Study on the inversion and spatiotemporal variation mechanism of soil salinization at multiple depths in typical oases in arid areas: A case study of Wei-Ku Oasis

作者：Jinming Zhang, Jianli Ding, Zihan Zhang, Jinjie Wang, Xu Zeng, Xiangyu Ge · 发表于：Agricultural Water Management · 年份：2025 · DOI：10.1016/j.agwat.2025.109542 · 被引用次数：25 · 研究领域：Soil Geostatistics and Mapping、Plant Ecology and Soil Science、Remote Sensing and Land Use

Soil salinization is a widespread issue in arid and semi-arid regions, severely threatening agricultural productivity and environmental sustainability. However, accurately modeling and predicting soil salinity at multiple depths over time remains a challenge due to complex interactions among environmental factors and limited ground observations. Understanding the spatiotemporal characteristics of soil salinity and its driving factors is essential for formulating more scientific and rational irrigation strategies and remediation methods. Taking the Wei-Ku Oasis, a typical arid region oasis, as an example, this study uses Landsat remote sensing imagery as the data source, incorporating soil salinity field measurements over a decade, employing the Bootstrap Soft Shrinkage(BOSS) algorithm to select feature variables, and building soil salinity inversion models at various depths through a Convolutional Neural Networks and Long Short-Term Memory networks (CNN-LSTM) framework. The spatiotemporal variation of soil salinity in the Wei-Ku Oasis is analyzed, and the driving mechanisms of soil salinity change in the study area are explored using the optimal geographic detector. Results indicate that: (1) The multi-depth soil salinity inversion models built with the CNN-LSTM framework exhibit superior predictive performance, with the 0–10 cm soil salinity prediction model achieving the highest accuracy, attaining an R² of 0.7 in the test set. The R² values for the test set of soil salinit...