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

Uncertainty analysis of evapotranspiration models is essential in hydrological modeling, particularly given the limited use of process-based models and ensemble algorithms for evapotranspiration estimation. In this study, the performance and uncertainty in two simplified process-based models (BTA and BTA-θ), and three classical ensemble algorithms (adaptive boosting, random forest, and extreme gradient boosting) in estimating evapotranspiration are assessed across various ecosystems. The results show that: (a) the BTA-θ model outperforms the BTA model across all ecosystems, and the integrated algorithms perform better than the two process-based models. (b) In the two process-based models, parameter k is insensitive to the outcomes, while parameters b, k1, and θ1 are highly sensitive. (c) The primary source of uncertainty in the ensemble algorithms is parameterization, the structure of the BTA model, and the input data of the BTA-θ model. The findings provide valuable insights for increasing the evapotranspiration estimation accuracy and reliability. This study helps to improve the accuracy and reliability of evapotranspiration estimation in terrestrial ecosystems.