Abstract:Accurate quantification of Terrestrial Evapotranspiration (ET) is crucial for understanding global water resource management amid climate change. However, against the backdrop of frequent global extreme climate events, existing ET products don't accurately reflect ground data under extreme climate conditions (high temperature and high vapor pressure deficit (VPD)) and no studies have addressed this uncertainty. This study, based on data from 150 flux towers (FLUXNET ET), utilizes four daily-scale ET product datasets (REA, GLEAMv3.6b, REA, ERA5) and ERA5 reanalysis meteorological data as inputs for Convolutional Neural Network (CNN), to explore ways to improve the accuracy of ET products under extreme climatic conditions, while three Machine Learning (ML) models were selected as comparisons. An additional 35 nearby sites provide independent validation for model applicability. The results show that the estimation accuracy of ET products under extreme climatic conditions has an average decrease of 37.5 % in the correlation coefficient (R2) and an average increase of 40 % in the root-mean-square-error (RMSE), whereas the CNN significantly improves this high uncertainty and outperforms the other ML models. The accuracy was significantly improved with the addition of ET products (R2 = 0.924, RMSE=0.323 mm/d) compared to having only meteorological data as input (R2 = 0.888, RMSE=0.369 mm/d). Under extreme climatic conditions, the R2 and RMSE of CNN are 0.786(171.6 % average improvement) and 0.448 (65.4 % average reduction) compared to ET products, with PBias ranging from -9% to 9 % and the most significant improvement in the America (region), EBF (land cover type), and Cwa (climate type). Furthermore, using ET products as inputs significantly enhances the CNN's estimation accuracy, yielding R2 and RMSE improvements between 10 % and 30 %. At the 35 independent validation sites, the CNN still performs well, with R2 and RMSE around 0.720 and 0.510 mm/d under extreme climatic conditions, and the accuracy within 500 km is even higher. This study enhances model precision, providing a valuable tool for optimizing water management and tackling climate change. These advancements help formulate precise irrigation strategies and water policies, particularly in extreme weather-prone regions, supporting ecological stability and mitigating water scarcity.

Application and Uncertainty Analysis of Data-Driven and Process-Based Evapotranspiration Models Across Various Ecosystems

Uncertainty in High and Low Flows Due to Model Structure and Parameter Errors

Uncertainty Analysis in Statistical Modeling of Extreme Hydrological Events

Investigating the Uncertainty and Transferability of Parameters in Swat Model under Climate Change

Identifying Precipitation Uncertainty in Crop Modelling Using Bayesian Total Error Analysis

Evapotranspiration Simulations in ISIMIP2a—Evaluation of Spatio-Temporal Characteristics with a Comprehensive Ensemble of Independent Datasets

Integrating partitioned evapotranspiration data into hydrologic models: vegetation parameterization and uncertainty quantification of simulated plant water use

Comparison of Satellite-Based Evapotranspiration Models over Terrestrial Ecosystems in China

Enhancing streamflow estimation by integrating a data-driven evapotranspiration submodel into process-based hydrological models

Improving High Uncertainty of Evapotranspiration Products under Extreme Climatic Conditions Based on Deep Learning and ERA5 Reanalysis Data

Comparison of physical-based, data-driven and hybrid modeling approaches for evapotranspiration estimation

Evaluating the uncertainty of terrestrial water budget components over High Mountain Asia

Parameter Analysis and Estimates for the MODIS Evapotranspiration Algorithm and Multiscale Verification

Effects of Conceptual Model Structure Uncertainties on Actual Evapotranspiration Simulation

Estimation of actual evapotranspiration from different ecosystems on the Tibetan Plateau based on a generalized complementary evapotranspiration theory model

Parameter estimation and uncertainty analysis in hydrological modeling

Uncertainty quantization of meteorological input and model parameters for hydrological modelling using a Bayesian‐based integrated approach

Reconciling Global Terrestrial Evapotranspiration Estimates From Multi‐Product Intercomparison and Evaluation

Using Precipitation, Vertical Root Distribution, and Satellite‐retrieved Vegetation Information to Parameterize Water Stress in a Penman‐Monteith Approach to Evapotranspiration Modeling under Mediterranean Climate

Uncertainty in Evapotranspiration from Land Surface Modeling, Remote Sensing, and GRACE Satellites

Integrating Temperature Gradient-Based 3T and Resistance-Based Models for Simulating Evapotranspiration and Its Components