A machine learning based algorithm using radar and optical remote sensing products to model soil salinity

Soil salinity is a global phenomenon affecting arid and semi-arid agriculture as well as the surrounding ecosystem, with recent global estimates revealing about 932.2 Mha areas (18% of global agricultural lands by estimate) suffering from various levels of soil salinization. This affects in reducing plant growth, reduced yields, and in sometimes crop failure. Despite this, no spatially distributed measurements of soil salinity are currently available at regional and global scales. This research develops a comprehensive framework to measure soil salinity using a Geographically Weighted Artificial Neural Network (GWANN) algorithm, targeting the use of radar remote sensing data (deformation and backscatter) from Sentinel 1A/1B, coupled with optical remote sensing data-derived salinity indices from LANDSAT, as well as some terrain (DEM, slope, etc.), and other hydrological variables (GLDAS soil moisture). Previous research in this topic entails an intense use of optical sensor derived salinity indices. Combining radar imageries with other relevant variables, along with the geographical weighting should further improve the ANN based model accuracy. The research would develop fine resolution maps of soil salinity, defined in terms of root zone (~1m) electrical conductivity (EC) estimates. The remote sensing-based soil salinity model is to be validated through K-fold validation technique using available insitu EC data from suitable pilot sites around the world (in spatiotemporal sync with Sentinel imageries). The broader aim of this ongoing research endeavor is to provide an ML-based approach for deriving soil salinity using remote sensing datasets, which could later be improved by using NASA's NISAR L band interferometric data (level 2). Furthermore, the methodological framework has potential to be implemented in other complex data scarce ecosystems (e.g., low-elevation deltaic mangrove areas of India/ Bangladesh).