Bayesian EO

Research

Agriculture

Gaussian Process Regression Model for Crop Biophysical Parameter Retrieval from Multi-Polarized C-Band SAR Data

Biophysical parameter retrieval using remote sensing has long been utilized for crop yield forecasting and economic practices. Remote sensing can provide information across a large spatial extent and in a timely manner within a season. Plant Area Index (PAI), Vegetation Water Content (VWC), and Wet-Biomass (WB) play a vital role in estimating crop growth and helping farmers make market decisions. Many parametric and non-parametric machine learning techniques have been utilized to estimate these parameters. A general non-parametric approach that follows a Bayesian framework is the Gaussian Process (GP). The parameters of this process-based technique are assumed to be random variables with a joint Gaussian distribution. The purpose of this work is to investigate Gaussian Process Regression (GPR) models to retrieve biophysical parameters of three annual crops utilizing combinations of multiple polarizations from C-band SAR data. RADARSAT-2 full-polarimetric images and in situ measurements of wheat, canola, and soybeans obtained from the SMAPVEX16 campaign over Manitoba, Canada, are used to evaluate the performance of these GPR models. The results from this research demonstrate that both the full-pol (HH+HV+VV) combination and the dual-pol (HV+VV) configuration can be used to estimate PAI, VWC, and WB for these three crops.

1. Ghosh, Swarnendu Sekhar, Subhadip Dey, Narayanarao Bhogapurapu, Saeid Homayouni, Avik Bhattacharya, and Heather McNairn. "Gaussian Process Regression Model for Crop Biophysical Parameter Retrieval from Multi-Polarized C-Band SAR Data." Remote Sensing 14, no. 4 (2022): 934.

Enhancing Plant Area Index Retrieval Using Gaussian Process Regression from Dual-Polarimetric SAR Data

In this paper, a Gaussian Process Regression (GPR) model is implemented to retrieve the Plant Area Index (PAI) of wheat and canola. Backscatter information from Sentinel-l dualpol GRD SAR data and in-situ measurements collected during the Soil Moisture Active Passive Validation Experiment 2016 (SMAPVEX16-MB) Manitoba campaign were used to calibrate and validate the proposed GPR model. A recently proposed pseudo scattering entropy, H c derived from dual-pol GRD SAR data has been used along with backscatter information to investigate the improvement in retrieval accuracy. Including the pseudo entropy parameter in the feature, space showed an improvement of 4.28% and 3.66% in the correlation coefficient (ρ) for wheat and canola respectively. Similarly, a decrease in nRMSE by 4% for wheat and 4.76% for canola was observed during PAI retrieval.

2.Ghosh, Swarnendu Sekhar, Narayanarao Bhogapurapu, Avik Bhattacharya, and Saeid Homayouni. "Enhancing Plant Area Index Retrieval Using Gaussian Process Regression from Dual-Polarimetric SAR Data." In 2023 International Conference on Machine Intelligence for GeoAnalytics and Remote Sensing (MIGARS), vol. 1, pp. 1-4. IEEE, 2023.

Forests

Gaussian process regression-based forest above ground biomass retrieval from simulated L-band NISAR data

Accurate mapping of forest above ground biomass (AGB) is essential for understanding changes in the rate of ecosystem processes such as biomass productivity, litter productivity, actual litter decomposition, and potential litter decomposition during secondary succession. They also play a vital role in evaluating forest carbon pools. This study presents a Matérn kernel-based Gaussian process regression (GPR) approach for biomass estimation integrating Synthetic Aperture Radar (SAR) backscatter intensities with LiDAR measurements. We use the backscatter intensities from five scenes of L-band Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) simulated NISAR data collected during the 2019 AM–PM NASA airborne campaign. The biomass map derived from LVIS (Land, Vegetation, and Ice Sensor) Lidar point clouds over the Lenoir landing site, Alabama, was used with these simulated NISAR data. We have utilized the GPR model to estimate the AGB of the entire forest alongside the major forest classes in the study area, namely deciduous, evergreen, and woody wetlands. To examine the dependency of the model on acquisition conditions, we calibrated and validated our proposed model utilizing scene combinations. The experiment shows that multiple-scene retrieval delivered improved AGB estimates compared to single-scene retrieval. We evaluated the efficacy of the GPR model for three AGB ranges, i.e., (i) 8 to 100 Mg ha−1, (ii) 8 to 230 Mg ha−1, and (iii) 8 to 470 Mg ha−1 at 20 m, 30 m, 50 m and 100 m spatial resolution, respectively. The results indicate that the RMSE incurred by the GPR model for all these AGB ranges seemed to reduce as we increased spatial extent or reduced spatial resolution from 20 m to 100 m. Further, we demonstrate the performance of the optimized GPR model to retrieve AGB within the biomass range of 8 Mg ha−1 to 100 Mg ha−1 for all the above-mentioned forest types. Finally, this research highlights the advantages of the Matérn kernel-based GPR model over other statistical regression models towards improved AGB mapping.

3. Ghosh, Swarnendu Sekhar, Unmesh Khati, Sandeep Kumar, Avik Bhattacharya, and Marco Lavalle. "Gaussian process regression-based forest above ground biomass retrieval from simulated L-band NISAR data." International Journal of Applied Earth Observation and Geoinformation 118 (2023): 103252.

Page updated

Google Sites

Report abuse