Global pattern and mechanism of terrestrial evapotranspiration change indicated by weather stations

TL;DR

This study uses weather station data and machine learning to estimate global terrestrial evapotranspiration, revealing its increase over time and identifying key climatic drivers, thus improving understanding of water cycle changes.

Contribution

It introduces a novel approach combining weather station data and machine learning for global ET estimation, enhancing accuracy and regional understanding.

Findings

Global ET increased from 493 to 522 mm yr-1 (2003-2019)

61.7% of stations showed ET increase between two periods

Main drivers of ET change include temperature, radiation, vegetation, and vapor pressure deficit

Abstract

Accurate estimation of global terrestrial evapotranspiration (ET) is essential to understanding changes in the water cycle, which are expected to intensify in the context of climate change. Current global ET products are derived from physics-based, yet empirical, models, water balance methods, or upscaling from sparse in situ observations. However, these products contain substantial limitations such as the coarse resolution due to the coarse climate reanalysis forcing data, the assumptions on the parameterization of the process, the sparsity of the observations, and the lack of global accuracy validation. Using estimates of ET based on the global weather station network and machine learning, we show that global ET ranged from 493 to 522 mm yr-1 and increased at the rate of 0.60 mm yr-2 from 2003 to 2019. Between the two periods of 2003-2010 and 2011-2019, 61.7% of stations showed an increase in ET. At the large river basin scale, the reliability of the produced ET in this study is comparable to gridded ET data and even higher in regions where weather stations are relatively dense and more representative. Correlation analysis and causal network analysis showed that the main drivers of ET long-term changes are changes in air temperature, radiation, vegetation conditions, and vapor pressure deficit. There is great variability in the causal mechanisms of ET change across vegetation cover and across seasons. This study highlights the promise of using weather stations to complement global ET and water cycle studies at the station scale.