The divergence between potential and actual evapotranspiration: An insight from climate, water, and vegetation change

Abstract

Recently, unprecedented extreme drought has appeared around the world. As the most direct signal of drought, evapotranspiration deserves a more systematic and comprehensive study. Further depicting their divergence of potential (ET_p) and actual evapotranspiration (ET_a) will help to explore the limitation of evapotranspiration. In this paper, the multi-source remote sensing datasets from the Climate Research Unit (CRU), Gravity Recovery and Climate Experiment (GRACE) and its follow-on experiment (GRACE-FO), the Global Land Data Assimilation System (GLDAS), and the Moderate Resolution Imaging Spectroradiometer (MODIS) during 2002 to 2020 were employed to explore the influence of meteorological, hydrological and botanical factors on ET_p, ET_a and their divergence - reduction of evapotranspiration (E_r) which represents regional vegetation and water limitations. According to the Pearson correlation analysis and the Boruta Algorithm based on Random Forest, the temperature is the first decisive promoter of evapotranspiration in the most area while the sparse vegetation is the primary or second determinant limiting the evapotranspiration in 61.84% of the world. In addition, the Coupled Model Intercomparison Project (CMIP6) data from 2030 to 2090 and the support vector machine regression (SVMR) model were applied to predict the future global ET_p, ET_a and E_r on the pixel scale. Predicted results of the model considering the water change not only can highly improve the model performance (with higher R²), but also can simulate the drought in Europe and the more intense ET_a in Africa. Thus, E_r proposed in this study provide a good reference for regional ET_a except for ET_p. The future evapotranspiration value derived by introducing the water storage changes into the machine learning model in this study is also valuable for climate change adaptation and drought warning.

Keywords: Actual evapotranspiration; Hydrological factors; Machine learning model; Multi-source remote sensing products; Potential evapotranspiration.