Controls of groundwater-dependent vegetation coverage in the yellow river basin, china: Insights from interpretable machine learning

Groundwater plays an important role in maintaining vegetation structure and ecological functions in arid regions. However, the relationship between groundwater depth (GD) and fractional vegetation cover (FVC) in a large region with various climate conditions remains poorly understood, primarily due to the complex influence of multiple environmental factors. Machine learning method, a powerful tool for identifying complex and nonlinear processes, has been used to rank explanatory variables on the regional-scale vegetation distribution, while missing the vegetation-groundwater relationship. We developed machine learning models via the extreme gradient boosting (XGBoost) to identify key controls of groundwater-dependent vegetation cover in the Yellow River Basin, China. Two XGBoost models, A and B, were constructed for shallow (GD <= 24.5 m) and deep (GD > 24.5 m) groundwater pixels, respectively, with the same sample number on the resolution of 1 km. Shapely additive explanations (SHAP) method is employed to assess the contributions of climatic, topographic, and edaphic features on FVC. Through an enhanced feature selection method using multicollinearity analysis and the Boruta algorithm, we found that the top four important features in both models were precipitation (P), saturated water content (SWC), air temperature (TA), and potential evapotranspiration (PET) during the growing season. The difference between models A and B indicates the influence of groundwater on vegetation. Shallower groundwater leads to smaller sensitivity of FVC to the precipitation when the monthly P in the growing season is smaller than 68.0 mm, and can also reduce the sensitivity of FVC to the air temperature when the average TA is lower than 13.5degree celsius in the growing season. This study suggests an effective method for recognizing the groundwater effect on the vegetation cover from mixed influences of environmental controls.