Crop phenology retrieval via polarimetric SAR decomposition and Random Forest algorithm

doi:10.1016/j.rse.2019.111234

Remote Sensing of Environment

Volume 231, 15 September 2019, 111234

https://doi.org/10.1016/j.rse.2019.111234 Get rights and content

Highlights

•
Volume scattering mechanism was extracted using polarimetric decompositions.
•
Random Forest algorithm was coupled with volume related polarimetric parameters.
•
Different machine learning approaches were evaluated for phenology retrievals.
•
Crop phenology retrieval methods were developed during SMAPVEX16-MB.

Abstract

Knowledge of crop phenology assists in making agricultural decisions such as appropriate irrigation and fertilization applications in order to optimize crop yield. The objective of this study is to monitor crop phenology using Synthetic Aperture Radar (SAR) polarimetric decompositions and a random forest algorithm applied to a multi-temporal RADARSAT-2 dataset, acquired during the Soil Moisture Active Passive (SMAP) Validation Experiment 2016 in Manitoba (SMAPVEX16-MB). The model-based and eigen-based polarimetric parameters are used to separate the vegetation and soil scattering contributions in the total radar signal. As the crop morphological shape and structure vary with phenological growth, our study assumes that the polarimetric parameters related to the volume scattering mechanism have the potential to track the crop phenology. The sensitivity of the polarimetric parameters to the ground identified crop phenology is analyzed for different crop types. For canola, a single polarimetric parameter is sufficient to characterize the crop phenology, due to the high volume scattering power and large temporal dynamic. For corn, soybean and wheat, combinations of multiple polarimetric parameters are required. For each crop type, the Random Forest algorithm trained using 60% of the data is used to retrieve the crop phenology. Performances are compared to Artificial Neural Network, Support Vector Machine Regression and k-Nearest neighborhood algorithms. The Random Forest algorithm provides the best phenology retrieval with significant (p-value < 0.01) spearman correlation coefficients (between the retrieved and ground identified phenology) of 0.93, 0.90, 0.85 and 0.91 for canola, corn, soybean and wheat, respectively. While a single polarimetric parameter demonstrates limited sensitivity to corn phenology, the retrieved phenology from the Random Forest algorithm using multiple polarimetric parameters agrees well with the ground measurements. Furthermore, the importance of different polarimetric parameters for phenology retrieval using the Random Forest algorithm is quantified for different crop types. These findings will be of interest in developing future analytical retrieval models.

Introduction

Knowledge of crop growth stage, or phenology, is integral to many aspects of precision agriculture. To be most effective, agricultural management decisions, such as irrigation and applications of fertilizers and pesticides, are often timed to specific crop phenology stages. Furthermore, understanding crop growth progression from germination to senescence, provides the agri-food industry and government organizations with vital information to better inform crop productivity forecasts in support of marketing and agricultural policy decision making (Ceglar et al., 2019). In addition, growth progression is affected by a changing climate with a warmer climate altering the duration of crop growth cycles as well as the timing of phenological events. Such variations necessitate adjustments to farming activities to mitigate any reductions in agricultural production (Pulatov et al., 2015). Consequently, knowledge of the spatial and temporal variability of crop phenology assists the agricultural sector in making the most appropriate farming decisions.

Polarimetric Synthetic Aperture Radar (PolSAR) signals are sensitive to crop morphological structure. As crops progress from one growth stage to the next, this structure changes, at times, dramatically. The availability of PolSAR data from satellites such as ALOS-PALSAR, RADARSAT-2 and TerraSAR-X, offers opportunity to investigate the potential of these data for crop phenology retrieval. Higher frequency SARs are advantageous for crop phenology monitoring. Scattering at shorter wavelengths is dominated by contributions from the vegetation layers, with less interference from the underlying soils. Thus, most crop phenology retrievals have focused on C– or X-band sensors.

Lopez-Sanchez et al. (2012) investigated retrieval of rice phenology using simple decision tree algorithms applied to dual-polarized TerraSAR-X data. Five phenology intervals (early vegetative, plant emergence, advanced vegetative, reproductive, and maturation) were discriminated based on the thresholds of two polarimetric observables derived from a dual-polarized target decomposition. Polarimetric RADARSAT-2 data were also used by Lopez-Sanchez et al. (2014) to retrieve rice phenology, using hierarchical decision trees based on thresholds of three polarimetric parameters (entropy, anisotropy and α angle) derived from the Cloude-Pottier decomposition. The same authors also simulated compact-polarized C-band SAR, reporting a similar retrieval accuracy as with the fully-polarized SAR data. Vicente-Guijalba et al. (2014) used a Kalman filter to dynamically assimilate a geometrical model and observation data to better infer rice phenology from TerraSAR-X time series. Yuzugullu et al. (2015) applied K-means clustering approaches to the TerraSAR-X co-polarized backscattering coefficients (σ_HH⁰, σ_VV⁰) and polarimetric phase differences, in order to group similar rice phenological structures. Morphology-based radar backscattering models were also proposed to estimate the phenology of rice from TerraSAR-X data (Yuzugullu et al., 2017).

Early investigations were focused on the retrieval of rice phenology, while other globally important crops such as canola, corn, soybean and wheat have not been intensively studied. Crop structural changes in response to phenological development vary widely among different crop types. Jiao et al. (2014) found that changes in canola growth stages were captured by the cross-polarized backscatter coefficient and the polarimetric entropy. Mascolo et al. (2015) used RADARSAT-2 data to distinguish three phenology intervals of onion, based on the cross-polarized backscatter coefficient along with the entropy or HH-VV coherence. Furthermore, the complex Wishart classifier was used to determine the phenology intervals of oat, barley, wheat and corn from RADARSAT-2 data, based on an automatic approach which makes use of the distances among the covariance matrices in a time series (Mascolo et al., 2016). This method did not require to predefine the phenology intervals as described in Lopez-Sanchez et al. (2012), thus providing a more generalized approach without the need for any polarimetric decomposition.

The aforementioned studies considered phenology retrieval as a classification problem, estimating phenology intervals rather than determining phenology advancement along a numerical scale. The Biologische Bundesanstalt, Bundessortenamt and CHemical (BBCH), as an example, is a widely used scale applied to phenology characterization of a wide range of crops (Bleiholder et al., 2001). For canola and spring wheat, linear regression equations were fitted between the polarimetric variables obtained from multi-temporal RADARSAT-2 data and crop height, Leaf Area Index (LAI) and Days-After-Seeding. Based on these linear equations, crop growth parameters were estimated and then used to infer the crop phenology (Canisius et al., 2018). However, this indirect retrieval of BBCH phenology is highly dependent on the relationships between the estimated growth parameters and BBCH for different crop types. Furthermore, a dynamic filtering algorithm was developed to recursively retrieve canola phenology using RADARSAT-2 and TerraSAR-X data (McNairn et al., 2018). These researchers used polarimetric parameters to calculate a novel growth state variable, which was then transformed to BBCH phenology.

Within this context, our study approaches crop phenology retrieval using polarimetric decomposition and Random Forest (RF) algorithms. The RF has been used to classify crops using airborne PolSAR data (Hariharan et al., 2018), and to estimate winter wheat growth parameters from dual-polarized Sentinel-1 data (Kumar et al., 2018). But to our knowledge, it has not yet been applied to crop phenology retrieval. In current study, multi-temporal RADARSAT-2 data and phenology measures taken during the Soil Moisture Active Passive (SMAP) Validation Experiment 2016 in Manitoba (SMAPVEX16-MB) campaign are used to develop an algorithm to retrieve phenology for canola, corn, soybean and wheat. According to Lopez-Sanchez et al. (2012), a single radar descriptor is not sufficient to retrieve rice phenology, as it is difficult to establish a direct relationship between phenology and one polarimetric observable. Thus, we focused on multiple polarimetric parameters related to the volume scattering mechanism. However, given potential correlations between polarimetric parameters, a sensitivity analysis to crop phenology was first undertaken, followed by an ensemble estimation using RF. This algorithm reduces redundancy among ensembles of polarimetric parameters, by averaging the estimates from multiple decision trees built using randomly selected polarimetric observables.

Section snippets

Study site

The SMAPVEX16-MB campaign was conducted over an agricultural test site (Fig. 1) measuring 26 km × 48 km and located in southern Manitoba (Canada). Numerous SAR experiments have been conducted over this site, spanning several decades (Bhuiyan et al., 2018). The site is dominated by annual agricultural production with a diverse crop mix that includes canola, corn, soybean, wheat and oats.

RADARSAT-2 data acquisitions

During the SMAPVEX16-MB campaign in 2016, Fine Quad-polarized Wide (FQW) RADARSAT-2 data were acquired on

Method

In this study, our goal is to retrieve and validate crop phenology from a time series of polarimetric RADARSAT-2 data. As the structure (shape, size and orientation of plant organs) varies among crops, this study assumes that polarimetric parameters with sensitivity to the volume scattering mechanism will be of most interest for determining phenology.

The flowchart in Fig. 4 shows the process for crop phenology retrieval. The coherency matrix [T3] was extracted from the RADARSAT-2 SLC products

Crop phenology analysis

The SMAPVEX16-MB campaign began when crops were in their early leaf development (BBCH10), finishing as crops were senescing (BBCH90). As field sampling did not always fall on the same day as the RADARSAT-2 acquisitions, a linear interpolation was used to create a temporally continuous curve of crop development. A constraint on the interpolation was that BBCH phenology for a given field always increased during the growth cycle.

Fig. 5 shows the temporal evolution of the interpolated BBCH

Conclusion

This study investigates the accuracy of crop phenology retrieval using C-band polarimetric parameters, coupled with Random Forest (RF), Artificial Neural Network (ANN), Support Vector Machine Regression (SVMR) and k-Nearest Neighborhood (KNN) algorithms. The polarimetric parameters responsive to multiple scattering events within the crop canopy volume were extracted using both model-based and eigen-based polarimetric decompositions.

Polarimetric parameters are most sensitive to the development

Acknowledgments

This work was supported in part by the Canadian Space Agency Class Grant and Contribution Program as part of the Canadian plan to spatial missions of soil moisture (14SUSMAPSH), and in part by the Natural Sciences and Engineering Research Council of Canada (RGPIN-2017-05533). The authors would like to thank the SMAPVEX16 funding agencies in Canada and US (Agriculture and Agri-Food Canada, United States Department of Agriculture, University of Manitoba, University of Guelph, Université de

References (26)

F. Canisius et al.
Tracking crop phenological development using multi-temporal polarimetric Radarsat-2 data
Remote Sens. Environ.
(2018)
A. Ceglar et al.
Improving WOFOST model to simulate winter wheat phenology in Europe: evaluation and effects on yield
Agric. Syst.
(2019)
X. Jiao et al.
Object-oriented crop mapping and monitoring using multi-temporal polarimetric RADARSAT-2 data
ISPRS J. Photogramm. Remote Sens.
(2014)
J. Liu et al.
Estimating crop stresses, aboveground dry biomass and yield of corn using multi-temporal optical data combined with a radiation use efficiency model
Remote Sens. Environ.
(2010)
H. McNairn et al.
Estimating canola phenology using synthetic aperture radar
Remote Sens. Environ.
(2018)
B. Pulatov et al.
Modeling climate change impact on potato crop phenology, and risk of frost damage and heat stress in northern Europe
Agric. For. Meteorol.
(2015)
L.a. Wang et al.
Estimation of biomass in wheat using random forest regression algorithm and remote sensing data
Crop J.
(2016)
H. Wang et al.
Potential of a two-component polarimetric decomposition at C-band for soil moisture retrieval over agricultural fields
Remote Sens. Environ.
(2018)
H.A.K.M. Bhuiyan et al.
Assessing SMAP soil moisture scaling and retrieval in the Carman (Canada) study site
Vadose Zone J.
(2018)
H. Bleiholder et al.
Growth stages of mono-and dicotyledonous plants, BBCH monograph

M. El-Hajj et al.

Irrigated grassland monitoring using a time series of TerraSAR-X and COSMO-SkyMed X-band SAR data

Remote Sens.

(2014)

M. El-Hajj et al.

Synergic use of Sentinel-1 and Sentinel-2 images for operational soil moisture mapping at high spatial resolution over agricultural areas

Remote Sens.

(2017)

I. Hajnsek et al.

Potential of estimating soil moisture under vegetation cover by means of PolSAR

IEEE Trans. Geosci. Remote Sens.

(2009)

Cited by (67)

Dense orchard landscape mapping based on image merging with skeleton prediction
2023, International Journal of Applied Earth Observation and Geoinformation
To address the difficulty of individual vine-plant segmentation in large orchards, this study proposes a segmentation model for individual canopies that combines the characteristics of tree deciduousness and natural growth patterns. This enables the landscape mapping of orchards under intensive planting conditions. In this study, winter and summer unmanned aerial vehicle (UAV) remote-sensing images of kiwifruit orchards were used for canopy segmentation. It involved: 1) target location identification of tree trunks in winter images and alignment of winter and summer image information; 2) semantic segmentation of branches in winter images; 3) screening of individual skeletons and prediction of branch growth; 4) removal of non-canopy parts of summer images using digital image processing to obtain canopy leaf segmentation results; and 5) segmentation of the extent of the individual plant canopies based on the predicted results of the individual plants. The branch precision rate for the final kiwifruit canopy at the plant level was 99.25% and the recall rate was 95.96%. The average recall rate of the canopy precision segmentation was 82.79%, the accuracy rate of which was 73.31%. This model does not involve plant height information and can implement instance segmentation of deciduous trees of the same species, with similar characteristics, and dense planting. The proposed solution addresses the problem of large-scale standardized planting of the same species, especially those that do not have obvious height or color differences and cannot be segmented using a digital elevation model.
Quantification of wetland vegetation communities features with airborne AVIRIS-NG, UAVSAR, and UAV LiDAR data in Peace-Athabasca Delta
2023, Remote Sensing of Environment
Arctic-boreal wetlands, important ecosystems for biodiversity and ecological services, are experiencing hydrological changes including permafrost thaw, earlier snowmelt, and increased wildfire susceptibility. These changes are affecting wetland productivity, species diversity, and biogeochemical cycles. However, given the diverse forms and structures of wetland vegetation communities, traditional wetland maps generated from lower spatial and spectral resolution satellite imagery lack community-level vegetation classification and miss spatially complex patterns. In this study, we built a cloud-based workflow to map wetland vegetation community of the Peace-Athabasca Delta (PAD), Canada, by leveraging high-resolution (5-m) airborne multi-sensor datasets, namely NASA's Airborne Visible/Infrared Imaging Spectrometer-Next Generation (AVIRIS-NG) and Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR), and a historical LiDAR archive. Validation of our classifications using ground references indicates that classifications derived from AVIRIS-NG have higher accuracies (≥87.9%) than either UAVSAR (65.6%) or LiDAR (75.9%) for mapping wetland vegetation communities. We also show improved classification accuracy when combining information from multiple sensors. In particular, incorporating AVIRIS-NG and UAVSAR datasets substantially reduced omission errors of wet graminoid and wet shrub classes from 29.6% to 20.5% and from 10.8% to 7.5%, respectively. Combining AVIRIS-NG and LiDAR datasets further improves overall accuracy (+2.2%) for most classifications, especially emergent vegetation, wet graminoid, and wet shrub. The best performing model, using features derived from all three sensors, achieved an overall accuracy of 93.5%. The framework established here can be used to leverage extensive airborne AVIRIS-NG and UAVSAR datasets collected across Alaska and northwest Canada to understand the spatial distribution of Arctic-Boreal wetland vegetation communities.
Exploring the effects of training samples on the accuracy of crop mapping with machine learning algorithm
2023, Science of Remote Sensing
Machine learning algorithms are a frequently used crop classification method and have been applied to identify the distribution of various crops over regional and national scales. Previous studies have underscored that the number of training samples strongly influences the classification accuracy of machine learning algorithms, resulting in extensive training sample collection efforts. This study, taking winter wheat as an example, challenges the above principle by selecting training samples with the time-weighted dynamic time warping (TWDTW) method and finds that the classification accuracy of machine learning algorithms highly relies on the representativeness and proportion of training samples rather than the quantity. With the increase of the representativeness of training samples, i.e. more comprehensively reflected the characteristics of winter wheat, the classification accuracy is continually improved. The best classification accuracy is further achieved when selecting the training samples of winter wheat and non-winter wheat according to the ratio of their statistical areas. On the contrary, only a slight difference was found in overall accuracy (91.26% and 90.74%), producer’s accuracy (86.33% and 86.65%) and user’s accuracy (97.37% and 96.01%) when using 1,000 and 10,000 training samples. Overall, this study demonstrates that the characteristics of training samples have a great impact on the classification accuracy of machine learning algorithms, and the training samples generated by TWDTW method are reliable for crop mapping.
Retrieval performances of different crop growth descriptors from full- and compact-polarimetric SAR decompositions
2023, Remote Sensing of Environment
This study investigates the different performances between Compact-Polarimetric (CP) and Full-Polarimetric (FP) SAR to retrieve multiple crop growth parameters including Vegetation Water Content (VWC), Leaf Area Index (LAI), height, and dry biomass. The objective is to study at which conditions the CP SAR can obtain comparable retrieval accuracy as FP SAR for specific crop types and growth descriptors. Polarimetric decompositions were used to extract CP and FP explanatory variables to quantify crop growth status. Then, the sensitivities of CP and FP variables to different crop descriptors were analyzed, revealing higher sensitivity to height than LAI, VWC, and dry biomass. In order to reduce the information redundancy of sets of CP and FP SAR variables, Partial Least Square (PLS) approach was used to develop new orthogonal SAR parameters, while Step-Wise Regression (SWR) was used to determine the optimal SAR parameters, followed by retrievals of multiple crop growth descriptors using the developed estimators. Validated by the ground measurements, the retrieval performances are found to be highly dependent on polarimetry dimension (CP or FP), crop types, and targeted crop descriptors. With crop growth and enhanced depolarization effects, the CP SAR can capture the increasing volume power and decreasing surface power but at a lower rate than FP SAR. The m-δ and m-χ decompositions provide similar scattering components which differ from the Random Volume Over Ground (RVOG) decomposition. The third Stokes parameter g₃ of the CP SAR data that indicates the strength of the circular polarization component is a common important parameter to infer the multiple crop parameters of interest. In agreement with the sensitivity analysis, better retrieval accuracies were obtained for height and dry biomass than VWC and LAI. For short vegetation such as soybean, full polarimetry is required to obtain the estimates of all four crop growth descriptors. On the contrary, for tall crop growth parameters, such as canola (height and dry biomass), corn (height, VWC, dry biomass), and wheat (LAI, VWC, dry biomass), the CP SAR features can provide comparable robust retrieval accuracy as the FP SAR features. This study deepens our insights into the CP SAR of the RADARSAT Constellation Mission (RCM) to timely monitor crop growth and health status.
Dual polarimetric decomposition in Sentinel-1 images to estimate aboveground biomass of arboreal caatinga
2023, Remote Sensing Applications: Society and Environment
Citation Excerpt :
In the Cloude-Pottier method, information on entropy (H), anisotropy (A) and alpha angle (α) is generated based on the decomposition of the full polarimetric (Cloude and Pottier, 1997) or dual polarimetric scattering matrix, which only provides H and α data based on the construction of the coherence matrix to estimate the wave depolarization, also known as H-α decomposition (Cloude, 2007). There is a greater application of the Cloude-Pottier decomposition in full polarimetric SAR data (Ji and Wu, 2015), in studies focusing on phenological monitoring of vegetation: Haldar et al. (2019), Wang et al. (2019) and Selvaraj et al. (2021), all using RADARSAT-2 images, for biomass estimates: Yang et al. (2019), Haldar et al. (2019) and Waqar et al. (2020a) using data from the same radar, and Golshani et al. (2019) and Waqar et al. (2020b), using images from ALOS/PALSAR-2. Nevertheless, dual polarimetric data also have important practical applications (Long et al., 2019), which is verified with the growing scientific production using the H-α decomposition using Sentinel-1 images in phenological studies or obtaining biophysical characteristics of the vegetation (Ataee et al., 2019; Nasirzadehdizaji et al., 2019; Gella et al., 2021; Umutoniwase and Lee, 2021; Weiβ et al., 2021).
The Caatinga is a typical biome of the Brazilian semiarid region, and there is a lack of studies to estimate its plant biomass using SAR images. Also, there are no studies at this topic that use information from polarimetric decomposition. Therefore, this study aimed to estimate aboveground biomass (AGB) of the arboreal Caatinga under different phenological conditions using entropy and alpha angle data obtained by different polarimetric filters by Dual Polarimetric Decomposition in Sentinel-1 SAR images. The forest inventory was carried out in 19 sample plots with an area of 30 × 30 m in the Caatinga of Sergipe state, Brazil, using an allometric equation to estimate the AGB. Sentinel-1 images were obtained in 3 phenological conditions of the Caatinga (Greenness, Intermediate and Dryness) to which polarimetric filters (BoxCar - BC, IDAN, Improved Lee Sigma - ILS and Refined Lee - RL) were applied and obtained the entropy (H) and alpha angle (α) attributes after Dual Polarimetric Decomposition. The attributes of each polarimetric filter were related to the AGB using linear regressions. The combination of attributes provided higher accuracy than individual estimates: data from Intermediate period provided higher accuracy for AGB estimates, using H of the IDAN filtering and α using ILS (R²: 32.05%). The attributes generated through Dual Polarimetric Decomposition in Sentinel-1 images showed a low relationship with the AGB of the Caatinga in the study area. Nevertheless, the study made it possible to indicate which polarimetric filters depending on the different leaf cover conditions in the Caatinga offer greater precision in relation to the measured AGB, serving as the basis for future studies.
Seasonal changes of leaf chlorophyll content as a proxy of photosynthetic capacity in winter wheat and paddy rice
2022, Ecological Indicators
Citation Excerpt :
Positive linear correlations of Vcmax25 with Na and Chla were found in previous studies. Observed data of forests demonstrated that Chla was a better proxy of Vcmax25 than Na (Croft et al., 2017; Wang et al., 2019; Chou et al., 2020). Our data showed that without differentiation of pre- and post-flowering stages, Na slightly outperformed Chla in estimating Vcmax25 for both winter wheat and rice.
Accurate representation of photosynthetic capacity and its seasonal variations is critical for modeling carbon sequestration of cropland ecosystems through photosynthesis. Previous studies indicated that the maximum carboxylation rate at 25 °C (V_cmax25) is the key determinant of photosynthetic capacity and can be mapped according to leaf nitrogen (N) and chlorophyll (Chl) contents. However, whether relationships of V_cmax25 with leaf nitrogen and chlorophyll contents in cropland ecosystems change over the whole growing season remain unclear. To address this question, we conducted intensive field campaigns in winter wheat and paddy rice farmlands of Eastern China, with leaf V_cmax25, the maximum electron transport rate at 25 °C (J_max25), N, and Chl measured every 7 to 10 days from the tillering stage to maturity. Results showed that V_cmax25 and J_max25 varied significantly during the growing seasons and maximized at flowering stages. For both winter wheat and paddy rice, Chl_a (Chl content at the area unit) showed significantly different relationships with V_cmax25 pre- and post- flowering while N_a (N content at the area unit) did not. When different empirical models were used for pre- flowering and post- flowering stages, V_cmax25 estimated from Chl_a was closer to observations (R² = 0.875, RMSE = 9.479 µmol m⁻² s⁻¹) than that estimated from N_a (R² = 0.565, RMSE = 17.718 µmol m⁻² s⁻¹). The findings in this study imply that the combination of remotely sensed leaf Chl_a content and phenology would improve the mapping of V_cmax25 for crops at regional and even global scales.

View all citing articles on Scopus

^☆: This work was supported in part by the Canadian Space Agency Class Grant and Contribution Program as part of the Canadian plan to spatial missions of soil moisture (14SUSMAPSH), and in part by the Natural Sciences and Engineering Resources Council of Canada (RGPIN-2017-05533).

View full text

Crop phenology retrieval via polarimetric SAR decomposition and Random Forest algorithm☆

Highlights

Abstract

Introduction

Section snippets

Study site

RADARSAT-2 data acquisitions

Method

Crop phenology analysis

Conclusion

Acknowledgments

Remote Sens. Environ.

Agric. Syst.

ISPRS J. Photogramm. Remote Sens.

Remote Sens. Environ.

Remote Sens. Environ.

Agric. For. Meteorol.

Crop J.

Remote Sens. Environ.

Assessing SMAP soil moisture scaling and retrieval in the Carman (Canada) study site

Vadose Zone J.

Growth stages of mono-and dicotyledonous plants, BBCH monograph

Irrigated grassland monitoring using a time series of TerraSAR-X and COSMO-SkyMed X-band SAR data

Remote Sens.

Synergic use of Sentinel-1 and Sentinel-2 images for operational soil moisture mapping at high spatial resolution over agricultural areas

Remote Sens.

Potential of estimating soil moisture under vegetation cover by means of PolSAR

IEEE Trans. Geosci. Remote Sens.