Modeling SAR Observables with Combining a Crop-Growth Model and a Machine-Learning

Our aim is to estimate Synthetic Aperture Radar (SAR) observables, such as backscatter in VV and VH polarizations, well the VH/VV ratio, cross-ratio (CR), interferometric coherence VV, from agricultural fields. In this study, we use Decision Support System for Agrotechnology Transfer (DSSAT) crop growth simulation model simulate parcel-level phenological parameters over 1500 parcels of silage maize Netherlands. The was calibrated using field data, including phases, leaf area index (LAI), above-ground dry biomass (AGB). simulations incorporate fine-resolution gridded precipitation data soil interaction between soil-plant-atmosphere genotype DSSAT. variables produced by DSSAT are then used inputs a Vector Regression (SVR) model. This trained SAR observables 2017, 2018, 2019, its performance evaluated independent fields each these years. results show close fit modeled observed C-band observables. importance vegetation estimation assessed. AGB showed significant backscatter. study demonstrates potential value combining models machine learning For example, SVR developed here could be an observation operator assimilation context constrain water dynamics