Deep learning for estimation of bio- & geophysical parameters from SAR data

With its unique amplitude and phase information, which is sensitive to subtle dielectric and morphological changes in the monitored target volume, SAR imaging has a pivotal role in estimating diverse bio- and geophysical parameters. However, predicting such varied parameters in a closed-form solution remains challenging due to the complex interaction between the SAR signal and monitored variables under dynamic and heterogeneous environmental conditions. In order to handle this situation, recently, many studies have begun to integrate deep learning (DL) models into estimation efforts to leverage SAR's unique sensitivity to changes in these variables.
This chapter first explains the underlying principles for the bio- and geophysical parameters estimation using SAR in conjunction with DL models, and then lists these parameters to highlight the diverse application areas where they excel. Next, learning strategies for those parameters are summarized, considering that the limited number of labeled data and learning representations with strong generalization capabilities are key factors in monitoring physical processes using SAR. Later, we demonstrate how DL architectures capture SAR's sensitivity to biophysical parameters by applying them to three different estimation problems: soil moisture, evapotranspiration, and crop yield. Finally, we put forward future study directions for having high spatiotemporal resolution bio- and geophysical parameters, using SAR and DL models.