Improving NDVI Estimation Accuracy from Synthetic Aperture Radar Data by Integrating Interferometric Coherence and Polarimetric Features

Normalized Difference Vegetation Index (NDVI) has been applied to diverse applications such as agricultural management, taking advantage of wide coverage and periodic nature of remote sensing. A significant limitation arises in obtaining accurate values under cloud cover. NDVI estimation from synthetic aperture radar (SAR) using microwaves that penetrate clouds has attracted interest as a promising solution to this problem. This study proposes a novel approach to NDVI estimation that overcomes the limitations of using like-polarized and cross-polarized backscatter coefficients, whose accuracy was limited by mixing of soil and vegetation scattering. Therefore, this study employs interferometric coherence and polarimetric-derived features to assess their potential to enhance the accuracy of NDVI estimation accuracy over backscatter coefficients. We trained UNet with a French wheat croplands dataset in order to assess their accuracy contributions. The results show interferometric coherence and polarimetric-derived features provide complementary information to backscatter coefficients, thereby enhancing the accuracy of NDVI estimation from SAR data.