Mechanistic analysis of soil salinity evolution under long-term mulched drip irrigation in a typical oasis irrigation district of Xinjiang of China

Soil salinization poses a significant challenge for sustainable development in oasis irrigation districts. This study focused on mulched drip-irrigated cotton fields in the downstream area of the Manas River Basin. Using machine learning and Shapley Additive Explanations (SHAP) algorithms, the long-term dynamics of soil salinity in the 0-100 cm soil profile and its key driving factors from 2013 to 2021 were analyzed. The results revealed that prolonged drip irrigation led to a notable decline in soil salinity within the 0-100 cm profile. In Field A, average salinity decreased from 5.58 g/kg to 2.39 g/kg, while in Field B, it declined from 11.39 g/kg to 3.88 g/kg. The random forest (RF) model exhibited robust predictive performance, achieving a coefficient of determination (R²) of 0.886 on the test dataset. The importance ranking of the influencing factors was as follows: groundwater depth>irrigation amount>groundwater mineralization>annual evaporation>annual precipitation>soil bulk density. SHAP analysis further revealed that irrigation amounts below 6500 m³/hm², groundwater depths exceeding 3.3 m, and groundwater mineralization levels below 8.75 g/L were associated with reduced soil salinity accumulation. The threshold ranges for mitigating salinity were identified as a groundwater depth between 3.3 and 3.5 m and an irrigation amount ranging from 6500 to 6800 m³/hm². The results can be a reference for effective soil salinity control strategies in oasis irrigation systems.