Understanding Evapotranspiration Driving Mechanisms in China With Explainable Machine Learning Algorithms

ABSTRACT Evapotranspiration (ET) is the second component of the water cycle, serving as a critical link among surface water, energy and carbon cycles. In general, ET mechanism studies mainly use statistical regression and numerical model sensitivity experiments. However, the use of a nonlinear explainable machine learning algorithm, which has become an effective tool for studying earth science, remains unexplored to study the ET mechanism. Thus, this study employs various explainable methods to quantify the effects of different climatic variables on ET changes across China and four sub‐regions (Arid, humid, transition and Qinghai‐Tibet Plateau regions). It was found that precipitation, temperature and leaf area index contributed the most to ET changes between 1981 and 2018. There was a clear spatial distribution of the dominant factor: Precipitation was the primary driver of ET changes over water‐limited regions, while air temperature dominated ET changes over energy‐limited regions. Meanwhile, we also compared the effect of each variable on ET with the changes of the respective variable and remaining variables. Air temperature would increase the effects of precipitation on ET changes, and air temperature's contribution was also amplified by precipitation. Our results confirm the effectiveness of using the explainable machine learning algorithm to study the hydrological cycle.