Carbon fluxes caused by land-use change contribute about 10% each year to anthropogenic CO2 emissions. These estimates have an uncertainty of typically +/- 50% even in recent years. The uncertainty of carbon sinks and sources due to land-use change is so large because the fluxes cannot be observed directly, given confounding influences on terrestrial carbon stocks by environmental changes such as the increase in atmospheric CO2. Therefore, dynamic global vegetation models and bookkeeping models have been used to quantify land-use change fluxes. In particular the latter have been the prime source of land-use flux estimates in global carbon budget analyses.

Bookkeeping models rely on information about carbon densities, response curves, and land-use transitions. This offers the opportunity of integrating observational data in estimates of global land-use change fluxes. Until now, however, global bookkeeping models have been using carbon densities on the level of country- or biome-averages that were derived from inventories or literature. Here we prepare the BLUE model (“Bookkeeping of Land Use Emissions”) for usage with observation-based assessments. BLUE already runs globally and applies spatially explicit information on land-use transitions. We now extend BLUE to apply spatially explicit information on carbon densities. This model extension alters the estimates of land-use change fluxes substantially in some regions, where land-use changes meet inhomogeneous vegetation. Our current analysis applies carbon densities from dynamic vegetation models, but opens ways to integrate biomass and soil carbon derived from Earth observations in a globally consistent framework to estimate global land-use change fluxes.

Area A - Board 72

Accurate measurement of Biomass is critical to understand terrestrial carbon cycle. Most of the recent research effort has focused on the Above Ground Biomass (AGB) mainly related to the forest ecosystem that represents the most dominant and dynamic component of the terrestrial carbon pool. The temporal variability in AGB due to forest losses and gain (e.g. fires, logging, land use changes and afforestation/reforestation) ideally requires a continuous monitoring. With the increased availability of space-borne data products related to the structure and function of terrestrial vegetation, an accurate analysis and mapping of the terrestrial carbon cycle from space is becoming possible. However, there is still no single technology that can directly, continuously and remotely assess AGB dynamics. For this reason, multiple satellite based dataset have to be integrated to support the generation of global AGB mapping. Specifically, two dataset are currently in existence describing AGB and dynamic nature of forest cover. On the one hand the ESA's GlobBiomass dataset provides a “static” map of AGB from remote sensing data ( i.e. optical, microwave, lidar) acquired for the year 2010 with a spatial resolution of 100m. On the other hand the Hansen’s Global Forest Change dataset provides “dynamic” maps of global forest extent and change from time-series analysis of Landsat images from 2000 to 2017 with a spatial resolution of 30m. Here we present a retrieval methodology combining ESA's GlobBiomass and Hansen’s Global Forest Change dataset to generate annual maps of AGB. The merging scheme consists of propagating through time a AGB map that refers to 2010 only using information of forest change. Different methods to translate forest change (i.e. gain and loss) to AGB have been tested and implemented. This study has been greatly facilitated by cloud computing infrastructures such as Google Earth Engine. The main working hypothesis is that there is a relationship between the change in AGB and the change in tree cover. Our maps therefore only represent the variations in AGB due to tree cover change. The final dynamic AGB dataset has a spatial resolution of 100m with an annual frequency from 2000 to 2017. The dynamic AGB dataset is here presented, reviewed with emphasis on achievements and limitations, and compared against changes at national scale from forest inventories. An outlook towards estimation of AGB with new orbiting instruments such as GEDI is also given.

Area A - Board 74

Carbon fluxes from land-use change are a large uncertainty within the global carbon cycle. One of the largest uncertainties comes from forest degradation – the removal of some trees where the resulting land cover retains enough tree cover to be classified as forest – which is widespread across the tropics. Current estimates suggest that the area affected by degradation could be two to ten times that of the area deforested annually. Such large ranges in estimates are partially due to the wide variety of degrading activities, from low intensity fuelwood extraction and sub-canopy agricultural activities to high intensity activities such as selective logging and fire, which each result in different levels of above ground biomass (AGB) removal. Furthermore, many remote sensing technologies are insensitive to degradation, due to the retention of tree cover, this limits out ability to estimate carbon emissions related to degradation. Due to the complex nature of forest degradation, methods which estimate changes in biomass over time can be used to identify areas of forest undergoing degradation and the associated carbon emissions. Here we test a novel methodology for monitoring forest degradation, which integrates L-band SAR data (from Alos Palsar/ Palsar 2) with ground based time series plot data to map changes in AGB over time. We tested this method in three sites across the tropics in Mexico, Ghana, and Indonesia, which each have different degradation pressures, climate regimes, and land use histories, to determine if a single methodology can be employed to measure degradation across the tropics. Saturation of L-band SAR data signal has been observed in high biomass forest, therefore, we assess the sensitivity of current L-band SAR satellite data, to establish the limitations of current technologies for measuring AGB changes related to degradation. This is particularly prudent in light of the new ESA BIOMASS mission to be launched in 2021, which should increase the sensitivity of degradation mapping in high biomass forests.

Area A - Board 75

In the ESA DUE GlobBiomass and CCI Biomass projects, we rely on the growing archives of SAR backscatter observations acquired at C- and L-band by past and current missions ENVISAT ASAR, Sentinel-1, ALOS PALSAR, and ALOS-2 for mapping the distribution of forest carbon stocks globally. Approaches utilizing the complementary information on forest above-ground biomass contained in multi-frequency radar backscatter have so far hardly been explored, in particular in the tropics. Noticeably, also multi-temporal observations of the backscatter were hardly considered in existing studies focused on the retrieval of biomass in the tropics, in spite of the improvements that had been achieved in other forest regions when using multi-temporal observations for retrieving biomass. At least regionally, it remains unclear i) what the actual potential and limitations of multi-temporal C- and L-backscatter observations for the mapping of above-ground biomass are, and ii) to which degree C- and L-band might complement the P-band data to be acquired by the upcoming ESA Biomass mission.
Based on a multi-temporal stack of air- and spaceborne backscatter images acquired at C-, L-, and P-band by Sentinel-1, ALOS-2, FSAR, and SETHI for AfriSAR campaign sites in Gabon (Lope, Mondah), we evaluated i) the sensitivity of backscatter at different frequencies to above-ground biomass under varying environmental conditions, and ii) the performance of a biomass retrieval using different frequencies separately or combined. Consistent with the approaches explored in GlobBiomass (see paper by Santoro et al.) for mapping biomass globally, a semi-empirical modeling framework was implemented for retrieving biomass from C- and L-band backscatter. In the case of P-band, empirical models were used.
The results suggest that i) P-band is most sensitive to above-ground biomass under all imaging conditions, ii) C- and L-band backscatter, provided multi-temporal observations are available, may achieve comparable accuracies as (few) P-band observations up to biomass levels of 200 to 300 t/ha, and iii) the benefit of combining frequencies is limited. With respect to the role of the imaging conditions, we find that wet season conditions are associated with reduced sensitivity to biomass at all frequencies and that in tropical areas characterized by persistently wet conditions (Mondah) only P-band provides acceptable sensitivity to biomass.
The results of this study reemphasize that P-band is the most suited frequency for mapping biomass and that the launch of the ESA Biomass mission can be expected to allow for improved biomass mapping across the tropics. Nonetheless, multi-temporal C- and L-band may provide almost equivalent information in biomass ranges up to 300 t/ha, at least for forests in drier regions of the tropics.

Area A - Board 76

Complex coherence plays an important role in the most application of PolInSAR. Most application estimate complex coherence using averaging in a neighborhood window, thus assuming stationarity. Accurate and reliable estimation of complex coherence, thus, is crucial to generate height maps. Several studies have been dedicated to optimizing complex coherence estimation. Estimated complex coherence has also been employed to classify SAR data in addition to mapping tree heights. The common core of these studies is to select homogeneous pixels following an appropriate similarity criterion. Such a criterion makes use of either polarimetric. Assuming stationarity and ergodicity Equation (1) can be estimated as

γ ̃=〈S_1 S_2^* 〉_Q/√(〈|S_1 |^2 〉〈|S_2 |^2 〉_Q )

Here, <. > is the averaging over a spatial ensemble denoted by $Q$ and $Q$ is the set of statistically homogeneous pixels (SHPs). be sufficiently large to ensure accurate estimation of gamma. $Q$ can be estimated in various ways. A conventional way is the Boxcar method inside a pre-defined rectangular window. Recently, there is an increasing attention towards using the adaptive selection of SHPs. For example, Refined Lee, wavelets, weighted averaging, adaptive Lee method, and the extended version of it are among the most popular methods. A recent study proposed Fixed-point (FP) method that uses the polarimetric scattering mechanism of pixels to select SHPs.
Most of these methods like the FP, use only polarimetric information and ignore the interferometric phase. To address this issue, a new method based on the complex covariance matrix including both polarimetric and interferometric information has been proposed. This method is called Double-Similarity (DS). The effect of its use, however, has not been examined on producing PolInSAR forest height map.
The objective of this study is to assess the accuracy of height estimation when using adaptive SHP selection methods for obtaining complex coherence. Double-Similarity and Fixed-Point have been applied in addition to Refined Lee and Boxcar methods. All methods were applied with varying window size.
Estimated complex coherence serves as input for the RVoG and RMoG_L models. The accuracy of the obtained height maps acquired from the RVoG and RMoG_L models is used as a criterion to evaluate the efficiency and reliability of the different complex coherence methods.
They obtain the complex coherence used by the RVoG and RMoG_L models to obtain tree height maps. Data are from L-band tomostack of the La Lope national park in Gabon with an average tree height of 30~m. The Double-Similarity (DS) method, which takes into account both polarimetric and interferometric information, in combination with the RMoG_L model had the most accurate height map with an RMSE of 2.50~m.
The Fixed-Point (FP) method with the RMoG_L model had an RMSE value of 3.70~m, the Double-Similarity (DS) method with the RVoG model had an RMSE of 6.70~m and the FP method with the RVoG model had an RMSE equal to 8.40~m. The two methods were compared with the Boxcar and Refined Lee methods as well. This comparison showed that the adaptive SHP selection methods improved height estimation accuracy by approximately 17\%. The study concludes that adaptive methods for obtaining coherence improve the height estimation accuracy, irrespective of the choice for a height estimation model. They, however, require longer computations.

Area A - Board 77

The aim of the study is to show potential of polarimetric analysis of ALOS-2 data for forest biophysical parameters retrieval within various forests of temperate zone. The approach is based on the fact that forests are very heterogeneous from the point of view of interactions between electromagnetic wave and illuminated object. It results in large diversity of scattering mechanisms within the particular radar pixels and large differences of scattering mechanism compositions for neighboring forest pixels. As a result, there are various relationships between particular forest parameters and back scattered radar signal for individual pixels within the forests. In this study the products of various polarimetric decompositions were used to select these pixels within the forests which are characterized by more or less similar composition of scattering mechanisms. The best results were obtained using relations between particular channels of Raney Decomposition. In the next step only these pixels were used for modeling of the relation between forest biophysical parameters and backscattered radar signal. The modeling was performed using Random Forest Method. The products were generated for the whole study areas and validated using randomly selected pixels which did not have to meet the criteria of the pixels selected for training of the model. As an input for modeling the original backscattering coefficient, selected products of polarimetric decompositions and optical Sentinel-2 data were used.
The research was carried out on six study areas located in different parts of Poland: Milicz, Piensk, Supraśl, Katrynka, Herby and Gorlice. The last one is located in mountainous region. There are various types of forests within test sites: deciduous, mixed and coniferous forest composed from various tree species located on various habitats. Two sets of ground truth data were used: field measurements for 500 sample plots of 25m radius for each study area and forest parameters products derived from Airborne Laser Scanning of 22 x 36 m pixel (forest stock volume, DBH, DBH basal area) and 5 x 5 m pixel (forest height). The reference data were collected in 2015 and 2016. Dual-polarimetric data from ALOS-2 satellite registered from 2014 to 2016 were available (four images per study area) and the set of Sentinel-1 data from these years. The polarimetric analysis was performed using SNAP and POLSARPRO software.
The obtained results show great potential of polarimetry and applied method of pixel selection for forest biomass retrieval. The maps of forest parameters obtained at the pixel level using proposed methods show much better the spatial changes of forest structure which can be seen e.g. on ortophoto than in the case of the modelling based only on backscatter coefficient. When ALS information was used as the reference (large number of reference points selected evenly for each stock volume class) it was possible to find one mathematical relation for selection of pixels used for modelling based on polarimetric Raney relations which was valid for all study areas. It gives the possibility to apply this method to a larger area. The obtained accuracies were similar for all study areas except of Gorlice study area which has mountainous character. The total accuracy of forest stock volume retrieval was close to the 90 m3 per hectar. The highest accuracies were obtained for the forests between 150 and 500 m3/ha (ca. 50-60 m3/ha). It was possible to obtain the results for the forests with stock volume up to more than 650 m3/ha without saturation of the signal. In the case of forest height the total accuracy was below 4 meters and the signal was not saturated for the heights of forest up to 33 meters. In the case of young forests the results obtained for forest height were much more precise than for stock volume. In the case of DBH basal area the best results were obtained for the young forests and then for the forests with very large stock volume. The least accurate results were obtained for medium stock volume forests.
All the research was performed within the project: “Remote sensing based assessment of woody biomass and carbon storage in forests” financed by National Centre for Research and Development from Poland.

Area A - Board 78

Information on the spatial distribution of forest structure parameters (e.g., aboveground biomass, vegetation height) are crucial for assessing terrestrial carbon stocks and emissions. In this study, we sought to assess the potential and merit of multi-temporal dual-polarised L-band observations for vegetation height estimation in tropical deciduous and evergreen forests of Mexico. We estimated vegetation height using dual-polarised L-band observations and a machine learning approach. We used airborne LiDAR-based vegetation height for model training and for result validation. We split LiDAR-based vegetation height into training and test data using two different approaches, i.e., considering and ignoring spatial autocorrelation between training and test data. Our results indicate that ignoring spatial autocorrelation leads to an overoptimistic model’s predictive performance. Accordingly, a spatial splitting of the reference data should be preferred in order to provide realistic retrieval accuracies. Moreover, the model’s predictive performance increases with an increasing number of spatial predictors and training samples, but saturates at a specific level (i.e., at 12 dual-polarised L-band backscatter measurements and at around 20% of all training samples). In consideration of spatial autocorrelation between training and test data, we determined an optimal number of L-band observations and training samples as a trade-off between retrieval accuracy and data collection effort. In summary, our study demonstrates the merit of multi-temporal ScanSAR L-band observations for estimation of vegetation height at a larger scale and provides a workflow for robust predictions of this parameter.

Area A - Board 79

Biomass retrieval from SAR is one of the main applications of radar remote sensing. Different sensor configurations have been tested in recent years. They are ranging from single polarization SAR data to fully polarimetric and tomographic SAR data sets either at one radar band or at different radar bands. However, it must be said that especially fully polarimetric or multi-band sensors are mostly limited to experimental campaigns. Operational space borne SAR missions offer SAR data at one radar band with mostly on like (HH/VV) and one cross-polarization (HV/VH).
Algorithms to retrieve biomass information from these data sets comprise simple regression models as well as sophisticated semi-theoretical radar backscatter models. Since the 1990s, SAR data were regularly acquired by the space agencies (ESA, JAXA, CSA). Latest with the advent of the ESA SAR sensors using the ScanSAR technique (ENVISAT mission with the ASAR instrument and the Copernicus Sentinel-1 satellite series with the C-SAR instrument), C-band data coverage of land surfaces increased substantially. Thus, multi-temporal to hyper-temporal data sets became available. They have been exploited successfully for biomass mapping from regional to global scales using change detection methodologies.
Even the SAR data acquired by the latest SAR sensors are still containing ambiguous information. Besides the geo-physical variable of interest (e.g. biomass), other environmental factors may have an influence on the backscattered signals. Soil moisture is one of these environmental factors. There are authors, who assume that even at C-band (radar band used by ESA SAR satellites since the 1990s) the water content of the soil layer below forests may have an effect on the radar backscatter at C-band.
Since 2015, the Department of Remote Sensing at the University of Jena maintains a measurement network with 240 sensors measuring soil moisture and soil temperature in a managed forest in Thuringia, Germany. In parallel, all Sentine-1a and b data sets acquired over the region are collected. Additional meteorological information will be used for data interpretation and analysis. We intend to present the results of our ongoing research activities and try to answer the question, if an influence of soil moisture below forest on C-band radar backscatter has been found for our test site region. This may also help with the interpretation and analysis of biomass maps produced by the ESA DUE GlobBiomass and the ESA CCI Biomass projects, where the department has been and is involved.

Area A - Board 80

The forest plays an important role in the global carbon cycle as reservoirs of carbon from the terrestrial ecosystem. Forest ecosystems account for approximately 72% of terrestrial carbon storage. Carbon absorbed or stored in forest trees is called biomass. Above Ground Biomass (AGB) is an essential parameter in monitoring climate change. The Laurisilva Forest, a World Heritage nature site, covers an area of approximately 15 000 hectares of the Madeira island (around 20% of the island). In this multistrata forest predominates endemic trees belonging to the Lauraceae family. Appraising AGB in Madeira Island is a challenge since it predominates in a zone where mist and fog frequently occur and where there is a high level of precipitation, with a minimum of 1700 mm/year and atmospheric humidity generally over 85%.
Traditionally, for the estimation of above-ground biomass in forest ecosystems is used a time-consuming, invasive and costly methodology with extensive in-situ data. This study intends to explore the use of high-resolution satellite imagery, such as synthetic aperture radar images (SENTINEL-1) and multispectral images (SENTINEL-2) deliverable by the European Space Agency (ESA) using the proposed methodology by Alboabidallah et al. (2017). The combination of vegetation indexes from optical images with tree height obtained by SAR images using interferometric techniques, allowed the use of allometric equations for the biomass estimation..
Two study areas were chosen: i) two Laurisilva Forest polygons composed mostly of trees of the family Lauraceae.; ii) two polygons composed of planted trees, not native mainly Oaks and European Beeches. For these areas (10mx10m) the number of trees, tree trunk diameter, tree height, tree crown volume and estimated biomass were measured in order to validate the results obtained with images.

Area A - Board 81

This paper investigates the benefits of integrating polarimetric radar variables with LiDAR measurements using machine learning approach in order to measure the forest canopy height. Multiple polarimetric variables are required as an input in order to ensure consistent height retrieval performance across a broad range of forest heights. We train the machine learning with LiDAR samples and different polarimetric variables based on small sample size collected across the images using a stratified random sampling approach. The trained machine was applied to the rest of the image using the same variables but excluding the LiDAR samples. The estimated height using this approach was validated versus LiDAR-derived height yielding good accuracy overall (r2=0.86, RMSE = 6.8 m). The concepts are demonstrated by means of airborne SAR and LiDAR acquisitions which were conducted as part of the ESA’s AfriSAR campaign over a tropical dense forest in Gabon in February 2016. The polarimetric SAR components are acquired by DLR’s F-SAS with P Band and LiDAR measurements by fine-beam LiDAR. The achieved results are validated versus reference data derived from airborne LiDAR measurements.
This technique has two key advantages over previously established approaches for forest height estimation: first, it does not require SAR to be in InSAR configuration as it uses polarimetric information only, and second, the full LiDAR coverage is not needed as it is based on LiDAR samples only. This results are significant for future space borne missions (GEDI, BIOMASS) which provide a wealth amount of LiDAR samples and polarimetric SAR measurements.

Area A - Board 82

Above Ground Biomass (AGB) is a key quantity for climate studies and especially to better assess carbon sources and sinks (Houghton et al. 2009). AGB is identified as an Essential Climate Variable by the Global Observing Systems Information Center (GOSIC) and the development of ESA's Biomass mission proves the need for a better monitoring of the AGB.
The vegetation optical depth estimated at C and X-bands from passive microwave (Liu et al.) was shown to be well correlated to AGB but the relation saturates for high AGB values. However, using passive microwaves VOD is a promising approach to study the temporal variation of AGB at coarse scale (~40 km). More recently, Rodriguez-Fernandez et al. (2018) studied the relationship of L-band VOD data from the ESA/CNES/CDTI SMOS (Soil Moisture and Ocean Salinity) mission. SMOS mission provides passive microwave measurements of the Earth surface emission at 1.4GHz since 2010, i.e. a 9 + year dataset. Over continental land, the retrieval algorithm (L-MEB, L- band Microwave Emission of the Biosphere) is designed to derive simultaneusly soil moisture and L-VOD using the multi-incidence angle capabilities of SMOS. SMOS has a coarse spatial resolution (products used are mapped 25 km grid) but an interesting revisit time of less than 3 days at the equator (daily at high latitudes).
Rodriguez-Fernandez et al. (2018) showed that L-VOD is well correlated with AGB for all biomes in Africa. They also showed that there are not significant signs of saturation of the L-VOD to AGB relationship even in the equatorial forest. In this communication we will extend the SMOS L-VOD to AGB study from Africa to the Globe and discuss how AGB time series derived from L-VOD can be used to study carbon stocks evolution at large scale on yearly basis at coarse scale. Finally, the use of disaggregation techniques combining SMOS L-VOD and high resolution radar, and their application to derive ABG at high resolution for dense forest areas such as Congo and Amazonia, will be discussed.

Area A - Board 83

ESA’s 7th Earth Explorer mission is BIOMASS, a fully polarimetric P-band SAR to be launched in 2022. The primary mission objective is to map forest biomass, forest biomass change (deforestation, forest degradation, forest regrowth) and forest height. In addition, BIOMASS has a number of secondary objectives including mapping of sub-surface geology, terrain topography under dense vegetation, glacier and ice sheet velocities and ionosphere properties. After a 3-month Commissioning Phase, BIOMASS will acquire tomographic SAR data in a 14-month period and polarimetric interferometric SAR data in a 4-year period.
BIOMASS calibration is challenging for several reasons [1]. The Biomass antenna is a deployable reflector with a 12 m diameter, which prevents characterization in a radio anechoic chamber. At P-band the ionosphere is also a major challenge for polarimetric calibration (channel imbalance and cross-talk). Finally, in combination with the global coverage requirements, the number and size of the required tomographic and interferometric baselines imply a repeat cycle of 7 months. Using a single transponder, the calibration would be too infrequent to always meet the calibration requirements. On the other hand, meeting the calibration requirements would require an excessive number of transponders to be distributed globally. Consequently, targets of opportunity must be used for complementary calibration, e.g. permanent scatterers or natural distributed target like ice sheets and deserts. Rain forests have been used as calibration targets, but for BIOMASS, rain forests are not sufficiently homogeneous and stable due to the large penetration depth at P-band.
For a natural distributed target to be applicable for radiometric calibration (or calibration validation) it must [1]
• have a high temporal stability
• have an extent comparable with the swath width (especially if it is also to be used for antenna pattern calibration)
• be homogeneous and/or its scattering pattern must be known such that incidence angle variations from passage to passage can be compensated for
• have a sufficiently high radar brightness, thereby ensuring an acceptable signal-to-noise ratio (SNR).
Prior knowledge of the absolute P-band backscattering coefficient does not exist for ice sheets and deserts. This information must be obtained in the BIOMASS Commissioning Phase, where BIOMASS will be calibrated using a transponder and a set of dedicated orbits. In the following 7 months the calibration (validation) will rely on the temporal stability of the distributed calibration targets. During this 7 month period the calibration targets will generally not be mapped with the same incidence angle as during the Commissioning Phase. This is why homogeneity and incidence angle dependency is required.
The homogeneity, stability and radar brightness of the backscattering coefficient of the Antarctic ice sheet has been addressed as part of an ESA study [1]. Due to the lack of P-band data, L-band data from ALOS/PALSAR and from the Aquarius scatterometer have been used. The backscattering coefficient turns out to have a large spatial variability across the Antarctic continent, but in a large region near Dome-C the backscattering coefficient is about –10 dB to –5 dB, which means that SNR is not prohibitive for BIOMASS calibration. Within a 4-year period the standard deviation of the backscattering coefficient is about 0.2 dB, and thereby this region is a good candidate for complementary radiometric calibration of BIOMASS. However, ionospheric scintillation is severe at high latitudes, and its impact in regard to BIOMASS calibration calls for additional investigation.

[1] S. Quegan, S., M. Lomas, M., K. P., Papathanassiou, J-S Kim, S. Tebaldini, D. Giudici, M. Scagliola, P. Guccione, J. Dall, P. Dubois-Fernandez and P. Paillou, “BIOMASS External Calibration Study”, Final Report, ESA/ESTEC Contract No. 4000116784/15/NL/CT, 2016.

Area A - Board 84

Accurate measurement and mapping of forest biomass is a critical component of carbon stock quantification, climate change impact assessment, suitability and location of bio-energy processing plants, and assessing fuel for forest fires [1]. Forest biomass is at present monitored globally using optical satellites (Landsat, MODIS, Sentinel-2), SAR (ALOS, PALSAR) and LiDAR technology. Despite this, biomass is still poorly quantified in most parts of the world, and the existing satellite sensors are not adequate to achieve a global biomass mapping and monitoring with sufficient accuracy.
A new and tantalising perspective of spaceborne observations of biomass on a global scale is offered by GNSS-Reflectometry, and specifically by the NASA CYGNSS Mission [2].
CYGNSS consists of a constellation of 8 microsatellites, launched in December 2016, which collects GPS reflections over both oceans and land, and within a latitudinal range of approximately +-38°.
Despite the wind speed being the primary scientific goal of CYGNSS, data are constantly collected everywhere over oceans and land, and they are freely available from https://podaac.jpl.nasa.gov. CYGNSS land data have been analysed within the project “Potential of Spaceborne GNSS-R for Land Applications” funded by ESA, showing among others a significant correlation with above ground biomass.
In this study, we attempt to exploit the potential of CYGNSS data for estimating the forest biomass on a global scale, by implementing a retrieval algorithm based on Artificial Neural Networks (ANN). ANN can be regarded as a statistical approach aimed at providing the minimum variance solution to the given problem: ANNs have been demonstrated able to reproduce almost any kind of input-output relationship and they have been successfully applied to several remote sensing problems [3].
The proposed algorithm is trained using a subset of the data derived ([4], [5]), then the biomass retrievals are validated on the remaining part of the dataset.
After validation, the algorithm is applied to generate maps of biomass over the latitudinal band covered by CYGNSS, with a resolution of 0.01°, which agrees well with the existing biomass maps [4-5].
The results open new perspective of measuring and monitoring biomass using CYGNSS and other existing (i.e. TechDemoSat-1) and forthcoming (i.e. FSSCAT) missions, and of integrating these valuable results from GNSS-R with the other standard measurement methods, to ultimately improve the available global maps of biomass. Notably, the high sampling of GNSS-R could also offer further insights into the temporal changes and variability of biomass, something that is not captured at present by any other existing satellite.




REFERENCES

[1] Kumar, L. and O. Mutanga, “Remote Sensing of Above-Ground Biomass”, Remote Sensing, 2017, 9, 935, doi:10.3390/rs9090935.

[2] Ruf, C., R. Atlas, P. Chang, M. Clarizia, J. Garrison, S. Gleason, S. Katzberg, Z. Jelenak, J. Johnson, S. Majumdar, A. O’Brien, D. Posselt, A. Ridley, R. Rose, V. Zavorotny (2015), “New Ocean Winds Satellite Mission to Probe Hurricanes and Tropical Convection,” Bulletin of American Meteorological Society, doi:10.1175/BAMS-D-14- 00218.1.

[3] Santi, E., S. Paloscia, S. Pettinato, G. Fontanelli. Application of artificial neural networks for the soil moisture retrieval from active and passive microwave spaceborne sensors. Int. J. Appl. Earth Observ. Geoinf. 2016, Volume 48, Pages 61–73. http://dx.doi.org/10.1016/j.jag.2015.08.002.

[4] Avitabile V, Herold M, Heuvelink G, Lewis SL, Phillips OL, Asner GP et al. (2016). An integrated pan-tropical biomass maps using multiple reference datasets. Global Change Biology, 22: 1406–1420.doi:10.1111/gcb.13139.

[5] Santoro, M., Beaudoin, A., Beer, C., Cartus, O., Fransson, J. E. S., Hall, R., Pathe, C., Schmullius, C., Shvidenko, A., Schepaschenko, D., Thurner, M., Wegmuller, U., “Forest growing stock volume of the northern hemisphere: an estimate for 2010 derived from Envisat ASAR data ” Remote Sensing of Environment, vol. 168, pp. 316-334, 2015.

Area A - Board 85

The boreal and temperate biome contains the largest forest areas of our planet (about 46% of the global forest area) and represents about one third of the Earth’s carbon sink. The temperate zone includes major part of densely populated areas and most intensively managed forests of the planet. The most severe climate change is expected in the boreal zone, and current climate change science considers boreal forests as a potential tipping element, particularly if critical scenarios of climate change (like RCP6.5 and RCP8.5) would be realized. All together requires availability of operative and reliable monitoring of dynamics of forest productivity (biomass) and impacts of forests on global carbon cycling. Such a system could only be developed based on a system integration of “multi” remote sensing information (particularly radars) and relevant ground based data.
The major objective of this work is to extend biomass retrieval algorithms developed in central Siberia for the Core CCI Biomass project to the pan-boreal region and test their viability. Central Siberia had been chosen due to the availability of excellent (albeit old for its major part) forest inventory data. The algorithm development in the Core Project is based on analysis of hyper-temporal Sentinel-1 SAR data for the years 2014 - 2017. Besides analysing the relation between biomass data and radar backscatter, it will also involve
derivation of interferometric coherence and their difference between unfrozen sommer and frozen winter conditions as well as on multi-temporal metrics.
The extension of the retrieval algorithms to the overall boreal and temperate biome is based on regionalization of the territory in homogeneous ecoregions (here - for the pan-boreal domain) and in-depth analysis at ten selected test areas, each of the size of one Sentinel-1 Interferometric Wide-Swath scene, which will be defined with associated partners.
Underestimation of AGB of high productive forests is a common problem in applications of short-wave radars. It requires special corrections of the signal behind the thresholds of saturation. In essence, such corrections could be only regional because they are dependent upon the diversity of numerous geographical, landscape and biometric specific features of forest ecosystems.
In this presentation we will report for a dedicated site how, 1) principles of such a regionalization will be suggested based on analysis of existing forest regionalizations and specific needs of radar applications; 2) draft version of
such a regionalization (upper level) will be suggested for the pan-boreal domain; 3) in-depth analysis of dependence of segmentation of radar imagery on characteristics of forest ecosystems will be provided; 4) systems of necessary auxiliary regional forest ecosystem models will be presented for selected test areas.

Area A - Board 86