Evaluation and simulation of terrestrial latent heat flux globally: A collaborative effort utilizing CMIP6 climate models and eddy covariance observations

Exchanging latent heat flux (LE) through evapotranspiration impacts the atmospheric thermodynamics and water cycle. The Earth System Models (ESMs) in the Coupled Model Intercomparison Project Phase6 (CMIP6) are vital to reproduce improved LE variations globally, albeit with significant uncertainties. Meanwhile, the rational attribution to regions of LE simulations is essential for informed water resources administration and climate control. This paper presents the first comprehensive evaluation of the ability of 49 ESMs to model global terrestrial LE during the 2000-2014 period based on 205 eddy covariance (EC) observations. Utilizing the Bayesian Model Averaging (BMA) method, we analyzed the spatial-temporal variation and attribution to regions of global LE during 1980-2014 by combining the top six models with EC observations. The results showed that most ESMs overestimated LE, with an average BIAS of 7.57 W.m(-2), covering -7-17 W.m(-2). Among them, the MIROC6 model evinced the highest predictive skill at various land cover types. Moreover, the BMA-based global average terrestrial LE showed low LE values in dry and cold regions of temperate and cold zones of middle and high latitudes but evidenced high LE values in hot and humid regions of low-latitude tropical zones. The inter-annual variations of the BMA-based global annual LE exhibited a significant linear increasing trend with a 0.027 W.m(-2) slope (P-value <0.05). Further attribution to regions analyses were concluded, considering that LE trends were the same as the temperature trends in the Northern Hemisphere, especially in the middle and high latitudes. These conditions are comparable to radiation in the equatorial regions, correlating substantially with precipitation in dry and semi-dry regions across Asia, Europe, and Africa. In addition, the advantages of CMIP6 ESMs on LE simulations mainly including bias, and interannual variability characteristics were also concluded compared with CMIP5 ESMs.