Nonlinear influences of climatic, vegetative, geographic and soil factors on soil water use efficiency of global karst landscapes: Insights from explainable machine learning

Abstract

Soil Water Use Efficiency (SWUE) represents a vital metric for assessing the relationship between carbon acquisition and soil moisture (SM) depletion in terrestrial ecosystems. However, the elucidation of time-lagged and cumulative effects, nonlinear influences, and indirect contributions of explanatory variables, including climate and vegetation characteristics, on SWUE in global karst landscapes remains limited. In this study, we analyzed the time-lagged and cumulative effects of climatic and biological factors on SWUE in global karst landscapes using the Autoregressive Distributed Lag Model. By comparing nine machine learning models, we further revealed the nonlinear effects, as well as the direct and indirect contributions of climatic, geographic, soil, and biological explanatory variables on SWUE across varying aridity, using the Random Forest Model, SHapley Additive exPlanations, Generalized Additive Model, and Partial Least Squares-Structural Equation Modeling (PLS-SEM). The findings suggested that precipitation and wind speed exert the most substantial time-lagged and cumulative impacts on SWUE in global karst landscapes, respectively. The Random Forest model outperforms eight other machine learning models, including CatBoost, LightGBM, and XGBoost, in accurately simulating SWUE. In global karst landscapes, SWUE was significantly affected by the positive contributions of evapotranspiration, leaf area index, and temperature, as well as the negative impacts of latitude and longitude. These influences exhibited varying degrees of nonlinearity across the aridity gradient. Using PLS-SEM based on the ‘geo-climatic-soil-biological’ cascade effect, it was found that gross primary production directly and significantly influences karst SWUE under both drought-prone and water-abundant conditions, significantly exceeding the impact of SM. Geographic, climatic, and biological factors indirectly influenced karst SWUE by affecting gross primary production. The impact of soil type, soil carbon and nitrogen content, and rootable depth on SWUE was minimal. This study enhances our understanding of carbon sinks and the water‑carbon cycle, providing valuable insights into resource use efficiency within karst environments.