Unlocking Soil Salinity Prediction With Remote Sensing Indices and Environmental Insights

Bibliographic Data

ID21649335
AuthorsTehseen Javed (0000-0003-2965-6006, College of Water Conservancy & Architectural Engineering Shihezi University Shihezi China), Zhang Jinzhu (College of Water Conservancy & Architectural Engineering Shihezi University Shihezi China, corresponding author), Li Wenhao (College of Water Conservancy & Architectural Engineering Shihezi University Shihezi China), Zhang Jihong (College of Water Conservancy & Architectural Engineering Shihezi University Shihezi China), Liu Jian (0000-0002-7759-7410, College of Water Conservancy & Architectural Engineering Shihezi University Shihezi China), Lin Haixia (College of Water Conservancy & Architectural Engineering Shihezi University Shihezi China), Zhenhua Wang (0000-0002-6902-1059, College of Water Conservancy & Architectural Engineering Shihezi University Shihezi China, corresponding author)
Year2025
Volume36
Issue15
Pages5168-5183
Publication date2025-09-01
Peer ReviewedYes
Open AccessYes
TypeARTICLE
VenueLand Degradation and Development (JOURNAL)
Journal identifiersISSN: 1085-3278 • E-ISSN: 1099-145X
PublisherWiley (PUBLISHER • GB)
DOI10.1002/ldr.5694
OpenAlexW4411519871
LanguageEN
Citations received2
References cited63

Despite advances in salinity prediction, a knowledge gap exists in accurately integrating remote sensing indices and environmental factors for effective management strategies. Therefore, this study examines the relationship between soil salinity (EC e ) and remote sensing (RS) indices, soil texture properties, and ecological features. Several statistical techniques, such as Pearson correlation, Geographically Weighted Regression (GWR), Principal Component Analysis (PCA), and SHapley Additive exPlanations (SHAP), were used to investigate the capability of these indices and indicators for the prediction of soil salinity. The study revealed that the Decision Tree (DT) showed the highest accuracy for soil salinity prediction among the machine learning models, while XGBoost exhibited lower predictive performance. Evaluating the environmental indices with ECe, the Normalized Difference Salinity Index (NDSI) showed the highest positive correlation with ECe ( r = 0.88), reflecting its effectiveness in salinity prediction. Moderate positive correlations were observed with the Soil Salinity Index (SSI, r = 0.65), while the Bare Soil Index (BSI, r = −0.85) and Soil‐Adjusted Vegetation Index (SAVSI, r = −0.76) demonstrated strong negative correlations. Soil physicochemical properties, including clay, silt, sand, organic carbon, and bedrock, exhibited weak relationships with ECe, with R 2 values consistently below 0.04, indicating limited predictive power. PCA analysis revealed distinct contributions of RS indices to ECe variability, with NDSI and SSI positively influencing salinity variability, whereas SAVSI contributed inversely, aligning negatively along PC1. SHAP analysis further reinforced the predictive dominance of RS indices, assigning the highest importance value to NDSI (0.61), followed by BSI (0.28) and SAVSI (0.08). In contrast, soil texture properties and organic carbon exhibited minimal significance, with importance values under 0.02. NDSI was further tested across low‐ and high‐salinity farms, consistently outperforming other indices. These findings highlight its advantage in improving salinity mapping management strategies and advancing precision agriculture/environmental planning through modern analytical approaches

Linear regression · Principal component analysis · Salinity · Silt · Soil salinity · Soil texture · Soil water · Statistics · Computer Science · Environmental Science · Mathematics · Remote Sensing in Agriculture · Soil and Land Suitability Analysis · Soil Geostatistics and Mapping · Geology · Soil Science

Tools

Open DOI