Li et al. (2025) Exploring the driving forces of soil salinity reduction using Random Forest and SHAP in water-saving oasis irrigation areas

Identification

• Journal: Irrigation Science
• Year: 2025
• Date: 2025-12-10
• Authors: Wenhao Li, Shuanglong Gao, Xiaoguo Mu, Yue Wen, Tehseen Javed, Zhenhua Wang
• DOI: 10.1007/s00271-025-01045-6

Research Groups

• College of Water Conservancy and Architectural Engineering, Shihezi University, Shihezi, Xinjiang, China
• Key Laboratory of Modern Water-Saving Irrigation of Xinjiang Production and Construction Group, Shihezi University, Shihezi, Xinjiang, China
• Key Laboratory of Northwest Oasis Water-Saving Agriculture, Ministry of Agriculture and Rural Affairs, Shihezi, Xinjiang, China
• Technology Innovation Center for Agricultural Water and Fertilizer Efficiency Equipment of Xinjiang Production and Construction Corps, Shihezi, Xinjiang, China

Short Summary

This study investigated the spatiotemporal dynamics and driving forces of soil salinity reduction in water-saving oasis irrigation areas of the Manas River Basin from 2013 to 2021, revealing a significant decrease in salinity primarily driven by declining groundwater depth.

Objective

• To characterize the spatiotemporal patterns of soil salinity in the 0–100 cm soil layer.
• To identify the dominant drivers of salinity change.
• To explore the mechanisms and interactions among these drivers.

Study Configuration

• Spatial Scale: Manas River Basin, northwestern Xinjiang, China (84° 44′~86° 50′ E, 43° 4′~46° 0′ N), covering an irrigated agricultural landscape. Data collected from 66 grid points (average sampling interval of 12 km) and 91 long-term observation wells. Soil samples were collected from 0–100 cm depth.
• Temporal Scale: 2013–2021, with specific analyses focusing on the rapid development phase of water-saving irrigation (2013–2014) and the implementation phase of total water use control (2020–2021).

Methodology and Data

• Models used:
  • Random Forest (RF) model for predicting soil salinity dynamics.
  • Shapley Additive Explanations (SHAP) for interpreting feature importance and contributions.
  • Structural Equation Modeling (SEM) to uncover structural relationships among variables.
  • Pearson correlation analysis for preliminary factor evaluation.
  • Tolerance and Variance Inflation Factor (VIF) tests for multicollinearity assessment.
  • Ordinary Kriging interpolation (using ArcGIS 10.8) for spatial data.
  • Semivariogram modeling (in GS+ 9.0) for geostatistical analysis.
  • Standard Normal Homogeneity Test (SNHT) for temporal homogenization of meteorological data.
• Data sources:
  • Annual irrigation reports from Shihezi City, Shawan County, and Manas County for total irrigation area (IAr), water-saving irrigation area (WSIA), groundwater extraction (UWDA), and surface water diversion (SWDA).
  • 91 long-term observation wells for groundwater depth (GD) and degree of mineralization of groundwater (DMG).
  • 9 meteorological stations and the Water-Saving Irrigation Experimental Station at Shihezi University for annual rainfall (AR) and evaporation.
  • Field soil samples collected at 66 grid points from five depth intervals (0–20 cm, 20–40 cm, 40–60 cm, 60–80 cm, and 80–100 cm) for electrical conductivity (EC) and salt content. Soil bulk density samples were taken at 0–10 cm, 10–20 cm, 20–30 cm, and 30–40 cm.
  • Electrical conductivity (EC) measured using a DDS-11A digital conductivity meter (0.00 μS·cm⁻¹ to 100.0 mS·cm⁻¹).
  • Salt content (g·kg⁻¹) determined via the drying residue method and calibrated with EC measurements.

Main Results

• From 2013 to 2021, the average soil salinity in the 0–100 cm layer of the Manas River irrigation area decreased by approximately 0.35 g·kg⁻¹.
• The area of moderately saline-alkali soils (4–6 g·kg⁻¹) decreased by 9%, and slightly saline-alkali soils contracted by 41%, while non-saline land expanded by 50%.
• Soil salinity exhibited a consistent declining trend, with no evidence of secondary salinization, but the rate of reduction decelerated over time (e.g., from 0.50 g·kg⁻¹·yr⁻¹ in 2013–2014 to 0.15 g·kg⁻¹·yr⁻¹ in 2020–2021).
• The Random Forest model accurately predicted soil salinity changes, achieving an R² of 0.82, a Mean Absolute Error (MAE) of 0.09 g·kg⁻¹, and a Root Mean Square Error (RMSE) of 0.11 g·kg⁻¹ on the validation set.
• SHAP analysis identified groundwater depth (GD) as the most influential factor driving salinity reduction from 2013 to 2021.
• The effects of surface water diversion amount (SWDA), irrigation amount (IA), and water-saving irrigation area (WSIA) on salinity change became progressively more pronounced over time.
• Structural Equation Modeling (SEM) confirmed that a reduction in GD was the primary direct driver of soil salinity decrease. Indirectly, WSIA expansion increased groundwater extraction, which in turn reduced GD and increased groundwater salinity.
• Increased annual rainfall (AR) and reduced evaporation also positively influenced desalination.
• Excessive reliance on groundwater to support WSIA expansion was associated with increased groundwater extraction and mineralization, posing a potential risk of secondary salinization, which was mitigated by sustained declines in GD.

Contributions

• Provides a comprehensive spatiotemporal analysis of soil salinity dynamics and its driving forces in water-saving oasis irrigation areas, specifically the Manas River Basin.
• Introduces a novel hybrid machine learning (Random Forest and SHAP) and statistical (Structural Equation Modeling) approach for robustly identifying and interpreting the complex drivers of soil salinity reduction.
• Quantitatively demonstrates the critical role of declining groundwater depth as the dominant factor in mitigating soil salinity in arid oasis agricultural systems.
• Offers valuable insights into the mechanisms and interactions among various environmental and management factors influencing salinity, including irrigation practices, groundwater dynamics, and meteorological conditions.
• Provides practical guidance for sustainable water and soil management in arid oasis agriculture, emphasizing the need for integrated groundwater management, optimized surface and groundwater allocation, and long-term monitoring to prevent secondary salinization.

Funding

• National Natural Science Foundation of China, Grant No. 52209064 (Study on Water and Salt Simulation and Multidimensional Critical Regulation Mechanism in Typical Arid Oasis Irrigation Area)
• National Natural Science Foundation of China, Grant No. 52279040 (Study on the Mechanism of the Effect of Three Water Transformation on Soil Salt Migration in Cotton Field with Long-term Drip Irrigation in Arid Area)

Citation

@article{Li2025Exploring,
  author = {Li, Wenhao and Gao, Shuanglong and Mu, Xiaoguo and Wen, Yue and Javed, Tehseen and Wang, Zhenhua},
  title = {Exploring the driving forces of soil salinity reduction using Random Forest and SHAP in water-saving oasis irrigation areas},
  journal = {Irrigation Science},
  year = {2025},
  doi = {10.1007/s00271-025-01045-6},
  url = {https://doi.org/10.1007/s00271-025-01045-6}
}