Abstract: Our understanding of water cycle dynamics is limited owing to uncertainties regarding global evapotranspiration (ET) and its drivers under climate change. This study aimed to improve the precision of global ET estimates using the remote sensing Penman–Monteith (RS-PM) model and reveal the driving factors of ET variation using geographical detectors. The RS-PM model accurately predicted the 8-d ET, with results comparable to those of the Global Land Surface Satellite (GLASS) algorithm. The RS-PM ET estimates for evergreen broadleaf forests and mixed forests had relatively high accuracy, whereas those for shrublands and wetlands had large uncertainties. The RS-PM model overestimated ET in high- and mid-latitude regions of the Northern Hemisphere and underestimated ET in low-latitude regions, the Southern Hemisphere, and regions between 90° and 180° E. Net radiation (NR) was the dominant factor driving global 8-d ET variation, whereas leaf area index (LAI) acted as the dominant driver of annual ET variation across most climatic zones. Precipitation was the main driver of annual ET in semi-arid and arid regions, in southern Africa, and most of Asia, while NR, temperature, and vapour pressure deficit (VPD) dominated annual ET in regions characterized by limited solar energy and those at high latitudes. The effects of LAI-VPD interactions had high explanatory power for ET variation in most regions; however, interaction effects varied across latitudes and climatic zones. This study provides a valuable reference for improving the RS-PM model to predict global ET and discovers discrepancies in the dominant factors on ET variation across global regions.

Key words: Evapotranspiration, Remote sensing, Climatic factor, Leaf area index, Interaction effect