Joint estimation of Plant Area Index (PAI) and wet biomass in wheat and soybean from C-band polarimetric SAR data

Retrieval of the Plant Area Index (PAI) and wet biomass from polarimetric SAR (PolSAR) data is of paramount importance for in-season monitoring of crop growth. Notably, the joint estimation of biophysical parameters might be effective instead of an individual parameter due to their inherent relationships (possibly nonlinear). The semi-empirical water cloud model (WCM) can be suitably utilized to estimate biophysical parameters from PolSAR data. Nevertheless, instability problems could occur during the model inversion process using traditional inversion approaches. Iterative optimization (IO) can have difficulty in finding the global minima while look up table (LUT) searches have a lower generalization capability. These challenges reduce the transferability of IO and LUT search inversions in computational efficiency and seldom account for the inter-correlation among the parameters. Alternatively, a machine learning regression technique with a regularization routine may provide a stable and optimum solution for ill-posed problems related to the inversion of the WCM. In the present work, the crop biophysical parameters viz. PAI and wet biomass are estimated simultaneously using the multi-target Random Forest Regression (MTRFR) technique. The accuracy of the retrieval method is analyzed using the in-situ measurements and quad-pol RADARSAT-2 data acquired during the SMAPVEX16 campaign over Manitoba, Canada. The inversion process is tested with different polarization combinations of SAR data for wheat and soybean. The validation used ground measured biophysical parameters for various crops, indicating promising results with a correlation coefficient (r) in the range of 0.6–0.8. In addition, the relationship between PAI and wet biomass using the multi-target and single output model is also assessed based on in-situ measurements. The results confirm that the inter-correlation between biophysical parameters is well preserved in the MTRFR based joint inversion technique for both wheat and soybean.