Machine learning-driven mapping of salt-affected soils: integrating laboratory analysis and environmental covariates to establish extent and spatial distribution
Year: 2026
Type: article
Abstract: Understanding spatial distribution of salt-affected soils (SAS) is essential for sustaining agricultural productivity, yet detailed information on their extent and characteristics in irrigation schemes remains limited. This study assessed the extent and types of SAS in three irrigation schemes by integrating remote sensing and soil analytical data. Soil samples were collected at 0–30 cm using a 250 m grid sampling approach. Soil pH, electrical conductivity of saturated paste extract (ECe), and exchangeable bases were analysed, while exchangeable sodium percentage (ESP) and sodium adsorption ratio (SAR) were calculated. Remote sensing indices derived using Google Earth Engine (GEE) were integrated with soil data to predict SAS indicators using Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting Machine (GBM/XGBoost), and Artificial Neural Network (ANN) models with k-fold cross-validation. RF generally achieved the better predictive performance, with R² values ranging from 0.83 to 0.91. Based on predefined thresholds, SAS affected 10.7%, 4.9%, and 3.4% of IR1, IR2, and IR3, respectively. Saline soils dominated IR1, saline-sodic soils IR2, and sodic soils IR3. These findings highlight spatial variability of SAS and the need for site-specific management. The resulting maps can guide targeted soil amendments and support rice-breeding programmes by identifying areas requiring salt-tolerant varieties.
Language: English
Subfield: Soil Science
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $1,060
OpenAlex ID: W7225420838