Seamless terrestrial evapotranspiration estimation by machine learning models across the Contiguous United States

Terrestrial evapotranspiration (ET) is an important control factor for water cycling and energy transport, serving as a crucial link between ecological and hydrological processes. Accurate estimation of ET is essential for enhancing efficient utilization of water resources, improving agricultural productivity, preserving ecosystems, and advancing climate change research. Despite its significance, high spatiotemporal resolution continuous ET datasets remain scarce. In ET estimation, machine learning methods have been widely adopted, with tree -based machine learning models gaining increasing attention due to their computational efficiency and reliability accuracy. However, research comparing the performance of these models remains relatively limited. In this study, we use data from flux observation sites and various remote sensing sources to explore the performance of four tree -based machine learning models in ET estimation across the Contiguous United States (CONUS). Our findings demonstrate the proficient performance of all four models in estimating terrestrial ET across CONUS. Particularly noteworthy is the outstanding performance of the extremely randomized trees (ERT) model, showing a high correlation (R 2 = 0.84), low bias (BIAS = -0.0003 mm/d), and low root mean square error (RMSE = 0.72 mm/d) with flux observation site data. Using this model, we successfully obtained a seamless terrestrial ET dataset (ERT_ET) with a spatial resolution of 1 km and multiple temporal resolutions (daily, 8 -day, monthly, and seasonal) from 2008 to 2018 across the CONUS. Compared to the MOD16 ET product, the ERT_ET outperforms with a higher R 2 by 0.40 and lower RMSE by 5.31 mm/8d, providing better performance in capturing detailed features. Moreover, our ERT_ET product is comparable to other widely used ET products (MOD16, PML-V2, and ETMonitor), further highlighting its reliability. These findings will contribute to studies in various fields, including global climate change, hydrological cycles, and drought monitoring.