Vertical Spatial Prediction of Soil Salinity and Management Zoning for Perambalur District of Tamil Nadu, India

Monitoring and managing soil salinity is vital due to its negative impact on agricultural production. Electrical conductivity (EC) at different profile levels is a baseline for assessing salinity and salt accumulation across soil horizons. Random Forest algorithm was applied to predict EC and Exchangeable Sodium Percentage (ESP) in Perambalur district at depths of 0–20 cm, 20–40 cm, and 40–80 cm, using 1159 soil samples obtained via conditioned Latin hypercube sampling. SCORPAN-based environmental covariates (32 nos) generate the model variables, with analysis of the percentage contribution of each environmental variable prediction process through Permutation Feature Importance analysis for identifying Elevation, Physiography, Geomorphology, Annual Mean Air Temperature, and Annual Mean Precipitation. Model performance was validated using a 70:30 holdback procedure and metrics (RMSE, NSE, and RPIQ). For EC, RMSE ranged from 0.20 to 0.28 dS m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup>, NSE from 0.40 to 0.58, and RPIQ from 1.06 to 1.26, respectively. For ESP, RMSE ranged from 3.40 to 3.54, NSE from 0.42 to 0.64, and RPIQ from 1.32 to 1.68. Using vertical predictions of EC and ESP, management zones were delineated via fuzzy C-means clustering at a 30 m resolution. These zones and spatial salinity distribution insights enhance the feasibility of targeted mitigation practices through informed policy decisions.