Wartenburger, Richard, Seneviratne, Sonia I., Hirschi, Martin, Chang, Jinfeng, Ciais, Philippe, Deryng, Delphine, Elliott, Joshua, Folberth, Christian, Gosling, Simon N., Gudmundsson, Lukas, Henrot, Alexandra-Jane, Hickler, Thomas, Ito, Akihiko, Khabarov, Nikolay, Kim, Hyungjun, Leng, Guoyong, Liu, Junguo, Liu, Xingcai, Masaki, Yoshimitsu, Morfopoulos, Catherine, Müller, Christoph, Müller Schmied, Hannes, Nishina, Kazuya, Orth, Rene, Pokhrel, Yadu, Pugh, Thomas A. M., Satoh, Yusuke, Schaphoff, Sibyll, Schmid, Erwin, Sheffield, Justin, Stacke, Tobias, Steinkamp, Joerg, Tang, Qiuhong, Thiery, Wim, Wada, Yoshihide, Wang, Xuhui, Weedon, Graham P., Yang, Hong, and Zhou, Tian
Actual land evapotranspiration (ET) is a key component of the global hydrological cycle and an essential variable determining the evolution of hydrological extreme events under different climate change scenarios. However, recently available ET products show persistent uncertainties that are impeding a precise attribution of human-induced climate change. Here, we aim at comparing a range of independent global monthly land ET estimates with historical model simulations from the global water, agriculture, and biomes sectors participating in the second phase of the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP2a). Among the independent estimates, we use the EartH2Observe Tier-1 dataset (E2O), two commonly used reanalyses, a pre-compiled ensemble product (LandFlux-EVAL), and an updated collection of recently published datasets that algorithmically derive ET from observations or observations-based estimates (diagnostic datasets). A cluster analysis is applied in order to identify spatio-temporal differences among all datasets and to thus identify factors that dominate overall uncertainties. The clustering is controlled by several factors including the model choice, the meteorological forcing used to drive the assessed models, the data category (models participating in the different sectors of ISIMIP2a, E2O models, diagnostic estimates, reanalysis-based estimates or composite products), the ET scheme, and the number of soil layers in the models. By using these factors to explain spatial and spatio-temporal variabilities in ET, we find that the model choice mostly dominates (24%–40% of variance explained), except for spatio-temporal patterns of total ET, where the forcing explains the largest fraction of the variance (29%). The most dominant clusters of datasets are further compared with individual diagnostic and reanalysis-based estimates to assess their representation of selected heat waves and droughts in the Great Plains, Central Europe and western Russia. Although most of the ET estimates capture these extreme events, the generally large spread among the entire ensemble indicates substantial uncertainties., Environmental Research Letters, 13 (7), ISSN:1748-9326, ISSN:1748-9318