Utilizing machine learning models for mapping salt affected soils of Surat and Bharuch, Gujrat, India
Digital Soil Mapping (DSM) provides an advanced approach for accurately assessing soil salinity, which is crucial for enhancing agricultural productivity in salt-affected regions like the Bharuch and Surat districts of Gujarat, India. In 2008, an extensive soil survey was conducted to address this issue, focusing on agricultural soils. Samples were collected and analyzed for pH, electrical conductivity (ECe), and exchangeable sodium percentage (ESP) from soil saturation extract. Nineteen bioclimatic variables from WorldClim2 were downloaded and resampled at 90-meter resolution. Additionally, twenty-four vegetation indices indicative of soil salinity were derived from Landsat data using Google Earth Engine, also at a 90-meter resolution. The study included land use and land cover (LULC) classification and utilized a digital elevation model (DEM) from SRTM, resampled at 90 meters. Terrain analysis using SAGA GIS generated nine topographic parameters, such as slope and aspect. Further soil covariates, including soil type, geomorphology, and proximity to the coastline, were computed. A total of sixty-five covariates were used to map the extent of salt-affected soils in the region. A Random Forest regression model was applied, with feature selection via recursive elimination identifying seven key covariates: canopy response salinity index, analytical hill shade, channel network base level, digital elevation model, valley depth, and geomorphology of soil. The model predicted that the highest salinity levels were concentrated in the western, coastal regions, while the eastern areas exhibited lower salinity. Approximately 286,000 hectares were identified as affected by varying degrees of salinity, with the distribution of net area across salinity classes as follows: for salinity levels < 4 dS/m, 580,044 hectares of land were affected; for 4–8 dS/m, 98,934 hectares; for 8–16 dS/m, 68,481 hectares; for 16–32 dS/m, 58,730 hectares; and for salinity levels > 32 dS/m, 27,180 hectares were impacted. The model shows a modest performance with an RMSE of 10.40, MAE of 3.92, and an R² of 0.19. An uncertainty map highlighted greater uncertainty in coastal areas compared to less saline regions. The study demonstrates the critical role of DSM in sustainable land management, as it facilitates targeted remediation strategies and informs land-use planning. By integrating multi-source geospatial data and advanced modeling techniques, DSM enhances the precision of soil salinity assessments, contributing to improved agricultural practices and land management in Gujarat.
