Predicting NDVI from SAR images toward Above Ground Biomass Estimation

Abstract

We propose a novel image-to-image translation method to accurately estimate the Normalized Difference Vegetation Index (NDVI) from Synthetic Aperture Radar (SAR) satellite images. The core innovation of this method lies in its ability to translate SAR images into NDVI maps with high accuracy, overcoming the limitations of traditional optical sensors which are often hindered by cloud cover and atmospheric conditions. By utilizing advanced machine learning techniques, this image-to-image translation method can effectively extract and enhance the vegetation signal from SAR data, providing a reliable alternative for continuous and comprehensive vegetation monitoring. In a case study involving the translation of NDVI values from SAR images of cabbage farms, our method outperformed competitors in a public competition and exceeded the average of the top five teams by 44%. This demonstrates the method's effectiveness in extracting and enhancing vegetation signals from SAR data, providing a reliable alternative for continuous and comprehensive vegetation monitoring. We have published both our image translation method with baseline methods and the geospatial-specific dataset at https://github.com/IBM/SAR-TO-NDVI. The high-precision NDVI maps generated through this method can also contribute to improving the estimation of Above Ground Biomass (AGB), which is essential for various environmental applications. Moving forward, we will discuss extensive validation studies to assess the performance and accuracy of this method across different regions and vegetation types in the presentation.