Uncertain Effect of Component Differences on Land Evapotranspiration

Abstract

Actual evapotranspiration (ETa) plays a key role in the redistribution of water, carbon and energy. The emergence of many ETa products has made uncertainty assessment increasingly important. The FLUXNET2015 dataset and 28 large watershed water balance datasets were used in this study. The monthly scale products of the ERA5, GLDAS, MERRA and PML terrestrial evapotranspiration models were evaluated from 2001 to 2017. The differences in composition and vegetation among the four ETa products were compared. The results showed that the site-scale evaluation performance was good, with R2>0.6 for approximately 64%~75% of the sites. Notably, overestimations were observed in 89.29% of the basins (bias>0), with significant overestimations in the Amazon, Ndavina, Yangtze and Zhujiang basins. There were large errors in the estimates of T and Es in the ERA5 product, related to the overestimation of Es and underestimation of T. The boundary between sea and land  was unclear in the PML products. Eo/ETa was overestimated, and there were clearly high values at the land margin (Eo peaks as high as 3803 mm/yr). The difference in evapotranspiration components had a considerable influence on the uncertainty of ETa. The vegetation types in the 4 ETa products for DBF***, EBF***, ENF***, MF***, GRA***, and CRO*** all exhibited significant differences at the P<0.001 level. This study contributes to product uncertainty analysis and the determination of ways to improve ETa products.

Keywords: land evapotranspiration, Uncertainty, component differences, multisource remote sensing drive