Global Assessment of Terrestrial Precipitation and Evapotranspiration in CMIP6 Simulations Using Observation-Based Estimates

Shanshan Sun ^1* Shiqin Xu ² Lingcheng Li ³ Yihua Lin ⁴ Hongbo Liu ¹ Viviana Maggioni ⁵ Yan Xu ¹ Congsheng Fu ⁶

• ¹ Yunnan University, Kunming, China
• ² King Abdullah University of Science and Technology, Thuwal, Makkah, Saudi Arabia
• ³ Pacific Northwest National Laboratory (DOE), Richland, Washington, United States
• ⁴ Institute of Atmospheric Physics, Chinese Academy of Sciences (CAS), Beijing, Beijing Municipality, China
• ⁵ George Mason University, Fairfax, Virginia, United States
• ⁶ Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences (CAS), Nanjing, Jiangsu Province, China

Rising concerns about climate change underscore the need to understand precipitation and evapotranspiration variability across multiple temporal scales. This study evaluates historical simulations from Phase Six of the Coupled Model Intercomparison Project (CMIP6) for precipitation (Pr), evapotranspiration (ET), and its components-soil evaporation (Es), transpiration (Et), and interception loss (Ei)-from 1981 to 2014, focusing on the temporal agreement of the mean seasonal cycle and interannual variability. We assess these variables using observation-based estimates from three Pr datasets (CRU4.0, GPCP v2.3, ERA5) and four land surface flux datasets (GLEAM v3.3a, GLDAS v2.0, ERA5-Land, MERRA-Land). Pearson's correlation coefficients (r) are used to identify "consensus regions". The results indicate that consensus regions of the mean seasonal cycle for Pr cover 92.9% of global land area, decreasing to 81.7% at the interannual scale. For ET and its components, the consistency of the mean seasonal cycle is observed over 79.0% of land area for ET, 55.5% for Es, 57.7% for Et, and 65.1% for Ei, with values dropping to 38.1%, 11.7%, 23.4%, and 21.2%, respectively, at the interannual scale.The multi-model means generally correlate better with observations than individual CMIP6 models. Across latitudes, Pr and ET exhibit the highest performance in reproducing the observed mean seasonal cycle, while Es and Et demonstrate the lowest performance. CESM2 shows the highest consistency in reproducing the mean seasonal cycle for Pr, while CMCC-CM2-HR4 performs best for ET and its components. Despite relatively high correlations with the observed mean seasonal cycle, the individual models and multi-model mean underestimates Pr in tropical regions and overestimates ET, Es, and Ei, while underestimating Et in general. The agreement between CMIP6 simulations and observational datasets deteriorates at the interannual scale. These findings highlight the need to improve Pr and ET simulations in CMIP6 models, particularly in tropics.