A Framework for Actual Evapotranspiration Assessment and Projection Based on Meteorological, Vegetation and Hydrological Remote Sensing Products

As the most direct indicator of drought, dynamic assessment and prediction actual evapotranspiration (AET) is crucial to regional water resources management. This research aims develop a framework for AET evaluation based on multiple machine learning methods multi-source remote sensing data, which combines Boruta algorithm, Random Forest (RF), Support Vector Regression (SVR) models, employing datasets from CRU, GLDAS, MODIS, GRACE (-FO), CMIP6, covering meteorological, vegetation, hydrological variables. To verify framework, it applied grids South America (SA) as case. The results meticulously demonstrate tendency identify decisive role T, P, NDVI in SA. Regarding projection, RF has better performance different input strategies According accuracy SVR pixel scale, dataset generated by integrating optimal two models. By using parameter inputs models jointly obtain output, become more reasonable accurate. can systematically comprehensively evaluate forecast AET; although products SA cannot calibrate relevant parameters, provides quite valuable reference drought warning allocating.