Crop Biophysical Parameter Retrieval Using Gaussian Process Regression from C-Band Polarimetric SAR Data

Crop biophysical parameters, such as the Plant Area Index (PAI), Wet-biomass (WB), and Vegetation Water Content (VWC), play a vital role in estimating crop yield. Over the last few decades, various biophysical parameter retrieval methods have come into existence. Studies have shown, weaknesses of physical and parametric methodologies super-sede their strengths. In this regard, non-linear non-parametric methods like the Gaussian Process Regression (GPR) have advantages over existing techniques. This study retrieved specific crop biophysical parameters over a Canadian test site using GPR employing different polarization combinations of the Synthetic aperture radar (SAR) data. We used the ground measurements of wheat and soybean of the SMAPVEX-16 campaign as the in-situ data. We characterized the performance of GPR by using statistical measures, such as the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the Pearson correlation coefficient. We noticed that besides the full pol (HH + HV + VV) combination, the performance of the dual-pol (HV + VV) configuration had shown encouraging results in estimating the biophysical parameters.