Spatiotemporal Dynamics and Evolution of Groundwater‐Dependent Ecosystems in the West Liao River Plain

ABSTRACT

Groundwater‐dependent ecosystems (GDEs) are essential to ecological stability in semi‐arid regions, yet their identification and mechanistic interpretation remain constrained by limited long‐term observations and insufficient model interpretability. Using the West Liao River Plain as a case study, we developed a machine learning framework for dynamic GDE mapping based on groundwater depth, vegetation maps, land use/land cover, multisource remote sensing and environmental variables. XGBoost, instead of random forest, was selected to reconstruct GDE distributions in 2004 and from 2009 to 2022. The spatiotemporal evolution and driving mechanisms of GDEs were analysed by integrating groundwater‐depth validation, SHAP interpretation, DBSCAN clustering, GDE dynamic type classification and logistic deviance partitioning. XGBoost showed good classification performance, supported by temporal generalisation tests and threshold sensitivity analysis. From 2009 to 2022, the GDE area showed an overall fluctuating decline followed by later recovery and was significantly negatively correlated with mean groundwater depth. SHAP identified 1 m soil moisture (SM), NDPI coefficient of variation (GS) and EVI coefficient of variation (GS) as the dominant predictors, with NDPI showing stronger discrimination under low vegetation cover. The modal groundwater depth was 3.87 m for GDEs and 3.94 m for non‐GDEs, suggesting that many GDE pixels may be in a mixed water‐use state transitioning from groundwater dependence to precipitation dependence. Deviance partitioning analysis showed that climate change dominated GDE degradation and recovery from 2009 to 2015, whereas from 2016 to 2022, degradation remained climate dominated but recovery showed a stronger signature of human activity control.