In this study, we examine the potential of continuous ground moisture monitoring over a mining site using a combination of in-situ soil moisture sensors and multi-sensor SAR images. We focus on examining and improving methodologies for surface soil moisture (or ground moisture) retrieval from SAR measurements focusing on detailed in situ reference observations for several key sediments types in the study area. The mining site represents a limestone quarry located in southeastern Finland. We hypothesize that sediment-specific well-calibrated models can be instrumental in improving soil moisture retrieval under different weather conditions to produce spatially explicit ground moisture estimates at high resolution compared to baseline approaches. Studied SAR data are represented by Copernicus Sentinel-1 C-band images, and methodologies will be expanded later to L-band images from ALOS-2 PALSAR-2 and the upcoming NISAR mission.

The demonstrated efficacy of interferometric synthetic aperture radar (InSAR) techniques has spurred the development of innovative SAR satellite missions like BIOMASS and NiSAR, poised for extensive application in forest monitoring. Nevertheless, prevailing methodologies for retrieving forest variables, including forest height and above-ground biomass, encounter substantial limitations. Traditionally, successful forest mapping necessitates a non-zero spatial perpendicular baseline, full polarimetry, and a relatively small (close-to-zero) temporal baseline. This study presents a novel approach for extracting forest biophysical variables by modeling the temporal decorrelation of repeat-pass InSAR coherence. We explore a hypothesis regarding the potential relationship between the temporal decorrelation of InSAR coherence and forest variables, such as tree height and aboveground biomass. This hypothesis is tested across diverse test sites in Finland, Canada, and Germany. Our findings suggest a viable means of extracting forestry information by quantifying the temporal decorrelation of C-Band InSAR coherence. We establish a clear connection between the temporal decay rate and crucial forest variables, such as forest above-ground biomass and tree height

In this study, we examine the potential of several self-supervised deep learning models in predicting forest attributes and detecting forest changes using ESA Sentinel-1 and Sentinel-2 images. The performance of the proposed deep learning models is compared to established conventional machine learning approaches. Studied use-cases include mapping of forest disturbance (windthrown forests, snowload damages) using deep change vector analysis, forest height mapping using UNet+ based models, Momentum contrast and regression modeling. Study areas were represented by several boreal forest sites in Finland. Our results indicate that developed methods allow to achieve superior classification and prediction accuracies compared to traditional methodologies and mimimize the amount of necessary in-situ forestry data.

Accurate mapping of forests is critical for forest management and carbon stocks monitoring. Deep learning (DL) is becoming more popular in Earth observation (EO), however, the availability of reference data limits its potential in wide-area forest mapping. To overcome those limitations, here we introduce contrastive regression into EO-based forest mapping and develop a novel semisupervised regression framework for wall-to-wall mapping of continuous forest variables. It combines supervised contrastive regression loss (CtRL) and semi-supervised cross-pseudo regression (CPR) loss. The framework is demonstrated over a boreal forest site using Copernicus Sentinel-1 and Sentinel-2 imagery for mapping forest tree height. Achieved prediction accuracies are strongly better compared to using vanilla UNet or traditional regression models, with relative root mean square error (rRMSE) of 15.1% on stand level. We expect that the developed framework can be used for modeling other forest variables and EO datasets.

ESA Forest Carbon Monitoring project (FCM) is developing Earth Observation based, user-centric approaches for forest carbon monitoring. Forest carbon accounting based on forest inventory requires precise and timely estimation of forest variables at various spatial levels accompanied by verifiable uncertainty information. In this paper, we present the algorithm trade-off and selection approach and preliminary results of the algorithm intercomparison exercise in the FCM project. The studies were performed over 7 European test sites located in Finland, Ireland, Romania, Spain and Switzerland, and one tropical forest site in Peru. EO datasets were represented by Sentinel-1, Sentinel-2, TanDEM-X and ALOS-2 PALSAR-2 imagery. Examined approaches include popular parametric and SAR/InSAR scattering physics based approaches, and nonparametric and machine learning approaches such as k-NN, random forests, support vector regression.

In this study, we evaluate the potential of deep learning models in predicting forest tree height in boreal forest zone using ESA Sentinel-1 and Sentinel-2 images. The performance of studied deep learning models is compared to several popular conventional machine learning approaches. The study area is located near Hyytiala forestry station in Finland, and represents a conifer-dominated mixed boreal forestland. Improved predictions were obtained when using combined optical and SAR data for all studied models. Our results indicate that UNet based models can achieve better accuracy in predicting forest tree heights (RMSE of$1.90m,\ \mathrm{R}^{2}$of 0.69), compared to traditional parametric and machine learning models with RMSE range of$2.27-2.41m$and$\mathrm{R}^{2}$range of 0.50-0.56 when satellite optical and radar data are combined.

In this study, we evaluate the performance of several imaging modes of X-band satellite SAR data in mapping various forest logging operations. We use stripmap and spotlight data acquired by DLR's TanDEM-X mission and ICEYE SAR satellites. Also, the potential of InSAR coherence data is evaluated. Experiments are performed over the study site located in the central part of Finland. Reference data are represented by forest management plans and harvester logging tracking reports. Our results indicate that not only clear-cutting, but also forest thinning operations can be detected using advanced imaging modes of satellite SAR data even at X-band, primarily using high resolution imaging modes and interferometric SAR capability.

Here, we examine a deep learning approach to perform land cover classification using country-wide SAR mosaics compiled using multitemporal Sentinel-1 imagery. We capitalize on our earlier study [1], demonstrating the suitability of deep learning models for land cover mapping using satellite C-band SAR images. A set of SAR mosaics compiled from consecutive Sentinel-1 IW mode acquisitions covering the whole territory of Finland was used in production of the whole-country land cover map. The imagery were used as an input to the state-of-the-art deep-learning model for semantic segmentation called FC-DenseNet. This model was pre-trained on the ImageNet dataset and further fine-tuned in this study. CORINE land cover map was used as a reference, and the model was trained to distinguish between 5 Level-1 CORINE classes. Upon the evaluation and benchmarking, we found that the FC-DenseNet model is able to achieve nearly 90% overall classification accuracy. These results indicate the suitability of deep learning approaches to support efficient operational wide-area mapping using satellite SAR imagery.

In this study, we assess the potential of long time series of Sentinel-1 SAR data in forest growing stock volume (GSV) estimation. The study site with 17,762 forest stands is located near the Hyytiälä forestry field station in Finland and represents the boreal coniferous forest. Altogether 96 images spanning more than three years of observations have been studied using linear and random forest regression approaches. Our analysis demonstrates considerable decrease in the prediction errors of GSV as the the number of input scenes increases. The use of feature extraction and dimensionality reduction techniques allows to achieve to nearly optimal performance already with 10 scenes. While the GSV prediction errors using individual Sentinel-1 scenes varied considerably from 86 to 93 m3/ha, the prediction accuracy with combined scenes improved to 76 m3/ha (44.9%) RMSE.

The ICEYE constellation features the first operational microsatellite based X-band SAR sensors suitable for all weather day-and-night Earth Observation. In this paper, we report on the status of the ICEYE Constellation and describe the characteristics of the first operational imaging modes.

To date, the potential of multitemporal interferometric SAR (InSAR) data in land-cover mapping has not been fully explored despite suitable time series increasingly acquired from SAR sensors. Here, we suggest to use an LSTM (Long Short Term Memory) based land-cover classifier to address this problem. Spatial context is preserved by using grey-level spatial dependencies and morphological profiles. Further, a 4-LSTM-based model was trained to capture the temporal dynamics of InSAR coherence. Altogether 39 Sentinel-1 interferometric coherence pairs acquired over Donana in Spain were used to evaluate the method performance. Achieved more than 90% overall accuracy indicates the strong potential of developed InSAR recurrent approach in improving differentiation between various land cover classes.

In this study, a set of bistatic interferometric SAR images acquired by the TanDEM-X mission are used along with fully polarimetric ALOS PALSAR data for the assessment of tropical forest properties in Nepal. Research to be presented at the conference concentrates on several scientific goals. First, location of interferometric scattering phase centre inside forest canopy is investigated using reference ALS measured canopy height model. Means for forest tree height extraction using both model based approaches (similar to Random Volume over Ground) and semi-empirical models are investigated and reported. It is shown, that forest tree height retrieval is possible with RMSE around 2.8 meters (R2=0.68). Secondly, correlations between ALOS PALSAR and tropical forest data are analysed, and AGB estimation using fully polarimetric SAR features is performed. Several statistical and non-parametric methodologies were tested and compared. Thirdly, forest AGB estimation is done using both TanDEM-X based tree height and L-band PolSAR features using statistical inversion framework.

Here, we investigate multiple ways of assimilating synthetic aperture radar (SAR) data to L-band Microwave Emission of the Biosphere (L-MEB) model to enhance the model performance over forested areas in the boreal zone. Land C-band satellite SAR backscatter data, X -band interferometric SAR coherence, as well as auxiliary data layers from forest authorities are used as a proxy in calculating the forest transmissivity, instead of traditionally used leaf area index (LAI) parameter. Our earlier experiments have shown, that when particularly ALOS PALSAR (L-band) and multitemporal composite Sentinel-l (C-band) data were applied, an improved agreement was achieved between the measured and simulated brightness temperatures (TBs) over forests. Here, we extend our analysis and examine data acquired by ALOS PALSAR, ESA Sentinel-l, and TanDEM-X mission of DLR, as well as several other auxiliary datasets on forest parameters. Our proposed model based approach indicates the potential of an SAR-based estimation of forest volume transmissivity and represents a viable way of active-passive microwave satellite data fusion and incorporating readily available reference data.

Single pass radar interferometry (sp-InSAR) is a well established technique for generation of digital elevation models (DEM). Differencing two DEMs acquired at different times can reveal topographic changes. However snow depth estimation by DEM differencing is still an ongoing topic in radar research: in contrast to snow free surfaces, the snow surface elevation is difficult to detect either because of microwave penetration into dry snow or because of the weak backscatter return from wet snow which significantly decorrelates the interferometric signal. In this study we demonstrate first results of wet snow depth estimation by differencing sp-InSAR DEMs acquired by the TanDEM-X satellite mission. The results show, in contrast to dry snow, a clear sensitivity to wet snow. However, additionally to a high vertical sensitivity of a few ten centimeters a very low noise-equivalent-sigma-zero (NESZ) is crucial for successful snow depth estimation.

In this study, bistatic interferometric Synthetic Aperture Radar (InSAR) data acquired by the TanDEM-X mission were used for automated classification of sea ice over the Baltic Sea, in the Bothnic Bay. A scene acquired in March of 2012 was used in the study. Backscatter-intensity, coherence-magnitude and InSAR-phase, as well as their different combinations, were used as informative features in several classification approaches. In order to achieve the best discrimination between open water and several sea ice types (new ice, thin smooth ice, close ice, very close ice, ridged ice, heavily ridged ice and ship-track), Random Forests (RF) and Maximum likelihood (ML) classifiers were employed. The best overall accuracies were achieved using combination of backscatter-intensity & InSAR-phase and backscatter-intensity & coherence-magnitude, and were 76.86% and 75.81% with RF and ML classifiers, respectively. Overall, the combination of backscatter-intensity & InSAR-phase with RF classifier was suggested due to the highest overall accuracy (OA) and smaller computing time in comparison to ML. In contrast to several earlier studies, we were able to discriminate water and the thin smooth ice.

In this study we compare semi-empirical interferometric coherence models, proposed in [1], for tree height estimation from TanDEM-X coherence scenes. The models are derived from Random Volume over Ground model, by applying simplifications and introducing empirical parameters at different complexity levels so that the models can be adapted to available ancillary data. Several different TandDEM-X interferometric scenes from Estonia are used to test the model performance in various conditions. All the results are compared with highly accurate canopy height models measured using airborne laser scanning. We demonstrate that models which are very simple to invert, produce accurate tree height estimates when the conditions are most favorable. Best results can be seen for winter images for frozen and dry snow conditions. Simple parametric sinc model can produce accurate tree height maps over large areas with pixel-wise deviation only few meters.

Vegetation optical depth and scattering albedo are crucial parameters within the widely used τ-ω model for passive microwave remote sensing of vegetation and soil. A multi-sensor data integration approach using ICESat lidar vegetation heights and SMAP radar as well as radiometer data enables a direct retrieval of the two parameters on a physics-derived basis. The crucial step within the retrieval methodology is the calculus of the vegetation scattering coefficient KS, where one exact and three approximated solutions are provided. It is shown that, when using the assumption of a randomly oriented volume, the backscatter measurements of the radar provide a sufficient first order estimate and subsequently lead to effective estimates of vegetation optical depth and scattering albedo acquired with the novel multi-sensor approach.

This letter proposes a novel way to assimilate synthetic aperture radar (SAR) data to L-band Microwave Emission of the Biosphere (L-MEB) model to enhance model performance over forested areas. L- and C-band satellite SAR data are used in order to characterize the forest transmissivity within the emission model, instead of the optical satellite imagery-based leaf area index (LAI) parameter. Examination of several combinations of satellite SAR data as a substitute for LAI within the L-MEB model showed that when ALOS PALSAR (L-band) and multitemporal composite Sentinel-1 (C-band) data are applied, an improved agreement was achieved between the measured and simulated brightness temperatures (TBs) over forests. The root mean squared difference between modeled and measured TBs was reduced from 6.1 to 4.7 K with single PALSAR scene-based transmissivity correction and down to 4.1 K with multitemporal Sentinel-1 composite-based transmissivity correction. Suitability of single Sentinel-1 scenes varied based on seasonal and weather conditions. Overall, this indicates the potential of an SAR-based estimation of forest volume transmissivity and opens a possible way of fruitful active-passive microwave satellite data integration.

Clear-cutting and logging operations are the most drastic and wide-spread changes that affects the hydrological and carbon-balance properties of forested areas. A long time series of Sentinel-1 images are used to study the potential for mapping logged areas in areas in boreal zone region and in tropical forest. In the first case study in southern Finland, the time series covered a full year starting in October 2014, in 200km-by-200-km study site. The Sentinel-1 images were acquired in Interferometric Wide-swath (IW), dualpolarized mode (VV+VH). All scenes were acquired in the same orbit configuration. In the second case study the potential of Sentinel-1 time series for mapping logged areas was studied over tropical forest in Mexico. Acquisitions were made in the time frame between November 2014 and September 2015. The temporal behavior of the C-band backscatter was studied for areas representing: 1) areas clear-cut during the acquisition of the Sentinel-1 time-series, 2) areas remaining forest during the acquisition of the Sentinel-1 time-series, and 3) areas that had been clear-cut before the acquisition of the Sentinel-1 time-series. Algorithms for mapping the spatial extent of logged areas were developed and tested, showing potential of long time series of Sentinel-1 data for successful delineation of clear-cuts despite high sensitivity to seasonal and weather conditions.

In this work we provide basic building blocks for semi-empirical models to be applied mainly for forest height extraction from X-band interferometric SAR images. The work uses Random Volume over Ground model as the main theoretical framework, and relies on the measurement data represented by over 3000 measurements points collected in Estonia in 2011 and 2012. Here we demonstrate that the best argument for empirical models which relate coherence and forest parameters is relative interferometric tree height (tree height divided by InSAR Height of ambiguity). Our results suggest that a very simple linear model with no additional a priori parameters can be used as a first approach for estimation of forest height. However, if more extensive dataset are available, a zero extinction model can provide improvement. Moreover, proposed semi-empirical models can also be used to predict forest properties related to forest extinction coefficient. All the derived model approximations are demonstrated by model simulations and verified with extensive dataset of forest measurements. Relation of semi-empirical parameters to physics based model parameters is discussed and the models accuracy is analyzed based on empirical dataset.

In this paper, we propose a new approach for improving boreal forest soil moisture estimation using L-band microwave radiometer. The effect is achieved by introducing improved description of forest canopy contribution from multisensor SAR measurements. Spaceborne L-band radiometer is a valuable tool for providing soil moisture estimates globally. Unfortunately, complex vegetation layer, such as forest, can hamper the accuracy of soil moisture retrieval leading to rather poor results particularly over boreal forest areas. Currently, the L-band Microwave Emission of the Biosphere (L-MEB) model adopted in the Soil Moisture and Ocean Salinity (SMOS) Level 2 Soil Moisture algorithm, uses Leaf Area Index (LAI) in order to to account for forest canopy contribution to total emission. However, it can argued that LAI presents poorly the actual structure of the coniferous forest. The LAI is calibrated to represent only the leaves, but at L-band, the main contribution to emission and attenuation is due to branches, while trunks and leaves have smaller effects. Here, we tested several combinations of spaceborne SAR data as a substitute of LAI in temperature brightness models for soil moisture retrieval. Particularly when L-band ALOS PALSAR stripmap data were used, the agreement between modelled and measured TB has improved from 0.46 to 0.55 in the L-MEB model.

In this study, we assess the potential of combining forest tree height derived from interferometric SAR data with satellite optical data for improving accuracy of forest stem volume mapping. Study site was located near the Hyytiälä forestry station in central Finland, with terrain representative of the boreal coniferous forest. As a primary interferometric data, several data takes of TanDEM-X data are used. Satellite optical data were represented by Landsat 8 image. The ground reference data were information on stand level from forest management plans. Firstly, forest tree height is estimated from TanDEM-X interfer ometric SAR coherency. Further, retrieved tree heights are combined with optical data for predicting forest stem volume using linear regression framework. Results of regression analysis performed demonstrate considerable improvement in terms of obtained accuracy figures of the combined stem volume estimation (RMSE = 34%, R2=0.57) compared with use of optical satellite data only (RMSE=40%, R2=0.43).

In this paper, potential of space-borne SAR data for monitoring selective logging operations in the tropical forest areas was assessed. Two separate studies were organized for this purpose. Both study sites were located in the northern part of the Republic of the Congo. In the first study, bi-temporal mosaics of ALOS PALSAR data were used in order to map areas affected by logging operations. Data were collected in 2007-2010. Development is in line with other studies and shows promising potential. In the second study, a time series of strip-map C-band SAR data for detecting and monitoring of selective logging activities was assessed. The technique primarily uses multi-temporal aggregation of orthorectified SAR imagery acquired before and after the forest disturbance, followed by the analysis of textural features of SAR backscatter temporal log-ratio image. This is the first successful demonstration of C-band SAR based mapping of selectively logged areas.

This is a selection of results of the North State project, that demonstrate how innovative methods applied to the new Sentinel data streams can be combined with models to monitor carbon and water fluxes for pan-boreal Europe.

This work investigates applicability of spaceborne repeat-pass interferometric SAR images for detecting mowing events on agricultural fields. Four pairs of one-day repeat-pass COSMO-SkyMed acquisitions were analysed and compared to in situ measurements of 11 agricultural grasslands to study the potential of X-band temporal interferometric coherence for detecting mowing events. Field works covered 11 test plots in Central Estonia with varying species composition and homogeneity. Temporal decorrelation due to changes in the grass height and wet biomass was analysed. A nonlinear relationship was observed between the wet biomass and temporal coherence, as well as between the grass height and the temporal coherence. Our results show that one-day temporal coherence decreases as the grass height and the wet biomass increases, until reaching a noise level at 25 cm and 400 g, respectively. The current study shows that detecting mowing event from multitemporal interferometric SAR images is a feasible technique and could be used for monitoring applications on the European level.

This paper evaluates performance of fully polarimetric SAR (PolSAR) data in several land cover mapping studies in the boreal forest environment, taking advantage of the high canopy penetration capability at L-band. The studies included multiclass land cover mapping, forest-nonforest delineation, and classification of soil type under vegetation. PolSAR data used in the study were collected by the ALOS PALSAR sensor in 2006-2007 over a managed boreal forest site in Finland. A supervised classification approach using selected polarimetric features in the framework of probabilistic neural network (PNN) was adopted in the study. It has no assumptions about statistics of the polarimetric features, using nonparametric estimation of probability distribution functions instead. The PNN-based method improved classification accuracy compared with standard maximum-likelihood approach. The improvement was considerably strong for soil type mapping under vegetation, indicating notable non-Gaussian effects in the PolSAR data even at L-band. The classification performance was strongly dependent on seasonal conditions. The PolSAR feature data set was further modified to include a number of recently proposed polarimetric parameters (surface scattering fraction and scattering diversity), reducing the computational complexity at practically no loss in the classification accuracy. The best obtained accuracies of up to 82.6% in five-class land cover mapping and more than 90% in forest-nonforest mapping in wall-to-wall validation indicate suitability of PolSAR data for wide-area land cover and forest mapping.

The paper describes evaluating the potential of peatland detection under forest canopy with L-band space borne quad-polarization data in the boreal forest zone. Particular emphasis was made on under what seasonal conditions this detection was possible using single SAR data-take. For this purpose multitemporal ALOS PALSAR imagery acquired over Kuortane test site in central Finland during 2007-2008 was used. Supervised classification experiments employing selected polarimetric features were performed using standard maximum likelihood approach and probabilistic neural network (PNN). Strong non-gaussianity effects were noted, with better performance demonstrated by PNN, utilizing non-parametric estimation of probability distributions of the respective polarimetric features. Suitability of several techniques aimed at compensating the presence of forest canopy was studied as well.

The paper describes algorithm development for tree height retrieval in the boreal forest zone from TanDEM-X interferometric imagery. A set of 8 TanDEM-X pairs was acquired during summer and autumn 2011 over southern Finland in order to evaluate the potential tree height retrieval performance for this space-borne instrument. Another focus of the study was evaluation of seasonal dependence of interferometric signature of boreal forest. The obtained results are compared to our previous studies on tree height retrieval with the airborne DLR E-SAR instrument in the same area. The obtained results show good potential of TanDEM-X in forest mapping when external terrain elevation model is available, though accuracy seems to be somewhat lower compared to airborne instruments due to increased noise.

This letter describes an algorithm development for the production of a large-scale fully polarimetric synthetic aperture radar (SAR) (PolSAR) mosaic using multitemporal Advanced Land Observing Satellite Phased Array type L-band SAR acquisitions. The PolSAR data were collected during the snow-melting season in 2007 over Finnish Lapland, resulting in considerable radiometric differences between mosaiced scenes originating at different dates. Several variants of polarimetric seam hiding between the original PolSAR images were proposed and evaluated in order to effectively eliminate stripes in the mosaic. The impact of such seam-hiding procedure on PolSAR classification performance was studied, along with the technical aspects of producing the PolSAR mosaic. The obtained results indicate the advantages of the considered seam-hiding procedures for producing homogeneous mosaics and obtaining consistent classification results in a single classification step.

Scattering phase center (SPC) location in boreal forests was studied in order to assist forest inventory with single- and quad-pol synthetic aperture radar (SAR) interferometry. Airborne X- and L-band interferometric SAR data collected by the DLR E-SAR instrument in southern Finland during the FINSAR campaign was used in the study. A simple Random Volume over Ground (RVoG) model was employed as the theoretical framework for inversion of forest parameters and interpretation of the obtained results. LIDAR measurements of the canopy height and terrain elevation were used as reference and auxiliary data. The RVoG model was found to satisfactorily explain the SPC location inside the canopy in boreal forests. We show that when using X-band, the height of the SPC is typically about 75% of the canopy height, as predicted by the RVoG model; however, the retrieved extinction was found to be rather low. The feasibility of highly accurate tree height inversion using single-polarization X-band interferometry (with RMSE approaching 1.5 m) is demonstrated using a digital terrain model. For this purpose, the traditional polarimetric interferometry SAR technique for phase center retrieval is modified to include a complementary LIDAR measured terrain model. At L-band, the phase center height was determined to be around 50% of the canopy height and even lower, indicating that the ground contribution is significant. Moreover, several simplified inversion approaches for tree height and extinction coefficient retrieval were considered based on several boundary cases of the RVoG model, describing the canopy frequently encountered in boreal forest environments. These analyses allowed developing a combined approach for simultaneous estimation of both forest height and extinction in the boreal zone when an accurate elevation model of the terrain is available.

Model-based approaches for decomposing polarimetric backscatter data from boreal forest are discussed in this paper. Several model-based decompositions are analyzed with respect for the most accurate estimation of the volume scattering component. A novel generalized model for description of the volume contribution is proposed when observed backscatter from forest indicates that media does not follow azimuthal symmetry case. The model can be adjusted to the polarimetric synthetic aperture radar (PolSAR) data itself, taking into consideration higher sensitivity of HH against VV backscattering term to the presence of canopy at L-band. The model is general enough to allow a broad range of canopies to be modeled and is shown to comply with several earlier proposed volume scattering mechanism models. It is afterward incorporated in the Freeman-Durden three-component decomposition, yielding an improved modification. The performance of the proposed modification is evaluated using multitemporal ALOS PALSAR data acquired over Kuortane area in central Finland, representing typical mixed boreal forestland. Several decompositions are also benchmarked in order to see how they satisfy physical requirements when decomposing covariance matrix into a weighted sum of individual scattering mechanism contributions. When using experimental data, the proposed decomposition is shown to better satisfy non-negativity constraints for the covariance matrix eigenvalues at each decomposition step with less additional PolSAR data averaging needed. Discussed decompositions are also evaluated for the accuracy of initial stratification based on dominating scattering mechanism using ground reference data.