is positive semi-definite (PSD) in this classifier. The proposed classification method establishes a bridge connecting multi-temporal polarimetric SAR data and crop classification maps. It uses the vertices and courses of time-varying feature curve to describe the crop phenological evolution. The features of different dates at the same phenological stage are rematched based on the curve shape, to ensure that the pairwise features are processed on the “the same phenological stage”, rather than “the same date”. It avoids the negative impact of uncertain phenological evolutions on crop classification. Furthermore, the combination of time-series alignment algorithm and SVM classifier not only activates the potential of time-series polarimetric SAR data to characterize the phenological evolution, but also improves the generalization of the classifier under the small number of samples. With 42 Sentinel-1 dual-polarization SAR images located in Gansu Province, China, the crop classification maps in 2018 and 2019 are generated respectively. The proposed method in this paper obtains the overall accuracies of more than 90% in both two years, and the changes of crop types from 2018 to 2019 are also in line with the actual crop rotation. Compared with traditional classification methods (SVM and the NN method with the DTW alignment), it is found that the proposed method has higher overall accuracy (OA) and better robustness in the case of small number of samples. The OA’s improvements of our method compared with the SVM method are 3% and 1% in 2018 and 2019, respectively. Such improvements are 14% and 12% respect to the NN method with the DTW alignment. This method can achieve more obvious improvement under the condition of less training samples and unaligned phenological sequences. Furthermore, the novel TWshapeDTW alignment is superior to the DTW and the time-weighted DTW alignment under the ppfSVM classifier. The OA’s improvement introduced by the TWshpaeDTW alignment respect to the DTW alignment can obtain 5% and 4% in 2018 and 2019, respectively.

Soil moisture is fundamental for several land applications that involve agricultural practices, such as irrigation management, crop growth monitoring, and the new precision farming approach. Thanks to the growing technological development in the field of Earth observation and remote sensing, Synthetic Aperture Radar (SAR) data can be used to estimate soil moisture with high accuracy and frequent temporal sampling. Unfortunately, the radar signal is influenced not only by changes in soil moisture, but also by the vegetation (i.e., plant height, growth stage, etc.) and other factors such as surface roughness. In this context, a possible solution could be given by the use of a polarimetric SAR decomposition, which could help disentangle the effects of vegetation from those of the soil by separating different scattering mechanisms, i.e., surface, double-bounce and volume. The objective of this study is to analyze which scattering mechanisms are mostly correlated to soil moisture or vegetation in order to evaluate their exploitation within a soil moisture retrieval algorithm. In fact, surface and double-bounce, which are known to be functions of the dielectric properties of the soil (surface) and soil-vegetation (double-bounce) are the contributions that could be used to estimate soil moisture after the vegetation effects has been removed (i.e., volume term) from the measured data [1]. A set of different polarimetric SAR decompositions, such as the Freeman-Durden three-components decomposition [2] or the Nonnegative Eigenvalue Decomposition [3], are applied to a time-series of geocoded covariance/coherency matrices (C3/T3). The measured C3/T3 matrices are derived from full-polarimetric L-band radar data collected by the SAOCOM-1A mission over five corn fields located in the Monte Buey site, Argentina, between October 2019 and February 2020. In particular, 9 descending SLC (Single Look Complex) images with a nominal spatial resolution of 10m x 6m in ground range and azimuth, respectively, are considered. The temporal evolution of the scattered powers related to each scattering contribution, along with the backscattering coefficients at different polarizations, will be evaluated with respect to in-situ data (i.e., soil moisture, plant height, crop growth stage) collected by the Argentinian Space Agency (CONAE) during the 2019-2020 season and concurrently with the SAOCOM acquisitions. Satellite-derived parameters, such as the Normalized Difference Vegetation Index (NDVI), will be also involved in the analysis as a proxy of the vegetation growth conditions. Then, the results obtained by using the polarimetric decomposition approach will be compared with the ones obtained by the application of the fully polarimetric model developed at Tor Vergata University. In fact, this electromagnetic model is able to simulate the radar backscatter along with the three different contributions (surface, soil-vegetation interaction, and volume) [4]. The in-situ data collected during the 2019-2020 field campaign will be used as a training dataset for the Tor Vergata model. Finally, a comparison will also be made with the results obtained by applying an empirical model, the Water Cloud Model (WCM), which is used to correct the vegetation effects on the radar backscatter [5]. [1] I. Hajnsek, T. Jagdhuber, H. Schon and K. P. Papathanassiou, "Potential of Estimating Soil Moisture Under Vegetation Cover by Means of PolSAR," IEEE Trans. Geosci. Remote Sens., vol. 47, no. 2, pp. 442-454, Feb. 2009. [2] A. Freeman and S. L. Durden, "A three-component scattering model for polarimetric SAR data," IEEE Trans. Geosci. Remote Sens., vol. 36, no. 3, pp. 963-973, May 1998. [3] J. J. van Zyl, M. Arii and Y. Kim, “Model-Based Decomposition of Polarimetric SAR Covariance Matrices Constrained for Nonnegative Eigenvalues,” IEEE Trans. Geosci. Remote Sens., vol. 49, no. 9, pp. 3452-3459, Sept. 2011. [4] M. Bracaglia, P. Ferrazzoli, and L. Guerriero, “A fully polarimetric multiple scattering model for crops,” Remote Sens. Env., vol. 54, no. 3, pp. 170-179, 1995. [5] N. Baghdadi, M. El Hajj, M. Zribi, S. Bousbih, “Calibration of the Water Cloud Model at C-Band for Winter Crop Fields and Grasslands,” Remote Sens. 2017, 9, 969.

In this study, the authors propose to detect agricultural events such as harvesting, crop planting or ploughing of a field. To do so, the method used is the temporal coherence of multitemporal and polarimetric data proposed by Ni et al., 2022 [1], which was already used for crop classification but not agricultural event detection. Detecting agricultural events, part of crop monitoring, is less developed than crop classification. In addition, accurate monitoring of the condition of a field is important, whether for government monitoring, disease control, irrigation management, or to find out which farming practices maximise the yield of a field, for example, by comparing the yield obtained from different sowing and harvesting dates. Moreover, it is important to precisely date these events in the context of the new common agricultural policy. A classic way to get this information is to send governmental or organisational agents to go from field to field and note the state of the field, whether it is cultivated or not, and what is growing there. This method is no longer relevant because of the new regulation, as it does not allow precise dating and cannot be applied to each field. An alternative is the use of optical images acquired by satellites. This method allows regular monitoring of fields on a national scale, like crop classification. Optical data are also often used to detect agricultural events, but the right dating detection of these events can be an issue. Indeed, when the weather is cloudy, sometimes during months, it is impossible to use optical data to analyse the fields as the clouds hide them and, therefore, to get an accurate dating. To study the fields in all weather conditions, the authors propose using radar data acquired by the Sentinel-1 satellite, as clouds do not block them. An RPG ground truth provided by the IGN is used to corroborate our results. The use of SAR data for crop classification or agricultural event detection is standard practice. To do so, we usually use the intensity returned by the fields for different dates and polarisations. Multiple methods exist to classify fields using machine learning or not according to the intensity returned over time. Moreover, an intensity variation can be due to an agricultural event and can therefore be detected. Adding interferometric coherence information to this intensity information improves the detection. To get this interferometric information, we analyse the variation in coherence between two acquisitions obtained with the same angle of view but at two different times. If the absolute value of the coherence is high, there has been little change between the two images. However, if the absolute value of the coherence is low, it means that there have been significant changes between the two acquisitions, which means that there is an agricultural event. An inherent issue with this method is that when we study an area of low intensity, the coherence obtained is necessarily low, so it is complex to study such areas. One is then generally limited to studying interferometric coherence only when one studies an area which, without temporal changes, will present a high coherence. Unfortunately, this is not the case for agricultural areas using C-band radar as for the Sentinel-1 satellite data. The idea is to use a method that allows information to be obtained even when the coherence is low. The method used here is the temporal coherence of multitemporal and polarimetric SAR data [1]. This method has already been used for agricultural classifications, and we apply it here to detect agricultural events. Firstly, we pre-process Sentinel-1 data using SNAP software to calibrate and deburst the SLC IW data. Then the polarimetric interferometric coherence matrix is calculated for each pixel in the image. This matrix is calculated using two acquisitions obtained with the same viewing angle but at two different times. These acquisitions have several polarisations, two (the VV and VH polarisations) in the case of the Sentinel-1 satellite data. These data are then used again in SNAP to perform the terrain correction of the images obtained. This stage is needed to study the results in areas where we have ground truth. We studied the temporal variation of the interferometric coherence obtained for each area with only VV or VH polarisation. After that, the obtained polarimetric interferometric coherence matrix is separated into a product of two terms, one related to coherent changes and the other to radiometric changes. The first term is only usable when the coherence is high, which is not the case in this study. However, the second term, named asymmetric coherence, contains information, and by studying it, we observe radiometric variations of the studied area. This asymmetric coherence has been studied for different pairs of unitary vectors chosen. These parameters allowed us to detect agricultural events confirmed by ground truth more efficiently than using only mono-polarisation interferometric coherence. In conclusion, we propose a new way to detect agricultural events in this study. We use temporal coherence of multitemporal and polarimetric data computed with Sentinel-1 satellite data to do so. This method improves detection compared to simple interferometric coherence, which uses only one polarisation. Moreover, this method seems promising for differentiating different agricultural events by improving the use of different unitary vectors. Indeed, in some cases, one can have events with a similar effect in terms of intensity, causing a decrease in coherence but a different polarisation effect. This study highlights the importance of utilising new and innovative methods in remote sensing to benefit agriculture. [1] J. Ni, C. López-Martínez, Z. Hu and F. Zhang, "Multitemporal SAR and Polarimetric SAR Optimization and Classification: Reinterpreting Temporal Coherence," in IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-17, 2022, Art no. 5236617, doi: 10.1109/TGRS.2022.3214097.

Verifiable and consistent measurement of forest carbon stocks and fluxes are necessary and they require to know where the biomass carbon is whether the vital functions of forests are changing, and what their future holds. Space agencies have made enormous investments in Earth Observation missions to map forest biomass across continents to support climate science and carbon markets. But satellites alone cannot produce accurate carbon maps–all maps vitally rely on field data collected by people and instruments to train their models and validate their products. To ensure satellites are producing reliable maps, it is needed frequently-acquired, high-quality field data. Furthermore, the challenge of acquiring ground biomass measurements is also one of environmental and social justice. The forests for which reference data are most needed, and the people depending on these forests, already suffer the worst impacts of climate change. Those in-country partners with unique forest expertise are key players in the fight against climate chaos, yet they are among the most disadvantaged globally. It follows that they need sustained support not only to collect data but to grow, train, and develop their own group's capacities. The GEO-TREES initiative proposes to fill this critical gap by building the world's first ground-based, standardized, open-access, equitably developed, reference forest biomass validation system to ensure that satellite observations accurately represent real forest carbon stocks, today and in the future. GEO-TREES is an ambitious world-wide network. It aims to establish at least 100 high-intensity forest biomass reference sites, to represent the main environmental and anthropogenic dimensions over which forests occur globally, and achieve greater sampling intensity in the critical tropics with an additional 210 lower-cost highly-distributed sites. Standing at the nexus of ecology and remote sensing, GEO-TREES builds on four principles: 1. Partnerships & engagement: To generate high-quality ground measurements, GEO-TREES partners with ecological and botanical specialists around the world. Partners are fully engaged and involved in every step of building the reference system. Without strong representation and fair funding of partners, particularly from the Global South, science capacity cannot be advanced, and the long-term sustainability of the GEO-TREES system would not be possible. 2. Innovative technologies: Ground measurement involves four integrated sets of measurements: forest inventory plots, airborne laser scanning, terrestrial laser scanning, and climate monitoring. GEO-TREES is based on established recommendations of the Committee on Earth Observation Satellites for validating biomass observations (https://lpvs.gsfc.nasa.gov/AGB/AGB_home.html), and will improve them based on new advances when necessary. 3. Long-term commitment: Forests are alive. Forest carbon stores change, sometimes rapidly, through space and time. Maintaining current, accurate estimates of carbon and biomass stocks requires continued long-term measurements. Long-term measurements also ensure the continued engagement and participation of partners throughout the system. 4. Open-access data: GEO-TREES is committed to equitably produced and openly shared global forest biomass reference measurements. High-quality, high-resolution maps of the world’s forests developed through the Earth Observation missions in partnership with GEO-TREES will be made open to all.

Terrestrial laser scanning (TLS) is being recognised as a key technology in forest monitoring by providing highly detailed in-situ measurements of 3D vegetation structure. While applications of TLS forest point clouds are manifold, including plant functional trait analysis and plant area index (PAI) estimation, it is a particularly valuable tool within the context of aboveground biomass (AGB) estimation (Calders et al., 2020) and how this is derived through spaceborne sensors. Here, we highlight (1) how TLS can be used as reference data for in situ AGB estimates; and (2) how TLS can be used to build digital twins to support cal/val activities of remote sensing. We demonstrate both applications using TLS data from a typical deciduous forest, Wytham Woods in the UK. 1. TLS as reference data for AGB measurements Through 3D reconstruction of individually identified trees using quantitative structure models (QSMs), TLS data allows for nondestructive estimation of tree volumes, which, combined with species specific wood-density, can serve as a highly accurate ground-truth reference for AGB quantification (Figure 1). Research has shown that TLS derived AGB is favourable both in terms of sampling strategy and AGB accuracy when compared to traditionally used allometric models (Calders et al., 2015). 2. TLS can directly support cal/val activities of spaceborne sensors through radiative transfer models Furthermore, these detailed QSMs can be parameterised as so-called digital twins which can serve as input for radiative transfer models (RTMs). RTMs are a powerful tool to establish the physically-based link between in situ structure and how it is monitored with satellite remote sensing, hence enabling the upscaling of AGB estimation. RTMs enhance our ability to monitor and understand the coupling between emitted, reflected or scattered electromagnetic (EM) waves, and a scene characterised by the structure and spectral signature of its biophysical components (e.g. leaves, bark, understorey). Figure 2 shows a case study of the first explicit digital forest twin derived from TLS data, which was further parameterized for optical RTM. This 1-hectare proof of concept RTM scene was generated for Wytham Woods, a deciduous forest near Oxford (United Kingdom) consisting of all 559 individual trees modelled explicitly from TLS, and showed the ability to simulate satellite data such as Sentinel-2 (Calders et al., 2018). 3. Outlook: integration of digital twins in microwave RTM Whereas optical RTMs have commonly been used in vegetation remote sensing to simulate optical (see Figure 2 as an example) or lidar data using 3D input scenes (Widlowski et al., 2015), microwave RTMs using 3D canopies as input is more recent (Tanase et al., 2019), partly driven by recent radar satellite missions, such as ESA BIOMASS, targeting vegetation specifically. However, currently established methods all are based on assumptions to generate their 3D input models and do not use an exact replica of real trees and forest, as is possible from TLS data. Ongoing and future work within the context of the SPACETWIN project (https://spacetwin.ugent.be/), focuses on enhancing our understanding on forest disturbances and recovery from space. SPACETWIN aims to integrate realistic TLS-derived digital twins of forest structure as 3D RTM input to build an emulator for optical, lidar and radar satellite data that will allow for near real-time interpretation of disturbances (Figure 3). This includes, among others, using the MIPERS4D RTM to simulate microwave (C-, L-, and P-band) backscatter, which can be parameterised using field-collected dielectric permittivity to represent the water content (Tanase et al., 2019). The combination of structurally accurate 3D digital twin forests with a parameterized microwave RTM would allow for a powerful instrument to facilitate the calibration and validation of remote sensing data and derived biophysical products such as biomass. References Åkerblom, M. et al. (2018) ‘Non-intersecting leaf insertion algorithm for tree structure models’, Interface focus, 8(2), p. 20170045. Calders, K. et al. (2015) ‘Nondestructive estimates of above‐ground biomass using terrestrial laser scanning’, Methods in ecology and evolution / British Ecological Society, 6(2), pp. 198–208. Calders, K. et al. (2018) ‘Realistic Forest Stand Reconstruction from Terrestrial LiDAR for Radiative Transfer Modelling’, Remote Sensing, 10(6), p. 933. Calders, K. et al. (2020) ‘Terrestrial laser scanning in forest ecology: Expanding the horizon’, Remote sensing of environment, 251, p. 112102. Tanase, M.A. et al. (2019) ‘Synthetic aperture radar sensitivity to forest changes: A simulations-based study for the Romanian forests’, The Science of the total environment, 689, pp. 1104–1114. Widlowski, J.-L. et al. (2015) ‘The fourth phase of the radiative transfer model intercomparison (RAMI) exercise: Actual canopy scenarios and conformity testing’, Remote sensing of environment, 169, pp. 418–437. Acknowledgements This research was funded by the European Union (ERC-2021-STG Grant agreement No. 101039795). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.

Precise descriptions of forest productivity, biomass, and structure are essential for understanding ecosystem responses to climatic and anthropogenic changes. However, relations between these components are rarely investigated, in particular for tropical forests. We developed an approach to simulate forest dynamics of around 410 billion individual trees within 7.8 Mio km² of Amazon rainforest (using the FORMIND forest model). We then integrated remote sensing observations from Lidar in order to detect different forest states and structures (caused by small-scale to large-scale natural and anthropogenic disturbances). Under current conditions, we identified the Amazon rainforest as a carbon sink, gaining 0.56 Gt C per year. We also estimated other ecosystem functions like gross primary production (GPP) and woody aboveground net primary production (wANPP), aboveground biomass, basal area and stem density. We found that successional states play an important role for the relations between productivity and biomass. Forests in early to intermediate successional states are the most productive and carbon use efficiencies are non-linear. Simulated values can be compared to observed values at various spatial resolutions (local to Amazon-wide, multiscale approach). Notably, we found that our results match other observed patterns (e.g. on forest productivity). We conclude that forest structure has a substantial impact on productivity and biomass. It is an essential factor that should be taken into account when estimating carbon budgets of the Amazon rainforest.

Africa’s terrestrial biosphere plays a critical role in the global carbon (C) cycle, dominating global C emissions from fire and significantly influencing interannual variation in atmospheric CO2 accumulation. However, Africa’s C-cycling remains highly uncertain; diminishing our capacity to make meaningful predictions of how African ecosystems are likely to respond to changes in climate and disturbance. C-stocks and changes are determined by the balance of inputs (e.g. allocated photosynthate) to plant tissues (e.g. wood) and its residence time. Both allocation and residence time remain uncertain across the globe. Recent studies have highlighted the association between uncertainty in C-stocks with uncertainty in allocation and residence time – providing a correlative link between stock changes and ecosystem traits. Therefore, a key knowledge-gap is information on how the major C-stocks are evolving over time. Satellite-based Earth Observation (EO) offers us the opportunity to begin to address this gap by providing estimates of above ground biomass (AGB) across the whole of Africa and, for the first time, annual estimates spanning decadal timescales. Here we present a model-data fusion analysis where we combine pixel level, i.e. local, EO information on leaf area with a new state-of-the-art map of African AGB spanning 2007-2017 with information on meteorology and disturbance to constrain an intermediate-complexity model of the African terrestrial C-cycle. We quantify the information content of moving from having a single to repeat AGB information across Africa. We will present our analysis of how ecosystem C-stock change is impacted and how these are underpinned by changes to ecosystem traits.

Land-use intensification and climate change are increasing the pressure on water availability and use around the world. Managed and unmanaged forests in different parts of the world are increasingly vulnerable to an increasing frequency of severe drought events caused by a changing climate. However, it is very difficult to quantify the baseline conditions of the growth and water stress on forests, owing to their remoteness, steep terrain and size. Planted forests have a key role in managing water resources sustainably, however, forest hydrological processes are poorly understood. In New Zealand, the government-funded 5-year Forest Flows research programme (www.forestflows.nz) is investigating these challenges with the novel integration of various terrestrial and remote sensing technologies. The overall goals are to capture data that identifies and quantifies the key drivers of forest hydrological processes spanning multiple scales for tree water use, watershed water retention, and watershed water release managed Pinus radiata (D. Don) plantations. Forest Flows has deployed a range of terrestrial sensors and remote sensing technologies to measure flux in groundwater, surface water, soil water, nitrate, tree growth and water use, and meteorological data in real time across five watersheds located across a range of climatic and physiographic regions. The sites are across a rainfall gradient with each site having different geological and soil properties. The collection of near real-time, big data from 1,717 terrestrial sensors required new approaches for data storage, management and analysis. This includes a combination of statistical, supervised and unsupervised machine learning, and process-based modelling approaches. Remote sensing measurements include multispectral camera and the SlimSAR L- & P-Band multi-frequency polarimetric SAR (Artemis, New York, USA) on an aerial platform, as well as Sentinel-2 and SMAP satellite acquisitions. SlimSAR is a novel combination of L- and P-Band measurements on the same unit. It is being deployed to take novel measurements of primarily soil moisture to 1m depth through a dense forest canopy on steep and complex terrain. Low-altitude airborne campaigns occurred in March/April and October/November 2022 and 2023 to provide data when the sites likely have high and low water stress, respectively. Individually these measurements provide valuable information, however, it is the fusion of temporally rich terrestrial data and spatially rich remote sensing data that will provide insights into forest hydrological processes. These data will also provide the ability to accurately scale measurements from point (tree), to watershed, to forest scales. This presentation will provide an overview of the Forest Flows programme and preliminary results - with a focus on SlimSAR retrievals for soil moisture, above ground biomass, and roots. We will present how periodic measurements with SlimSAR relates to forest hydrological processes.

SAR polarimetry faces many obstacles to its use, which is still not widespread: the technological cost is often accompanied by a decrease in swath size, or more often by a degradation in resolution. Worse still, many of the so-called second-order parameters, which allow additional information to be extracted from the random structure of the signal, require estimates that are often carried out spatially, once again to the detriment of the resolution. But recently, the easy access to time series allows us to consider a new development of polarimetry, especially for urban environments. We have named this discipline "PolTimeSAR" in 2019 [Colin, 2019]. The possible benefits expected from polarimetry can be considered for several functionalities: classification for land use, estimation of physical parameters (dielectric permittivities, orientations of anthropic objects, biomass), and detection (of buildings, changes, vehicles, permanent points for deformation estimation). In this study, we are mainly interested in the detection in time series, either for: - The detection of Permanent Scatterer, useful for estimating deformations, or obtaining a density of points belonging to artificial structures [Hanjsek and Desnos, 2021]. - The detection of changes, whether they are persistent as for the setting up of a construction site, or punctual in time as for a vehicle appearing only at a single date. To handle this functionality, we propose two new formalisms, particularly adapted to the Sentinel-1 VV/VH acquisition mode: - The temporal coherence matrix of the received wave, a 2 × 2 complex matrix obtained by averaging of the direct product of the Jones vector and its Hermitian conjugate - The multivariate coefficient of variation [Albert and Zhang, 2010] applied either to amplitude polarimetric vectors or Stokes vectors 1- Wave coherence When recording a time series for two polarimetric channels, copolar and cross polar, as is the case for the Sentinel-1 images for example, or for Tandem-X (HH,HV) time series, we can consider that the signal corresponds to the reception of a Jones vector, for a fixed polarization emission. In this case, the time series can be considered as a succession of Jones vector records of the scattered wave. The random aspect of the fluctuation requires a ́statistical study of the variations and correlations between the components (Ex(t), Ey(t)) of the electric ́field. We propose to perform this analysis through the coherence/covariance matrix which introduces the notion of coherence related to the re-emitted wave. It is a 2 × 2 matrix obtained by the time average of the direct product of the Jones vector and its Hermitian conjugate. We propose the diagonalization of this coherence/covariance matrix of the received wave, as well as the description of the eigenstates thus obtained, interpreted as polarization ellipses. 3 parameters are proposed: orientation of the ellipse, ellipticity, and randomness of the eigenstates, inspired by the scatering diversity in [Praks et aL, 2009]. The results obtained allow the detection of polarimetrically deterministic targets in a very reliable way, without using spatial estimation, and with a result totally independent of the total intensity of the wave. 2 - The coefficient of variation The coefficient of variation is a statistical parameter defined as the ratio between the standard deviation and the mean. In radar time series, it is used for two purposes: the detection of permanent scatterer and the change detection [Colin and Nicolas, 2020]. In both cases, it has been shown that having a diversity of polarizations improves the detections. Indeed, deterministic objects often have a preferred orientation axis, and in this case, a preferred polarimetric response as well. In [Colin, 2019], it was shown that over continental areas, only total polarimetry could detect a maximum of changes. The removal of a polarimetric channel is accompanied by the loss of at least 25% of detections. For this reason, in order to extend the algorithms that rely on the coefficient of variation to detect changes to a polarimetric scenario, the maximum of the coefficients of variation calculated individually in each of the available polarimetric channels is used. Here, however, we go further in this approach, and propose the analysis of the different existing multivariate coefficients of variation, and its extremums, whether for the detection of Permanent Scatterers, or for the detection of changes. Four mathematical formulations exist and are proposed. We apply these 4 formulations to several scenarios: - a dual pol HH/VV case. In this case, only the two modules are used, and the acquisition cannot be considered as describing a polarimetric state. - a dual pol HH/HV case. In this case, it is possible to use both modules. But it is also possible to consider the vector of dimension 2 as the Jones vector of the scattered wave. This Jones vector can then be transformed into a Stokes vector containing 4 real components. We show that the use of the multivariate coefficient of variation allows a substantial gain in the detection of changes or bright spots, compared to the combination of individual detectors, with different performances depending on the type of detection considered [Aerts, 2015]. Albert, A., & Zhang, L. (2010). A novel definition of the multivariate coefficient of variation. Biometrical Journal, 52(5), 667-675. Aerts, S., Haesbroeck, G., & Ruwet, C. (2015). Multivariate coefficients of variation: comparison and influence functions. Journal of Multivariate Analysis, 142, 183-198. Colin Koeniguer, E., & Nicolas, J. M. (2020). Change detection based on the coefficient of variation in SAR time-series of urban areas. Remote Sensing, 12(13), 2089. Hajnsek, I., & Desnos, Y. L. (Eds.). (2021). Polarimetric synthetic aperture radar: principles and application (Vol. 25). Springer Nature. Colin Koeniguer, E. (2019). POL-timeSAR: Benefits of polarimetry on change detection in time-series. Communication presented during POLINSAR 2019 event: https:// polinsar2019.esa.int/files/presentation199.pdf, January 2019 Praks, J., Colin Koeniguer, E, & Hallikainen, M. T. (2009). Alternatives to target entropy and alpha angle in SAR polarimetry. IEEE Transactions on Geoscience and Remote Sensing, 47(7), 2262-2274.

Rising temperatures in the Arctic are leading to recent changes in permafrost regions. The monitoring of permafrost state and dynamics benefits from the large-scale capabilities of airborne and spaceborne remote sensing as well as the opportunity to image regions that are difficult to access. In this regard, the German Aerospace Center (DLR) conducted an experimental airborne SAR campaign aiming at characterizing SAR responses over permafrost landscapes and assessing to which extent parameters of interest, such as soil moisture and soil layer properties, can be estimated in such regions [1]. The campaign comprises two data sets acquired in the Canadian Arctic in August 2018, when the upper layer of the soil is thawed, and in March 2019, when the ground is entirely frozen. Smaller (X- and C-band) and larger (L-band) wavelengths were used. This study focuses on a particular test site encompassing the Trail Valley Creek catchment (Northwest Territories), which is a well-studied area of continuous permafrost covered by tundra vegetation [2]. Previous polarimetric analysis shows an overall decrease in backscattering power of several dB from summer to winter and a relative increase of the surface contribution (entropy/alpha analysis) [3]. Simultaneously, landscape features remain clearly recognizable at all wavelengths in winter. As permafrost parameters (vegetation, soil moisture, soil type, winter snow depth) are suspected to be related with another [2], unambiguous interpretation of the signal is challenging. Adding another dimension (interferometry) related to phase center height and therefore signal penetration depth is a step towards better interpretation of the signal. As several baselines were flown over Trail Valley Creek test site and fully polarimetric data was acquired, polarimetric SAR interferometry observables can be retrieved. A particular observation of interest is the coherence region, namely the set of coherences that can be obtained from all possible pairs of polarimetric acquisitions at a given baseline. Pol-InSAR coherence regions are characterized by their shape and extent in the complex plane. In particular, their extent in amplitude and phase in the complex plane are of interest, as they can be related to the vertical distribution of scatterers within a given resolution cell. In summer, overall small vegetation (below 1m) covers the soil, and little penetration into the ground is expected as the upper layer of the soil is thawed. In winter, the soil is entirely frozen, covered by snow, and the vegetation is partially frozen. Larger penetration into the frozen soil is expected at larger wavelengths. These effects could influence the coherence region features. An analysis and a comparison of the Pol-InSAR coherence region parameters for the winter and summer datasets over the Trail Valley Creek catchment will be presented. Several frequencies will be considered and the influence of landcover will be assessed by discriminating between different vegetation types. [1] I. Hajnsek, H. Joerg, R. Horn, M. Keller, D. Gesswein, M. Jaeger, R. Scheiber, P. Bernhard, S. Zwieback, “DLR Airborne SAR Campaign on Permafrost Soils and Boreal Forests in the Canadian Northwest Territories, Yukon and Saskatchewan: PermASAR”, POLINSAR 2019; 9th International Workshop on Science and Applications of SAR Polarimetry and Polarimetric Interferometry, 2019 [2] I. Grünberg, E.J. Wilcox, S. Zwieback, P. Marsh, and J. Boike, “Linking tundra vegetation, snow, soil temperature, and permafrost”, Biogeosciences, 17(16), 4261-4279, 2020 [3] P. Saporta, A. Alonso González, I. Hajnsek, “A temporal assessment of fully polarimetric multifrequency SAR observations over the Canadian permafrost”, EUSAR 2022; 14th European Conference on Synthetic Aperture Radar, VDE, 2022

1. Introduction Peatlands are the most efficient terrestrial carbon store; in less than 3% of land area globally, they store 600 Gt of carbon, almost as much as the atmosphere holds [1]. In addition to their crucial role in the global carbon cycle, peatlands provide numerous ecosystem services, including fresh water provision, flood prevention, and increased biodiversity [2]. Many peatlands have been damaged by human activities such as drainage for agriculture or forestry and now require effective restoration, management, and monitoring [3]. Traditionally, monitoring takes place in the field, but remote sensing can be a highly cost-effective alternative with wide spatial coverage [4]. Synthetic Aperture Radar (SAR) can be particularly effective as it can capture images regardless of weather conditions or solar illumination [4]. Existing research with SAR backscatter [5] and Interferometric SAR [6] has shown potential, but polarimetric SAR (PolSAR) – specifically its potential to monitor peatland dynamics and (using longer wavelengths) water table depth (WTD) – has received less study. This research addresses this gap, using multitemporal SAR data from across Scotland to explore PolSAR’s potential. L-band PolSAR has shown ability to detect peatland subsurface water flows – particularly in combination with the Touzi Decomposition [7] – and its ability to predict WTD will be explored using ALOS-2 imagery. C-band SAR will be used to explore the relationship between WTD and Surface Soil Moisture (SSM) derived from Sentinel-1. Finally, the Cumulative Sum (CuSum) time series analysis technique will be applied to Sentinel-1 backscatter to explore its potential for automatic identification of peatland changes and disturbances – such as those due to natural causes like degradation or anthropogenic causes like restoration. 2. Methodology 2.1. ALOS-2 Five quad-pol ALOS-2 products covering parts of Central Scotland during 2016 and 2017 were acquired. 34 WTD loggers were covered by at least one product when active – most at Flanders Moss, a raised bog in Stirlingshire. Across all products, there were 93 instances of a WTD logger being covered by a product and having WTD data available from that product’s acquisition date. For each of these instances, the pixels surrounding the WTD logger were processed using every combination of six polarimetric decompositions – H-A-Alpha, Huynen, Pauli, Touzi, van Zyl, and Yamaguchi – four speckle filters – Boxcar, IDAN, Improved Lee Sigma, and Refined Lee – and four speckle filter sizes – 5x5, 7x7, 9x9, 11x11, and equivalents for IDAN. For every processing combination, we calculated nine weighted means of all pixels within 50m of the WTD logger, weighted – using coefficients from 0.001 to 0.5 – exponentially by distance from the logger. Finally, for each polarimetric decomposition parameter, the set of weighted means that had the highest R2 when regressed with WTD were input into the Relaxed Lasso to produce a linear predictive model of WTD. To validate this model, Leave-One-Out Cross Validation (LOOCV) was used with datasets split prior to selection of sets of weighted means based on R2. 2.2. Sentinel-1 A time series of Sentinel-1 imagery covering Flanders Moss between 2017 and 2022 was acquired. From this imagery, SSM was estimated using a backscatter intensity and change-detection based algorithm by TU Wien [8]. After outliers – caused by heavy rainfall and snow – in the resulting time series were removed, mean SSM was compared to mean WTD. 2.3. Disturbance Detection (Sentinel-1) We acquired Sentinel-1 imagery covering a group of neighbouring bogs in the Scottish Borders between 2017 and 2022. Using this imagery, a time series of VH/VV backscatter ratio was produced, and the CuSum technique applied. CuSum calculates the cumulative sum of the difference between the backscatter of an upcoming image and a reference of expected backscatter. Where there are significant similarities between an upcoming image and the expected reference, the cumulative sum is low and tends to zero. However, where there are significant differences, the cumulative sum is higher, and if these differences persist, the cumulative sum will continue to increase. 3. Results and Conclusions 3.1. ALOS-2 When trained and tested on the full dataset, our model had an R2 of 0.66, an MAE of 3.96 cm and an RMSE of 4.82 cm. It had 17 predictor observables, with the Touzi Decomposition especially influential: 13 of the 16 Touzi parameters considered were included, while just 4 of the 15 other parameters considered were included. Cross-validated performance was lower; R2 was 0.26, MAE was 6.06 cm, and RMSE was 7.31 cm. In all, these results show a clear physical connection between ALOS-2 signal and WTD, and demonstrate the potential of PolSAR for WTD monitoring. However, we hypothesise that this connection is affected by various characteristics such as vegetation and microtopography that differ from site to site, and that, therefore, a universal statistical model is not appropriate. In future, we will use more physical parametrizations and will classify sites based on various characteristics before training different models for different classes of sites. 3.2. Sentinel-1 When compared, the time series of mean SSM and mean WTD across Flanders Moss were found to generally follow the same trajectory between 2017 and 2022. However, the trajectories diverged during 2020 and the first half of 2021, and future work could investigate the cause of this divergence. 3.3. Disturbance Detection (Sentinel-1) All sites showed some diversion from zero – indicating some level of change – with one site – Leadburn – seeing significant divergence from zero. This large divergence at Leadburn began in 2020, coinciding with the start of restoration activities there – including tree removal and ditch blocking. In future, statistical thresholds could be set, enabling automatic near real-time disturbance detection using CuSum. 4. References [references in attached PDF]

The world has witnessed many devastating wildfires in recent years due to human-induced climate change [1]. According to the National Wildland Fire Situation Report, there were 5,449 forest fires in Canada, with a 10-year average of 5,900 annually. The total burned area recorded by the Canadian Interagency Forest Fire Centre was 1,610,216 ha in 2022. Identification of fire location and extent, especially with SAR’s unique weather-resilient ability, offers unique potential to support firefighting operations including prevention, preparedness, response and recovery. Moreover, information on the progression of burnt areas supports enhanced decision making for wildfire emergency response, by improving situational awareness. In this regard, optical sensors such as MODIS, VIIRS, Sentinel-2, and Landsat are often used for preliminary mapping and boundary identification of burned areas [2]. However, dense cloud cover over forests freqeuntly makes it impossible to get information about conditions on the ground. As a result, Synthetic Aperture Radar (SAR) data are used in wildfire mapping [3]. Tanase et al. [4] investigated the sensitivity of multifrequency X-, C- and L-band SAR data for burn severity estimation. Later different orthonormal projections of covariance information obtained from SAR data were utilized to reduce target characterization ambiguities [5, 6]. Dey et al., [7] have proposed a spectrum analysis of the target scattering parameter by projecting the coherency matrix onto different non-orthogonal scattering mechanism bases. This polarimetric spectrum is sensitive to the geometry of the targets and can be a highly effective technique for target characterization and classification. This study utilizes this polarimetric spectrum to detect burned forest areas using ALOS-2 full polarimetric L-band data captured on May 28, 2016 over Fort McMurray, Province of Alberta, Canada which burned in 2015. The random forest classifier produced 85 % overall accuracy, which is 10 % higher than using the only three eigenvalues of the coherency matrix (T3). The classified map using the polarimetric spectrum after a post-processing step and the ground truth map are shown in Fig. 1.

The strong scientific interest and the accompanying rapid development of machine learning, in particular deep learning, has led to a significant improvement in automatic image interpretation in recent years. Research generally focuses on classification or segmentation of optical images, but there are already several successful approaches that apply deep learning techniques to the analysis of PolSAR or Pol-InSAR images. While the success of learning-based methods for the analysis of optical images has been strongly driven by public benchmark datasets such as ImageNet and Cityscapes, which contain a large number of annotated training and test data, comparable datasets for the PolSAR and especially the Pol-InSAR domain are almost non-existent. This conclusion and the demand for large and representative expert-annotated benchmark datasets for the SAR community is also reached by Zhu et al. (2021) in their analysis of the current state of deep learning based SAR image analysis. To fill this gap, this paper presents a new, large-scale multi-frequency Pol-InSAR benchmark dataset for training and testing learning-based methods, which will be made available to the SAR community. This dataset is intended to improve the development of new approaches or the adaptation and improvement of existing ones. Furthermore, a defined division of the data into training and testing sections will ensure the comparability of approaches of different works. For the segmentation of PolSAR images, there already exist a few annotated datasets that are frequently referenced in the respective literature. These include the PolSF dataset, which contains PolSAR images from various spaceborne and airborne systems over San Francisco, the E-SAR dataset from Oberpfaffenhofen, and the Flevoland dataset, which contains an AIRSAR image depicting agricultural areas. The described datasets have several limitations that make them inadequate for evaluating deep learning image segmentation networks. One weakness is the small amount of data, which is usually insufficient for training deep networks. Another problem is the lack of complexity of the segmentation task due to generic classes that are too easily distinguishable or spatial distributions of classes that are too simplistic, regular, and resembling each other. As a result, very high classification performances can already be achieved by simple classifiers, so that a comparison of more sophisticated classifiers, which are necessary for more challenging real-world tasks, is not possible. Another disadvantage of existing datasets is that the division into training and test areas is not fixed, which prevents the comparability of research that uses these datasets for evaluation. Our annotated, multi-frequency Pol-InSAR dataset named Pol-InSAR-Island provides an extensive database as well as a challenging segmentation task consisting of 11 land cover classes. The Pol-InSAR data of the dataset were acquired in April 2022, on behalf of the Lower Saxony Water Management, Coastal Defence and Nature Conservation Agency (NLWKN), with the airborne F-SAR system (Reigber, 2020) of the German Aerospace Center (DLR) during a measurement campaign over the German Wadden Sea. The basics necessary for this measurement campaign in terms of data acquisition and processing were developed within the GeoWAM project (Pinheiro, 2020; Schmitz, 2022). The data acquisition was performed simultaneously in S- and L-band. Interferometric analysis is enabled by imaging the area two times with a time offset of 12 minutes and a vertical baseline of 12m. The Pol-InSAR-Island dataset does not include all flight data of the measurement campaign, but only data sections that cover the island Baltrum. Based on co-registered interferometric image pairs the coherency matrix T6 is calculated for each pixel, which is obtained from the scalar products of the Pauli scattering vectors. Subsequently, this representation is projected from the slant-range to the ground-range geometry. The geocoded products has a ground resolution of 1m x 1m. The island of Baltrum is home to many different biotopes and vegetation species and is surrounded by the North Sea and tidal flats. The diverse land cover results in the target classes of the demanding Pol-InSAR-Island dataset. For a precise annotation of the data, an existing biotope type map, published by Marine.Daten.Infrastruktur.Niedersachsen (MDI-NI) and Marine.Daten.Infrastruktur.Deutschland (MDI-DE), is used, which is a result of the Trilateral Monitoring and Assessment Program (TMAP) processing from 2013. In this map the island Baltrum is classified into 40 different biotope types. Since the biotope map was generated several years prior to the SAR measurement campaign, significant changes in the extent of the different biotope types are to be expected, making the map not sufficient for accurate annotation. Therefore, the map is revised within a semi-automatic process. Automatically, transition areas between different biotope types are removed and strongly similar biotope types are grouped into one class each. The decision about the grouping of biotope types is based on the semantic context and an analysis of the class separability based on polarimetric features. The analysis of class separability is performed visually based on a projection of the polarimetric features into a two-dimensional feature space using the UMAP algorithm. Manual improvements to the resulting annotation are made by visually interpreting the SAR images and simultaneously acquired optical images. The Pol-InSAR-Island dataset is split into two spatially disjoint subsets, using a chessboard grid that alternately defines training and test patches. Such a split pattern was chosen to ensure that all 11 classes are represented in both subsets and additionally mapped at different incidence angles. Overall, this results in an extensive labeled multi-frequency Pol-InSAR dataset that provides a challenging segmentation task. The public availability of the dataset can make a crucial contribution to the further development and comparability of learning-based segmentation methods for Pol-InSAR images in the future.

The main objective of the Biomass mission is to determine the global distribution of above-ground forest biomass (AGB) in order to reduce the major uncertainties in the calculations of carbon stocks and fluxes associated with the terrestrial biosphere. To achieve this goal, it will carry, for the first time in space, a fully polarimetric P-band synthetic aperture radar (SAR). After the launch in 2024, there will be a commissioning phase of 6 months, followed by an 18-month phase during which there will be SAR tomography coverage. In the following interferometric phase, Pol-InSAR coverage will be performed every 9 months until the end of the 5-year mission. In dense tropical forests, PolInSAR dual-baseline ground notching and TomoSAR imaging capabilities will be used to reduce ground scatter in forests, which will improve signal sensitivity at high AGB values. The most important finding of the innovative TomoSAR technique is that the AGB range up to 450 Mg/ha could be estimated using tomographic intensity, with a sensitivity of around 50 Mg/ha per dB. This unique ability of Biomass to estimate high biomass values of dense forests will meet the imperative requirement for reliable information about the spatial distribution of tropical forest biomass and the carbon loss from disturbances. In the last decade, studies were carried out to generate pan-tropical biomass maps by combining field inventories with multi-source remote sensing data, most of them not optimum for measuring the high range of AGB in tropical forests. On the other hand, due to the dynamic forest cover and land use changes in the tropics, higher spatial and temporal resolution systems than BIOMASS are more adapted to the estimation of area losses due to deforestation and forest degradation, and higher frequency SARs are more suitable for estimating carbon gain from forest regeneration in the tropics. In this context, mutual gains will be made by combining BIOMASS data with data from other missions that provide complementary information on forest biomass, structure, height, and changes. The paper will analyse the synergy of Biomass with contemporary missions with different sensors ( L and C-band SARs, Lidar, L-band radiometer) and will present the advantages of combining data from these different missions. Currently, research is ongoing to gain knowledge on the synergy of P-band BIOMASS with other sensors, such as a) using structure parameters from Lidar (GEDI) footprint to ingest into BIOMASS TomoSAR and PolInSAR, b) using simultaneous experimental tower observations at L and P-band to relate the seasonal effects on both frequencies (in terms of backsattering intensity change and temporal decorrelation), or c) to gain knowledge on the synergy between P-band SAR backscatter and L-band radiometer emissivity (SMOS). It is already recognised that the combination of BIOMASS with other missions is expected to enhance the biomass distribution and its changes, in particularly those induced by the Land Use Land Cover change required for calculating carbon uptake and emissions. This is considered in the following topics: - the synergy with L-band SAR (NISAR and ALOS-4) to estimate AGB at the low range of biomass will improve the accuracy of AGB estimated by BIOMASS in the low range of biomass (< 50 Mg/ha), including that of regenerating forest, - detecting and estimating the area of deforestation and forest degradation using high temporal and spatial resolution Sentinel-1 data (e.g., in the TropiSCO project) will improve BIOMASS disturbance products for the carbon loss in tropical forests. Furthermore, the high AGB range of BIOMASS could be used to "calibrate" the AGB of dense tropical forests in existing biomass maps (e.g., CCI Biomass, SMOS VOD Biomass), to improve the estimation of the long-term biomass change based on these products. In summary, to manage carbon stocks, mitigate climate change, and understand the role of tropical forests in the global carbon budget, the community will benefit during the lifetime of BIOMASS, from overlapping operational time from other missions to develop harmonized and accurate biomass and biomass change products.

The abstract has been accepted as oral presentation to the Polinsar&Biomass Conference. The conference will be held at Espace Vanel, Toulouse, France

To support both the climate modelling and carbon cycle science communities, the ESA Climate Change Initiative (CCI) Biomass project is producing maps depicting the global distribution of woody above-ground biomass at 100 m spatial resolution. In the first three years of the project, maps have been produced for the years 2010, 2017, and 2018based on advanced versions of the retrieval algorithms developed in the frame of the predecessor ESA GlobBiomass project(Santoro et al., 2021). The project relies onmulti-temporal stacks of spaceborne C- and L-band radar data acquired by the ESA C-band SAR missions ENVISAT ASAR(Wide-Swath mode, WS) and Sentinel-1(Interferometric Wide-Swath mode, IW) and JAXA’sL-band SAR missions ALOS-1 PALSAR (Fine Beam Dual-Polarization mode, FB) and ALOS-2 PALSAR-2(Fine Beam Dual-Polarization and ScanSAR modes, WB) for mapping above-ground biomass. In addition, the mapping of above-ground biomass considers spaceborne LiDAR (ICESAT GLAS, ICESAT-2) data to support the modeling of multi-temporal C- and L-band radar backscatter with information on forest structural differences reflected in varying allometric relationships between forest height, density, and above-ground biomass. In the fourth year of the project, the maps for 2010, 2017, and 2018 are being reproduced and new maps are produced for the years 2019 and 2020 using a further refined retrieval algorithm and a significantly enhanced database of ALOS-1 PALSAR and ALOS-2 PALSAR-2 L-band backscatter imagery. Improvements of the retrieval algorithm are founded in an updated global database of allometric functions describing the interrelationships of canopy density, height, and above-ground biomass. The database was compiled using ICESAT GLAS and ICESAT-2 spaceborne Lidar information on canopy density and heightand 1) national forest inventory statistics on above-ground biomass released by 106 countries per administrative (provinces, states, counties) or ecological unit (e.g., broadleaf or needleleaf forests), or 2) national averages reported in the FAO Forest Resources Assessment for 2020 (https://fra-data.fao.org/WO/fra2020/home/) for 94 countries. The global data base of L-band backscatter imagery, which so far comprised annual mosaics of ALOS-1 and ALOS-2 FB imagery and per-cycle mosaics of ALOS-2 ScanSAR data over the tropics, was complemented with all individual acquisitionsof ALOS-1 PALSAR and ALOS-2 PALSAR-2 in Fine Beam mode. In support of the CCI Biomass project, JAXA has provided access to all individual ALOS-2 PALSAR-2 Fine Beam acquisitions, which are processed by the CCI Biomass consortium to radiometrically-terrain-corrected level. In addition, JAXA has released all ALOS-1 PALSAR Fine Beam images acquired between 2007 and 2010 at radiometrically-terrain-corrected level. The availability of muti-temporal observations is crucial to overcome limitations of previous releases of the CCI Biomass products associated with the use of single observations of L-band backscatter provided by the annual mosaics produced by JAXA. An independent validation based on in situ plots distributed across the major forest biomes, albeit not with a systematic sampling design, as well as intercomparisons with airborne LiDAR-derived biomass maps available for various sites in South and North America, Africa, Europe, Southeast Asia, and Australia so far confirmed the continuous improvement of the growing time series of CCI Biomass maps with each annual release of a new set of maps, albeit the maps still present regional biases and a per-pixel uncertainty of 30-50%.The independent validation of the maps to be released in year 4 of the project has not been completed by the time of writing this abstract. The quantification of annual and decadal changes in above-ground biomass is a critical component of CCI Biomass. Based on the above-ground biomass maps for different years, CCI Biomass has released annual to decadal change products. The change maps are accompanied by quality flags indicating the reliability/probability of the reported changes. However, quantification of pixel-level changes beyond those flagged as probable in the released change products is currently discouraged. Intercomparisons of the biomass maps and associated uncertainties highlighted the limitations to the estimation of biomass on a global scale, which could be categorized as signal- or processing-dependent. The signal-dependent limitations relate to the varying sensitivity of C- and L-band backscatter to biomass as well as the insufficient characterization of forest structure in the modeling locally. Processing-dependent limitations were a consequence of local or systematic imperfections in the pre-processing of the available radar data to radiometrically terrain-corrected level. Furthermore, radar data acquired by different satellite missions had to be used for the three different epochs and this posed restrictions on the inter-annual harmonization of global maps. This was largely attributed to the different acquisition modes and inconsistent multi-temporal acquisition plans, resulting in a strongly varying number of observations as well as seasonal coverage between years.Future activities in CCI Biomass will therefore seek to improve the inter-annual consistency of above-ground biomass estimateswhen (re-)producing existing maps andnew maps for additional years, i.e.2015/16,2022, and potentially 2007-2009. References Santoro, M., Cartus, O., Carvalhais, N. et al. (2021) The global forest above-ground biomass pool for 2010 estimated from high-resolution satellite observations. Earth System Science Data 13: 3927–3950. doi: 10.5194/essd-13-3927-2021

In this presentation, we report results of Earth Observation (EO) data and methods intercomparison conducted within ESA Forest Carbon Monitoring project. The studies were performed over six European test sites located in Finland, Ireland, Romania and Spain. EO datasets were represented by satellite radar images including Sentinel-1 time series, ALOS-2 PALSAR-2 and TanDEM-X images, and multispectral optical Sentinel-2 images. Examined approaches include popular parametric and physics-model based approaches, and nonparametric machine learning approaches such as k-NN, random forests, support vector regression and deep learning models. The ESA Forest Carbon Monitoring project is developing EO based, user-centric approaches for forest carbon monitoring. The approaches are meant to support monitoring of forest biomass and carbon resources in near-future operational monitoring systems. To respond to the needs of various stakeholders, forest carbon accounting based on forest inventorying requires precise and timely estimation of forest variables at various spatial levels accompanied by verifiable uncertainty information [2, 3]. Satellite remote sensing data combined with in-situ forest measurements is a cost-effective means for producing forest attribute maps and forest estimates on various areal levels [1]. Used satellite image sets include satellite optical and imaging radar data augmented by forest plot or forest stand-level measurements. The primary focus of these analyses was to compare performances of several combinations of EO data and methods in forest structural variable prediction to support carbon stock estimation. Forest variables include growing stock volume, above ground biomass, tree height, diameter at breast height and tree species. Forest types include semi-natural coniferous and broadleaf boreal, temperate and Mediterranean forests as well as Eucaliptic plantations. Model-based framework was used to infer forest variables. Among studied methods, few were suitable for simultaneous prediction of several forest structural variables, while others were used to predict only a single forest attribute. Many of the approaches were data source agnostic and can use several data sources simultaneously (e.g. both SAR and optical), while others rely on specific type of SAR or InSAR data. The candidate methods included well-known machine learning and nonparametric approaches such as support vector machines (SVR), random forests (RF), multiple linear regression (MLR) with optional transformations of response variables and k-NN method. Physics-based SAR and interferometric SAR models were tested as well, particularly in scenarios enabling minimization of reference datasets. Additionally, an in-house semi-supervised Probability method was examined with a goal of combined use of satellite optical and SAR data. Potential of semi-supervised deep learning models [4] and model transfer in predicting forest variables was also examined. Complementarity of various optical and SAR datasets for forest mapping was also studied, as well as importance of various EO image features in retrieval of forest attributes using feature selection techniques (sequential feature selection, lasso, ranking using random forest and mutual information measure). Our results indicate that Sentinel-2 and combined Sentinel-2 and Sentinel-1 data were the most important data combination for predicting tree species proportions. For other structural variables, most centrally GSV and forest height, the best predictions were provided by combining radar and optical datasets, with a key role of Sentinel-2 and TanDEM-X datasets. Use of Sentinel-1 data only led to predictions in range of 50-80% RMSE on forest plot level (20-30m). For practically all sites, combining Sentinel-1 with Sentinel-2 improved prediction accuracy by a small margin of 2-4% units compared to using Sentinel-2 data alone (30-60% RMSE), indicating it is useful to combine the two Copernicus datasets. Combination of Sentinel-2 and TanDEM-X data gave strong improvement (20-40% RMSE), particularly for forest height and growing stock volume. For several studied prediction methods and test sites, using all data bands simultaneously provided the best performance. With non-parametric approaches, such as kNN and Probability method, excluding more noisy bands improved the prediction in several cases. Use of feature weighting in prediction can be useful to overcome the issue. Over the majority of test sites, MLR provided modest accuracies with strong smoothing and saturation effects on predicted forest variables. Basic InSAR/SAR models (such as WCM and RVoG) required additional fine-tuning to achieve accuracy levels similar to supervised approaches. However, they proved robust when lacking reference data. Nonparametric methods such as kNN and Probability have demonstrated similar performance levels and were suitable for multivariate prediction of forest attributes. They favoured smaller dimensionality of feature space and appear very sensitive to non-representative data. RF was somewhat superior to SVR with both approaches yielding the best possible predictions after finetuning their hyperparameters. RF and SVR demonstrated the best possible predictions for several forest variables. Further improvement in forest variable prediction on pixel-level was gained using deep learning approaches, that normally require training/pretraining using fully segmented labels. Our early results indicate that particularly for tree height, accuracy of 20-25% RMSE is feasible on 10 m pixel level. Effective ways of model transfer require further extensive studies to enable wide-area mapping in those areas where fully segmented training labels are not available. References 1. R. E. McRoberts and E. O. Tomppo, “Remote sensing support for national forest inventories,” Remote Sensing of Environment, vol. 110, no. 4, pp. 412–419, 2007. 2. M. Herold, S. Carter, V. Avitabile, et al., “The role and need for space-based forest biomass-related measurements in environmental management and policy,” Surveys in Geophysics, vol. 40, no. 4, pp. 757–778, 2019. 3. J. Miettinen, S. Carlier, L. H ̈ame, et al., “Demonstration of large area forest volume and primary production estimation approach based on Sentinel-2 imagery and process based ecosystem modelling,” Intl. J. Remote Sensing, vol. 42, no. 24, pp. 9467–9489, 2021 4. S. Ge, H. Gu, W. Su, J. Praks and O. Antropov, "Improved semisupervised UNet deep learning model for forest height mapping with satellite SAR and optical data," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 5776-5787, 2022.

The ESA 7th Earth Explorer BIOMASS mission will provide measurements of forest aboveground biomass (AGB), height and disturbances. Estimates of forest AGB, height, and disturbances are crucial information, particularly in the tropics, to improve our understanding of the global carbon cycle and to reduce uncertainties in the calculations of land carbon stocks and fluxes. The forest disturbance products from BIOMASS were defined as yearly products at a resolution of approximately 0.25 ha. Wall-to-wall estimates of AGB over multiple years, with an accuracy of 20% at 4 ha, would probably allow for the detection of forest disturbances and regrowth, so that the AGB loss could be directly inferred from change in AGB maps. However, the detection of forest disturbances at 0.25 ha pixel size does not allow to depict all disturbed areas, as the size of disturbed areas decrease in time as emphasized by [1] in Amazonia. In addition, fast detection, e.g., weekly, cannot be provided through the BIOMASS mission although useful for numerous users. For more detailed information about forest disturbance (i.e., refined temporal and spatial resolution), the alternative is to use more frequent and higher resolution disturbance detection with operational satellite missions (Sentinels, Landsat, etc). Carbon loss assessment is expected to be improved using forest disturbance maps from high temporal and spatial resolution data, and AGB maps from BIOMASS and others sensors. In tropical forests, synthetic aperture radar (SAR) data are preferably used for disturbance detection, such as Sentinel-1 data acquired every 6 to 12 days at 10 m pixel size, and soon NISAR data. In fact, forest disturbances monitoring using Sentinel-1 has been developed for operational applications. For example, the so-called TropiSCO operational system has been developed by the CESBIO, CNES and GlobEO to provide maps of forest disturbances using Sentinel-1 data and is funded from 2021 through the Space Climate Observatory [2,3]. In this study, we demonstrate the ability of the TropiSCO alert system to detect forest loss at the large scale, accurately and in a timely manner. We produced forest loss and carbon loss maps using SAR Sentinel-1 data, every week whatever the meteorological conditions, at a 10 m pixel size and 0.1 ha minimum mapping unit, from 2018. The maps have been produced over Vietnam, Cambodia and Laos in Southeast Asia, Gabon in Africa, and French Guiana, Suriname, and Guyana in South America. The maps are updated daily through a full automatized process on the CNES high performing computers. The forest loss maps produced were thoroughly validated [4,5]. according to the recommendations made by Olofsson et al. [6]. The estimated user accuracy was 0.95 for forest disturbances and 0.99 for intact forest, and the estimated producer’s accuracy was 0.90 for forest disturbances and 0.99 for intact forest, with a minimum mapping unit of 0.1 ha, in Southeast Asia. The resulting maps, updated daily, are browsable and downloadable via the platform https://www.tropisco.org. Forest loss and carbon loss maps shall be produced over the whole Congo Basin and Amazonia by the end of 2023. In this study, we show the three main recent advances related to the TropiSCO initiative: 1) The forest loss and carbon loss maps recently developed over the Congo Basin. 2) A new advanced alert system dedicated to Amazonia, taking advantage of the two methods from [7] and [8] to map forest loss in Amazonia in an optimal manner. We show the first results, including spatial accuracy assessment and comparisons with other existing alert systems [9]. 3) A new alert detection probabilistic method taking advantage of various sources of data, in particular SAR at various frequencies and optical data. This method aims at using all available data in the near future for forest disturbances detection, i.e., Sentinel-1 and -2, Landsat, NISAR and BIOMASS. [1] Kalamandeen et al. (2018). Pervasive rise of small-scale deforestation in Amazonia. Scientific reports, 8(1), 1-10. [2] https://www.tropisco.org/ [3] https://www.spaceclimateobservatory.org/ [4] Ballère et al. (2021). SAR data for tropical forest disturbance alerts in French Guiana: Benefit over optical imagery. Remote Sensing of Environment, 252, 112159. [5] Mermoz et al. (2021). Continuous Detection of Forest Loss in Vietnam, Laos, and Cambodia Using Sentinel-1 Data. Remote Sensing, 13(23), 4877. [6] Olofsson et al. (2020). Mitigating the effects of omission errors on area and area change estimates. Remote Sensing of Environment, 236, 111492. [7] Bouvet et al. (2018). Use of the SAR shadowing effect for deforestation detection with Sentinel-1 time series. Remote Sensing, 10(8), 1250. [8] Doblas et al. (2022). DETER-R: An operational near-real time tropical forest disturbance warning system based on Sentinel-1 time series analysis. Remote Sensing, 14(15), 3658. [9] Reiche et al. (2021). Forest disturbance alerts for the Congo Basin using Sentinel-1. Environmental Research Letters, 16(2), 024005

The Soil Moisture and Ocean Salinity (SMOS) satellite is the first instrument performing systematic L-band observations from space. Its passive radiometer measures de thermal emission from the Earth at L-band (1.4 GHz, 21 cm). It has been specifically designed to estimate soil moisture, but thanks to its multi-incidence angle capabilities, it allows to estimate simultaneously the optical depth. The L-band vegetation optical depth (L-VOD) is mainly due to the water molecules contained in the ligneous parts of the vegetation and can be used to infer vegetation parameters such as the above ground biomass. L-VOD has been shown to be more sensitive to biomass than higher frequency VOD (Rodriguez-Fernandez et al. 2018). In this presentation we will compare L-VOD to higher frequencies VOD and to optical vegetation indices. For instance, we will show that L-VOD provides complementary information to other vegetation variables to understand post-fire recovery in different biomes (Bousquet et al. 2022). As already mentioned, L-VOD is highly sensitive to biomass and, in spite of a low spatial resolution, the high SMOS temporal revisit allows to study the evolution of carbon stocks since 2010. However, uncertainties are large and not always taken into account in applications studies. We will discuss such uncertainties and in addition we will present an innovative way to estimate biomass from SMOS observations without using the L-VOD (Salazar-Neira et al. 2023). Using neural networks it is actually possible to retrieve biomass from SMOS brightness temperatures. The results show that these estimations have a larger dynamical range than those obtained from VOD and lower uncertainties. In addition, using a multi-year reference data set (ESA Climate Change Initiative Biomass) it has been possible to obtain a more robust estimation that can be used to extrapolate in time. Bousquet, E., et al. (2021). SMOS L-VOD shows that post-fire recovery of dense forests is slower than what is depicted with X-and C-VOD and optical indices. Biogeosciences. Rodríguez-Fernández et al. The high sensitivity of SMOS L-Band vegetation optical depth to biomass, (2018) Biogeosciences, 15, 4627-4645 Salazar-Neira, Rodriguez-Fernandez, Mialon et al. Above ground biomass estimation from passive microwaves brightness temperatures using neural networks, (2022), IEEE JSTARS, submitted

A comparative study on the X and C-band data application for the crop phenology development estimation Pasternak Marta, Kamila Pawłuszek-Filipiak Expanding population and climate change are affecting food production and are therefore a very crucial concern nowadays. Considering Synthetic Aperture Radar (SAR) technology which can capture images during the night and without cloud coverage, this technology seems to be a perfect alternative for passive remote sensing in vegetation monitoring. Thus, this work focuses on the examination of various SAR-delivered indices from C- and X-band SAR sensors in the context of monitoring crop phenological development. The study was performed in the agricultural region between Jelcz-Laskowice, Nowy Dwór, and Piekary villages in the eastern part of Lower Silesia in Poland. The investigation was conducted on a total of 30 fields including 7 maize fields, 7 wheat fields, 5 rapeseed fields, 6 potato fields, and 5 rye fields. On dates ranging from March 27th, 2021, to September 21st, 2021, 18 field visits were conducted to gather ground truth information via field investigation. The field visits were scheduled to coincide as closely as feasible with the Sentinel-1/TerraSAR-X anticipated flights. The phenological phases of the plants were assessed according to the official BBCH scale (Biologische Bundesanstalt, Bundessortenamt und CHemische Industrie). The ground data were utilized to confirm the actual consistency between the radar remote sensing signal and real data. For the experiment, 38 Sentinel-1 images were collected from relative orbit number 73 between May 25th, 2021 and November 20th, 2021 in the Interferometric Wide Swath mode. Moreover, 12 TerraSAR-X images were collected from relative orbit number 17 between June 17th, 2021 and November 29th, 2021 to examine the impact of SAR wavelength on the performance of crop development monitoring. Since TerraSAR-X is not freely available, hence it was not possible to establish a consistent and dense time series as in the case of the Sentinel-1 mission. Backscattering coefficients in VV, VH polarizations, interferometric coherence, and vegetation radar indices were calculated for C- and X-band data. Noise associated with the decorrelation effect was filtered out by using different temporal filters of Savitzky-Golay and moving average filters. Sentinel-1 data presents a very high consistency with ground truth data. A pretty good degree of correlation between the C-band signal and the development of the majority of plants was already attained without the use of temporal filtering. The TerraSAR-X (X-band) data, however, did not show such a significant correlation, which may be due to two factors: the short number of images and the poor capability of the radar signal to penetrate vegetation. Analyses performed in this research showed the greatest effectiveness of the VH coefficient calculated for C-band data (Sentinel-1) in combination with temporal filtering using a moving average with a time window equal to 9 observations. This combination made it possible to obtain the highest degree of consistency with ground truth observation of phenological phases (Pearson correlation coefficient = 0.88).

Climate change and global warming caused by the emission of greenhouse gases, especially carbon dioxide, have led to increased attention on the storage of organic carbon on land in terrestrial ecosystems. On the other hand, land use activities have led to the depletion of organic carbon reservoirs. By predicting land use change, it is possible to determine the extent of expansion and destruction of organic carbon reservoirs and guide the changes made in the appropriate directions. In recent years, the cellular automata (CA) model has been widely used in predicting land use changes due to its dynamic nature and unique characteristics in modeling the natural and physical effects of the earth's surface. The main goals of this research were to predict land use changes using a combination of CA and Markov chain and their effect on the amount of carbon storage. To model land use changes, satellite images were selected for the period 2000 to 2020 in the Sanjabi-Ravansar catchment with an area of about 57000 hectares in Kermanshah province in the west of Iran. Image classification was done using the support vector machine (SVM) method. After modeling, the land use changes were simulated using the CA model for the mentioned periods. Logistic regression and Markov chain transition matrix were used for simulation. Overall accuracy and kappa coefficient were used to evaluate classification results. The amount of soil organic carbon for 2020 was estimated using soil sampling and measuring soil organic carbon in a Latin hypercube soil sampling system. Finally, interpolation of soil organic carbon quantity was done with kriging regression and support vector machine with the help of auxiliary variables of topography and vegetation. Then, changes in soil organic carbon stocks were compared to the land use change map. Results of this research indicated that there is noticeable diversity in land use management in all land uses (forest, pasture, and agriculture) in the study area. The land use change from the potential land use to the land uses with positive and negative scenarios was very diversely done on a wide scale in the watershed. Also, there were no changes in the type of land use in some land uses. But there were changes in the exploitation pattern with the approach of increasing or decreasing the pressure on the soil. Most scenarios of pressure increase in the use of irrigated agriculture and scenarios of pressure reduction in the use of forest had occurred. The overall accuracy of land use classification in 2000 and 2020 was 0.83 and 0.81, respectively, and the Kappa coefficient was 0.78 and 0.77, respectively. The results show an increase in the area of residential areas, barren lands, and forests by the amount of 966.22, 2901.93, and 1003.71 hectares respectively in 2020. The area of rangeland decreased by about 2638.48 hectares and agricultural lands increased by the amount of 9337.29 hectares compared to the 2000 year, respectively. The amount of carbon storage is estimated to be increased in forest area in study periods in about 0.2 -0.9 percent. We faced a decrease about 0.1 to 1.5 percent in the amount of soil carbon storage in the rangeland and agricultural land.

In recent years’ groundwater over-exploitation and groundwater level decline damage humans and environment and causes land subsidence as well, which has been a problematic issue in arid and semi-arid areas such as Iran. Most of agriculture and drinking water demands is supplied by groundwater resources. As a consequence, annual decreasing of groundwater level is observed in most regions of Iran. The negative groundwater budget in last two decades in several aquifers has caused the land subsidence in some areas. The impacts of both geological and hydrodynamic characteristics of the aquifer on the land subsidence values were considered in this research. The main goal of this paper is to assess the land subsidence related to aquifer characteristics. Qazvin plain as one of the largest agricultural areas in Iran was selected as a case study, since its experience both groundwater declines as well as subsidence. In this study the Interferometric Synthetic Aperture Radar (InSAR) technique used to estimate subsidence by using Sentinel-1 satellite from 2015 to 2021. InSAR is a powerful tool for measuring ground surface deformation, displacement, topography and etc by combination of radar images in different times. In this method, an interferogram has been calculated by subtracting the phase of two radar images contains different phase contribution mainly due to topography. The purpose of this paper is map the subsidence in aquifer. Qazvin is one of the industrial and agricultural provinces of Iran that 76% of water consumption comes from groundwater. The results of this paper show that there is a direct relationship between groundwater characteristics and land subsidence.

Dry forests and savannas have highly heterogeneous horizontal and vertical structure of woody vegetation along with high temporal variability in moisture and phenology. This presents challenges to SAR-based approaches to mapping canopy height and biomass that are distinct from those encountered in tropical and temperate forests. Dense time-series of C-band (Sentinel-1A/1B) and, in future, L-band (NASA/ISRO NISAR) provide a pathway to reduce the impact of signal noise and environmental conditions on mapping and extract additional information more directly related to height, using repeat-pass Interferometric SAR (InSAR) techniques. As part of a NASA Carbon Monitoring System (CMS) investigation, this work explores the potential information in Sentinel-1(S1) C-band InSAR temporal decorrelation and Global Ecosystem Dynamics Investigation (GEDI)-derived relative height metrics for canopy height estimation. The 12-day InSAR correlation (𝛾) [1] and seasonal median were analyzed over a wide range of canopy heights and woody ecosystems, with a focus on dry forests and savannas in Australia, India, and South Africa. This study validated canopy height estimates over three different countries with contrasting canopy structures. A modified random volume over ground (RVoG) model is used to estimate canopy height from InSAR decorrelation [2-4]. To enable large-area applications, the algorithm was developed on the NASA-ESA Multi-Mission Algorithm and Analysis Platform (MAAP). A tile-based approach to large-area mapping is employed using the Military Grid Reference System. A total number of approximately 31.8 million GEDI samples were used for the validation study across all three test sites. Results showed a correlation coefficient (r) ranging from 0.65 to 0.8 with RMSE ranging from 2.76 m to 4.39 m at 500 m resolution (Figure 1). Poorer performance of the inversion was observed for South Africa compared to those achieved for Australia and India. These poorer results are likely due to the predominantly shorter-statured and sparser woody vegetation compared to Australia and India, which results in a lower range of observed canopy height and lower InSAR correlation. The results for India showed improved performance since these represent a more diverse mixture of diverse forest types. The strongest model performance was observed over Australia. However, these represent the most diverse range of conditions, from rainforest to savannas and low open shrublands. Furthermore, a low inversion accuracy was observed over non-forest/agricultural areas due to loss of coherence from agricultural activities, an effect that can be detected and masked out using ancillary maps of active crop areas [5]. Further, an underestimation of canopy heights was observed for trees higher than 25-30 m. This underestimation is expected due to the limited sensitivity of the C-band for taller canopies. Our preliminary country-scale results show that the accuracy of the proposed approach is well within the acceptable range across diverse forest and woodland types. The modified RVoG model is being improved through local parameterization with GEDI and ICESat-2 canopy heights, enabling parameters to vary spatially following predefined strata or local smoothing constraints. The application of this model on the MAAP has the potential to generate wall-to-wall canopy height maps over large areas, which could open new pathways into high spatial resolution mapping of aboveground biomass. In summary, this study provides new insights into the applicability and limitations of using C-band InSAR data for canopy height estimation in dry forest and savanna studies. The limitations with C-band could be overcome partly, if not completely, through joint retrieval with InSAR time-series from longer wavelength instruments like L-band, such as those proposed for the upcoming NISAR, ALOS-3/PALSAR-4, and ROSE-L missions.

In this work, we propose the use of several change detection technique applied to satellite SAR data to detect and map the huge landslide which affected Ischia island on November 26th, 2022. It was activated along the flank of Mt. Epomeo on the NW sector of the island by a strong flooding event which severely impacted on the hamlet of Casamicciola Terme causing several damages to the roads, building collapses, and also 12 casulaties. To detect the local landslide phenomenon we exploit SAR images acquired from Cosmo-SkyMed second generation missions which are charactrized by a ground resolution of about 2x2m. Then, the applied change detection techniques are based on the intensity, the coherence and the reflection symmetry property of the SAR signal. Then, the performance of all of them are discussed according to the features of the detector and the investigated scenario and additional analysis have been performed to quantitatively establish the best one in such a case. The detection and the mapping of the area affected by this kind of phenomena is important and can be very useful to support both the stakeholders for the first aid and the Civil Protection for the post-crisis management.

On February 6th, 2023, a 7.8 magnitude earthquake hit the southern and central regions of Turkey, as well as the northern and western regions of Syria. This earthquake was one of the largest earthquakes ever recorded causing extensive damage to the buildings and infrastructure in the affected regions and more than 50000 casualties. The disaster management authorities have been struggling to assess the damages and prioritize rescue and relief operations due to the widespread nature of the damages. In this context, remote sensing using Synthetic Aperture Radar (SAR) can provide valuable insights into the extent and severity of the damages. This study presents the application of multi-polarimetric and multi-frequency SAR change detection methods for detecting damages in the aftermath of the Turkey earthquake. Within this context, a quantitative analysis of earthquake-induced damages is performed using single and dual polarimetric SAR imagery collected before and after event using X-band CosmoSKY-MED 2nd generation and C-band Sentinel-1 dataset. Two change detection approaches have been exploited. The first one relies on identifying changes in the scattering properties of the target objects. The second one uses the Intensity Correlation Difference (ICD) which estimates the changes in the spatial distribution of the scatters, and their SAR intensity value, within a user-defined window. The joint use of SP and DP methods together with different resolution images is also addressed to both improve the change detection and to reduce false detections. Preliminary results demonstrate the effectiveness of the SAR change detection methods in detecting damages to buildings and infrastructure. The detected damages are verified with the very high-resolution satellite optical images collected by Maxar. The study also highlights the importance of the case study for disaster management authorities to prioritize rescue and relief operations in the affected regions. In conclusion, the SAR change detection methods provide a powerful tool for detecting damages in the aftermath of the 2023 Turkey earthquake. The case study highlights the importance of remote sensing in disaster management and the potential of SAR change detection methods for providing valuable insights into the extent and severity of the damages. The study results can help the disaster management authorities in prioritizing the rescue and relief operations and in planning the reconstruction efforts in the affected regions.

Often space agencies and researchers in the Polarimetric Synthetic Aperture Radar (PolSAR) community are divided on the added advantage of using quad polarimetric SAR data in comparison to dual polarimetric SAR data. This leads to a further question of whether the diagonal of the polarimetric covariance matrix, which contains intensity only information, is sufficient for Earth Observation (EO) applications from a radar polarimetry point of view. Note that the freely available Ground Range Detected (GRD) Sentinel-1 product from ESA is only intensities and yet is an important operational EO and monitoring tool. In this study, we take the first steps to understanding: ``what value the off-diagonal terms in the covariance matrix provide over its diagonal counterparts from the perspective of sea ice applications?”. Our findings suggest that the availability of just one key off-diagonal term can significantly boost segmentation of sea ice scenes. We demonstrate this using a C-band quad polarimetric RADARSAT-2 dataset covering different sea-ice types and revisiting a polarimetric SAR image segmentation technique first introduced in POLINSAR 2013 using the Extended Polarimetric Feature Space (EPFS). We choose the polarimetric parameters from the EPFS. However, we do not include one of the features whose computation relies on availability of Single Look Complex (SLC) product rather than being directly computed from the Multi-look Complex (MLC) SAR data. The rest of the EPFS features are the Geometric Brightness (GB) which is similar in description and behavior to SPAN (total power - sum of the diagonal entries), the co-polarization and the cross-polarization ratios, and lastly, the co-polarization correlation coefficient. The diagonal entries of the polarimetric covariance matrix directly correspond to intensity features in the different polarimetric channels. Except for the co-polarization correlation coefficient, all the other features, both polarimetric and intensity, are log-transformed for better performance. The complex-valued co-polarization correlation coefficient is split into real and imaginary parts to expand the polarimetric feature set. Then a Gaussian Mixture-based Clustering is performed with the polarimetric features on one hand, and the intensity features on the other, under identical process parameters. In this study, we compare the selected EPFS features and the intensity features for segmentation of polarimetric SAR image containing different sea ice types. Using the above-mentioned RADARSAT-2 data set, a 66% enrichment in features (5 real polarimetric features from 3 real intensity features) leads to a 100% increase in the number of distinct clusters automatically determined by the image segmentation algorithm. Further analysis shows that in most cases, we obtain at least two distinct clusters derived from polarimetric features for every cluster derived from intensity features, without corrupting the visually identifiable distinct sea ice segments in the scene from the Pauli RGB image.

Having an impact on both the environment and on the society, the dynamics of glaciers and ice sheets are a key indicator of climate change. In order to accurately calculate mass balances for climate models and improve the process understanding of ice dynamics and hydrology more information about the subsurface structure of glaciers and ice sheets is required. Polarimetric and interferometric SAR data can provide such information due to their sensitivity to different subsurface structures, dielectric properties and due to the penetration of the microwave signals into the ice. [1] A particular pattern, related to the surface and subsurface ice properties, was found at the Russel glacier, also known as K-Transect, in western Greenland in an experimental airborne SAR dataset. The scattering structure of a glacier can look uniquely on SAR images, depending on the area. Either they could be described as homogeneous areas with uniform layers or as quite heterogeneous areas that differ strongly from each other. First analyzes demonstrate very heterogenic and complex scattering patterns in the ablation zone of the Russell glacier. In L- and P-band, these features are more pronounced due to a deeper signal penetration into the ice sheet [2]. Based on polarimetric and tomographic analyses in [3], the idea of two different ice types is introduced, whose glaciological origin is not yet fully understood. To further understand the striking variability within the test site, three smaller representative areas were chosen based on a Pauli decomposition, each with surface (sample area 1), volume (sample area 2), and dihedral scattering (sample area 3) as the most prominent scattering mechanism. During a first polarimetric investigation at L-band, various differences between the sample areas could be detected. Sample area 2 shows a high scattering entropy and a high mean alpha angle and sample area 3 displays even higher values, which implies that more mechanisms contribute to the total backscatter whereas sample area 1 shows surface scattering as a dominant scattering mechanism. In a first interferometric analysis at L-band, the coherences are higher in sample areas 1/3 and the lowest in test site 2 due to volume correlation, however all test sites show a high variability within their respective areas. The phase center depths show a clearly different behavior for the area 1 and 2. A shallow penetration can be seen in test site 1 with phase center depths of around 0m to -10m, which aligns with higher coherences and surface scattering as the main scattering mechanism. Due to the high amount of volume scattering, area 2 shows a depth of up to -50m in HV polarization. Sample area 3 appears to be in a similar range as area 1 but also applies the highest variance within its respective section. In order to gain a better conceptual understanding of these complex scattering mechanisms and their vertical structures in the ice subsurface, which also manifest themselves in distinct Pol-InSAR coherence regions, further interferometric and tomographic analyzes will be presented. [1] Fischer, Georg, Modeling of Subsurface Scattering from Ice Sheets for Pol-InSAR Applications (2019) PhD Thesis ETH Zurich [2] Parrella, Giuseppe et al., Model-Based Interpretation of PolSAR Data for the Characterization of Glacier Zones in Greenland (2021) IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Vol. 14 p. 11593-11607 [3] Pardini, Matteo et al., A Multi-Frequency SAR Tomographic Characterization of Sub-Surface Ice Volumes (2016) European Conference on Synthetic Aperture Radar

Gradual desertification, a consequence of climate change, is impacting agricultural activities across many countries in the Sahel region of northern Africa [1]. In response to this, efforts have been made across the region to adapt and so provide resilience to the communities most vulnerable to this. This study focuses on Senegal, a country that relies heavily on agricultural irrigation systems, particularly in the northern region based around the Senegal River Valley where rice is the primary crop. For those sites distal from defined irrigation works, there is a greater reliance on the ability of the soil to retain the necessary moisture levels over time to support crop production following the rainy season (June-October). With changing precipitation patterns and increasing land surface temperatures, studying changes in such soil moisture, particularly deep soil moisture, is critical to being able to assess the impact on rice crop productivity at such sites [2]. For regions outside the Senegal River Valley, the interplay between available groundwater and soil moisture plays an essential role in the ability of communities to support local agricultural activities. Whilst it is possible to monitor changes in groundwater and infer variations in soil moisture through the using GRACE mission, the poor spatial resolution of the latter limits its ability to be used at the localised scales involved [3]. The BIOMASS mission offers the possibility of being able to map and quantify such deep moisture changes. In contrast to existing orbiting assets (SMAP and SMOS), its use of P-band SAR will permit the estimation of soil moisture at greater depths particularly so in such an arid/semi-arid environment [4]. Combined with contemporaneous estimates of NDVI and other vegetative index time-series at equivalent spatial scales using data from platforms such as MODIS, Landsat and Sentinel-2, relationships between soil moisture and crop production can be discerned, and trends identified that could be used to flag the most vulnerable locations in the regions under scrutiny. By implementing this proposed method, the impact of climate change and desertification in the Sahel region could be remotely monitored from space using BIOMASS in conjunction with existing assets, in particular those of the Copernicus Programme. This would provide policymakers and researchers with valuable information regarding the state of the region's groundwater and soil moisture status, which are critical resources for the agricultural sector and administrative water management decisions. Changes in the groundwater and soil fertility, along with the spread of desertification, are a severe strain on the food supply chain in the region. We suggest that P-band SAR observations with BIOMASS have significant potential for monitoring environmental and agricultural changes in the wider Sahel region caused by climate change, and could aid in stabilizing the food-supply chains for its almost 1 billion inhabitants. 1. OECD (2022), Environmental Fragility in the Sahel, OECD Publishing, Paris 2. Rigden, A.J., Golden, C., Huybers, P. (2022), Retrospective Predictions of Rice and Other Crop Production in Madagascar Using Soil Moisture and an NDVI-Based Calendar from 2010–2017. Remote Sens., 14, 1223. https://doi.org/10.3390/rs14051223 3. Sadeghi, M., Gao, L., Ebtehaj, A., Wigneron, J.P., Crow, W.T., Reager, J.T., Warrick, A.W., (2022), Retrieving global surface soil moisture from GRACE satellite gravity data, Journal of Hydrology, 584, 124717. https://doi.org/10.1016/j.jhydrol.2020.124717 4. Quegan, S. et al, (2019), The European Space Agency BIOMASS mission: Measuring forest above-ground biomass from space, Remote Sensing of Environment, 227, 44-60. https://doi.org/10.1016/j.rse.2019.03.032

Background Forest vertical structure describes the distribution of leaves, branches and stems along vertical direction. The temporal change of this distribution is caused by the seasonal development of canopies and regulates the exchange of carbon, water and energy between the land surface and the atmosphere. Dynamics in vertical forest structure can be measured at stand-scale using phenological cameras or terrestrial laser scanning. It is still a challenge to continuously monitoring forest vertical structure dynamics at large scale. Global Ecosystem Dynamics Investigation (GEDI) is a spaceborne waveform lidar, which can measure forest vertical structure with high resolution. However, GEDI footprints are separated with 60 m along track and 600 m across track. There is also no fixed revisit time for most GEDI acquisitions since GEDI footprints from different passes usually do not cover same locations. We hypothesize that this mismatch between the need to monitor the vertical and dynamics of forest structures and the temporal coverage of GEDI can be bridged with observations from Synthetic Aperture Radars (SAR). Sentinel-1 is a spaceborne C-band (6 cm wavelength) SAR system with larger coverage than GEDI and fixed revisit time (6 or 12 days). Sentinel-1 signal mainly interacts with leaves and branches in the canopy, is sensitive to changes of canopy structure and might penetrate into deeper layers of the canopy. Here we aim to explore Sentinel-1 observations for wall-to-wall mapping of seasonal dynamics in vertical canopy structures at regional scales in temperate forests. Therefore, our objective is to investigate the sensitivity of Sentinel-1 variables to seasonal changes of vertical GEDI structure metrics. Methods Because GEDI has much denser observations around 51º N, we chose two deciduous broadleaved forests (Hainich and Hohe-Schrecke) and two evergreen coniferous forests (Tharandt and Tambach-Dietharz) in Germany as study sites. Canopy cover profiles from the GEDI Level 2B product from 2019 to 2023 were used to represent forest vertical structure since it is more sensitive to leaf on/off. The canopy cover at each height layer (from ground to canopy top with 5 m interval) was calculated from the original cumulative canopy cover. GEDI footprints were filtered using criteria: 1) Maximum relative height (RH100) is between 25 m and 30 m; 2) observations were acquired using power beam; 3) under-canopy terrain slope is smaller than 5 degrees; 4) beam sensitivity is larger than total canopy cover; and 5) forest type. Spatially-distributed GEDI observations within the entire forest patch were combined into one time series, which we assume to represent the temporal dynamics of the entire forest patch with homogenous vertical structure. For the processing of Sentinel-1 data, we used Alaska Satellite Facility’s HyP3 platform to generate Sentinel-1 Radiometrically Terrain Corrected (RTC) backscatter from 2019 to 2023, and 6-day interferometric coherence (VV polarisation) from 2019 to 2021. Coherence is introduced because it indicates the similarity between two SAR acquisitions, and changes in forest vertical structure affect the coherence. We assume the spatial resolutions of RTC (30 m) and coherence (40 m) are comparative to GEDI footprint (25 m) considering the geolocation inaccuracy of GEDI footprints and the homogeneity of forests. Sentinel-1 VH and VV backscatter, VV/VH ratio and the coherence with the smallest time lag to GEDI acquisition date (the time lag threshold is 10 days) were extracted. For each study site, the Sentinel-1 time series was calculated through averaging the Sentinel-1 data at all GEDI footprints for each GEDI acquisition date. We compared the correlation between Sentinel-1 time series with GEDI canopy cover time series. The influence of Sentinel-1 viewing direction was analysed through comparing results from ascending and descending orbits in Hainich and Hohe-Schrecke. Sentinel-2 derived Copernicus phenology metrics were used as auxiliary data for interpreting GEDI and Sentinel-1time series. Results In broadleaved forests, the total canopy cover shows significant seasonal patterns. For each year, it increases from 0.2 to a peak around 0.8 and then decrease to 0.2 again. Canopy cover at 15-20 m and 20-25 m layers show the largest change (around 0.2 between the minimum and maximum value) while 5-10 m and 25-30 m layers show the smallest change. Sentinel-1 VH polarisation has seasonal changes while VV polarisation does not. VV/VH ratio shows clearer periodically changes than VH since some noise is reduced after the division. The total canopy cover has stronger correlation with VV/VH ratio than VH backscatter. Canopy cover between 10 m and 25 m is more correlated with VV/VH than layers below 10 m and above 25 m. Coherence is around 0.7 in spring and winter, and drops to 0.3 in summer. Although the coherence is negatively correlated with canopy cover, the relationship is less significant than the relationship with VV/VH. For Hainich, the positive correlation between 10-20 m canopy cover and VV/VH ratio is robust in both ascending and descending case. The coherence from descending orbit also shows negative correlation with canopy cover. In coniferous forests, the total canopy cover is between 0.4 and 0.6, and it has less temporal change than in broadleaved forests. The 15-20 m and 20-25 m layers has more canopy cover all the year but the magnitude of change is similar as in the 10-15 m layer. Both Sentinel-1 VH and VV backscatter increases and decrease with the changes in total canopy cover while VV/VH is relatively stable in each year. The relationship with VH and VV backscatter is not significant for canopy cover at height layers. The coherence time series in Tharandt has stronger seasonal pattern than that in Tambach-Dietharz while it is noisier than in broadleaved forests. There is no relationship between coherence with neither total canopy cover nor height-level canopy cover. For Tambach-Dietharz, no significant difference was found when compare canopy cover with Sentinel-1 metrics from descending orbit. Conclusions Our results show the potential of using Sentinel-1 VV/VH ratio to monitor the temporal dynamics of canopy cover at 10-20 m layers in 30 m high broadleaved forests. This may contribute to quantifying the dynamics of vertical forest structure and improve our understanding on the role of forest in terrestrial ecosystems.

SAR data has been utilised recently to estimate orchard biophysical parameters. Researchers indicated how the backscatter values vary with the different biophysical parameters in jute- a typical case for non-rice crop. The regression model works best when the crop/vegetation is in the peak vegetative stage as there is minimal soil contribution due to dense crop canopy. Radar Vegetation Index (RVI) is correlated with VWC (Vegetation Water Content) for vegetation types having a large dynamic range and used to retrieve crop biophysical parameters. NDVI has been used to correlate with many parameters like nitrogen content of the plant, crop canopy of horticultural crops, etc. Canopy Cover (CC) in horticultural crops in relation to NDVI indicated an appreciable R2 value of 0.95 while SAVI gave a R2 value of 0.89. This indicated that differences in soil are not a major source of error for NDVI as NDVI is sensitive to soil brightness differences. Crop height in corn has a high correlation with Radarsat-2 data for the early stages. Vegetation indices have often been used in estimation of biomass. SAR data is found to be sensitive to plant biomass. Sentinel-1 and 2 has often been used to study vegetation biomass. The complimentary nature of SAR to optical datasets has been demonstrated for mango orchard biophysical parameters assessment. The microscale pigment related information compliments the macroscale structural changes and is a novel attempt in this domain. A backward regression analysis was devised for estimation of the mango tree biophysical parameters like canopy radius, DBH, tree height, etc., synergistic use of SAR and optical data performed well towards this. Parameters related to canopy architecture like height, DBH, canopy radius can be predicted with high accuracy using sensitive indices from both the sensors. The amalgamation was way more successful as SAR data is sensitive to dielectric and canopy architecture while optical data makes use of reflectance for capturing the microscale pigment properties. Biomass generated from allometric equation were found to be closely related to SAR and optical data. Overall, this study clearly indicates the successful applicability of high-resolution satellite data to mango orchards biophysical parameters.Correlation between the optical based indices and backscatter values was investigated. Simple linear regression between the field parameters, indices, and backscatter values was performed. Multiple regression analysis was performed between every biophysical parameter and the indices. In this objective, the focus is to establish a relationship between biophysical parameters and indices. The study site is located in northern India covering the Siyana tehsil of Bulansdshahr district of Uttar Pradesh and a little portion of the nearby Hapur district. The boundary are by 7755'25.67"E in the west, 78 12' 34.82"E in the east, 28 47' 0.14"N in the north and 28 25' 48.37"N in the south. The region comes under the upper Gangetic Plains Zone 5 as per the agroclimatic zone classification by the Planning Commission. The average rainfall in the Siyana tehsil is 702mm along with an average elevation 216.14m above mean sea level. The area is highly suitable for agriculture, owing to the all-time water availability from the perennial Ganges flowing along the eastern side. For MLR, backward stepwise regression analysis was performed. Backward regression was used in the study as the number of variables were less than the sample size. All the SAR and optical indices and backscatter coefficients were used to predict each biophysical parameter. If found R2 greater than 0.45, the optimum significance F (

The timely and reliable identification of changes in urban areas is crucial for city planning and building safety. While optical remote sensing images have been used for change detection in urban environments, they are often affected by cloud cover and illumination conditions. Synthetic Aperture Radar (SAR) is an active microwave coherent imaging radar that can be used for all-weather and day-and-night imaging. Multi-polarization and multi-temporal SAR imagery can play an important role in change monitoring due to their unique scattering characteristics and data availability at regular intervals. In this study, we developed a more robust and permanent change detection algorithm for urban environments using multi-temporal and polarimetric SAR data acquired from the JPL/NASA Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) and Sentinel-1. The most promising polarimetric indicators were identified. The results showed that the Shannon Entropy and the determinant of polarimetric covariance matrix were the most robust parameters in detecting changes in the urban environment. Fitting a hyperbolic tangent model function to the multi-temporal polarimetric parameters significantly reduced non-essential changes and provided further information regarding whether buildings were constructed or destroyed and the date of change occurrence. We also demonstrated the potential usefulness of dual-polarization SAR observations for robust change detection after disaster events. The study highlights the potential of SAR for change detection in urban environments and suggests that dual-polarization with an adequate number of multi-temporal SAR datasets could be used for robust change detection.

This work presents a novel method for assimilation of meteorological data, SAR-derived Surface Soil moisture (SSM) and interferometric SAR (InSAR)-derived Ground Surface Motion (GSM) for monitoring peatlands condition and hydrology in the Forth Valley, Scotland. We use SAR imagery from the Sentinel-1 SAR satellite and meteorological observations obtained from short-latency ground weather stations. We present preliminary findings that qualitatively analyse the relationship between SSM, GSM, PET, Net rainfall and peatland Water Table Depth (WTD) measurements captured by a network of ground water loggers. We also present the GSM in peatland sites in good condition and compare it with GSM in sites where it is known that bare peat exists. We show that degraded peatland show no signs of hydrology-driven seasonality and presents a negative trend in surface motion indicating subsidence.

As forests play an important role in the carbon cycle and biodiversity, their accurate description at large scale is essential. Local in-situ measurements are now often complemented by airborne or spaceborne LiDAR measurements that enable the characterization of forest structure at larger scale. However, even though the availability of up-to-date LiDAR data has improved for many regions in the last decade, this kind of data still shows temporal or spatial gaps. Consequently, these data sets do not allow for a global assessment of the current forest state. In this study, we investigate the potential of multi-frequency SAR for retrieving different LiDAR-derived structural forest parameters as a complementary tool to provide information on current and global forest structure. Our study area is situated in a deciduous broadleaf forest in central Germany. For describing forest structure, we used both airborne and spaceborne derived metrics. ALS metrics were derived from a leaf-off point cloud with a point density of 20 pts/m² provided by the Thuringian State Office for Land Management and Geoinformation and consist of fractional cover based on vegetation return numbers and standard deviation of the height distribution of the returns, aggregated to a 25 m x 25 m grid. Spaceborne metrics consist of GEDI L2B data at 25 m footprint dimension, non-gridded product[1], including total canopy cover, plant area index, forest height diversity and relative height 98% (canopy height). As comparative SAR data for forest structural metrics retrieval, we used radiometrically terrain-corrected backscatter data from TerraSAR-X HH, Copernicus Sentinel-1 VV and VH and ALOS-2 PALSAR-2 HH and HV data, acquired at corresponding dates to the LiDAR data and resampled to a 25 m grid size. We perform multivariate regression and evaluate which SAR dataset or combination thereof is best suited for retrieving the different forest structural parameters in temperate broadleaved forests, considering the different frequencies and available polarizations. Furthermore, as GEDI data includes leaf-off and leaf-on conditions, a comparison of both phenological states on the regression metrics of SAR data with GEDI structural parameters is also performed. Our preliminary results show the strongest correlation coefficient between GEDI cover and ALS fractional cover (r = 0.40), as well as between GEDI relative height, GEDI forest height diversity and ALS standard deviation (r = 0.45 and 0.41, respectively). Generally, higher correlations are observed at L-band HV and X-band HH data for all considered metrics. Moreover, ALS shows a slightly stronger correlation than GEDI, possibly because of its higher horizontal and vertical accuracy and original spatial resolution. Whereas the GEDI metrics canopy cover and plant area index display the highest correlations at L-band HV under leaf-on conditions, GEDI forest height diversity conversely shows a stronger correlation with the L-Band signal at leaf-off, similarly to ALS standard deviation. Further work will investigate the results with respect to the complementarity of the different frequencies for retrieving the different metrics. [1] Dubayah R, Tang H, Blair JB, Armston J, Luthcke S, Hofton S and Blair J, 2021, GEDI L2B Canopy Cover and Vertical Profile Metrics Data Global Footprint Level V002. NASA EOSDIS Land Processes DAAC; https://doi.org/10.5067/GEDI/GEDI02_B.002

New NASA spaceborne lidar missions – the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) and the Global Ecosystem Dynamics Investigation (GEDI) – have been acquiring science in parallel from 2019 to 2023. Neither were designed for measurement of non-forest vegetated ecosystems - ICESat-2 was designed to quantify ice elevation change, and GEDI was designed to measure vegetation structure and aboveground biomass across Earth’s tropical and temporal forests. GEDI and ICESat-2 provide near-direct but sample measurements of vegetation structure. Using these samples allows a local calibration of Synthetic Aperture Radar (SAR) based models for large area mapping of vegetation structure and aboveground biomass. To date, however, the calibration and validation of GEDI's algorithms and products for estimating canopy structure and aboveground biomass have been focused on temperate and tropical forests. Similarly, much of the research on extracting information about vegetation structure information from ICESat-2 has been in temperature and nutrient limited environments (e.g., boreal forests). Performance of both instruments for estimation of vegetation structure parameters are largely unexplored in water limited savanna ecosystems that are often dominated by sparse and short-statured vegetation. In this study, we present an evaluation of the efficacy of GEDI and ICESat-2 for the estimation of vegetation height and canopy cover across global savannas. Our analysis is conducted across pilot sites in South Africa and Australia that cover dry, temperate and tropical savannas. We examine measurement error in GEDI Level-2A/2B as well as ICESat-2 ATL03/ATL08 data products by colocation and comparison with equivalent reference estimates simulated from Airborne Laser Scanning (ALS) data. By accounting for uncertainty related to the mission sampling design, the performance and consistency of GEDI and ICESat-2 vegetation metrics at different height and cover ranges are shown, highlighting the impacts of filtering on science data quality and quantity, and the impact of differences between the ICESat-2 photon-counting and GEDI waveform measurement configurations. This study contributes to developing improved GEDI/ICESat-2 vegetation structure metrics that can be used to train SAR-based models. Refinements of algorithms focused on savanna and shrubland ecosystems may further extend the limits of detection for low stature vegetation and and reduce uncertainties in GEDI and ICESat 2 estimates. Taking advantage of the availability of both mission datasets on the ESA-NASA Multi-Mission Algorithm and Analysis Platform (MAAP) platform, we demonstrate how gridded metrics from both missions can integrated with data from current and future satellite missions (e.g., NASA-ISRO NISAR) for wall-to-wall mapping of canopy height.

Methods to extract scattering power components and their importance for land cover applications have been extensively studied in radar polarimetry. In radar polarimetry, the scattering power decomposition helps better characterize the scattering mechanism from a target. With Full-Polarimetric (FP) Synthetic Aperture Radar (SAR) data, the target decomposition techniques are broadly categorized into coherent and incoherent decomposition [1, 2]. Coherent decomposition techniques utilize information from the 2 × 2 complex scattering matrix S. In contrast, incoherent decomposition techniques are based on second-order statistics from the 3 × 3 covariance C or the coherency matrices T [3]. However, in a dual-pol SAR configuration, one does not have complete polarimetric information, limiting the discrimination of scattering characteristics from a trihedral and a dihedral target. Therefore, obtaining the scattering powers from these targets is challenging by applying existing decomposition techniques applicable to full-pol data to dual-pol SAR data. Nonetheless, the Complex Radar Cross Section (C-RCS) is distinct for a trihedral and a dihedral target. In this study, we measure the total polarized power returned from these targets to formulate two indices that characterize scattering from built-up (i.e., “dihedral-like”) and “surface-like” targets. In the literature, the built-up and “surface-like” targets are prototyped with dihedral (or narrow dihedral) and trihedral (or sphere, flat plate) coherent target models, respectively. Hence, one can adequately characterize such targets by measuring their polarized return. Motivated by the advancements, this study introduces a novel method to obtain the scattering power components from the dual-pol SLC and GRD SAR data. The idea is conceived from the two proposed indices: DpRBI [4] and DpRSI, that characterize scattering from the built-up (i.e., “dihedral-like”) and “surface-like” targets, respectively. For representation convenience, we denote the DpRBI and DpRSI index derived from the SLC SAR data as DpRBIS and DpRSIS, respectively. Similarly, DpRBI and DpRSI index derived from the GRD SAR data are denoted as DpRBIG and DpRSIG, respectively. We then utilize these two indices as a measure of similarity for a given pixel, either to a “dihedral-like” or a “surface-like” target. We then used the Scattering Power Factorization Framework (SPFF) proposed by Ratha et al., [5] to obtain the three scattering component powers for each pixel. We associate the two indices: DpRBI and DpRSI, as a measure of similarity for a given pixel to the built-up (“dihedral-like”) or “surface-like” targets, respectively. For example, a pixel with a high value of DpRBI and a low value of DpRSI is more comparable to built-up targets and vice-versa. The SPFF is based on the convex splitting of unity to get the weights from the two proposed indices: DpRBI and DpRSI. The dominancy of the scattering components (i.e., either “dihedral-like” or “surface-like”) is preserved in the SPFF. The total scattered power ⟨g0⟩ is split to obtain the three scattering power: “dihedral-like” (Pd−l), “surface-like” (Ps−l), and the residue power Pr. The proposed method is applied to dual-pol Sentinel-1 SLC and GRD data acquired over Barcelona, Spain, and Milan, Italy. Through visual inspection, we found the “dihedral-like” power (Pd−l) to be high for the built-up areas for the two cities. Similarly, over the waterbodies and the vegetation areas, the “surface-like” (Ps−l) and the diffuse powers (Pr) are found to be high, respectively. For quantitative assessment, we selected different land covers, viz., urban area (U), waterbody (W), and vegetation cover (V). In U, the values of DpRBIS(0.92 ± 0.07) and DpRBIG(0.95 ± 0.09) are found to be high compared to low DpRSIS(0.09 ± 0.05) and DpRSIG(0.11 ± 0.06) values. Consequently, we observed the Pd−l power (≈ 90) to be dominant compared to the other scattering components for both SLC and GRD SAR data. In contrast, for W, high values of DpRSIS(0.96 ± 0.001) and DpRSIG(0.95 ± 0.001) are found compared to low DpRBIS(0.08 ± 0.005) and DpRBIG(0.005 ± 0.004) values. It is due to the dominant “surface-like” scattering from W that resulted in a high Ps−l component. We observed the Ps−l component to be greater than 90 over the W. Over V, we observed both the DpRBI and DpRSI values to be low. The mean values of DpRBIS, DpRBIG, DpRSIS, and DpRSIG are found to be 0.32±0.17, 0.18±0.08, 0.20±0.06, and 0.14±0.06, respectively. One should note that both indices quantify the polarized information. Therefore, their values will be low for depolarizing targets like V. In cases where both the DpRBI and the DpRSI values are low, the residue power, i.e., diffused power, will be high. Hence, we observed the Pr (>70) to be dominant over V.

Radar observations that combine polarimetry and interferometry allow an enhanced characterisation of the natural environment and its processes. Fully polarimetric imaging provides data on scattering mechanisms present in the resolution cell and thus on the structure of the natural medium. Furthermore, differential interferometry allows the monitoring of displacements which occur on the line-of-sight component, focusing on changes caused by movements of the natural environment. Another key factor in radar measurements for characterization of natural processes is the revisit time: monitoring some of the Earth’s dynamic processes requires dense temporal sampling and long time series durations. For this reason, ground-based radar systems are a suitable choice for applications with high temporal sampling requirements as they can provide repetition times on the order of minutes and virtually unlimited time series [1]. In contrast, current spaceborne synthetic aperture radar (SAR) Earth Observation systems cover large areas in a single pass but their revisit time is, at best, within the region of several hours, making them unsuitable. KAPRI is a ground-based portable, frequency-modulated continuous-wave (FMCW) , Ku-band polarimetric radar interferometer. It is able to acquire both monostatic and bistatic full-polarimetric measurements with high spatial and temporal resolution. The bistatic configuration makes use of two independent KAPRI devices, one operating as the primary transmitter/receiver and the other as the secondary receiver. A new processing pipeline as well as a calibration method were developed to address these differences between bistatic and monostatic configurations [2]. The bistatic configuration is of special interest as less extensive research of its applications in the context of natural processes has been done if compared to the monostatic case, in part, due to the added challenges coming from separating the transmitter and the receiver [3]. KAPRI’s bistatic regime sets the scene for new investigations given that had not been possible by employing the monostatic mode, such as the measurement of 3-D deformation vector fields or the investigation of possible presence of unique scattering events exclusively detectable in bistatic configurations. These bistatic capabilities of KAPRI were recently leveraged for the first characterization of the coherent backscatter opposition effect (CBOE) in dry snow using a terrestrial bistatic radar system [4]. A better understanding of bistatic radar measurements, which can be efficiently provided by ground-based systems, lays the foundations for future SAR space missions and pushes forward the present knowledge on snow landscape areas and their role on climate. [1] Baffelli, S., Frey, O., Werner, C., & Hajnsek, I. (2018). Polarimetric calibration of the ku-band advanced polarimetric radar interferometer. IEEE Transactions on Geoscience and Remote Sensing, 56(4), 2295–2311. https://doi.org/10.1109/TGRS.2017.2778049 [2] Stefko, M., Frey, O., Werner, C., & Hajnsek, I. (2022). Calibration and Operation of a Bistatic Real-Aperture Polarimetric-Interferometric Ku-Band Radar. IEEE Transactions on Geoscience and Remote Sensing, 60, 1–19. https://doi.org/10.1109/TGRS.2021.3121466 [3] Moccia, G. Salzillo, M. D’Errico, G. Rufino, and G. Alberti, “Performance of spaceborne bistatic synthetic aperture radar,” IEEETrans. Aerosp. Electron. Syst., vol. 41, no. 4, pp. 1383–1395, Oct. 2005 [4] Stefko, M., Leinss, S., Frey, O., & Hajnsek, I. (2022). Coherent backscatter enhancement in bistatic Ku- and X-band radar observations of dry snow. Cryosphere, 16(7), 2859–2879. https://doi.org/10.5194/tc-16-2859-2022

Human Aspirations And Populations Are Proportionately Increasing And In Turn Settlements In Sensitive Zones Of Nature By Encroaching On The Heart Of Natural Space Have Increased Drastically. Habitation In Hilly Regions Is Increasing Rapidly; Consequently, Mankind Has Become Prey To Disasters Like Earthquakes, Floods, Landslides, Etc. But Human Scientific Temperament Consistently Strives To Beat Natural Adversity And Minimizes Human And Economic Loss By Detecting, Monitoring, And Predicting Hazards Causing Ground Deformations. Satellite-Based Microwave Remote Sensing Using Interferometric Synthetic Aperture Radar (InSAR) Technique Has Been Proven A Noble Technique For Land Displacement Study That Occurred Due To Landslides. In General, Ground-Based InSAR And WSN-IoT-Based Ground Sensors Are Installed For In-Situ Detection Of Ground Displacement That Occurred Due To Landslides. But Ground-Based InSAR Has A Narrow Coverage And WSN-IoT-Based Sensors Setup Gets Damaged When a Landslide Happens And Causes Disruption In The Continuous And Periodic Analysis Of The Landslide-Prone Region. Hence Such Setups Are Not Cost-Effective For Monitoring A Large Geographical Area And Are Unsuitable For Inaccessible Regions. On The Other Hand, InSAR-Based Time-Series Analysis Made It Easy To Study The Inaccessible Region With Wider Geographical Coverage Along With Periodic Monitoring. Moreover, Easy Public Access To Sentinel-1 Satellite Data Provided By Copernicus Open Access Hub For Further Processing Makes Satellite-Based Remote Sensing Studies Affordable For Researchers. Satellite-Based Remote Sensing For Real-Time Analysis In Critical Cases Of Land Displacement Has Not Been A Preferred Methodology Because Of High Temporal Decorrelation Due To The Long Revisit Time. However, For Monitoring A Large Geographical Area And Broader Coverage Of Inaccessible Terrain Satellite-Based Remote Sensing Is The Only Viable Solution. With The Latest Development, InSAR Supported With Differential Interferometric Sar (DInSAR) And Persistent Scatterer Interferometric Sar (PSInSAR) Techniques Overcome This Temporal Decorrelation To Some Extent And Hence Are Also Suitable For mm-Level Accuracy In Land Deformation Studies. InSAR Technique Facilitates The Detection Of Displacement That Occurred Due To Landslides By The Time-Series Analysis Of Multidate Images Of Sentinel-1 Satellite Data. The Fine-Quality Interferogram Generated By Processing Master And Slave Images (Taken At An Interval Of 6 Days) With The Help Of SNAP (ESA Open-Source Software) Has Several Fringes That Are Observed To Detect The Displacement That Occurred Due To The Landslide. The Main Task Which Is Proposed In This Paper Is The Prediction Of Rainfall-Induced Landslides By Regression Analysis Between The Antecedent Rainfall And Corresponding Land Displacement. Antecedent Rainfall Data Obtained By The Meteorological Department For Every Week Is Fed As The Input Experimental Data In A Machine Learning-Based Predictive Algorithm And Corresponding Displacement Which Is Analyzed From Interferogram Study Is Treated As Output. Thus Linear Regression Is Performed On Displacement Data Recorded At Every 6-Day Interval (Revisit Time) Of A Landslide-Prone Region With Corresponding Weekly Rainfall Data To Get The Cumulative Rain Displacement. Similarly, Further Feeding Experimental Data In The Form Of Antecedent Rainfall Data And Corresponding Displacement Of Several Months To The Predictive Algorithm Optimizes The Accuracy Of The Predictive Algorithm And Facilitates The Prediction Of Landslide Corresponding To Antecedent Rainfall For The Selected Landslide Prone Region. The Regression Model Is Used To Generate An Early Warning Signal For Rainfall Induced Landslides Based Upon The Careful Threshold Selection Using Expert Knowledge And Experimental Data. Thus, Forecasting Land Displacement By Generating Early Warning Levels i.e., Green, Yellow, Orange, And Red As Per The Degree And Severity Of Land Displacement Is Possible. Further, In 2023 NISAR satellite Will Be Launched And Its Design Permits Complete Global Coverage In A 12-Day Exact Repeat To Generate Interferometric Time Series And Perform Systematic Global Mapping Of The Earth's Changing Surface. Hence this Will Help In a More Accurate Analysis Of Land Deformation Due To Landslides. Thus, After The End Of The Sentinel-1 Service, NISAR Mission Data would Facilitate Our Future Work In The Development Of A More Accurate Landslide Early Warning System.

Multitemporal interferometric synthetic aperture radar (MT-InSAR) shows broad application prospects in geodetic Earth surface deformation observation with the advantages of millimeter accu-racy and long time-series monitoring ability. PS-InSAR is widely used for obtaining time-series deformation at the original spatial resolution by selecting persistent scatterers (PSs) with stable backscattering over a prescribed time interval. Conventional PS-InSAR performance depends on the PS density which is high in urban areas that have abundant man-made structures but poor in suburban districts that have many natural features. Thus, DS-InSAR has been developed to increase the measurement scatterer (MS) density, while maintaining the resolution of PS-InSAR. The essence of DS-InSAR technique is to preprocess the distributed scatterers (DSs) into the PSs and then solve them together in time series. DS refers to a group of adjacent pixels sharing a similar scattering mechanism with many small scatterers of similar size, which are likely to be affected by temporal and geometrical decorrelation. Decorrelation can be reduced by phase optimization of a certain number of neighboring pixels, namely the homogeneous pixels (HPs). Therefore, the HPs identification and the phase optimization are two key steps of DS preprocessing. (1) HPs identification: There are many HP identification methods, whether using single-polarization data or multi-polarization data. The HPs obtained by single-polarization methods are considered the statistical homogeneous pixels (SHPs). As the name implies, the SHP is identified by the statistical test between the center pixel and the remaining pixels within the predefined window one by one. Considering the advantages of multi-polarization data in describing the physical structures and dielectric properties of ground objects, HP selection methods based on multi-polarization data have inevitably been proposed. Polarimetric homogeneous pixels (PHPs) were first proposed by using a likelihood ratio test to determine the polarimetric stationarity. The similarity between PHPs can be characterized by clustering the interconnected pixels with similar scattering properties, which are assessed by the Cloude-Pottier eigenvalue-eigenvector decomposition with the Wishart distance measure. However, the above methods only use the polarimetric statistical information without considering the polarimetric interferometric SAR (PolInSAR) information. There is no doubt that such methods are unqualified to identify the accurate HPs, which should have not only similar backscattering properties but also consistent interferometric phases. The polarimetric interferometric homogeneous pixels (PIHPs) identification methods were proposed by using the PolInSAR coherency matrix, of which the large data dimension seriously increases the burden of similarity measurement. Besides, the similarity test indicator of this method is a trade-off between interferometry and polarimetry, of which the weight assignment is uncontrollable. Additionally, the PIHP identification based on the hypothesis statistic test relies on the rationality of the statistical model. To sum up, the existing PolInSAR similarity measurements have disadvantages of large computational burden, uncontrollable weight alignment, and uncertain statistical model, limiting the development of high-precision PIHP identification. (2) Phase optimization: Phase optimization is important for retrieving the phase history, which determines the phase quality and affects the MS density. As polarimetric SAR (PolSAR) data can be used to characterize the scattering process of a target, the decorrelation noise could be reduced by searching for the optimal scattering projection vector corresponding to the maximum temporal coherence or the minimum amplitude dispersion. The existing methods can be divided into the following: 1) a simple selection of existing polarimetric channels (e.g. BEST [6] and CMD-PolPSI [7]); 2) an exhaustive search of scattering vectors (e.g. ESPO [8]) and 3) eigen-decomposition of the polarimetric coherency matrix with an iterative search of scattering vectors (e.g. ESM [9]). These methods have been successfully applied for monitoring urban areas. However, only a single optimal scattering vector with the highest coherence can be obtained, of which the scattering mechanism is ambiguous. Considering the above limitations, we propose a novel DS preprocessing method for MT-PolInSAR deformation monitoring. First, the PolInSAR similarity measure for deformation monitoring is proposed, which combines polarimetric intensity, interferometric coherence, and phase. It considers static (polarimetric) and dynamic (interferometric) homogeneity together and can easily control the weight of two kinds of homogeneity. By choosing proper weighting factors, the accurate deformation spatial distribution can be depicted. Second, a novel object-oriented PIHP identification method based on BPT segmentation is proposed. Considering the advantage of hierarchical data representation, an image can be adaptively separated into multiple homogeneous regions with different sizes, which is beneficial for describing complex deformation scenes. Based on the BPT frame, a new PIHP identification method is proposed, which is not only useful for the reduction of the speckle noise effect but also useful for the reduction of interferometric phase noise. Third, a new PolInSAR phase optimization method is presented in which the Sum of Kronecker Product (SKP) decomposition method is applied to DS with PIHPs. Unlike existing polarimetric optimization methods, the proposed method considers polarimetric and interferometric coherence information simultaneously, resulting in separation of the polarimetric scattering process for each target and the maximum diversity for the corresponding phase center locations. Physical reliability of the phase optimization solution is thereby ensured. The performance of the novel method is evaluated using 30 quad-polarized C-band Radarsat-2 SAR images over Kilauea Volcano, Hawaii. There are mainly five kinds of ground objects in the test area, including the evergreen forest, bare land, grassland, shrub, and impervious surface. The results show that the DS number of our method is three times that of the traditional single-polarization (HH) method and twice that of the traditional quad-polarization method (ESPO). The overall coverage rate of final MS points of our method is over 70%, including the vegetation areas and bare land. It proves that our method provides a higher density of MS points and a higher quality of DS interferometric phase with a temporally stable phase center than traditional methods.

In this research study, we aim at demonstrating some preliminary results on the use of deep learning methods for Sentinel-1 polarimetric information reconstruction. The Sentinel-1 platform is able to acquire with one of four acquisition modes, namely, Stripmap (SM), Interferometric Wide Swath (IW), Extra Wide Swath (EW), Wave (WV). In the SM acquisition mode, S-1 provides uninterrupted coverage with a resolution of 5m-by-5m at a swath width of 80 km. SM is only used to image small islands and in exceptional cases, supporting emergency actions. On the other hand, in the WV acquisition mode, a single stripmap image (acquired at either VV or HH polarization) is acquired with an alternating elevation beam, generating 20 by 20 km vignettes at regular intervals of 100 km, at a 5m-by-5m resolution. WV mode at single-VV polarization is the default mode for acquiring data over the ocean. The use of PolSAR Data/Image Reconstruction from Single-Pol to Dual/Quad Pol has been already demonstrated e.g., by studies employing UAVSAR data. These methods can be useful e.g., for the generation of improved QuickLook Products, in order to provide users with an improved, lower-resolution, information for qualitative assessments, as well as for potential quantitative analysis after complete/robust validation. As a general principle, Polarimetric decompositions are optimal and designed for Quad-Pol data; nevertheless, Decompositions such as H-alpha(-Anisotropy) formulations are also derivable for the Dual-Pol cases, performing worse (than Full-Pol) for discriminating the three canonical scattering mechanisms from an isotropic surface, horizontal dipole, and isotropic dihedral, but still resulting as useful in some scenarios (forest, urban, ship). In this study we aim at demonstrating the use of Deep Learning models for dual-pol/ H-alpha for dual-pol information reconstruction from single-pol, exploiting dual-pol SM data for the unsupervised learning of GANs (Generative Adversarial Networks) designed to reconstruct the missing cross-pol and H-alpha for dual-pol information. After robust validation, the developed deep learning models can be used to potentially improve the WV mode VV single-pol information (default for WV mode) and to potentially provide improved quicklook information to users (also considering that WV mode products currently have no default quicklook/colorization scheme).

Sentinel-1 satellite provides free access to dual-polarization (VV and VH) images. The integration of information from both VV and VH channels in polarimetric persistent scatterer interferometry (PolPSI) techniques is expected to enhance the accuracy of ground deformation monitoring as compared to conventional PSI techniques, which utilize only the VV channel for Sentinel-1. Persistent scatterer (PS) and distributed scatterer (DS) points play a crucial role in the PSI techniques. PSs with high phase qualities are commonly found in urban areas. As a complementary for PSs, DS points whose phase is affected by noise are commonly present in rural areas. In this study, the identification and selection of PS and DS is based on an optimal channel created by combining the two polarimetric channels. PS candidates are selected through the amplitude dispersion (DA) criterion. To jointly utilize both PS and DS points, an adaptive speckle filtering based on the selection of homogeneous pixels (HP) was applied to the coherency matrix. Then, DS candidates were identified by using the average coherence criterion. Finally, using both PS and DS points, the Coherent Pixels Technique (CPT) was employed as the Persistent Scatterer Interferometry (PSI) processing method. In order to analyze how the introduction of the VH channel helps improve the deformation measurement results, an experiment over Barcelona in Spain was carried out. The dataset consists of 189 dual-polarization SAR images acquired between December 2016 and January 2021. A wide variety of scenarios are present in this region, i.e., airport, harbor, and urban areas which exhibit diverse orientations of streets and buildings with respect to the acquisition geometry. Additionally, ground deformation is expected over some areas due to settlement of recent constructions and in the harbor. Regarding PS, there are two cases in which the VH data contribute to improve the PS density. The first corresponds to scatterers that are oriented with respect to the incidence plane. The VH amplitude value of those scatterers are higher than VV channel. The second case appears more frequently than the first case and corresponds to pixels in which the VH amplitude is low but stable. Through the application of PolPSI technique, the VH channel can contribute to the selection of high-quality pixels by reducing the presence of peaks and fluctuations present in the VV channel, thus enabling the selection of pixels with good quality which would not have been identified if only VV data were processed (Luo, et al., 2022). Instead of increasing the density, the contribution of VH channel for the identification of DS points is associated with a more accurate selection of HP. The polarimetric information enables the differentiation of pixels that belong to different targets but have similar amplitude values in the VV channel. This results in a more reliable deformation measurement, as the HP group becomes more accurate. A comparison with experimental data and all cases (single- and dual-pol) serves to illustrate and evaluate the performance of PolPSI in this domain. Reference: Luo, J., Lopez-Sanchez, J. M., De Zan, F., Mallorqui, J. J., & Tomás, R. (2022). Assessment of the Contribution of Polarimetric Persistent Scatterer Interferometry on Sentinel-1 Data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15, 7997-8009.

【Objective】Using the five bands (P/L/S/C/X) of quad-pol SAR data acquired by the High-resolution Airborne System, this study analyzed the response law and sensitivity of different band signals to forest stem volume. Moreover, the potential of estimating stem volume based on both single-frequency and multi-frequency PolSAR data was evaluated.【Method】The study was conducted in Chaocha Forest Farm, located in Genhe City, where a total of 69 plots with an area of 25m × 25m were surveyed. The stem volume of each plot was calculated by applying species-specific allometric equations, and then scaled-up to the entire study area using relationships established with LiDAR height metrics. Stratified sampling was employed to collect a total of 196 samples. The Water Cloud Model was utilized to analyze the response law of SAR backscattered intensity to stem volume across different bands, and the dynamic range and saturation point were quantified. Furthermore, multiple polarimetric decomposition components were extracted and their sensitivity to stem volume was analyzed using correlation coefficients. On this basis, random forest and support vector regression algorithms were used to perform feature selection and regression modeling.【Result】The Water Cloud Model analysis revealed that the dynamic range of the longer wavelength (P/L) is higher compared to the shorter wavelength (S/C/X). In particular, the saturation point for the P band exceeds 160 m3/ha, whereas it does not surpass 100 m3/ha for the other bands. The correlation analysis results indicated that the correlation between the P band, L/S bands, and C/X bands and stem volume decreases in order, with values above 0.6, between 0.3-0.4, and below 0.3, respectively. When estimating stem volume using single-frequency data, the accuracy for the P band was 73.79%, whereas for the other bands it did not exceed 60%. The incorporation of L or S band improved the accuracy of P-band estimation by 2%, while the contribution of C or X band was considered negligible. The highest accuracy for stem volume estimation could be achieved by combining all the bands, resulting in an accuracy of 77.62%.【Conclusion】Taking into account factors such as signal dynamic range, saturation point and correlation, the P band is the most sensitive to forest stem volume, which is significantly better than the other bands. The L/S bands are second in sensitivity, while the C/X bands are the least sensitive. When estimating forest stem volume using PolSAR data, the P band should be the first choice. In addition, the joint use of the L or S band can provide a good gain in estimation performance.

PolSAR (Polarimetric Synthetic Aperture Radar) images are usually represented by scattering matrix S, nevertheless in order to study its statistical behavior, it is more convenient to represent these images by the lexicographic basis vector, from which the covariance matrix Z can be derived. The Z are complex Hermitian positive definite matrices, which have a natural Riemannian metric tensor. Once represented by its covariance matrices, the multilook PolSAR images can be handled by Wishart distribution, whose parameters are the covariance matrix, where |Z| is related to image brightness, and the number of looks L, which is related to signal to noise ratio. The goal of this work is to present an unsupervised classification of multilook PolSAR images. One of the unsupervised classification is the data clustering, whose main goal is to divide the data set into subsets, based on the sample's similarity. Distances are the most predominantly metric to quantify the similarity between two samples. Since the PolSAR images can follow the Wishart distribution, the selected contrast estimator was the Hellinger stochastic distance, which measures the similarity between two Whisharts. In this work we propose an unsupervised clustering algorithm named Bisecting Stochastic Clustering (BSC). This algorithm is a combination between the Stochastic Clustering (SC) algorithm and the hierarchical divisive clustering algorithm. The BSC algorithm is a top-down procedure, it starts with all samples in an unique cluster that are successively split into two new sub-clusters. This algorithm is mainly divided into three steps: 1. the initial parameter determination, 2. the cluster bi-partitioning procedure, and 3. the choice of a suitable cluster to split. The SC algorithm is a technique based on K-means, which uses stochastic distances as a similarity metric. This algorithm can be trapped in a local minimum, which leads to incorrect clustering results. Therefore, the choice of good initial parameter candidates is essential for the clustering quality. For this reason we propose the use of an initial seed estimation. The method used for initial parameter determination is named Riemann Principal Direction Divisive Partitioning (RPDDP). The RPDDP is a new proposed algorithm, whose goal is to perform the bi-partition of a dataset. This algorithm estimates the dataset covariance matrix under the Riemann geometry, in order to find the principal component, which is used to separate the input data in two sub-clusters. From the RPDDP two estimated sub-clusters are derived, which are used as the BSC initial parameters. Once the initial parameters are determined, the BSC second step is performed by the SC algorithm. The BSC builds a dendrogram in order to represent the dataset splitting. Each sub-cluster, or “node”, links two successor sub-clusters in the dendrogram. When three or more nodes are available in one dendrogram level, the algorithm needs to choose a node to split. The BSC third step uses the information gain as the node choice rule. The information gain is based on the decrease in entropy after the dataset splitting, i. e., it represents the amount of entropy removed between clusters. The information gain is computed to all nodes in a given level. The candidate node with the highest information gain is the winner and will be split in the next bisecting step. In order to validate the proposed algorithm, four PolSAR images were analyzed. The images are from three different scenarios: the Brazilian Caatinga, the Amazon forest and urban area. Hence, our ambition is to analyze the BSC algorithm with various scattering mechanisms. The images used in this work are in L and C bands, and full quad-polarized. Two of the images are provided by the Spaceborne Imaging Radar-C/X-SAR, and the other two images are generated by the PALSAR. The images are named: the Bebedouro SIR-C in C band, the Bebedouro SIR-C in L band, the Tapajós ALOS/PALSAR, and the San Francisco ALOS/PALSAR. The proposed algorithm was analyzed in both quantitative and qualitative ways. The confusion matrix is an external cluster validation method used to quantitatively characterize the image classification accuracy. The qualitative analysis seeks for features that are observed and can not be measured with a numerical result in order to validate the BSC algorithm classified image. In this work the qualitative analysis handles the cluster dendrogram, and the scattering mechanism. One of the advantages of the BSC algorithm is that there is no need for inputs such as the number of classes. The algorithm can run until the stop criteria is reached and generate a suitable number of classes. The dendrogram, information gain, and scattering mechanisms information helps on the unknown classes characterization. For instance the BSC classified the San Francisco image and generated seven classes more than the predicted on the truth image. Three out of seven classes were created in the water region, meaning that the ocean may have different classes due to, for example, sea level, pollution, and so on. In general the L bands images had superior classification results in comparison with the C band image. The real PolSAR image classification qualitative analysis ratified the results of experiment 3 and showed that the BSC-R had the best accuracy result. The L band's images had an average accuracy result of approximately 90 %.

For many years, target characterization in radar polarimetry has been a subject of great interest. The availability of full polarimetric Synthetic Aperture Radar (SAR) data for Earth and planetary observation has prompted academics to create novel methods to extract pertinent scattering properties. These descriptors offer useful information on scattering from basic to complicated targets. To quantify target information, Cloude et al.~\cite{cloude1996review,cloude1997entropy} proposed an eigenvector-eigenvalue based analysis of the $3 \times 3$ coherency matrix. By introducing a new parameterization of the eigenvectors, referred to as the $\alpha-\beta$ model, where $\alpha$ is the scattering-type parameter, and $\beta$ is the target orientation angle, they suggested extracting average parameters. They extracted the mean scattering-type parameter $\overline\alpha$ using this method from fully polarimetric SAR data. The polarimetric entropy $H$ (in the von Neumann sense) was developed as a basis-invariant quantity to characterize the level of statistical disorder. These two variables are frequently utilized in the literature for target characterization and categorization~\cite{lee2009polarimetric}. Touzi~\cite{touzi2006target} addressed the need to evaluate the robustness of the $\alpha-\beta$ model under a change of wave polarization basis. A new scattering vector model was proposed derived from the con-diagonalization of the Kennaugh-Huynen scattering matrix projected into the Pauli basis. Two parameters, $(\alpha_{s}, \Phi_{\alpha_{s}})$ were introduced in the new scattering vector model. Here $\alpha_{s}$ is the angle of the symmetric scattering vector defined in the trihedral–dihedral basis, and $\Phi_{\alpha_{s}}$ is the phase difference between the vector components in the trihedral–dihedral basis. These parameters are widely used in literature for several applications~\cite{touzi2009phase, bhattacharya2011polarimetric, li2012polarimetric}. Ratha et al.~\cite{ratha2019polsar} utilized the geodesic distance (GD) measured over the unit sphere centered at the origin in the space of real $4\times4$ matrices. This unit sphere contains all the normalized Kennaugh matrices, equivalent to the second-order information expressed by the $3 \times 3$ coherency (or covariance) matrices. The geodesic distance is suitably used to obtain three roll-invariant parameters: scattering-type $\alpha_{\text{GD}}$, helicity $\tau_{\text{GD}}$, and scattering purity $P_{\text{GD}}$. Furthermore, the authors also proposed a $(P_{\text{GD}},\alpha_{\text{GD}})$ unsupervised classification scheme to cluster full polarimetric SAR images into eight zones. Unlike the parameterization of the eigenvectors, Dey et al.~\cite{dey2020target, dey2021four} proposed a scattering-type parameter $\theta_{\text{FP}}$, and a scattering asymmetry parameter (i.e., helicity) $\tau_{\text{FP}}$ for full polarimetric (FP) SAR data. The method utilizes the coherency matrix elements (or the Kennaugh matrix elements) and the Barakat degree of polarization~\cite{barakat1983n}. This method does not utilize the eigendecomposition of the coherency matrix and relates the Huynen parameters~\cite{huynen1970} to characterize target information. Dey et al. similarly extended the notion of $\theta_{\text{FP}}$ to dual co-polarimetric (i.e., HH-VV) and compact polarimetric (i.e., RH-RV or LH-LV) SAR data. They utilized them along with $\tau_{\text{FP}}$, and $H$ for many applications~\cite{dey2020isprs, dey2021rice}. This work proposes an alternate scattering-type parameter that essentially discriminates between corresponding targets (viz., dipole--quarter wave and dihedral--helices) that has an identical value of the Cloude $\alpha$ parameter. To derive this alternate scattering-type parameter, we define the mean $m$ and the standard deviation $s$ of the complex eigenvalues of the $2 \times 2$ scattering matrix $\mathbf{S}$. In radar polarimetry, $\mathbf{S}$ describes the polarization state of the backscattered radiation by representing the polarimetric backscattering coefficients of the two co-polarized and the two cross-polarized signals. It encompasses complete polarimetric information about backscattering from targets for each pixel. Let us define the normalized scattering matrix $\mathbf{\underline{S}} = \mathbf{S}/\|\mathbf{S}\|$. Then, using $\mathbf{\underline{S}}$, let us define the normalized Graves matrix $\mathbf{\underline{G}} = \mathbf{\underline{S}}^{*}\mathbf{\underline{S}}$. The Graves power density matrix, also called the polarization power scattering matrix, is a $2 \times 2$ Hermitian and positive semi-definite. Therefore, we express the two quantities, $m$ and $s$ using the two normalized $2 \times 2$ matrices, $\mathbf{\underline{S}}$ and $\mathbf{\underline{G}}$. We define a $2 \times 1$ real vector $\mathbf{u}_{\text{T}}$ using $|m|$ and $s$ that can suitably represent elementary targets. We propose an alternate scattering-type parameter $\overline{\alpha}_{\text{A}}$ by utilizing the norm of the difference between the measured target vector $\mathbf{u}_{\text{T}}$ and that of a dipole. Therefore, one can note from Table~\ref{tab:urban_tabel} that $\overline{\alpha}_{\text{A}}$ suitably discriminates between corresponding targets with an identical value of the Cloude $\alpha$ parameter.

Natural disturbances such as windthrows, wildfires, and insect outbreaks are intensifying around the globe and compromise forest carbon stocks, timber production, and other ecosystem services. These rapid ecological transformations require regular monitoring and assessment, often combining different ground and remote sensing data. We focused here on the Czech Republic, which has become Europe`s disturbance epicentre due to the country-wide outbreak of spruce bark beetle Ips typographus. The outbreak started in 2018 and has persisted, damaging up to 30 mil. m3 of wood annually. The emerging disturbance patterns range from a single tree and patch mortality to areas large hundreds of hectares covered by the mosaics of standing dead trees and salvaged stands. This structural complexity and the rapid progress of the disturbance make its mapping and appraisal challenging. The study presents a new approach based on the fusion of optical and SAR remote sensing data for monitoring the outbreak between 2015 and 2021 across the whole of the Czech Republic. The study used extensive ground-truth data to train the deep learning-based classification algorithms. Landsat 8 data (30 m resolution) and Sentinel-1 A SAR data (20 x 5 m resolution, center frequency of 5.40 GHz) were pre-processed in the Google Earth Engine platform (GEE). Atmospheric correction and cloud masking were applied to the Landsat 8 collection of images, and vegetation indices such as Normalized Differential Vegetation Index (NDVI), Normalized Differential Water Index (NDWI), and Enhanced Vegetation Index (EVI) were generated. For Sentinel 1 SAR data, radiometric calibration, speckle filtering, and terrain correction were applied to reduce noise and distortions. SAR variables and polarizations, such as Ratio Vegetation Index (RVI), backscatter, and co-polarization, were extracted. The ground-truth data described forest cover change within the studied period. A difference of two Digital Surface Models with a centimetre accuracy from the periods 2016-2018 and 2018-2020 were used. These data were acquired in the frame of the regular airborne monitoring of forest conditions in the Czech Republic: the dataset included circa 200 000 polygons with an area of approximately 600 000 ha. The python API in the Google Earth Engine (GEE) platform was used to develop and train the classification model. The API allowed us to interact with the GEE platform using the Python programming language, providing a flexible environment for model development. We utilized Google Collab Notebooks, a cloud-based Integrated Development Environment (IDE), which provided powerful computing resources and pre-installed software libraries. TensorFlow, a widely used deep-learning framework, was used for implementing and training the model using Python in the GEE platform. The classification model was represented by a Multilayer Perceptron Neural Network. The input layer consisted of the predictors derived from SAR (polarimetric properties, RVI, and backscatter) and optical (NDVI, NDWI, EVI, Band 5, Band 7) remote sensing data. Several architectures were tested, with hidden layers consisting of fully connected layers with a rectified linear unit (ReLU) activation function. The output layer neurons had a sigmoid activation function, which produced binary classification outputs (disturbed – not disturbed). To optimize the model`s performance model, we used Hyperparameter optimization by implementing grid search in the Keras Tuner package to find the best combination of hyperparameters, i.e., the number of neurons in each hidden layer, activation function, dropout, and learning rates. A combination of hyperparameters that produced the best validation accuracy was used. Combining Landsat and SAR variables as forest disturbance predictors produced an overall training accuracy of 88% and validation accuracy of 85%, outperforming single-sensor data-based classifications. By comparing the predicted changes for the whole of the country against all ground-truth data, we achieved a precision of 0.66 (a proportion of correctly classified and all classified disturbed areas), a recall of 0.83 (a proportion of correctly classified and all reference disturbed areas), and an F1 score of 0.71 (a harmonic mean of precision and recall). As expected, the accuracy was higher in large, disturbed patches. The total disturbed areas in the country and within the administrative districts (n = 88) show a high correspondence with the annual forestry statistics based on the forest owners` reporting (in terms of m3 of salvaged wood). The developed framework was proved suitable for annual disturbance mapping with accuracy and spatial detail suitable for supporting management planning and disturbance management. Further research will extend this framework toward forest biomass change mapping, using an extensive plot-based National Forest Inventory dataset as ground-truth data.

In recent years, the demand for synthetic aperture radar (SAR) remote sensing data has increased manifold due to its all-weather operational resources monitoring. Compact polarimetry (CP) offers a trade-off between coverage, hardware requirements and polarimetric information content compared to full polarimetric (FP or QP) SAR data. India’s EOS-04 (RISAT-1A), launched on 14th Feb. 2022, offers SAR imaging capabilities in diverse polarization modes that includes conventional linear-pol (single, dual and quad), compact-pol modes and ScanSAR full-pol in C-band. This is a successor mission of the RISAT-1 SAR imaging system with better calibration and noise-correction for target characterization. Much research has been conducted on using CP data either by using simulated CP (SCP) data from full-pol or real CP data from RISAT-1, ALOS-2 and RCM SAR systems. However, limited research has been conducted comparing real CP data with FP data from the same SAR system. In this paper, we have used RISAT-1A single look complex (SLC) data of full-pol and real compact data in FRS-1 mode for comparative classification assessment over the Mumbai metropolitan region of India. The Full-pol and compact mode (RH/RV, right circular transmit-linear receive) data were acquired on 10 and 12 Nov. 2022 respectively at the same incidence angle, but in different look direction. The study area is dominated by land features such as settlements, water bodies and vegetation cover. For classification purposes, settlements are further subdivided into oriented and non-oriented urban types depending on the radar look direction. Vegetation cover is sub-classified into the reserved forest and mangrove forest. There are also wetlands near the creeks. The training and test regions of interest (ROI) were generated using ground truth information for classification and accuracy assessment purposes. The polarimetric SAR covariance matrix follows the complex multivariate Wishart distribution. Hence, Wishart supervised algorithm has been applied for classifying the targets. The classification comparison is performed for four polarization modes, i.e., QP, real CP, SCP and linear dual-pol (HH+VV, HH+HV and VV+VH) over the study area. The overall accuracy assessment suggests that QP offers the highest classification accuracy (90.75%) compared to any other polarization combinations for the given classes, which is approximately 5.5% higher than the second best polarimetric combination i.e. SCP. Water bodies are classified accurately in all the polarization combinations with 100% accuracy. Vegetation features have shown higher classification accuracy in full-pol than other polarization combinations. However, urban (non-oriented) areas have shown similar accuracy (83%) in quad-pol and real-CP SAR data. The real CP classification accuracy for the same urban feature is ~12% higher than simulated CP data. Interestingly, the overall classification accuracy of simulated CP (85.03%) and real right-circular CP (84.37%) are almost similar. Further investigation is required in terms of noise-levels, transmission circularity and calibration parameters of real and simulated CP SAR data while performing such comparisons. The overall accuracy results indicate that the QP offers the highest classification performance, followed by CP (real and simulated) and dual linear co-pol (HH+VV) data. The dual-linear cross-pol combinations (HH+HV and VV+VH) have the least classification accuracy over the given study area.

Detection of Acacia Xanthophloea spp using Sentinel-1( Single Look Complex) and Machine Learning Approaches Anne OSIO, Kenya and Samson AYUGI, Kenya Key words: Synthetic Aperture Radar, Object Detection, Machine Learning, Change Detection, flooding SUMMARY Automatic mapping of land cover types on flooded Riparian landscapes is one of the most challenging problems in remote sensing. Although the Object-based approach has been embraced in Land Use Land cover studies few studies have applied it to riparian vegetation discrimination. In order to improve on the accuracy and efficacy of any classification, new approach to data collection and extraction has become increasingly necessary. The study investigated an Object based classification based on Sentinel-1 (SAR) of Single Look Complex data using three classifiers namely Naive Bayes, Decision Tree and Random Forest. Four land cover types i.e. Acacia Forest, Built-up, Grasslands, Water and other were successfully retrieved. Change detection was noted on the North western strands having reduced from an area of 212.7 hectares in 2008 before the floods to an area of 64.64 hectares in 2019 after floods. ALOS-1 Level 1.1 was used as reference image captured in 2008 before floods and Sentinel 1 SLC data captured in 2015, 2017 and 2019 after the floods, Acacia Forest strands were captured to have been degraded especially on the North western part of Lake Nakuru. Classification based on Machine learning on Polarimetric Matrix generated bands C11, C22 and their ratio were used to obtain the results as follows; Naive Bayes 91.1% , Decision Tree 94.1% while Random Forest took the lead by 94.4%. One-way Analysis of Variance (ANOVA) was used to compare variations among the group algorithms. The results show that there was no significance difference between and among the group means at F 2,15 =0.529 at p-value 0.60 and α

This study aims to investigate the synergies of fusing high-resolution TanDEM-X InSAR data and the upcoming BIOMASS PolInSAR data (lower resolution) for large-scale forest height mapping. A new merging algorithm was proposed and validated by TanDEM-X InSAR data and simulated BIOMASS PolInSAR data from an airborne SAR campaign. TanDEM-X and TerraSAR-X constellation forms the first single-pass interferometer in space, allowing for the first time the acquisition of InSAR and PolInSAR data without the impact of temporal decorrelation disturbance globally. Single-pol, dual-pol, and quad-pol TanDEM-X data have all been demonstrated to be capable of estimating forest parameters in different forest scenarios [1-3]. Due to the weaker penetration capability compared with low-frequency SAR as well as lower SNR and coverage capability compared with single-pol mode, the TanDEM-X PolInSAR forest height inversion is considered to be a suboptimal choice for global-scale forest height mapping [2]. Single-pol TanDEM-X InSAR data have been widely used for forest height mapping based on the well-known two-layer model – random volume over ground (RVoG) model– with or without the assistance of external LiDAR measurement as a priori information [2-4]. When the external LiDAR DTM is available as input to calculate the sub-canopy ground phase for the RVoG model, the TanDEM-X inversion achieves the highest accuracy compared with only coherence amplitude-based inversion. However, large-scale DTM data is not available for many countries and regions. The upcoming ESA BIOMASS mission brings opportunities to solve this limitation. As the 7th Earth Explorer Mission within the frame of the ESA Earth Observation Program, the BIOMASS mission will carry, for the first time, a P-band polarimetric interferometric SAR payload in space, flying as repeat-pass acquisition, to accomplish the mapping of sub-canopy topography (DTM), forest height, and a full range of the world’s above-ground forest biomass [5-6]. However, one of the biggest challenges for forest structure parameter retrieval is the highly reduced spatial resolution due to the limited bandwidth of 6 MHz [6-7]. Such limitation also decreases the interferometric coherences and polarimetric behavior in single interferometric pair. The expected BIOMASS-derived DTM would be approximately 100m resolution and is still too coarse to directly calculate the ground phase on the support of the TanDEM-X InSAR inversion. In this paper, a new fusion algorithm of TanDEM-X and BIOMASS data is proposed. Firstly, a coarse resolution DTM can be obtained by BIOMASS PolInSAR data using the coherence TLS line-fitting method on the basis of RVoG model [8]. Then, in order to generate a high-resolution resolution (12m) ground topography map, the high-resolution TanDEM-X InSAR-derived DEM and lower-resolution BIOMASS PolInSAR DTM are merged using a wavelet-based method [9]. More specifically, the high-resolution TanDEM-X DEM is decomposed into high-frequency and low-frequency signals by wavelet decomposition. Since the sub-canopy topography is expected to be continuous and smooth and should belong to the low-frequency signals, the low-frequency signals retrieved from TanDEM-X DEM are replaced by the lower-resolution BIOMASS DTM signals. Afterward, a high-resolution DTM can be obtained by an inverse wavelet transform. Finally, the DTM is input to calculate the RVoG model parameter, i.e., the sub-canopy ground phase, and the TanDEM-X InSAR data can be used to inverse forest height and extinction (or ground-to-volume amplitude ratio, if we fix the extinction) with DTM aided. To evaluate the method performance, a TanDEM-X single-pol interferometric pair (acquisition on 2011.02.27) and a set of simulated BIOMASS PolInSAR data were introduced to validate the proposed method. The simulated BIOMASS 6MHz data are characterized by an azimuth resolution of 12.5m and range resolution of 25m, which is simulated using the P-band full-resolution data from the BioSAR-2008 airborne campaign and is released by the DLR [10]. The result shows that the forest height inversion from TanDEM-X/BIOMASS fusion method yields a higher accuracy than the TanDEM-X-only inversion. An amplitude-based inversion using the SINC model [1,11-12] was adopted for the TanDEM-X-only inversion since no external data is introduced for calculating the phase. The TanDEM-X SINC model-based method presents an overestimation due to the ignorance of both extinction and ground-to-volume amplitude ratio contribution, while the forest height from the TanDEM-X/BIOMASS fusion method shows a high correspondence with respect to LiDAR measurements (Fig. 1). Based on the comparison with the LiDAR reference, the root-mean-square error (RMSE) of forest map from proposed fusion method is 2.77m under the resolution of 50m×50m window, with an R2 of 0.84. This study demonstrates that the fusion of TanDEM-X and BIOMASS data has great potential for large-scale forest parameter mapping and global biomass estimation.

Forest disturbances, like logging and wildfires, determine a loss in terms of woody biomass and modify Earth carbon balance. Earth observation capacity to monitor forest disturbance and regeneration plays a fundamental role in supporting ecosystem management and surveillance. Procedures for monitoring and classifying forest disturbances using satellite data have improved recurrently in last decades, with the development of many algorithms that exploit dense time series at high spatial resolution. Characterization of disturbance in forest ecosystems, like identification of event occurrence dates and loss in terms of biophysical parameters (i.e. biomass, leaf area index) are essential features available for public authorities and practitioners, that could serve information needs related to altered ecosystems. Identification, classification and disturbance characterization can be generated using methodological approaches accounting for differences between pre- and post-event conditions, observed by satellites. Even though parameters related to photosynthetic activity (e.g. leaf area index, canopy cover density) are available from Earth Observation time series for both pre- and post- disturbance events, biomass products availability is still limited. At present, Earth Observation biomass measurements are available from ESA Climate Change Initiative (CCI) Biomass project, being last distributed product referencing to year 2018. Global Ecosystem Dynamics Investigation Lidar (GEDI), a NASA project started in 2018, consists of high resolution laser ranging of Earth’s forests and topography and provides information to measure forest structure. On one hand, the ESA CCI Biomass product is not able to provide information about post-event, on the other hand GEDI provides point information with very limited probability of revisit in time. Forthcoming ESA Biomass and the NASA-ISRO NISAR will provide unprecedented forest biomass measurement, with revisit time and spatial resolution suitable for forest disturbance monitoring. In the meanwhile, alternative methods should be used in order to estimate biomass loss from forest disturbances. A biomass downscaling approach is here presented, with the aim to estimate potential woody biomass removal after disturbance events takes place. The proposed approach is applied on logging and wildfires test cases, makes use of vegetation biophysical parameters estimated at 20 m spatial resolution to downscale ESA CCI Biomass product, with a specific focus on post-event conditions. Results show the capability to characterize disturbances in forest ecosystems by estimating potential biomass removal, and highlight biomass products capacity in providing information for operational services addressing ecosystems monitoring and surveillance.

One of the major challenges faced by society and the environment in this century is climate change, which has led to various impacts, such as the severe drought in the Po River basin. This condition has been primarily caused by increasing temperatures and a shortage of rainfall. The Po River District Basin Authority announced in June 2022 that this is the worst water crisis experienced in the last 70 years. Monitoring the river basins continuously can be crucial for understanding the evolution of the issue and aiding in the formulation of regional or national water resource management strategies. Decision-makers can rely on cost-effective and dependable data on the river basins through the analysis of time-series satellite-based measurements. Within this context, remote sensing plays an important role. An optical sensor is a potential tool for its ease of interpretation. Nevertheless, adverse meteorological conditions, such as cloud cover and solar illumination, can significantly impact optical radiation, making it challenging to extract information at times. On the other hand, radar sensors offer continuous observations in almost all-weather conditions and cover a wide area. Synthetic Aperture Radar (SAR) can be particularly advantageous for monitoring river basins due to its high-resolution imaging capabilities. State-of-the-art approaches, aimed at exploiting the SAR imagery to monitor the water extent of inland water basins, are mainly based on the use of SAR imagery collected at different times, i.e., a time series analysis. The first step consists of land/water discrimination; while the second step consists of using image processing techniques to monitor water-area variations in the basins. However, monitoring river basins with single-polarization SAR data can be complicated because of speckle noise, which can make interpreting images difficult, and the potential limitations in distinguishing between water and land. Therefore, using polarimetric information can be crucial for effective water basin monitoring. In recent years, the growing number of multi-polarization SAR missions has resulted in the development of new value-added products across various domains such as forests, land, urban areas, and oceans. The main goals of this study are to develop multi-polarimetric and multi-temporal methods to effectively monitor the change of surface water area of the Po River, which is the longest river in Italy. The selected test site was strongly affected by drought in the Summer of 2022. Hence, monitoring it can be fundamental to understanding his emplacement evolution as well as the effects of climate change. The proposed methodology is hereinafter described. First, a dual polarimetric procedure, applied along the whole time series of Sentinel-1 SAR imagery, is developed to generate binary images where land and river are clearly distinguished. Then, the extent of the water coverage is estimated by using the pixels labeled as “water” in the binary image. Results are obtained by processing a set of 88 dual polarimetric (DP) SAR data collected at C-band from the Sentinel-1 mission over the section of the Po River close to the town of Piacenza, Emilia Romagna, from June 2015 to September 2022 (one per month). Preliminary results show that: a) SAR dual-polarimetric approach can effectively monitor the water body changes that occurred over the river Po basin in an effective way; b) the increase in temperatures together with the decrease in rainy days can likely be the reason behind the drought phenomenon verified in the Summer of 2022 over the Po River; c) SAR monitoring can be a useful tool for helping on understanding effects caused by the climate change.

The Forest Carbon Monitoring project (https://www.forestcarbonplatform.org/) for the European Space Agency is developing Earth Observation (EO) based user-centric approaches for forest carbon monitoring. Different stakeholders have a common challenge to monitor forest biomass, but specific requirements vary between users. Policy-makers need information to make better decisions; public organizations need information for national and international level reporting. Companies require means to respond to increasing monitoring requirements, and tools for carbon trading. To support forestry stakeholders in these requirements, the project has developed a prototype of a monitoring platform which offers: • A selection of monitoring approaches designed for forest biomass and carbon monitoring for varying large and small area requirements. • Cloud processing capabilities, unleashing the potential of the increased volumes of high-resolution satellite data and other large datasets for forest biomass and carbon monitoring In this poster, we will highlight the key achievements and findings of the project. The project started with an extensive review of policy needs and users’ technical, methodological and data requirements. Project user partners were interviewed for detailed requirements. This information was reflected against the current state-of-the-art of EO based forest carbon monitoring methods to identify the potential and limitations of EO based forest biomass monitoring. During the first year of the project, different approaches for data processing, forest structural variable estimation and biomass estimation were tested and evaluated. Key findings include e.g. (1) the benefits for optical and radar data fusion, particularly with X-band radar data improving height estimation, (2) the benefits and capabilities of multivariable methods (such as kNN and Probability) in production of forest structure variable estimations and (3) the potential of deep learning approaches. Robustness of the methods and cost of the datasets were also considered important aspects for operational monitoring platform. During the second year of the project, three different types of demonstrations were conducted and validated: • Local demonstrations designed to meet private company and other small area requirements. • Provincial to national demonstrations aimed primarily at administrative agencies, often using National Forest Inventory (NFI) based approaches. • Continental demonstration, aiming to meet the needs of international organizations and other communities requiring continental level information. Altogether eight demonstrations were conducted, seven in Europe and one in Peru. Each user organization received a set of output products (according to their requirements) from three main product groups: 1. Forest structure variable products 2. Biomass and growth products 3. Change products Altogether 79 products with area coverage varying from 1 to 750 Sentinel-2 tiles were delivered to the users. The output products were produced with k-NN (Alt 2001), Probability (Häme et al. 2001), PREBAS (Minunno et al. 2019), BIOMASAR (Santoro et al. 2021) and Autochange (Häme et al. 2020) methods. All of these methods were implemented and the processing was conducted in the Forestry TEP (https://f-tep.com/) cloud processing platform. Input satellite datasets included Sentinel-1, Sentinel-2 and PALSAR-2 data, in most demonstrations using a combination of both optical and radar data. In addition, field plot measurements were used for model training and validation. The output products include (1) European wide Growing Stock Volume (GSV), Above Ground Biomass (AGB) and Below Ground Biomass (BGB) maps for 2020 and 2021, (2) European wide biomass clearance map 2020-2021, (3) Above Ground Biomass (AGB), Below Ground Biomass (BGB) and Stem Volume Increment (SVI) maps of Finland for 2017 and 2019 and (4) numerous local to provincial level maps of forest structure variable (diameter, basal area, height, conifer proportion etc), biomass and clearance in demonstration areas in Ireland, Romania, Spain and Peru. A visualization and downloading portal will be set up for viewing and accessing publicly available output products. References: Alt, H. (2001) The Nearest Neighbor. In: Alt, H. (eds) Computational Discrete Mathematics. Lecture Notes in Computer Science, vol 2122. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-45506-X_2 Häme, T., Stenberg, P., Andersson, K., Rauste, Y., Kennedy, P., Folving, S., & Sarkeala, J. 2001. AVHRR-based forest proportion map of the Pan-European area. Remote Sensing of Environment, 77(1), 76-91. Häme, T., Sirro, L., Kilpi, J., Seitsonen, L., Andersson, K. and Melkas, T. (2020) A hierarchical clustering method for land cover change detection and identification. Remote Sensing 12: 1751. Minunno, F., M. Peltoniemi, S. Härkönen, T. Kalliokoski, H. Mäkinen, and A. Mäkelä. 2019. A. Bayesian calibration of a carbon balance model PREBAS using data from permanent growth experiments and national forest inventory. Forest Ecology and Management 440: 208-257. Santoro, M., Cartus, O., Fransson, J.E.S., 2021. Integration of allometric equations in the water cloud model towards an improved retrieval of forest stem volume with L-band SAR data in Sweden. Remote Sensing of Environment 253, 112235.

Agricultural land plays a central role in the climate change issue through greenhouses emissions, but cropland can be a source or a sink of carbon, depending on management and crop species. A set of storage practices can thus contribute to the increase of carbon in the soil (e.g., straw restitution to the soil, cover crops …), and consequently, constitute a useful lever for reducing the increase of atmospheric CO2. Quantifying the carbon stored or released following the implementation of different agricultural practices is therefore a major challenge for the sustainable and rational management of agricultural land. In this context, the combined use of satellite imagery and agrometeorological modelling allows the estimation of key descriptors of surface conditions, variables that are necessary for quantifying the components of the carbon budget (e.g., biomass, yield, CO2 fluxes) over large scales (e.g., landscape, regions). Recently, such an approach has been implemented and validated on different crops (wheat and sunflower) using experimental setups gathering various data such as net CO2 flux measurements by eddy-covariance, destructive biomass sampling or yield collections. The model used, namely SAFY-CO2, was constrained by the LAI time series derived from optical images. In addition to the daily estimates of the different CO2 flux components (i.e., net ecosystem exchange, gross primary production, and ecosystem respiration), the model allows an estimation of biomass as well as grain yield. In this approach, the accuracy levels associated with the simulated variables of interest are conditioned by the assimilation of dense and regular observations of the vegetation status. The absence of observations of the surface under cloudy conditions can then constitute a limitation of the approach, especially during certain phenological stages where the vegetation shows strong changes. The combined use of optical and radar satellite images then provides an alternative, and can be a valuable advantage when these signals provide complementary information as demonstrated in the literature for several crops as wheat, sunflower, rapeseed. In this study, optical and radar data are assimilated into SAFY-CO2 model to estimate biomass and CO2 fluxes. Dataset is based on a network of plots (12 to 15 plots per year) regularly monitored during three successive years (2018, 2019, 2020) with phenological stages and biomass measurements, as well as CO2 fluxes measured on a plot equipped with an eddy-covariance setup in 2009, 2011, 2013 and 2018. This database allows to test different assimilation schemes (of optical and/or radar satellite data) and their associated impact on the modelling performances over contrasted climatic conditions.

Planet Labs is delivering a suite of satellite-based Planetary Variables to assist governments, NGOs and customers in informed decisions across a variety of sectors and industries, including sustainable agriculture, ecological management, and emergency response. All Planetary Variables, i.e., soil water content, land surface temperature, vegetation biomass proxy, forest and woodland carbon, have been identified by the World Meteorological Organization, the United Nations, and space agencies as Essential Climate Variables that critically contribute to the characterization of the Earth’s climate and environment. Planet's Biomass Proxy product offers a timely, accessible, and analysis-ready relative measure of the above-ground crop biomass daily at 10m spatial resolution. Based on Earth observations from Sentinel-1 and Sentinel-2 constellations, the Biomass Proxy is a cloud-free vegetation monitoring product for agriculture and pasture management at scales varying from the intra-field scale to national scales. The Biomass Proxy is currently used by agribusiness companies to characterize and monitor the different components of agricultural ecosystems including the early detection of crop growth anomalies, the optimization of agricultural practices and agrochemical input efficiency, or the scheduling of field activities such as crop harvesting, grass mowing and grazing practices. Therefore, the Biomass Proxy plays an increasingly crucial role in agricultural decision support systems. In this presentation, we will focus on recent key achievements including product evaluation, agricultural applications, and algorithm refinement in response to customer feedback. Understanding the context, objectives and constraints that Planet’s customers face is critical to design and develop the most appropriate information for decision making. Moreover, a key requirement for generating a product that will be used by a broad user community is accurate knowledge of the various sources of uncertainty that arise from the retrieval and the measurement. We will use several pilot case studies that are representative of various agricultural systems and climate, to demonstrate and quantify the ability of the biomass proxy to monitor in near real time the above ground plant fresh biomass of irrigated and rainfed corn and soybean fields in Nebraska, and to forecast yield early in the growing season of crops associated with a wide range of agricultural practices including conventional and organic certified treatments, regenerative agriculture and biodiversity, no-tillage and crop rotation. Using high-quality ground-based measurements, the biomass proxy was able to explain around 90% and 70% of the variance of the in-season crop fresh biomass and final crop yield, respectively, with a relative standard error lower than 15% for corn and wheat fields.

MDA has been providing subtle forest change services using RADARSAT-2 for the last seven years to assist clients in compliance monitoring of forest cover change resulting from natural causes, illegal logging or encroachment. Recently, efforts have been made to extend this service to quantify changes in forest biomass. In order to estimate forest biomass change using remote sensing data, a crucial step is to classify observations into different types of forests. High resolution optical data can be used to accurately classify different forest types; however, data can be affected by cloud cover which limits our ability to monitor large areas regularly. Unlike optical sensors, SAR sensors have the unique advantages of being able to acquire images day and night, and repeatedly capture the images in the same imaging geometry with the same beam mode, which is necessary for robust change detection. Combined with wide imaging swaths, this is particularly beneficial for the operational monitoring of biomass changes over large regions. One of the main challenges when utilizing single polarization SAR data is that different forest types can present similar radar signatures which limits our ability to separate different forest classes. In this paper, we present two MDA methods developed to classify different forest types: 1) texture-based and 2) histogram-based. The methods compute forest classification using different combinations of polarizations from SAR sensors which include RADARSAT2, Sentinel-1 and SAOCOM. We found that results from combining multiple polarizations provide better estimates than those from a single polarization. We discuss the advantages and disadvantages of each method and summarize their versatility in forest classification.

Forest monitoring is high on the scientific and political agenda. Global estimations of forest biomass and biomass change are needed as essential climate and ecosystem variables. New innovative space instruments especially designed for the task have been launched recently and more are expected in the near future. These space-based instruments require ground-based estimates for algorithm calibration and product validation. The Forest Observation System – FOS (http://forest-observation-system.net/) that has been developing since 2006 was recently merged with the GEO-TREES (https://geo-trees.org) initiative and became a data component of it. The Data.GEO-TREES is an international cooperation to maintain a global in-situ forest biomass database to support earth observation and to encourage investment in relevant field-based observations and science. The GEO-TREES aims to link the Remote Sensing (RS) community with ecological and forest inventory teams working on the ground for a common benefit. The benefit of GEO-TREES for the RS community is the partnering of the most established ecological research teams and networks that manage permanent forest plots; to overcome data sharing issues and introduce a standard biomass data flow from tree level measurement to the plot level aggregation served in the most suitable form for the RS community. Ecologists benefit from the GEO-TREES with improved access to global biomass information, data collecting protocols, gap identification and potential improved funding opportunities to address the known gaps and deficiencies in the data. GEO-TREES is an open initiative with other networks and teams most welcome to join. The online database provides open access for both metadata (e.g. who conducted the measurements and where) and actual data for a subset of plots where the authors have granted access. A minimum set of database values includes: principal investigator and institution, plot coordinates, canopy height and above ground biomass of trees. Plot size is 0.25 ha or large. Large plots are divided by sub-plots (0.125 or 0.25 ha) in order to observe the variability in height and biomass. The database will be essential for validating and calibrating satellite observations and various models. Comparison of plot biomass data with available global and regional maps (incl. GlobBiomass, CCI Biomass, NASA JPL, GEDI gridded, ICESat-2 boreal) shows wide range of uncertainties associated with biomass estimation. This study was partly supported by the European Space Agency via projects IFBN (4000114425/15/NL/FF/gp).

The objective of this paper is to propose a multi-baseline polarimetric synthetic aperture radar interferometry (PolInSAR) method for robust forest parameter retrieval and analyze its feasibility and efficiency under the observation configuration of ESA’s BIOMASS mission [1-2]. As the 7th Earth Explorer Mission within the frame of the ESA Earth Observation Program, the BIOMASS mission will carry, for the first time, a P-band polarimetric interferometric SAR payload in space, flying as repeat-pass acquisition, to accomplish the mapping of a full range of the world’s above-ground forest biomass, as well as support the national scale forest inventory and global carbon flux calculations. However, besides the temporal decorrelation caused by a 4-days repeat cycle, one of the biggest challenges for forest structure parameter retrieval is the highly reduced spatial resolution due to the limited bandwidth of 6 MHz. Such limitation leads to decreased interferometric coherences and polarimetric behavior in single interferometric pair. In addition, during the PolInSAR phase of BIOMASS mission observation mode was designed as a three-pass acquisition (i.e., three-image acquisition, combined as dual-baseline or three-baseline PolInSAR) and the limited number of available baselines will further reduce the performance and stability of traditional multi-baseline method [2]. Although several multi-baseline PolInSAR methods have been proposed and evaluated with high-resolution airborne SAR data during the past two decades, there is still strong demand for the development of robust and fast multi-baseline algorithms dedicatedly suitable for space-borne dataset characterized by limited bandwidth and limited number of available baselines. In this paper, a robust three-baseline PolInSAR method was developed by introducing rigorous multi-baseline coherence optimization. More specifically, the multi-baseline PolInSAR scattering response can be represented by a 3N×3N (in the case of N-image acquisition) coherency matrix. After rewriting and whitening the matrix, joint diagonalization was applied to all possible polarimetric interferometric cross-covariance matrices, and optimal Equal Scattering Mechanisms (ESM) coherence sets as well as correlation matrices with ESM structure can be computed [3-4]. Subsequently, these correlation matrices can be utilized to estimate the coherence line under the context of the random volume over ground (RVoG) model [5], and the optimal coherences were used as observables to invert forest parameters (forest height, extinction coefficient, ground to volume ratio) with a constraint of fixed sub-canopy ground height. In this sense, the ground phase is estimated by three baselines jointly and using the whole available observation information of multi-baseline PolInSAR instead of performing classical line-fitting separately, which should increase the stability of line parameters estimation in case of losing resolution and coherence. In addition, estimating PolInSAR optimal coherences using three-baseline jointly, instead of one after the other, is advantageous to capture and exploit the cross-correlations and intrinsic coherence between the three baselines, which is beneficial for mitigating the interferometric coherence loss and reducing the instability of polarimetric properties in case of low resolution. To analyze its performance on BIOMASS configuration, three-image BIOMASS simulation data (SLC azimuth resolution × range resolution: 12.5m×25m) were emulated using P-band full-pol SAR data acquired over the Paracou tropical forest site (France) during the TropiSAR 2009 airborne campaign [6]. By doing this, both the limitation of 6 MHz bandwidth and temporal decorrelation were taken into account for simulation. Based on the analysis of optimal coherences, the reduction of coherence was found to be mitigated compared with linear coherence estimation as well as other coherence optimization methods. The three-image inversion results show that, although the performance appears worse than the 125 MHz airborne data, the proposed algorithm yields an acceptable accuracy of forest height (Fig. 1). Based on the comparison with the LiDAR reference forest map, the root-mean-square error (RMSE) is 2.56m under the resolution of 200m×200m, representing 9.6% error of the mean height, which could satisfy the expected index of forest height retrieval accuracy for BIOMASS mission (≤20% error for 200m×200m resolution forest map). This study demonstrates that the limited bandwidth of BIOMASS spaceborne data has a certain impact on PolInSAR forest structure parameter retrieval, and the proposed method is still able to work acceptably under these circumstances and meet the requirements of the BIOMASS mission.

The National Aeronautics and Space Administration (NASA) and the European Space Agency (ESA) are both developing future L-band SAR missions to address key science questions and application needs relevant to solid Earth, ecosystems, cryosphere, and hydrology. The NASA-ISRO SAR (NISAR) mission is a dual-frequency L-/S-band SAR satellite scheduled for launch in late 2023 and recently identified as a pathfinder for the NASA Earth System Observatory. NISAR will acquire global dual-polarimetric L-band data with 20 MHz range bandwidth every 12 days (6 days ascending and descending), delivering unprecedented dense time-series at L-band and new Geocoded Single Look Complex products. The Radar Observation System for Europe at L-band (ROSE-L) is one of the ESA’s High-Priority Candidate Missions scheduled for launch after 2028 with the goal of augmenting the Copernicus constellation to address important information gaps and enhance existing Copernicus services and related applications. In the current design, ROSE-L is a two-spacecraft system that will operate in the Sentinel-1 orbit and be phased to achieve a repeat interval of 6 days. Both NISAR and ROSE-L are designed to make repeated observations from a narrow orbital tube in order to generate time-series with nominal zero interferometric baselines. While this design choice has several benefits, it cannot address the retrieval of vegetation vertical structure recommended by the 2017-2027 Decadal Survey for Earth Science and Applications [1], for which long cross-track interferometric baselines are required [2]. This paper analyzes the performance of tomographic mission concepts involving satellites flying in formation with NISAR or ROSE-L in order to augment the observation capabilities of these missions through multi-baseline measurements. In particular, receive-only co-fliers are attractive thanks to their simplified hardware architecture and to the ability to coherently combine their images without relying on a tight cooperation with the mothership SAR satellite. The performance analysis is carried out with emphasis on various tomographic algorithms (e.g., backprojection or HistTomo [3] algorithms) adopted to generate tomograms, and by varying the radar instrument and formation parameters in pre-defined distributed formations with one to six satellites with non-zero cross-track baselines. Three multi-static modes are considered for each scenario: SISO (single-inputs-single-output), SIMO (single-input-multiple-output), and MIMO (multiple-input-multiple-output). Performances are derived from closed-form equations where available as well as from point-target and distributed-target simulations using the Distributed Aperture Radar Tomographic Sensor (DARTS) [4] Trade Study Tool (TST) supported by real data. Using the DARTS TST provides the advantage to analyze the global system performance taking into account orbital, radar signal, and scene characteristics that would be too complex for compact and closed-form analytical models [5]. The flowchart of a sample performance evaluation ingests a realistic orbit file consisting of orbit parameters for six platforms matching the 12-day repat cycle of NISAR or ROSE-L. Additional inputs include the antenna patterns, the platform and radar parameters, and the grid definition to specify the geographic boundaries and the spatial resolution for the simulation. Gridded waveforms and canopy heights from the GEDI L2/L3 data set are also provided as an input to the simulation. The SAR signals are generated and processed using various tomographic algorithms from which the tomograms are constructed. The constructed tomograms are examined to compute the performance metrics such as resolution and integrated side-lobe ratios, which are then compared among the various scenarios and input profiles for final assessment. A task currently in progress is the conversion of tomograms into bio-physical quantities such as biomass, which will be then compared with existing and pre-ingested biomass maps derived from other sensors (e.g., GEDI). [1] A. Reigber, and A. Moreira, “First demonstration of airborne SAR tomography using multibaseline L-band data,” IEEE Transactions on Geoscience and Remote Sensing, 38(5), pp.2142-2152, 2000. [2] A. Donnellan, D. Harding, P. Lundgren, K. Wessels, A. Gardner, M. Simard, C. Parrish, C. Jones, Y. Lou, J. Stoker, J. Ranson, B. Osmanoglu, M. Lavalle, S. Luthcke, S. Saatchi, and R. Treuhaft, “Observing Earth’s Changing Surface Topography and Vegetation Structure: A Framework for the Decade,” NASA Surface Topography and Vegetation Incubation Study, Mar. 2021. [3] G. H. X. Shiroma M. Lavalle, "Digital Terrain, Surface, and Canopy Height Models From InSAR Backscatter-Height Histograms," in IEEE Transactions on Geoscience and Remote Sensing, vol. 58, no. 6, pp. 3754-3777, June 2020. [4] M. Lavalle, I. Seker, J. Ragan, E. Loria, R. Ahmed, B. Hawkins, S. Prager, D. Clark, R. M. Beauchamp, M. S. Haynes, P. Focardi, N. Chahat, M. Anderson, K. Matsuka, V. Capuano, and Soon-Jo, “Distributed Aperture Radar Tomographic Sensors (DARTS) to Map Surface Topography and Vegetation Structure,” in 2021 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), July 2021. [5] I. Seker, and M. Lavalle, “Tomographic performance of multi-static radar formations: Theory and simulations,” Remote Sensing, 13(4), p.737, 2021.

Tomographic SAR surveys require illumination under slightly different viewpoints to form a data-stack containing multiple SAR images of the same area. The data-stack is then focused via digital signal processing to produce a collection of voxels that represent the backscattered energy in three dimensions, thus allowing direct imaging of the interior of the illuminated media. Space agencies, in the first place, have increasingly invested in SAR tomography, funding both airborne and ground-based campaigns aimed at evaluating the potential of this technique and its feasibility in the context of Earth Observation [1-12]. Despite the many successful experimental campaigns, however, spaceborne tomography of natural scenarios is yet to come. This is largely due to the fact that that the vertical resolution provided by SAR tomography is inherently linked to the number of available viewpoints, which - for the case of a single satellite - corresponds to flying multiple passes over a given area. As different passes are usually collected at time lags on the order of days, the acquisitions are affected by temporal decorrelation from any temporal change on the scale of a fraction of a wavelength. For this reason, the one mission currently appointed to gather systematic tomographic acquisitions of forested areas is the forthcoming ESA Earth Explorer BIOMASS, for which temporal decorrelation is predicted to be successfully mitigated by the three-day revisit time and the long wavelength. It is then clear that the success of spaceborne TomoSAR is crucially linked to the possibility of flying formations of SAR sensors. Concrete signs of the practical possibility to fly SAR formations have appeared in recent years, since advances in electronics and antenna technologies have made SAR payload compatible with small satellites [13-15]. A measure of the interest raised by this new scenario is easily found in the recent literature [16]-[19], in the rising of commercial companies, and in new programs funded by space agencies to develop applications based on small satellites [20]. In this context, several works have appeared to propose new concepts based on formations of SARs operating in Multiple Input Multiple Output (MIMO) mode, such that each sensor acts both as transmitter (Tx) and receiver (Rx) [16-19]. Operating in MIMO mode allows for the creation of virtual sensors placed in the mid-point between the Tx and the Rx, resulting in a multiplication of viewpoints available for tomographic imaging. Still, the position of the resulting virtual sensors is critical, as high-quality tomographic imaging sets strict requirements about sensor spacing and overall array length. Accordingly, the concept of MIMO SAR Tomography is inherently linked to a careful design of the flight formation. Interestingly, the problem of finding the longest uniformly-spaced linear array is well known in literature on array theory [21], and algorithms to optimize the array layout. Assuming all sensors act as Tx and Rx, an array of N physical sensors can generate up to N(N+1)/2 virtual sensors. This value is to be considered as a theoretical upper bound, in that the possibility to find a solution is not guaranteed. Conversely, in [21] a simple algorithm is proposed to design the physical array to generate a uniform virtual array of length L = (N+1)*(N+3)/4-1 (to be rounded up or down for even values of N). In the context of this work our interest is only focused on the case of few sensors, as any of them corresponds to a satellite. Accordingly, we decided to solve the problem of optimal sensor deployment by running an exhaustive search. As a result, we get that a formation of 4,5,6 physical MIMO satellites could generate as many as 9,13,17 virtual satellites. Still, a fundamental problem needs to be solved in the context of SAR. Indeed, a tacit assumption in array literature is that the echoes associated with different transmissions can be perfectly separated at the Rx. To do that, transmission of orthogonal waveforms is generally assumed. In the case of SAR, however, achieving truly orthogonal waveforms is only possible when coupled to complex hardware solutions such as on-board digital beamforming [22], which would be difficult to marry with low-cost small satellite technology. Following the above discussion, in this work we analyze the implications of implementing MIMO SAR Tomography by assuming a simple Frequency Division Multiplexing (FDM) access scheme, where all satellites transmit simultaneously on different frequency bands and receive the echoes scattered by the Earth’s surface in all transmitted bands. In so-doing, a formation on N satellites produces N^2 distinct SAR images, with imaging performance that might largely vary depending on formation deployment and the bandwidth allocated to each transmission. As a consequence, obtaining high quality imaging using FDM MIMO TomoSAR requires the establishment of new criteria to optimize the formation, as well as new approaches to SAR focusing. Both topics will be studied both theoretically and practically, by studying fundamental performance bounds and addressing processing algorithms. Moreover, we will derive new criteria for designing the SAR formation and show the benefit of careful optimization. The work will be supported by results from numerical simulations. Preliminary results indicate the concrete possibility for a coherent combination of all N^2 images to obtain both 2D high-resolution SAR images and tomographic imaging with performance comparable to the best airborne and ground-based campaigns.

Synthetic Aperture Radar Tomography (TomoSAR) exploits the use of several co-registered SAR images, acquired with different tracks (later called TomoSAR stack), to synthesize an aperture in elevation. Spectral analysis methods recover the vertical backscattered Power Spectrum Pattern (PSP) allowing the 3D imaging of an illuminated scene [1]. The resulting tomograms present ambiguities located at a distance inversely proportional to the separation between baselines [2]. Therefore, the larger the TomoSAR stack, the better is the ambiguity rejection. In practical scenarios, the size of the TomoSAR stack is constrained to a revisit time, since temporal decorrelation problems may occur [2]. An additional limitation to the number of passes in the stack is the feasibility of individual missions to perform them. In order to ease this issue, we use a deep neural network to synthesize Single Look Complex (SLC) SAR images from an “artificial” baseline, i.e. a sensor path with a line-of-sight not acquired in a specific TomoSAR stack. Deep Learning methods use a cascade of nonlinear processing units to obtain features of a dataset as humans do, i.e., learn by example. This knowledge is later used to map an input into a desired output. A U-net encoder-decoder network architecture [3] is employed with five encoder blocks and five decoder blocks. The contractive path (encoders) half the number of features at each step, while the expansive path (decoders) double them. In our case, the network takes six of seven baselines as an input and estimates the remaining baseline. We envision a scenario where multiple TomoSAR stacks are required to cover a region of interest. The goal is to acquire a large stack only for a part of the scene. This data can subsequently be leveraged to train the network to synthesize a subset of the stack. For the remaining parts of the scene, only a smaller stack has to be acquired as it can be extended by SLCs synthesized by the network. As a proof of concept, we limit ourselves to a single tomographic stack that is divided into train and test regions by a split in azimuth. Experiments are performed using the 2015 UAVSAR mission from NASA/JPL over Munich. The training and testing subsets have a size of 14000x8000 pixels. Further experiments are done using FSAR data from DLR over Froschham in Germany. Preliminary results show the correct estimation of both the phase and the amplitude in the targeted SLC. The amplitude is evaluated with a scatter plot between the original and the estimated data. The phase is assessed by performing interferometry between adjacent baselines; in one case using both original SLCs and in the other using one original SLC and the artificial SLC. [1] A. Moreira, P. Prats-Iraola, M. Younis, G. Krieger, I. Hajnsek and K. P. Papathanassiou, "A tutorial on synthetic aperture radar," in IEEE Geoscience and Remote Sensing Magazine, vol. 1, no. 1, pp. 6-43, March 2013, doi: 10.1109/MGRS.2013.2248301. [2] G. D. Martín-del-Campo-Becerra, S. A. Serafín-García, A. Reigber and S. Ortega-Cisneros, “Parameter selection criteria for Tomo-SAR focusing,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 14, pp. 1580–1602, Jan. 2021. [3] O. Ronneberger, P. Fischer and T. Brox, "U-Net: Convolutional networks for biomedical image segmentation", MICCAI, vol. 9351, pp. 234-241, November 2015.

Tomographic Synthetic Aperture Radar (TomoSAR) provides a direct and model free approach to estimate microwave vertical reflectivity profile when a large(r) number of interferometric / tomographic acquisitions are available [1]. In a multi-frequency framework each frequency band requires a different set of tomographic flights with a suitable baseline distribution. However, while several SAR missions at different frequencies are today, or will be in next future, able to provide interferometric measurements, ESA’s BIOMASS mission is going to provide the tomographic data at P-band. [2]. The question that rises in this context is how far the tomographic reconstructed vertical reflectivity profiles at P-band can be used to support the inversion of interferometric measurements at different frequencies. It is known, however, that multi frequency acquisitions add additional information on the forest structure which is not visible at one frequency band only. This paper investigates the possibility of exploring Polarization Coherence Tomography (PCT) [3] to transfer tomographic reconstructed reflectivity profiles at P-band to other frequencies (e.g. at X- and L-band) when interferometric data at these frequencies are available. PCT, in contrast to TomoSAR, allows to reconstruct a vertical profile from interferometric measurements at only a few (even a single) spatial baselines. PCT considers a profile decomposed in a set of basis functions. As the choice of basis impacts directly the quality of the profile reconstruction, it is important to investigate different types of basis functions. Legendre polynomials [3] is a good choice for lower forest heights, as its higher order components are negligible at smaller heights. However, for taller forests, we propose to consider a different set of basis which is more adapted to the particular forest and certain frequency band. The basis derivation relies on the availability of some pre-calculated (preferably high vertical resolution) profiles and is derived from the eigenvectors of the corresponding profile covariance matrix [4][5]. The selected frequency specific basis, in our case P-band, is then used for inverting the PCT profiles at X- and L-bands. First results show some correlation of PCT inverted components between different bands and imply a possibility to establish a transfer function across them. AFRISAR 2016 campaign in Gabon by DLR’s airborne system FSAR [6] will be used for the concept demonstration. In light of the upcoming repetition of the campaign in April 2023, the application of the proposed multifrequency approach will also be considered to detect changes in forest structure. REFERENCES [1] A. Reigber und A. Moreira, „First demonstration of airborne SAR tomography using multibaseline L-band data,“ IEEE Transactions on Geoscience and Remote Sensing, Bd. 38(5), pp. pp. 2142-2152., 2000 [2] S. Quegan, The European Space Agency BIOMASS mission: Measuring forest above-ground biomass from space. Remote Sensing of Environment, 227, 44-60, 2019 [3] S. R. Cloude, „Polarization coherence tomography,“ Radio Science, vol. 41, no. 4, 2006. [4] R. Guliaev, V. Cazcarra-Bes, M. Pardini and K. Papathanassiou, “Forest Height Estima-tion by Means of TanDEM-X InSAR and Waveform Lidar Data,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 14, p. 3084-3094, Feb. 2021. [5] R. Guliaev, M. Pardini and K. Papathanassiou, "A Comparison of Function Bases for Polarization Coherence Tomography in Forest Scenarios." EUSAR 2021 [6] I. Hajnsek, M. Pardini, et al.,, "Technical assistance for the development of airborne SAR and geophysical measurements during the AfriSAR campaign, Final technical report, 2016

Changes of 3D synthetic aperture radar (SAR) reflectivity profiles in forest scenarios are effected by changes of both the 3D distribution of structural elements (induced by growth, management, logging, mortality, disturbance etc.) and their dielectric properties (induced by rainfall, droughts, daily water cycle, seasonality, etc.) [1]-[2]. Their parameterization at different spatial and temporal scales in terms of changing 3D reflectivity components and their unambiguous separation and interpretation are critical missing aspects. Time series of tomographic (TomoSAR) reconstructions would provide a way to image directly 3D reflectivity changes [3]-[4], but the limited vertical resolution and / or the imaging algorithm can introduce interpretation ambiguities. A different way to proceed is to consider the corresponding TomoSAR coherence matrices. Reflectivity changes can then be represented and separated in terms of subspace directions common to two TomoSAR coherence matrices at different times [5]. Experimental results suggest that the components with maximum reflectivity increase and decrease are enough to characterize the change between the respective profiles [5]. Not only, but each of them can be well modelled by means of a Dirac-delta profile, and their characterization requires only a height information. As a consequence, the reconstruction of the TomoSAR reflectivity change can be performed by means of a reduced number of interferometric acquisitions. Concepts of this kind become attractive for spaceborne implementations, in which suitable baseline distributions to realize significant height resolution and radiometric sensitivity repeatedly in time may not be achievable. In this context, the objective of this work is to understand how far (even single-baseline) polarimetric interferometric (Pol-InSAR) acquisitions at different times can be used to separate and characterize reflectivity change components provided an initial TomoSAR observation. Understanding and establishing this link would allow on the one hand the interpretation of changes in Pol-InSAR measurements, and on the other hand the direct exploitation of their (higher) sensitivity to reflectivity changes. First of all, differences between subspace-based change characterization methodologies of the kind reported in [6]-[8] after an appropriate extension to TomoSAR as in [5] including polarimetric information will be addressed. The application of the developed scheme to a TomoSAR / Pol-InSAR case is then further analysed. The analysis will be aimed at discussing primarily the effect of the vertical wavenumber(s) in the Pol-InSAR acquisition(s), the allowed vertical resolution of the reconstructed reflectivity changes, and the potential of the changing components to describe different types of changes (structural or dielectric). The analysis will be supported by real airborne data acquired by the DLR’s airborne platforms E/F-SAR over the Traunstein forest (South of Germany) in relevant campaigns with temporal separations from hours up to months with underlying daily, weekly and seasonal reflectivity changes. References: [1] A. Moreira, P. Prats-Iraola, M. Younis, G. Krieger, I. Hajnsek, K. P. Papathanassiou, “A Tutorial on Synthetic Aperture Radar," IEEE Geoscience and Remote Sensing Magazine, vol. 1, no. 1, pp. 6-43, March 2013. [2] M. Pardini, J. Armston, W. Qi, S. K. Lee, M. Tello, V. Cazcarra-Bes, C. Choi, K. Papathanassiou, R. Dubayah, L. Fatoyinbo, “Early Lessons on Combining Lidar and Multi-baseline SAR Measurements for Forest Structure Characterization,” Surveys in Geophysics, vol. 40, pp. 803-837, Jul. 2019. [3] V. Cazcarra-Bes, M. Tello-Alonso, R. Fischer, M. Heym, and K. Papathanassiou, “Monitoring of Forest Structure Dynamics by Means of L-Band SAR Tomography,” Remote Sensing, vol. 9, no. 12, p. 1229, Nov. 2017 [4] M. Pardini, K. P. Papathanassiou, F. Lombardini, “Impact of Dielectric Changes on L-Band 3-D SAR Reflectivity Profiles of Forest Volumes,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 12, pp. 7324-7337, Dec. 2018. [5] M. Pardini and K. P. Papathanassiou, "Separation and Characterization of 3D Reflectivity Changes in Forest Scenarios using SAR Tomography," EUSAR 2022; 14th European Conference on Synthetic Aperture Radar, Leipzig, Germany, 2022, pp. 1-6 [6] A. Alonso-González, C. López-Martínez, K. P. Papathanassiou and I. Hajnsek, "Polarimetric SAR Time Series Change Analysis Over Agricultural Areas," in IEEE Transactions on Geoscience and Remote Sensing, vol. 58, no. 10, pp. 7317-7330, Oct. 2020 [7] A. Marino, M. Nannini, “Signal Models for Changes in Polarimetric SAR Data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-18, 2022. [8] S. R. Cloude, "A Physical Approach to POLSAR Time Series Change Analysis," in IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1-4, 2022.

Tropical forests are particularly important. Although they only cover about 6% of Earth’s surface, they are home to approx. 50% of the world’s animal and plant species. Their trees store 50% more carbon than trees outside the tropics. At the same time, they are one of the most endangered ecosystems on Earth: about 6 million of hectares per year are felled for timber or cleared for farming. Compared to the other components of the carbon cycle (i.e. the ocean as a sink and the burning of fossil fuels as a source), the uncertainty in the land local carbon stocks and the carbon fluxes are particularly large. This is especially true for tropical forests, which remain poorly characterized compared to other ecosystems on the planet. More than 98% of the land use change flux should be due to tropical deforestation, which converts carbon stored as woody biomass (of which around 50% is biomass) into emissions. In the frame of the ESAs BIOMASS mission, selected in May 2013 as the 7th Earth Explorer mission to meet the pressing need for information on tropical carbon sinks and sources through estimates of forest height and biomass a first airborne campaign over tropical forest in Gabon was conducted. The campaign, called AFRISAR took place over four forest sites in Gabon where two acquisitions at different season where made, the first one was conducted by ONERA (SETHI system, July 2015) and the second one by DLR (F-SAR system, February 2016). After 7 years a second campaign, called GABONX, is conducted exactly over the same test site and two further one selected by the Gabonese partners (the space agency AGEOS and the Ministry of Forest) with the same configuration. This time the campaign will serve also the mission objectives of ROSE-L, with its six days repeat-pass cycles acquiring fully polarimetric and multi-baseline data sets in L- and P-band. The GABONX campaign is scheduled for April 2023 for data acquisition and will deliver information about the change occurring within the 7 year in terms of forest height retrieved in L- and P-band. The main purpose of this work is to present an overview of DLRs GABONX campaign, and to discuss first results about forest height and concepts of forest height changes Matteo Pardini, Marivi Tello, Victor Cazcarra-Bes, Konstantinos P. Papathanassiou, Irena Hajnsek, L- and P-Band 3-D SAR Reflectivity Profiles Versus Lidar Waveforms: The AfriSAR Case, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Year: 2018 | Volume: 11, Issue: 10 | Journal Article | Publisher: IEEE

The TomoSense experiment was conceived to provide the scientific community with unprecedented data to study the features of radar scattering from temperate forests, comprising tomographic and fully polarimetric SAR surveys at P-, L-, and C-band, acquired in mono- and bistatic mode by simultaneously flying two aircraft. The TomoSense dataset is complemented by a detailed forest census, Terrestrial Laser Scanning (TLS), and Airborne Lidar Scanning (ALS) products. All campaign activities were finally and successfully closed in fall 2021, resulting in an amount of over 1800 SAR images at different polarizations, frequency bands, and acquisition modes. A significant part of data processing activities was dedicated to interferometric and tomographic calibration of SAR data, necessary to finely estimate platform motion and achieve accurate tomographic focusing Calibration activities resulted in the generation of finely coregisted, phase calibrated, and ground steered complex SAR image stacks at all frequency bands, which were included in the final data delivery to ESA. Calibrated image stacks were afterwards processed to generate multi-frequency mono- and bi-static Tomographic cubes representing forest scattering in three dimensions, intended to serve as the basis for all subsequent scientific analyses and also included in the final data delivery. Scientific analyses were conducted by investigating the connection of Tomographic cubes to biophysical parameters. Essential to these activities was the availability of a large amount of biophysical information from independent measurements, including field-works, TLS, and ALS. Based on the large amount of results produced by the study, the following conclusions are drawn. P- and L-Band: Both P and L band are observed to provide sensitivity to the whole vegetation layer, in that both frequencies allow for a clear detection of terrain and forest canopies. Moreover, both frequencies are robust w.r.t. temporal decorrelation over few hours, resulting in the possibility to produce high-quality tomographic imaging from repeat-pass campaign data. Interestingly, bistatic data at L-Band are observed to contain weaker contributions from the terrain level than mono-static data. As a result, the Ground-to-Volume backscatter power ratio (G2V) is systematically larger for monostatic measurements by up to 4 dB, depending on polarization and local conditions (understory, topography). Forest height could be successfully estimated using repeat pass mono-static P- and L-Band data, repeat pass bi-static L-Band data, as well as using only simultaneous interferograms from each bistatic pass. The investigation of tomographic data is investigated in depth in the work by Bennet et al., to be presented at this workshop [1]. Interestingly, the use of normalized tomographic indicators like fractional volume intensity was observed to provide sensitivity to AGB as well, the best result being assessed in slightly over 20% on the aggregated forest class using bistatic L-Band data. C-Band: Tomographic analysis of C-Band data indicates that the residual coherence in repeat-pass interferograms is mostly determined by scattering from the ground level, whereas the signal from the forest canopy is nearly impossible to detect because of temporal decorrelation. Analysis of interferograms formed by mono- and bi-static data collected in the same flight reveals that tall forests decorrelate almost immediately because of wind gusts, thus confirming the results from the BorealScat experiment. However, tomographic processing allowed for exploiting vestigial coherence from the vegetation, resulting in the possibility to detect forest canopies in parts of the image and observe a good match w.r.t. Lidar height. Overall, results obtained at this site indicate that C-Band waves care capable of penetrating down to the ground level. This finding provides an element in support of the feasibility of C-Band tomography of temperate forests, clearly provided that acquisitions are taken at a temporal baseline of a few tens of milliseconds. The TomoSense data-base: one important product of the study consists in the creation of a data-base intended to serve as an important basis for studies on microwave scattering from forested areas in the context of future studies on Earth Observation missions. The data-base comprises complex SAR images and tomographic cubes at different levels of processing, as well as ALS-derived maps of forest height and AGB, forest census, and TLS profiles. Complex SAR images in the data-base are already finely coregistered, phase calibrated, and ground steered, in such a way as to enable future researchers to directly implemented any kind of interferometric or tomographic processing without having to deal with the subtleties of airborne SAR data. In addition to that, the data-base comprises tomographic cubes representing forest scattering in 3D both in Radar and geographical coordinates, which are intended for use by non-Radar experts. Overall, we emphasize that the amplitude of the TomoSense data-base is such that a number of questions had to be left unexplored within the time of the study, and in particular those concerning the joint use of different tomographic observables. For this reason, it is our opinion that the TomoSense data-base will represent a most valuable tool for student and researchers in the next years. [1] Patrik J. Bennet, Lars M. H. Ulander, Mauro Mariotti D’Alessandro, and Stefano Tebaldini: “TomoSAR Sensitivity to Temperate Forest Above-Ground Biomass at P- and L-band in the TomoSense ESA Campaign”, submitted at the PolInSAR-BIOMASS 2023 workshop.

This paper compares two techniques for estimating forest height and structure using synthetic aperture radar (SAR) data, namely SAR tomography (TomoSAR) and phase histogram (PH). Using multiple SAR images, TomoSAR can synthesize a new aperture in cross-range direction resulting in three-dimensional resolution capabilities. In this principle, TomoSAR allows for a direct imaging of the three-dimension (3D) backscattered power of the vegetation layer, from which biophysical parameters such as forest height and terrain topography can be extracted[1]–[4]. Differently from TomoSAR, the PH technique assigns each pixel to a specific height bin based on the phase-to-height factors, and provides backscattered power distribution about the forest structure by accumulating the amplitudes of pixels fall in the same bin within a given spatial window[5], [6]. In principle, PH technique can provide tomography-like products by exploiting a single interferometric pair. Yet, it has to be kept in mind that the phase/height relation is only valid for isolated scatterers, which is indeed an assumption that cannot be strictly verified in forested areas. In this paper, we compare the two techniques by applying both for the analysis of data from the ESA campaign TomoSense, flown in 2020/21 to support research on remote sensing of forested areas by tomographic SAR[7]. The dataset analyzed includes monostatic repeat-pass L-Band data acquired at the Eifel National Park, North-West Germany, by flying about to 30 trajectories in two headings. The SAR dataset is complemented by 3D structural canopy measurements made via terrestrial and airborne LiDARs, from which forest height and above-ground biomass (AGB) maps are produced. SAR tomography is implemented by Capon beamforming using up to 30 passes, which results in the possibility to image the forest at a fairly fine resolution (say consistently better than 5 m). The PH technique is implemented by forming a “super-interferogram”, obtained by juxtaposing different spatial regions from different interferograms such that the height of ambiguity (HoA) is consistently within the desired range, since flight trajectories were not stable enough to provide an acceptable HoA over the whole imaged scene. A first comparison between TomoSAR and the PH technique is shown in figure 1. In this case, the backscattered power profile obtained by tomography is better defined. Yet, The PH profile is observed to follow the same overall trend, and exhibits correlation with LiDAR forest height. Estimations of forest height are considered in figure 2. TomoSAR produced an average root mean square error (mRMSE) of 2.63 m with respect to LiDAR forest height, with stable performance within every forest height bin. The PH technique is observed to produce a larger RMS error of 4.72 m, which exhibits an increase as the forest grows higher. Although preliminary, these results seem to confirm that the products generated by SAR Tomography are inherently more accurate. This is of course not surprising, since: TomoSAR uses a much larger data volume than the PH technique. Moreover, the physical model connecting phase and height is strictly only valid for isolated targets, which is not exactly the case of forested areas. Still, it must be noted that the validity of this model is expected to be tightly linked wavelength and spatial resolution. Accordingly, as a new method to obtain the vertical structure of the forest through a few of interferograms, the PH technique is indeed an alternative method for providing forest height products that can be considered in future spaceborne satellite missions. For future work proposals, we plan further research to investigate the PH technique using different wavelength and studying the impact of spatial resolution.

Introduction The new ESA TomoSense campaign consists of multi-frequency (P, L, and C) airborne tomographic SAR acquisitions over Kermeter forest in northern Germany. Particularly, the L- and C-band SAR surveys were acquired via flying two separated aircrafts operated in a bistatic interferometric configuration for each pass, where one antenna transmits radar signal towards illuminated targets while the echoes are simultaneously received by two antennas with certain spatial separation[1]. Tomographic imaging was enabled by flying repeatedly at different heights. Such an observing mode represents a promising candidate for future low-frequency tomographic spaceborne missions to investigate forest scenarios, since a simultaneous acquisition of an identical scene allows to avoid fast temporal decorrelation [2] for forest scenario and mitigate atmospheric propagation artifacts rooted in repeat-pass SAR surveys. Thus, the L-band TomoSense data could motivate a series of studies concerning advanced SAR technologies, and support the scientific applications [3] for low-frequency spaceborne missions. In any case, a fundamental requirement is the accurate measurement of sensor-to-target distance within the synthetic apertures along flight and/or baseline directions. In TomoSense dataset, two key sources of errors pose the major challenges: 1. The uncertainties in navigational data provided along with SLC products: the accuracy of modern navigational data is still inadequate for high-quality InSAR and TomoSAR in repeat-pass manner concerning the varying position along the flight direction [4-6], and the location of one flight line and/or platform with respect to another [7-8]. 2. A potential presence of clock-drift perturbations [1,9,10] as no dedicated communication link [11,12] between two separated platforms was employed for time synchronization in the acquisition of TomoSense L data. Above errors give rise to phase errors and geometrical mismatches in the SAR interferograms, and eventually result in degraded tomographic performances ranging from spurious sidelobes to totally defocusing. A dedicated calibration approach is therefore developed to compensate the disturbances using natural scatterers and to obtain clean tomographic reconstructions. Methodology Our proposed calibration approach adopts three steps to compensate the potential clock drift, residual motion errors of dual platforms varying with respect to flight direction, and positioning errors of dual platforms with respect to baseline direction: The first calibration step is carried out based on multi-squint InSAR (MSInSAR) for each single-pass bistatic interferometric pair: the clock errors in monostatic SLC images have been compensated as the same USO is used for modulation and demodulation. This offers an opportunity to capture and compensate the clock drift in the bistatic SLC image with respect to the reference monostatic image via MSInSAR. It is noted the signal of MSInSAR also consists of the component related to residual baseline error between two platforms, which can be separately inverted using a different spatially-varying signal model[1]. In this step, only bistatic data are corrected. The second step is to retrieve (N-1) passes of time-varying trajectory corrections for both platforms out of all coherent interferometric pairs driven by small-baseline: a traditional way of doing so is to calibrate each pass with respect to a reference pass one pair by one pair[13], leading to a suboptimal performance in this tomographic survey due to the coherence losses in the pairs with large geometrical separation or long temporal gap. A joint calibration approach is therefore proposed to retrieve (N-1) passes of corrections for bi-platform motion errors from M+K coherent repeat-pass pairs, where (N-1)

Started in July 2018 near the Paracou research station in French Guiana, the TropiScat-2 experiment has been set-up in order to achieve two main objectives. Firstly, to extend the time series of P-band tomographic data previously acquired by the AfriScat and TropiScat experiments (respectively in Ghana from 2014 to 2017, and in French Guiana from 2010 to 2013), and secondly to explore the synergies between P-band and other measurements, with a specific focus on the mitigation of temporal changes which impacting the retrieval of forest height of Above Ground Biomass (AGB). With less than 10 percent of missing or non exploitable data (mainly due to electrical failure or maintenance defaults during the covid periods), the first objective has been successfully achieved, approaching soon with its fifth anniversary the duration planned for the BIOMASS mission, providing thereby an unique database to assess the temporal changes for this case of preserved tropical dense forest (cf.[1,2]) Regarding the second objective, we mainly focus on synergies between observations at P-band and other radar frequencies potentially (quasi-)concomitant with BIOMASS acquisitions, such as C and L bands through the current or future Sentinel-1 and NiSAR missions, characterized both by unprecedented revisits (less than 12 days). To better analyse the effects of acquisition time and geometry, L-band and C-band measurements acquired every 15 min (as for the P-band sequences) are also part of the new capabilities brought by TropiScat-2, considering at C-band a specific new array of 12 horn antenna. Thanks to the high revisit coherences, we were able to highlight and quantify the link between temporal decorrelation and evapotranspiration, particularly pronounced at C-band (cf.[3,4]). Specific patterns of coherence decrease could be also clearly distinguished between dry and rainy seasons, making possible the prediction of P-band decorrelation from L-band zero-baseline coherences. Although diurnal and seasonal patterns could be also characterized through the backscattering coefficients at P, L and C bands, a functional link between them is not obvious to generalize due to many periods without correlation, hence the need of further analysis and the reconfiguration of the C-band antenna array in July 2021, in order to benefit from tomographic imaging capabilities (VV and VH polarisations) which will contribute to better assess the main components of the C-band backscatter. More recently, a L-band radiometer (@TerraTech,cf.[5]) has been also installed thanks to ESA fundings, providing the opportunity to assess the potential future synergies between BIOMASS and spaceborne radiometer such as SMOS or SMAP, especially to better dissociate the interlinked effects of dry and fresh biomass on backscatter and emissivity. Although the instrument calibration is still pending, we could highlight the impacts of seasonal effects on the diurnal amplitudes of V and H brightness temperatures, also estimated by infrared measurements provided by a thermal imaging camera (8-14µm). Furthermore, the scene characterization gave rise to several research work, especially for the modeling of forest structure thanks to terrestrial and drone based laser measurements (cf.[6]), the modeling of tree and branches motions thanks to in situ accelerometer and the modeling of dielectric permittivity changes thanks to the use of a non-intrusive system (referred to as Wood Penetrating Radar,cf.[7]). Such measurements related to geometrical or dielectric dynamic structure, as well as meteo and fluxes (esp. CO2 and H20) data available from the Guyaflux tower (eddy covariance system,cf.[8]), are not only essential for data analysis but also for the parameterization of EM models, and the dynamic spatialization of these results through simulators like MIPERS-4D (cf.[9]). As a core topic for an optimized use of the future BIOMASS data, on-going work related to this dynamic spatialization will be also presented (cf.[9]), justifying also our strong motivations to pursue the TropiScat-2 experiment to integrate a wider range of variability in the observing conditions, especially after the last two years (2021-2022) of precipitation records. [1] S. El Idrissi Essebtey, L. Villard, P. Borderies, T. Koleck, B. Burban and T. Le Toan, "Long-Term Trends of P-Band Temporal Decorrelation Over a Tropical Dense Forest-Experimental Results for the BIOMASS Mission," in IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-15, 2022, Art no. 5102415. [2] L. Villard, P-A Bou, S. El-Idrissi, T. Koleck, L. Ferro-Famil, S. Tebaldini & T. Le Toan, "Insights of Tower-based experiments regarding current and prospective acquisition scenarios for the BIOMASS mission", ESA POLinSAR Symposium, 26-30 April 2021. [3] S. El Idrissi Essebtey, L Villard, P. Borderies, T. Koleck, J.P. Monvoisin, B. Burban and T. Le Toan, "Temporal Decorrelation of Tropical Dense Forest at C-Band: First Insights From the TropiScat-2 Experiment" in IEEE GRSL, vol. 17, no. 6, pp. 928-932, June 2020. [4] S. E. I. Essebtey, L. Villard, P. Borderies, T. Koleck, B. Burban and T. L. Toan, "Comparative Study Of Temporal Decorrelation At P, L And C-Bands: First Insights From The Tropiscat-2 Experiment," 2020 Mediterranean and Middle-East Geoscience and Remote Sensing Symposium (M2GARSS) , Tunis, Tunisia, 2020, pp. 246-249. [5] Houtz, Derek, Reza Naderpour, and Mike Schwank. 2020. "Portable L-Band Radiometer (PoLRa): Design and Characterization" Remote Sensing 12, no. 17: 2780. [6] G.Vincent, Antin C, Laurans M, Heurtebize J, Grau E, Durrieu S, Dauzat J. "Exploring LAI seasonality at multiple spatial scales in a tropical wet forest using high frequency repeated lidar overflights", accepted communication to ESA to Living Planet Symposium, 23-27 May 2022, Bonn. [7] F. Demontoux, L. Villard, JP Wigneron, T. Koleck, A. Mialon, T. Le Toan, Y. Kerr “Toward P and L band permittivity models of wood derived from non-intrusive in-situ measurements.” Oral communication at ESA POLinSAR/Biomass Symposium, 26-30 avril 2021. [8] https://paracou.cirad.fr/website/experimental-design/guyaflux-tower [9] M. A. Tanase, L. Villard,et al "Synthetic aperture radar sensitivity to forest changes: A simulations-based study for the Romanian forests", Science of The Total Environment, Volume 689, 2019, Pages 1104-1114, ISSN 0048-9697. [10] L. Villard, S. El Idrissi, C. Girod, T. Koleck, L. Ferro-Famil, B. Burban and T. Le Toan, "Scaling Effects regarding the Applicability of Tower-based Experiments Results relating Microwave Backscatter to Meteorological Observations : study case of TropiScat-2 over a tropical dense forest", accepted communication for the ESA Living Planet Symposium, 2022.

Although seasonal snow cover is widely recognized as a critical water resource, accurate SWE assessment is still challenging when considered at an operational level, especially in regions characterized by complex topography [1]. Current retrieval methods are either based on SAR polarimetry of differential interferometry (DInSAR). The former rely on specific assumptions to model the snowpack, [1], [2], [3],[4], [5], which can turn out to be problematic in heterogeneous areas. For DInSAR the resulting accuracy is critically dependent on the capability to compensate for topography and tropospheric delays at local scale, which can be quite a challenging task in mountain regions [9]. In addition, the amount of SWE at a given date is calculated by integrating differential measurements, which might introduce propagation errors. Direct measurement of absolute SWE at a given date is enabled by SAR Tomography (TomoSAR), which leverages multiple across-track baselines to provide direct imaging of the snowpack. Experimental works on the use of tomography demonstrated that X- and Ku-Band waves provide sensitivity to scattering from both the air/snow and snow/terrain interfaces in dry snow, as well as the possibility to image the multi-layered structure of the snowpack [10], [11], [12]. On this ground, the principle enabling SWE retrieval from tomographic observations can be understood in simple terms by studying the position where a scatterer within the snowpack appears when tomographic processing is carried out by assuming propagation in vacuum. This methodology was successfully applied in [12], demonstrating not only the retrieval of bulk density, but also the possibility to study snow density within each detected layer. Within this paper, we aim at investigating the extent to within which the result in [12] can be reproduced by using a formation of SAR satellites [13]. We will proceed by running a sensitivity analysis to test candidate spaceborne configurations. We will distinguish between two main models of dry snow. The layered snow model describes the situation observed in [10], [11], [12], where both the air/snow and snow/terrain interfaces can be detected and their heights assessed. By transparent snow, we refer to the DInSAR model discussed above, where the effect of the snow is limited to an increase of the optical path. Consistently with published literature [1-12], we expect the layered and transparent snow models to be verified at high (X- and Ku-Band) and low (L-Band) frequencies, respectively. Still, we do not a-priori exclude that the transparent snow model can apply at high frequency [8]. We assume that wet snow does not allow for wave penetration [11], and therefore we will not consider that case. In the transparent snow case, snow depth can no longer be measured, due to the assumed absence of detectable scattering from the snowpack. Yet, it can be showed that the difference is approximately proportional to density, allowing the inversion in both cases. Preliminary results indicate that accurate SWE retrieval is possible at the condition that apparent heights are measured to within an accuracy of few centimeters. Despite these difficulties, our preliminary conclusion is that the presented concept has a potential for accurate and fine scale SWE retrieval, which would help fill current Earth Observation gaps. REFERENCES [1]. Tsang. L, et al. Global monitoring of snow water equivalent using high-frequency radar remote sensing, The Cryosphere, 16, 3531–3573, 2022, https://doi.org/10.5194/tc-16-3531-2022 [2]. H. Rott et al., "COREH2O: High-resolution X/Ku-band radar imaging of cold land processes," 2013 IEEE International Geoscience and Remote Sensing Symposium - IGARSS, 2013, pp. 3479-3482, doi: 10.1109/IGARSS.2013.6723578. [3]. A. Patil, G. Singh, C. Rüdiger, S. Mohanty, S. Kumar and Snehmani, "A Novel Approach for the Snow Water Equivalent Retrieval Using X-Band Polarimetric Synthetic Aperture Radar Data," in IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 5, pp. 3753-3763, May 2021, doi: 10.1109/TGRS.2020.3016527. [4]. G. Singh et al., "Snowpack Density Retrieval Using Fully Polarimetric TerraSAR-X Data in the Himalayas," in IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 11, pp. 6320-6329, Nov. 2017, doi: 10.1109/TGRS.2017.2725979. [5]. S. Leinss, G. Parrella and I. Hajnsek, "Snow Height Determination by Polarimetric Phase Differences in X-Band SAR Data," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 7, no. 9, pp. 3794-3810, Sept. 2014, doi: 10.1109/JSTARS.2014.2323199. [6]. T. Guneriussen, K. A. Hogda, H. Johnsen and I. Lauknes, "InSAR for estimation of changes in snow water equivalent of dry snow," in IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 10, pp. 2101-2108, Oct. 2001, doi: 10.1109/36.957273. [7]. Rott, Helmut & Nagler, Thomas & Scheiber, Rolf. (2003). Snow mass retrieval by means of SAR interferometry., Proceedings of the FRINGE 2003 Workshop (ESA SP-550). 1-5 December 2003, ESA/ESRIN, Frascati, Italy. Editor: H. Lacoste. Published on CDROM., id.29 [8]. S. Leinss, A. Wiesmann, J. Lemmetyinen and I. Hajnsek, "Snow Water Equivalent of Dry Snow Measured by Differential Interferometry," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 8, no. 8, pp. 3773-3790, Aug. 2015, doi: 10.1109/JSTARS.2015.2432031. [9]. Tarricone, J., Webb, R. W., Marshall, H.-P., Nolin, A. W., and Meyer, F. J.: Estimating snow accumulation and ablation with L-band InSAR, The Cryosphere Discuss. [preprint], https://doi.org/10.5194/tc-2022-224, in review, 2022 . [10]. S. Tebaldini and L. Ferro-Famil, "High resolution three-dimensional imaging of a snowpack from ground-based sar data acquired at X and Ku Band," 2013 IEEE International Geoscience and Remote Sensing Symposium - IGARSS, 2013, pp. 77-80, doi: 10.1109/IGARSS.2013.6721096 [11]. O. Frey, C. L. Werner, R. Caduff and A. Wiesmann, "A time series of SAR tomographic profiles of a snowpack," Proceedings of EUSAR 2016: 11th European Conference on Synthetic Aperture Radar, 2016, pp. 1-5 [12]. Rekioua et al., Snowpack permittivity profile retrieval from tomographic SAR data, Comptes Rendus Physique, 2017 [13]. C. Angelone, V. Mancini, M.L. Tampellini, L. Maioli, G. Montano, D. Giudici, P. Guccione, F. Gerace, P. Gabellini, S. Falzini, A. Monti Guarnieri, A. Fedele, F. Tataranni, A. Montuori, S. Natalucci, “SATURN – A Synthetic Aperture Radar CubeSats Swarm Mission for Earth Observation”, International Astonautical Congess 2022

Snow water equivalent (SWE) is an important physical parameter in the hydrological cycle, thus identified by the IPCC AR6 for studying climate science [1]. Many airborne campaigns and field work (e.g. NASA’s SnowEx and Finnish NoSREx) have been carried out with international satellite mission concepts proposed (e.g. ESA’s CoReH2O and Canada’s TSMM) to monitor a variety of snow characteristics [2] by using active and passive microwave payloads. Microwave radar or SAR signals can penetrate snowpack with day/night and all-weather-observing capability. The backscatter power of SAR signals at various frequencies (L/C/X/Ku-band), polarizations and incidence angles has been widely used with sensitivities to snow parameters: grain size, density and depth [2, 3, 4]. However, it has been demonstrated that the differential phase of a pair of SAR signals or SAR interferometry (InSAR) is related to SWE change with a sub-wavelength sensitivity (e.g., cm at L-band and mm at C-band) [5, 6, 7, 8]. The theoretical foundation for SWE retrieval using repeat-pass InSAR phase measurement was first established by Guneriussen et al. [5]. It was then applied to C-band spaceborne repeat-pass InSAR dataset from ERS with a short temporal baseline of 3 days over Austrian Alps [6] and North Slope of Alaska [7] and with a long temporal baseline of 35-175 days over Tianshan Mountains [8], as well as higher frequency (X/Ku-band) dense time series from a ground-based radar [9]. So far, the only low frequency (L-band) demonstration of this technique used airborne InSAR data: 4-month pair acquired by DLR’s E-SAR [6] as well as ∼12-day pair by NASA/JPL’s UAVSAR [10]. However, there is a lack of spaceborne low-frequency (P/L-band) demonstration of this technique for large temporal baseline of few months, which captures the whole season of snow accumulation with deeper penetration through snowpack. For short temporal baselines of few days, phase unwrapping is straightforward due to scene-wide high coherence [5, 6, 7]. However, for large temporal baselines of few months, previous phase unwrapping approaches become problematic due to strong temporal decorrelation, where large connected regions with very high (e.g. > 0.5) coherence can be unwrapped but a large amount of confined high-coherence regions surrounded by scattered low-coherence areas cannot be accurately unwrapped and have to be masked out [8]. In this work, we apply this approach and demonstrate a novel processing routine with four components (two-step phase unwrapping [11], phase noise reduction, atmosphere correction, and SWE inversion) for spaceborne repeat-pass InSAR data (e.g. NASA’s NISAR, China’s Lutan-1 and ESA’s ROSE-L) with large temporal baselines. The approach is validated using JAXA’s L-band ALOS-2 InSAR data with a temporal baseline of 4 months capturing the entire phase of snow accumulation in the US state of Colorado, as well as another InSAR pair from ESA’s C-band Sentinel-1 with a time span of 24 days capturing the majority of the snow accumulation. The results will be further validated by comparing against the in-situ snow measurements available from NASA’s SnowEx [12] and USDA’s SNOTEL [2] over the mountainous areas in Colorado, US [13]. ∗This work was financially supported by the Ministry of Science and Technology through the National Key R&D Program of China under grant number 2022YFB3903300 and grant number 2022YFB3903301. The ALOS-2 data was provided by JAXA EORC through Yang Lei’s participation as PI in the 3rd Research Announcement on the Earth Observations (RO-RA3).

The Snow Water Equivalent (SWE) is a measure of the amount of water that is stored within a snowpack and is thus an essential parameter for hydrological and climate models. Accurate SWE estimates have a particular importance in regions where people rely on snowmelt for their fresh water supply or power generation. Precise estimates of SWE can be obtained from in-situ measurements. However, those are only carried out on limited number of test sites and have therefore a smaller spatial coverage than remotely sensed data. Synthetic Aperture Radar (SAR) has the ability to monitor large areas with a high spatial resolution. In addition, microwaves can penetrate into dry snow and are therefore sensitive to snow parameters such as snow depth, density, anisotropy and SWE. Repeat pass SAR interferometry (DInSAR) can be used to retrieve the SWE change between two acquisitions [1], [2]. Due to the fact that snow has a different dielectric constant than air, radar waves are refracted in the snow pack. The difference in the optical path length, which results from a SWE change between the two measurements, induces an interferometric phase difference that can be related to SWE. One limitation of this method is the 2π interval of the interferometric phase. This results into the fact that only a limited range of SWE changes can be retrieved before phase wrapping occurs. In addition, polarimetric measurements can also give insights into snow properties and can be used to invert the snow depth from the Copolar Phase Difference (CPD) between the VV and HH channel because of the anisotropic structure of snow. When assuming a snow density and anisotropy of the snow pack, the CPD can be related to the fresh snow depth [3] and help to predict the phase wraps of the interferometric phase. The objective of this study is to use interferometric and polarimetric SAR measurements to estimate SWE. For that, dual polarized TanDEM-X data in the VV and HH channel over a test-site in Svalbard are utilized. Due to the short wavelength of X-band, phase wraps of the interferometric phase are one of the main limitations for the DInSAR SWE retrieval. By using the snow depth information retrieved from the CPD. In almost all cases the interferometric phase wraps predicted by the polarimetric CPD are in agreement with phase wraps predicted from in-situ measurements. However, due to the fact that the temporal baseline is 11 days, snow microstructure changes between the DInSAR measurements are affecting the CPD and limiting the performance. The changes in the snow anisotropy, as measured by the CPD, also have an influence on the interferometric phase for the VV and HH channel. Therefore, the aim is to exploit the polarimetric and interferometric measurements simultaneously using Polarimetric Interferometric SAR (PolInSAR) techniques. For that purpose, the multitemporal coherence region [4] will be analysed for the timeseries and its information content will be compared to the previously described approach of treating the single-pol DInSAR phase and the polarimetric CPD phase separately. The validation with in-situ snow measurements will then show the potential benefit of PolInSAR techniques for a SWE retrieval. [1] T. Guneriussen, K. A. Hogda, H. Johnsen, und I. Lauknes, „InSAR for estimation of changes in snow water equivalent of dry snow“, IEEE Trans. Geosci. Remote Sens., Bd. 39, Nr. 10, Art. Nr. 10, Okt. 2001, doi: 10.1109/36.957273. [2] S. Leinss, A. Wiesmann, J. Lemmetyinen, und I. Hajnsek, „Snow Water Equivalent of Dry Snow Measured by Differential Interferometry“, IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., Bd. 8, Nr. 8, Art. Nr. 8, Aug. 2015, doi: 10.1109/JSTARS.2015.2432031. [3] S. Leinss, H. Loewe, M. Proksch, J. Lemmetyinen, A. Wiesmann, und I. Hajnsek, „Anisotropy of seasonal snow measured by polarimetric phase differences in radar time series“, 2015. [4] J. Ni, C. Lopez-Martinez, Z. Hu, und F. Zhang, „Multitemporal SAR and Polarimetric SAR Optimization and Classification: Reinterpreting Temporal Coherence“, IEEE Trans. Geosci. Remote Sens., Bd. 60, S. 1–17, 2022, doi: 10.1109/TGRS.2022.3214097.

Radar polarimetry is a popular tool for observation of natural environments, including snow and ice-covered environments, as polarimetric observations can yield information about the observed targets' physical properties. Most polarimetric observations are carried out in the monostatic mode, i.e. a configuration where the transmitter and the receiver are co-located. However, bistatic polarimetric measurements (where the transmitter and the receiver are spatially separated) are also of scientific interest, as they can provide access to a larger number of polarimetric scattering matrix observables [1,2]. This is due to the reciprocity principle, which dictates that the scattering matrix elements of the cross-polarized polarimetric channels are equal in the monostatic configuration (HV=VH). This principle brings the total number of polarimetric observables from 7 (in the bistatic case) down to 5 (in the monostatic case). One of these two additional bistatic observables is the phase difference between the two cross-polarized channels (XPD, HV-VH). It is always equal to zero in the monostatic case, however in the bistatic case it can acquire a non-zero value, and could potentially yield information about the medium structure, similarly as the co-polarized phase difference (CPD, HH-VV) does in the monostatic case. It is thus of interest to explore the bistatic scattering behaviour of natural media to investigate whether the reciprocity principle remains valid, and if not, to investigate the relation of the additional bistatic polarimetric parameters to the ones observed in the monostatic mode. Currently, the availability of full-polarimetric bistatic datasets is limited. The only currently operating bistatic spaceborne radar system for Earth observation is TanDEM-X, which operates with a small bistatic angle of less than 1 degree. While observations under this bistatic angle can already yield interesting observables [3], the large-angle bistatic parameter space remains relatively unexplored. Experimental campaigns focusing on large bistatic angle values are usually carried out with airborne or ground-based sensors [4,5]. KAPRI (Ku-band Advanced Polarimetric Radar Interferometer) [6,7] is a ground-based full-polarimetric real-aperture radar system capable of performing simultaneous monostatic and bistatic acquisitions with a large bistatic angle (up to 60 degrees), with kilometer-scale spatial coverage and revisit time of several minutes. In August 2021 and March 2022 we carried out two observation campaigns with KAPRI, observing the Jungfraufirn area of the Great Aletsch Glacier [8]. The observations revealed large differences in polarimetric properties of the observed snow cover between the two seasons, and between the monostatic and bistatic observations. In this contribution, we will present the results of full-polarimetric observations of the snow cover at Ku-band, and an intercomparison of the acquired monostatic and bistatic datasets. We will present an analysis of both the spatial and the temporal variation of the observed polarimetric parameters, and discuss the connection of these parameters' values to the medium properties. Particular attention will be given to polarization phase differences, and the relation between their values in the two acquisition geometries. The CPD exhibits relatively smooth spatial variation in summer when penetration depths are assumed to be short due to liquid water content, while it exhibits rapid spatial variation and phase wrapping in winter when penetration depths are likely longer. The XPD exhibited a predicted zero value in the monostatic dataset, however in the bistatic dataset a non-zero value was observed in both summer and in winter. Possible interpretations of these observed values of the polarization phase differences will be discussed. REFERENCES 1] Etienne Everaere, Polarimetry in bistatic configuration for ultra high frequency radar measurements on forest environment, Ph.D. thesis, Ecole Polytechnique, 2015. [2] Jong-Sen Lee and Eric Pottier, Polarimetric radar imaging: from basics to applications, CRC press, 2009. [3] Marcel Stefko, Silvan Leinss, Othmar Frey, and Irena Hajnsek, “Coherent backscatter enhancement in bistatic ku- and x-band radar observations of dry snow,” Cryosphere, vol. 16, pp. 2859–2879, 2022. [4] Pascale Dubois-Fernandez, Hubert Cantalloube, B Vaizan, Gerhard Krieger, Ralf Horn, Michael Wendler, and V Giroux, “Onera-dlr bistatic sar campaign: Planning, data acquisition, and first analysis of bistatic scattering behavior of natural and urban targets,” Radar, Sonar and Navigation, IEE Proceedings -, vol. 153, pp. 214–223, 1 2006. [5] Karlus A C De Macedo, Gerard Masalias, Hugo Keryhuel, and Christian Trampuz, “Airborne stereo formation campaign in support of the harmony mission demonstration,” in 7th Workshop on RF and Microwave Systems, Instruments & Sub-systems + 5th Ka-band Workshop, 2022. [6] Simone Baffelli, Othmar Frey, Charles Werner, and Irena Hajnsek, “Polarimetric calibration of the ku-band advanced polarimetric radar interferometer,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, pp. 2295–2311, 4 2018. [7] Marcel Stefko, Othmar Frey, Charles Werner, and Irena Hajnsek, “Calibration and operation of a bistatic real-aperture polarimetric-interferometric ku-band radar,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–19, 2022. [8] Marcel Stefko, Othmar Frey, and Irena Hajnsek, “Snow characterization at Ku-band with a bistatic polarimetric ground-based radar,” in IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium. 7 2022, pp. 4256–4259, IEEE.

Polarimetric and (multi-baseline) interferometric SAR techniques are promising tools to investigate the subsurface properties of glaciers and ice sheets, due to the signal penetration of up to several tens of meters into dry snow, firn, and ice. Two different lines of research were addressed in recent years. The first is based on PolSAR, which provides not only information about the scattering mechanisms, but also has the uniqueness of being sensitive to anisotropic signal propagation in non-scattering layers of snow and firn. The second line of research is related to the use of Pol-InSAR and TomoSAR to retrieve the 3D location of scatterers within the subsurface. So far, the different SAR techniques were mainly assessed separately. In the field of PolSAR modeling efforts have been dedicated to establish a link between the co-polarization HH-VV phase difference (CPD) and the structural properties of firn [1]. CPDs have then been interpreted as the result of propagation effects due to the dielectric anisotropy of the firn volume. This modeling approach establishes a relationship between the measured CPD and firn density, firn anisotropy and the vertical backscattering distribution in the subsurface of the glacier or ice sheet. By assuming bulk values for density and anisotropy and employing a constant signal extinction for the vertical backscattering function, i.e. a uniform volume, a first attempt to retrieve firn thickness from PolSAR data was presented [1]. In the fields of Pol-InSAR and TomoSAR for the investigation of the subsurface scattering structure of glaciers and ice sheets, recent studies were concerned with the estimation of the vertical backscatter distribution, either model-based or through tomographic imaging techniques. The complexity of (Pol-)InSAR models for the retrieval of subsurface structure information is mainly limited by the available observation space. Thus, constant signal extinction volumes [2], with additional Dirac deltas to represent refrozen melt layers [3] and variable extinction volumes [4] have been modelled and used to retrieve information about the subsurface. With TomoSAR, the imaging of subsurface features in glaciers [5], and ice sheets [4][6][7] was demonstrated and the effect of subsurface layers, different ice types, firn bodies, crevasses, and bed rock was recognized in the tomograms. Those studies on PolSAR, Pol-InSAR and TomoSAR techniques made promising steps towards a subsurface structure information retrieval on glaciers and ice sheets, but the direct relationship to geophysical parameters, which are useful for the glaciological community, is only limited. The most promising way to achieve this goal is the combination of PolSAR and Pol-InSAR/TomoSAR techniques in order to exploit the synergies between the individual methods and by integrating the models and algorithms into one common framework. A first experiment in this direction was the integration of TomoSAR vertical scattering profiles into the PolSAR firn anisotropy model [8]. This allowed the inversion of the firn density from experimental airborne F-SAR data over Greenland, which would be a promising geophysical information product. However, this experiment enabled only the inversion of a single bulk density value for the entire penetration depth and was based on hundreds of samples of CPD measurements and TomoSAR profiles across a wide range of incidence angles, which is of limited applicability. Therefore, this study will continue this investigation, assessing to which degree an inversion is possible on reduced and more feasible observation spaces. Open questions are, which number and range of incidence angles as well as which and how many baselines are required, and whether depth-varying densities could be retrieved. The results should give an indication if such an approach might be feasible with future spaceborne SAR systems. [1] G. Parrella, I. Hajnsek, and K. P. Papathanassiou, “Retrieval of Firn Thickness by Means of Polarisation Phase Differences in L-Band SAR Data,” Remote Sensing, vol. 13, no. 21, p. 4448, Nov. 2021, doi: 10.3390/rs13214448. [2] E. W. Hoen and H. Zebker, “Penetration depths inferred from interferometric volume decorrelation observed over the Greenland ice sheet,” IEEE Transactions on Geoscience and Remote Sensing, vol. 38, no. 6, pp. 2571–2583, 2000. [3] G. Fischer, K. P. Papathanassiou and I. Hajnsek, "Modeling Multifrequency Pol-InSAR Data from the Percolation Zone of the Greenland Ice Sheet," IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 4, pp. 1963-1976, 2019. [4] G. Fischer, M. Jäger, K. P. Papathanassiou and I. Hajnsek, "Modeling the Vertical Backscattering Distribution in the Percolation Zone of the Greenland Ice Sheet with SAR Tomography," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 12, no. 11, pp. 4389-4405, 2019. [5] S. Tebaldini, T. Nagler, H. Rott, and A. Heilig, “Imaging the Internal Structure of an Alpine Glacier via L-Band Airborne SAR Tomography,” IEEE Transactions on Geoscience and Remote Sensing, vol. 54, no. 12, pp. 7197–7209, 2016. [6] F. Banda, J. Dall, and S. Tebaldini, “Single and Multipolarimetric P-Band SAR Tomography of Subsurface Ice Structure,” IEEE Transactions on Geoscience and Remote Sensing, vol. 54, no. 5, pp. 2832–2845, 2016. [7] M. Pardini, G. Parrella, G. Fischer, and K. Papathanassiou, “A Multi-Frequency SAR Tomographic Characterization of Sub-Surface Ice Volumes,” in Proceedings of EUSAR, Hamburg, Germany, 2016. [8] G. Fischer, K. Papathanassiou, I. Hajnsek, and G. Parrella, “Combining PolSAR, Pol-InSAR and TomoSAR for Snow and Ice Subsurface Characterization,” in Proceedings of the ESA POLinSAR Workshop, Online, Apr. 2021.

Sea ice topography plays a crucial role in understanding atmosphere-ice-ocean interactions. Mapping sea ice elevation and roughness is also vital for ship navigation in polar oceans by providing information on ice deformation and safe routes. While laser altimeters offer very high-spatial resolution (< 1m), their small spatial coverage and long revisit time (e.g., 91 days for ICESat-2) limit their effectiveness for consistent observation of sea ice. Synthetic aperture radar (SAR) has become a highly valuable tool for monitoring polar regions since it can provide all-weather day/night imagery on a large spatial scale. Single-pass SAR interferometer such as TanDEM-X offers for the first time the chance to generate a sea ice digital elevation model (DEM) despite the natural dynamics of sea ice [1]. In the abstract, sea ice elevation refers to the height of sea ice, including the snow cover, relative to the local sea surface height. The DEM generated from interferometric SAR (InSAR) is subject to bias because it measures the phase center height, which is affected by microwave penetration into the snow layer over sea ice. In the case of X-band, the penetration into younger ice (YI), such as first-year ice and new ice, is negligible due to the highly saline surface [2]. However, for older ice (OI) that undergoes desalination, such as multi-year ice with dry snow cover, the penetration depth can be up to 1m [2]. Therefore, it is necessary to classify sea ice into different categories to account for the significant variation in penetration bias for accurate sea ice elevation retrieval. SAR polarimetry provides sufficient information on the scattering processes and thus is a valuable tool to characterize sea ice [3, 4]. With the polarimetric SAR Interferometry (Pol-InSAR) technique, a two-layer-plus-volume model [5] is developed to estimate the penetration depth over old and deformed sea ice with snow cover. In order to investigate sea ice topography in the Antarctic, Operation IceBridge (OIB) and TanDEM-X Antarctic Science Campaign (OTASC) was successfully carried out in the western Weddell Sea in October 2017 [6]. The coordinated acquisitions of optical DEM from OIB aircraft and the SAR imagery (VV and HH polarizations) from the TanDEM-X satellite provide a unique opportunity to characterize sea ice topography in the Antarctic. This study proposes a new method for estimating sea ice elevation over YI and OI from dual-polarization interferometric SAR images. The method incorporates sea ice classification and DEM generation. In the first step, we use a random forest classifier with selected co-polarimetric features as input to categorize sea ice into YI and OI based on their penetration bias. In the second step, we generate sea ice DEM separately for YI and OI. For YI, the standard InSAR processing is conducted to retrieve the elevation as the penetration depth is negligible. For OI, we develop a two-layer-plus-volume model [5] that combines InSAR processing with a correction for penetration bias to generate the sea ice DEM. The effectiveness of the proposed method is demonstrated through validation using the OTASC dataset covering a 200×19km area in the Weddell sea, where the root-mean-square error between model-derived DEM and the reference data (optical DEM) is 0.24m. It indicates that the proposed method successfully compensates for the penetration bias and achieves precise elevation mapping over YI and OI. Furthermore, we apply the method to 162 SAR images, which covered 12 sectors (each corresponding to around 500×20km) in the Weddell Sea and the Ross Sea, to retrieve the sea ice elevation and macroscale roughness (standard root mean height). The retrieved sea ice elevation shows positive skewness and fits well with the Exponential Modified Gaussian probability density function (PDF) for all sectors. We also investigate the variation of sea ice elevation and roughness relative to the distance to the coast and associate the variation with sea ice classes obtained from Ice Chart. These variations will be interpreted with geophysical parameters to better understand the sea ice formation and dynamics in our ongoing study. Reference [1] Wolfgang Dierking, Oliver Lang, and Thomas Busche, “Sea ice local surface topography from single-pass satellite InSAR measurements: a feasibility study,” The Cryosphere, vol. 11, no. 4, pp. 1967–1985, 2017. [2] Martti Hallikainen and Dale P Winebrenner, “The physical basis for sea ice remote sensing,” Microwave remote sensing of sea ice, vol. 68, pp. 29–46, 1992. [3] Lanqing Huang and Irena Hajnsek, “Polarimetric behavior for the derivation of sea ice topographic height from TanDEM-X interferometric SAR data,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 14, pp. 1095–1110, 2021. [4] Son V Nghiem, Lanqing Huang, and Irena Hajnsek, “Theory of radar polarimetric interferometry and its application to the retrieval of sea ice elevation in the Western Weddell Sea, Antarctic,” Earth Space Sci., vol. 9, no. 5, pp. e2021EA002191, 2022. [5] Lanqing Huang, Georg Fischer, and Irena Hajnsek, “Antarctic snow-covered sea ice topography derivation from TanDEM-X using polarimetric SAR interferometry,” The Cryosphere, vol. 15, no. 12, pp. 5323–5344, 2021. [6] Son Nghiem, Thomas Busche, et al., “Remote sensing of Antarctic sea ice with coordinated aircraft and satellite data acquisitions,” in IGARSS. IEEE, 2018, pp. 8531–8534.

In this study, we investigate polarimetric sea ice features computed from quad-pol synthetic aperture radar (SAR) images collected over the fast ice area on the north-eastern coast of Greenland during the CIRFA-2022 cruise with RV Kronprins Haakon in the period April 22nd to May 9th, 2022. On this cruise, Radarsat-2 and ALOS-2 quad-pol SAR images were acquired co-incident with detailed in-situ measurements characterizing the physical surface properties. The fast ice, which is stationary sea ice locked by grounded icebergs, is ideal for studying relationships between observed polarimetric SAR features and the physical properties of sea ice. SAR backscatter from arctic sea ice is resulting from a combination of several scattering mechanisms, in general categorized as surface scattering, volume scattering, and double bounce scattering. Previous analysis of polarimetric decompositions of sea ice scenes using well-established algorithms shows that the power of the double bounce component often is an order of magnitude below the contribution from surface and volume scattering, and that high double bounce is restricted to isolated features. There is also experimental evidence that, for some ice types, electromagnetic backscatter has a high degree of depolarization. Depolarization is in some decomposition models mostly attributed to volume scattering. However, for sea ice backscattering it is anticipated that depolarization can also be caused by surface phenomena such as rough and highly deformed ice, as well as by volume type scattering from brine inclusions and inhomogeneities inside the ice column. There is still uncertainty regarding which polarimetric features relate to which physical properties of the ice. From the literature, there seems to be consensus that the availability of HH- and VV-polarization, additional to HH- and HV-polarizations, enables an improvement of ice type separation. However, this gain is primarily attributed to the signal intensities only. When it comes to the phase difference and correlation coefficient between the HH- and VV-polarized signals, there is still debate if a consistent “global” relationship between the parameters and ice types or ice properties exist, although single cases has been reported in the literature showing that the phase difference can improve separation of thin and thick ice. Based on records of coincident in-situ and quad-pol data acquired during the CIRFA-2022 cruise, this paper addresses questions related to the relation between physical sea ice type properties and SAR polarimetric features containing a diverse set of ice types, including multi-year ice, leveled and deformed first-year ice, thin young ice, leads and open water. The presentation will briefly review our analysis methodology and high-light some particularly interesting preliminary results.

Existing research in automatic vessel detection using Synthetic Aperture Radar (SAR) mainly concentrates on large, metallic ships. This project focuses on the detection of small, non-metallic targets, i.e. inflatable rubber vessels which are often found in distress fully loaded with migrants in the central Mediterranean Sea. The physical attributes of such kind of targets, namely its small size and height and the absence of materials of high dielectricity, decrease the detection capabilities. To gain a better understanding of the target’s backscattering properties ant to increase detection capabilities we use multi-platform SAR data. The science developed in this work builds on the shoulder of previous research (Lanz et al. 2020, Lanz et al. 2021) which we have carried out on inflatable vessel detection. Our data collection campaign resulted in SAR imagery of a 12m long rubber inflatable vessel identical in construction and material to those which are often used by human traffickers to send migrants to cross the central Mediterranean Sea. The data acquisition was done in a small lake near Berlin, Germany, which functioned as a test bed. The experiment setting is such to emulate the backscattering of 80 passengers on board. In experiments with a ground SAR system we identified wet clay pebbles, packed in 30x40cm air-tight plastic bags, as the best choice to simulate humans in SAR. The collection comprises single, dual and quad-polarimetric data and covers a variety of different sensor and scene parameters. For detector testing we used different combinations of dual-pol StripMap TerraSAR-X (TSX) data. With the high-resolution single-pol imagery from ICEYE we shed some light on the spatial distribution of the scattering intensity inside the vessel and in its close vicinity. The quad-pol Cosmo SkyMed data enabled us to apply a variety of coherent and incoherent polarimetric decompositions (Pauli, Cloude-Pottier, Yamaguchi and Cameron) to get insight about the different scattering mechanisms occurring on the inflatable. Different combinations of incidence angles, polarimetric channels and vessel orientations relative to the LoS support the effort of analysing their impact on the vessel scattering and its detectability. To facilitate the testing of vessel detector algorithms in a more realistic situation of different sea states, we put together a collection of dual-pol TSX data of wave heights between 0.5m and 6.5m respecting two different wind directions (up/Down, cross relative to the LoS), different polarizations and incidence angles. We combined this data with the inflatable backscattering by copying the vessel pixels in the open ocean background, respecting the main sensor parameters. Our detector testing included the intensity-based CA-CFAR detector, six Polarimetry-based and one sub-look-based detector following a qualitative evaluation of their detection capabilities regarding the special target. Amongst the polarimetry detectors are the Depolarization Ratio Anomaly Detector (iDPolRAD) (Marino et al. 2016), the Geometrical Perturbation-Polarimetric Notch Filter (GP-PNF) (Marino, 2013), the Polarimetric Match Filter (PMF) (Swartz et al. 1988, Novak et al. 1989), the polarimetric symmetry (PolSym) (Nunziata, 2012), the polarimetric entropy (PolEntr) detector and the Polarimetric Whitening Filter (PWF) (Novak et al. 1989). The sublook-based detectors include the sublook correlation method (Marino et. al. 2015). This work offers the innovative method of how we emulated the images of the ocean with a fully occupied inflatable, the detector comparison and the detector fusion. Further, we tested the surface scattering detecting variant of the iDPolRAD, called ‘SiDPolRAD’(Lanz et al. 2021), and the iDPolRAD with co-pol (HH VV) data, adding dihedral scattering and Bragg scattering to the range of detectable scattering mechanisms of these detectors. We introduce a double bounce detection algorithm that uses the T22 element of the coherency matrix and we tested the combination of detectors to enhance the detection capabilities for the inflatable at higher sea states. For TSX-StripMap dual cross-pol data the fusion of iDPolRAD with the SiDPolRAD delivered good results, however it could not outperform the PolSym and the PWF (Pd=90%, Pfa

Hybrid polarimetry is a dual polarized SAR which transmits circularly polarized wave (Right Circular or Left circular) and receives coherent linearly polarized wave and hence termed as Hybrid compact Circular Polarimetric SAR system. This unique architecture of polarimetric data collection stands in line with full polarimetric system with wide swath coverage, suitable PRF, larger incidence angle range and reduced down link data volumes in comparison with tradition fully polarimetric system. ISRO’s Earth Observing Satellite (EOS-04) is follow-on Radar Imaging Satellite (RISAT-1) intended to serve operational users. In order to cater to user’s main objective of having high-quality quantitative data with extended swath coverage and spatial resolutions, EOS-04 was launched on 14th February 2022 from SHAR, India’s Spaceport. EOS-04 is equipped with on-board Hybrid polarimetric architecture and Full polarimetry apart from conventional single/dual polarization to support gamut of land and ocean applications. RISAT-1 launched in 2012 was first of its kind in Earth Observation SAR missions to operate in Hybrid polarimetry and stood as a pre-cursor to many International SAR communities. However, EOS-04 has come up with Hybrid Polarimetric Medium Resolution ScanSAR (MRS) and Course Resolution ScanSAR (CRS) with 8 beams and 12 beam configurations covering a swath of 163 km and 223km respectively to offers polarimetric measurements and generate information products to the user community. Methodology: The methodology implemented on EOS-04 Hybrid polarimetric Single Look Complex (SLC) data corresponding to Stripmap and ScanSAR in performing validation of polarimetric parameters using point target and distributed targets are presented. ScanSAR SLC data containing burst wise complex data for each polarization is considered to concatenate the beams based on the Number of bursts, first valid line, number of valid samples, Minimum range of burst buffer, Start and End Time of beam(seconds) etc. Many intricacies were handled for extracting and processing of ScanSAR beam-wise SLC data, while mosaicking in slant range domain to form a complete scene. The polarimetric portraits resulting from the Stokes vector of the backscattered field and its child Stokes parameters are derived and validated over point target data like trihedral and dihedral corner reflectors, Active Radar Calibrator and Homogenous distributed targets to comprehend the Surface, Double and Volumetric scattering mechanism. The Stokes child parameters are strident tools when polarized intensity is used in the place of total power (which contains polarized and unpolarized power) for their normalization. M-Delta and M-Chi hybrid polarimetric decompositions are implemented on EOS-04 Hybrid polarimetric data to decipher the characteristics of various theme-specific targets towards image classification w.r.t calibrated and uncalibrated data. Results and Discussion: The polarimetric portraits like Degree of Circularity, Degree of linear polarization, Circular Polarization ratio (CPR), Relative Phase, Degree of Polarization are evaluated for scenario related to calibration targets (viz ARC, trihedral and Dihedral corner reflectors) and validated for Stripmap mode initially. In light of ScanSAR burst mode of operations, however these polarimetric parameters are validated for Cal pass acquired over NRSC-ISRO cal site and tabulated Figure 1(right). Based on still further planned acquisitions, the validation especially in the overlapping regions of SAR beams in Azimuth and Range directions can be performed to evaluate the impact of thermal noise and banding effect. Hence, the dynamics of phase value in over-lapping and non- over-lapping regions of beams for the same type of target in the scene are studied. The analysis results pertaining to In-orbit testing phase data products are highlighted here. To address many applications which were earlier limited by swath coverage and off-nadir angles, this paper brings out calibration and validation methodology and results of EOS-04 ScanSAR mode of hybrid polarimetric data and its exploitation that benefits user community in a holistic manner to perform time-series analysis.

Estimation of soil moisture is one of the main important factors in the field of agriculture. It is useful for crop yield prediction, vegetation growth monitoring, drought prediction, irrigation management. Radar backscatter being sensitive to soil roughness, crop cover and soil dielectric properties can be used to retrieve soil moisture up to certain accuracy. In this study, semi empirical soil moisture retrieval model such as Oh, Dubois, Oh-2004, X-Bragg have been compared with the novel method for estimating soil moisture using θ(FP) scattering parameter with and without vegetation correction at major phenological stages of four crops i.e., wheat, corn, pasture and soyabean at different dates. The L-Band UVSAR Full-pol data and ground truth data collected during SMAPVEX12 campaign over Winnipeg, Manitoba, Canada has been used for validating the results. The paper compares time series measured soil moisture and the estimated soil moisture from the above model based on the inversion rate, root mean squared error (RMSE) and the correlation coefficient with the increase in the vegetation water content.

Soil moisture is one of the most relevant geophysical variables playing a key role in a vast range of applications such as environmental monitoring (e.g., climatology, hydrology, drought development and fire risk assessment). Additionally, it represents a significant guiding factor for agricultural activities, especially for smart irrigation and crop yield estimation [1]. In this context, Synthetic Aperture Radar (SAR) has become, over the years, one of the most valuable sources of information for accurate and continuative estimation of soil moisture with high spatial and temporal resolutions, in almost any weather conditions, over agricultural areas. However, SAR-derived soil moisture retrievals are affected by several factors such as topography, incidence angle, soil roughness and vegetation cover, which is responsible for additional signal attenuation and scattering mechanisms over the soil. Concerning the algorithms for soil moisture estimation, Artificial Intelligence (AI), and in particular Artificial Neural Networks (ANNs), have been demonstrated to be an important instrument for Earth Observation (EO) applications designed to retrieve information from Remote Sensing (RS) data. In general, when using AI for geophysical parameters estimation, there is a risk that the physical meaning behind the ANN mapping criteria becomes challenging to understand. Nevertheless, if properly trained, AI algorithms can reproduce the physics associated with the phenomenon we are observing, as shown in our previous work [2]. This study aims to synergically adopt electromagnetic data modelling and an ML algorithm design and development to separate the different scattering contributions. More in detail, the components associated with the ground and the vegetation cover are distinguished so that only the former can be used for more effective soil moisture retrieval algorithms. For this purpose, the “Tor Vergata” electromagnetic model developed by Bracaglia et al. [3] was adopted to generate a set of simulated Mueller matrices at L-band together with the backscatter components addressing the different canonical scattering mechanisms (double bounce, volume, and surface scattering) for VV, HH and HV polarizations. Being validated with several experimental data, the “Tor Vergata” model provides the possibility of simulating data with different values of vegetation- and soil-related variables (plant height and soil moisture/roughness) while setting the sensor configuration variables such as frequency and incidence angle [2]. Different conditions of maize plant height, soil roughness and soil moisture were considered to generate the simulated data. Moreover, to motivate the choice of isolating and using the soil backscatter for soil moisture estimation, a preliminary analysis on the sensitivity of simulated total backscatter and the soil-related component (in HH and VV polarizations) to soil moisture was performed. For this analysis, the HV polarization was not considered since it is not relevant for soil condition study, at least for simulated data. As a first result, it is possible to observe that the total backscatter is less sensitive to soil moisture variations if compared to the soil component. Besides, the Pearson correlation coefficient is higher when the soil backscatter component is considered. Also, some metrics, such as data standard deviation and the slope of the linear interpolating line, suggest that adopting the soil backscatter component would help in retrieving soil moisture. Then, the simulated dataset was used to train a properly designed ANN model taking as input the elements of the Mueller matrix and using the backscatter soil-related components (soil and double bounce scattering) in the selected polarizations as targets. The selection of the double bounce as target is motivated by the fact that this component is strictly influenced by soil properties (e.g., soil roughness and soil moisture). For this purpose, the sensitivity of the double bounce component will be evaluated. Moreover, the HV polarization was included, as target, to allow further analysis concerning the capability of the ANN in separating the backscatter components in all three polarizations. In order to assess the performance of the proposed methodology, the ESA BelSAR2018 campaign, providing full-polarimetric SAR data at L-band, acquired with aerial surveys and in-situ soil moisture measurements, is considered to derive the correlation between the soil moisture itself and the soil-related backscatter estimated by the trained ANN model. Moreover, the sensitivity of those components to the soil moisture will be compared with the sensitivity shown by the outcomes of canonical polarimetric decompositions (e.g., Freeman Durden [4], Generalized Freeman Durden [5], Pauli [6], Yamaguchi [7]). In this regard, it will be shown how the proposed approach can help in estimating soil moisture by removing the vegetation influence, demonstrating that the isolated soil-related backscatter components retrieved by the ANN are more sensitive to soil moisture if compared to the total received signal. [1] Xing, M.; Chen, L.; Wang, J.; Shang, J.; Huang, X. Soil Moisture Retrieval Using SAR Backscattering Ratio Method during the Crop Growing Season. Remote Sens. 2022, 14, 3210. [2] L. G. Papale, F. Del Frate, L. Guerriero and G. Schiavon, "A Physics-Based ML Approach for Corn Plant Height Estimation with Simulated Sar Data," IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, Kuala Lumpur, Malaysia, 2022, pp. 4159-4162. [3] M. Bracaglia, P. Ferrazzoli, and L. Guerriero, “A fully polarimetric multiple scattering model for crops,” Remote Sens. Env., pp. 170-179, 1995. [4] A. Freeman and S. L. Durden, “A three-component scattering model for polarimetric SAR data,” IEEE Trans. Geosci. Remote Sens., pp. 963-973, May 1998. [5] Jong-Sen Lee and Eric Pottier, Polarimetric Radar Imaging: From Basics to Applications, CRC Press, 2009 [6] S.R. Cloude, “Polarisation: Applications in Remote Sensing”, Oxford University Press, ISBN 978-0-19-956973-1, 2009. [7] Y. Yamaguchi, A. Sato, W. Boerner, R. Sato and H. Yamada, “Four-Component Scattering Power Decomposition With Rotation of Coherency Matrix,” IEEE Trans. Geosci. Remote Sens., pp. 2251-2258, June 2011.

Soil moisture plays a crucial role in sustaining a healthy ecosystem by supporting plant growth and survival, improving crop yields, and ensuring good crop quality. It also helps bind soil particles together, reducing the risk of erosion. Given the critical importance of soil moisture in maintaining the environment and supporting human activities, it is crucial to have reliable information on soil moisture content across broad regions. While testing soil moisture with portable equipment is convenient at the local level, conducting such tests on a large or global scale can be both expensive and time-consuming. Therefore, it is imperative to develop effective and efficient methods for soil moisture monitoring and mapping, such as using remote sensing techniques, to obtain timely and accurate soil moisture information over large areas. Microwave remote sensing is a powerful tool for directly observing and regularly sampling soil moisture over a significant portion of the Earth's land surface from air or orbit. Microwave measurements have an advantage in that they are not significantly impacted by cloud cover and fluctuating surface solar illumination. Synthetic Aperture Radar (SAR), which penetrates the Earth's surface and detects the backscatter or reflection of radar waves, is particularly sensitive to surface dielectric properties. As the amount of backscatter is related to the dielectric constant of the soil, which in turn is related to the soil moisture content, radar data provides a useful means for estimating soil moisture at global and local levels. However, surface roughness and vegetation can affect the accuracy of soil moisture retrieval from SAR data. Moreover, theoretical scattering models are complex, and semi-empirical models have limitations across a wide range of scenarios. Therefore, such models may require additional support to accurately depict the non-linear relationship between natural targets and SAR features. Machine learning algorithms are particularly useful for capturing these non-linear correlations between output and input parameters and outperform traditional linear correlations. This study aims to assess soil moisture and dielectric retrieval over agricultural land using machine learning algorithms and decomposition techniques. Specifically, features were extracted from three Polarimetric decomposition models on simulated NASA-ISRO SAR (NISAR) L-Band radar images. To train the models, a total of 20 input polarimetric parameters, including the freeman-durdan, H/A/Alpha, and van Zyl decompositions were used. The machine learning algorithms namely Random Forest, Neural Network, K-Nearest Neighbors, Multiple Linear Regression, Decision Trees, Stochastic Gradient Decision, and XGBoost Regression were tested to determine their efficacy in retrieving soil moisture and dielectric values. Soil moisture and dielectric values were retrieved from wheat, corn, and soybean fields. An open-source software named SMF v1.0 (Soil Moisture Framework v1.0) was developed during this research. This framework automates the process and integrates machine learning techniques on one platform, which is the first of its kind showcased in this paper. The study found that the Random Forest regression technique produced the most precise soil moisture estimations for wheat fields, with R² of 0.87, RMSE of 0.03, and MAE of 0.02. Additionally, the results for real dielectric constant retrieval for the wheat field were R² 0.87, RMSE 1.35, and MAE 0.99. These findings suggest that machine learning algorithms can be an effective tool for estimating soil moisture and dielectric values. The research provides essential recommendations for future missions such as ROSE-L and NASA-ISRO SAR missions (NISAR).

The Kuiseb River is one of the major ephemeral rivers of western Namibia, setting the northern limit of the Namib Sand Sea and outflowing in the Atlantic Ocean at Walvis Bay. Such ephemeral rivers are of highest importance for the country, since they are related to recent past climatic conditions and represent potential water resources to sustain humans, vegetation, and wildlife in the hyper-arid Namib Desert. Using high resolution radar images from the Japanese ALOS-2 satellite, we mapped for the first time numerous channels hidden under surface aeolian sediments: while non-permanent tributaries of the Kuiseb River appear north of its present-day bed, a wide paleochannel system running westward, assumed by previous studies, could be clearly observed in the interdune valleys in the south. Radar-detected channels were studied during fieldwork in May 2019, using both subsurface ground penetrating radar profiles and high resolution drone-generated digital elevation models. It allowed us to confirm the existence of the “Paleo-Kuiseb” drainage system, remnant of the Holocene history of the Kuiseb River moving northward under the progression of the Namib Sand Sea [Paillou et al., 2020]. As the dynamics of such ephemeral river systems is still not fully understood, we also investigated potentials of multi-wavelength satellite data to monitor the dynamics of the Kuiseb River. We acquired time series of multi-spectral optical and near infrared imagery together with synthetic aperture radar scenes from the Sentinel 1 and 2 satellites operated by ESA. It allowed us to monitor several years of changes in surface and sub-surface properties of the Kuiseb River vicinity, providing information on the hydrologic and vegetation dynamics. Strong variations in remote sensing data were observed during March–April 2017 and June–July 2018 in a tributary of the Kuiseb River, rain events causing its reactivation. However, evidences of reactivation were also observed during a major flood in 2021, while no rain occurred on the studied area. This is related to the dynamics of the aquifer of the Kuiseb River: the flood recharges the alluvial aquifer and the rising water table level produces a signal that is detected by the satellite radar sensor [Normandin et al., 2022]. Finally, we considered a promising new technique to map soil moisture in the vicinity of the Kuiser River: the closure phases model developed by [De Zan, 2018]. The closure phases computed from three SAR interferograms are in principle immune to all simple propagative effects like target displacement, delays in atmospheric propagation, and topographic effects: this should allow in desert areas to monitor surface electrical properties (dielectric constant, radar penetration depth) with are due to changes in soil moisture. This information should help us to better understand the relationships between flood events and the dynamics of the Kuiseb aquifer. Using times series of Sentinel 1 SAR images acquired in 2021, we produced moisture maps which show the changes of the soil moisture over the alluvial plain north of the Kuiser river: these are consistent with rain and flood events. References Ph. Paillou, S. Lopez, E. Marais, K. Scipal, “Mapping Paleohydrology of the Ephemeral Kuiseb River, Namibia, from Radar Remote Sensing”, Water, 12 (1441), 2020. C. Normandin, Ph. Paillou, S. Lopez, E. Marais, K. Scipal, “Monitoring the Dynamics of Ephemeral Rivers from Space: An Example of the Kuiseb River in Namibia”, Water, 14 (3142), 2022. F. De Zan, G. Gomba, “Vegetation and soil moisture inversion from SAR closure phases: First experiments and results”, Remote Sensing of Environment, 217 (562-572), 2018.

1. Introduction This article presents a methodology for detection, monitoring and tracking of water hyacinth (an aquatic weed) using Polarimetric data from Sentinel-1 satellite. The study area is the Winam Gulf of Lake Victoria, Kenya. Water Hyacinth (WH) whose scientific name is Eichhornia crassipes is one of the dreadful and noxious invasive species with single plant having the ability to produce 14 x 107 daughter plants and cover a 1.4 Km2 area in a year. WH causes severe effects to the environment where it blooms such as increasing in the cases of malaria, falarisis, decreased fish populations, hampers recreation, and navigation. Native to South America, the WH was introduced to Lake Victoria in 1989 established itself by spreading along the shores of the lake and forming mats that covered areas of more than 70 square miles 100 years after introduction . WH was identified among the most dangerous invasive species by the International Union for Conservation of Nature’s (IUCN) due to its adverse socio-economic impacts and difficulty in elimination from water bodies. Remote sensing offers less expensive alternative to monitoring of WH. Remote sensing provides the capability to distinguish and quantify WH infestation from a color Infrared imagery from multiple sites . Mapping of WH at a large scale has been demonstrated using optical data from Landsat-8. Cavalli, et al. The mapping of aquatic weeds using Landsat, ENVISAT/MERIS and TERRA/ASTER and field measurement has been demonstrated before. NDVI from Sentinel-2 was used to monitor water hyacinth and Chl-a in Lake Tana in Ethiopia. With optical data, cloud coverage is a potential challenge to continuous and all-weather imaging. Synthetic Aperture Radar (SAR) data on the other hand are collected at any time of the day and independent of the weather conditions of a place thus offering varied application. Additionally, polarimetric information can be used to discriminate different scattering mechanisms happening in the scene. Specular scattering (mirrorlike) on surfaces such as water where less signal is scattered back to the sensor leads to darker image tones in SAR images enabling the classification of open waters. Additionally, when waves are present on the lake water, the backscattering can be modelled as a Bragg Surface. Vegetation produces volume scattering that appear brighter in all the polarization channels. Previous classification of vegetation, wetland area, and aquatic macrophytes achieved producers’ accuracy of over 85% using SAR data. 2. Methodology Sentinel-1 is a dual-polarimetric from the Copernicus programme provided by the European Space Agency (ESA). Interferometric Wide Single Look Complex (IW SLC) data with a 5m by 20m spatial resolution is used for analysis. The initial temporal resolution was 6 days until the failure of sentinel 1A leading to a resolution of 12 days. 438 datasets were analyzed for a period between January 2017 to November 2022. These products are SLC dual-polarization channels of VV and VH where V-Vertical linear, and H-Horizontal linear. The datasets were grouped in three separate frames based on the orbit and coverage over the location. The methodology has three main processing steps that is Pre-processing, co-registration, and detection. Pre-processing and Co-registration was done using ESA SNAP graph builder. SLC contains amplitude and phase information of the electromagnetic wave. The polarimetric covariance matrix including co-polarization VV, cross-polarization VH, and their correlation VV*VH were evaluated. The change detection method of Optimisation of Power Difference (OPDiff) is applied in this research on the polarimetric covariance matrices. The power difference decomposes the scene to maximal scattering mechanism that has been either added or removed. OPDiff is achieved by finding a linear combination of polarimetric channels thus providing the smallest or highest difference. The output of this is eigenvalues and eigenvectors. The change detector compares differences against acquisitions when there was no water hyacinth with the region. In this work, the minimum difference in power was evaluated from eigenvalues of the matrices and a threshold set. The minimum difference is able to reject the false alarms coming from the wave surface scattering which is polarized and dominant in the first eigenvector. Previous research established methodology to be the best performing among a series of other methodologies evaluated. 3. Results and Conclusion The change detector was applied to each image in the stack and a threshold was set following a statistical rational fitting the distribution of the OPDiff in areas without change. Finally, a water hyacinth heatmap was produced showing the ratio of occurrence of water hyacinth in a single pixel in the analysis period. Kisumu city bay area was covered in WH for approximately 12% of the period between 2017 and 2022 November (estimated 7 months cumulatively of clogging). The analysis on annual WH coverage showed an increasing trend between 2017 and 2018 and decline in 2019, and steady rise in 2020, while declining in 2021, to 2022. The largest extend covered by the water hyacinth was recorded in 2018 at 212.2 square kilometers in the Winam Gulf. The upward trend began in March of 2018 from 7 km2 rising to the peak of 212.2 Km2 in December. This steadily declined in areas until November 2019. The increase was also recorded in 2020 steadily rising between March and November peaking in September maximum recorded coverage of 36 Km2. 2021 and 2022 recorded general low coverage area with coverage of less than 24 Km2. This increasing trend coincided with the increase amount of rainfall above the long term mean of the catchment draining into Winam gulf. Monitoring the movement of the WH showed great variability within the areas. The movement is both from East to West and West to East. The tracking of the WHH is challenging due to the 12 days repeat cycle and the high variability of the WH movement. Using Sentinel-1 data, we demonstrated how we can use polarimetric data to quantify and monitor the dynamics of the WH within Lake Victoria. 4. REFERENCES A. Marino and I. Hajnsek, “A Change Detector Based on an Optimization With Polarimetric SAR Imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 8, pp. 4781–4798, 2014, doi: 10.1109/TGRS.2013.2284510.