High uncertainty of evapotranspiration products under extreme climatic conditions

Terrestrial Evapotranspiration (ET) is an important process for understanding regional or global water, energy, and carbon cycles, and with the development of satellite observations and increased research investment, grid ET products covering a broad spatial extent are becoming more readily available. However, as global warming becomes a reality and extreme climate events occur worldwide, existing studies do not go far enough to verify that these grid products can still be used under extreme climate conditions. This study evaluates nine global ET products, including land surface reanalysis products (GLDAS_CLSM, GLDAS_NOAH, ERA5, and FLDAS), remote sensing (RS) products (GLEAM_v3.6b, MOD16A2, and PMLv2), and multi-source data fusion products (REA and Synthesized), using observations from 153 flux towers worldwide. The objective is to evaluate their performance in estimating ET under extreme climatic conditions (high temperature, high vapor pressure deficit (VPD), and drought). The results indicate that the estimation accuracy of all ET products is significantly reduced under extreme climatic conditions, showing high uncertainty, and this impact is most severe in the Americas (AM) region. Overall, multi-source data fusion products showed the best estimation performance and were less affected by extreme climatic conditions. Among the remote sensing modeling products, GLEAM_v3.6b showed the best performance, while MOD16A2 has the lowest estimation accuracy. Land surface reanalysis products were most affected by extreme conditions, with CLSM and NOAH showing similar performance and ERA5 having the largest errors (RMSE_ERA5 = 1.699 mm/d, MAE_ERA5 = 1.294 mm/d). The ET products show significant error fluctuations and overestimation (PBias > 0.5) in most of North America, and there is a decline in simulation accuracy in arid and semi-arid regions near 30°N, with most ET products showing overestimation. The most significant errors were observed in cropland areas (CRO) and deciduous broadleaf forest areas (DBF), with significant overestimation in mixed forest areas (MF). The results of this study provide valuable insights for researchers in selecting ET products under extreme climatic conditions and encourage product developers to consider uncertainty under such conditions, thereby improving product accuracy.