Triple collocation-based multi-source evaporation and transpiration merging

The main objective of this study is to combine three reanalysis-based evaporation (E) and transpiration (T) es-timates (i.e., ERA5-Land, GLDAS, and MERRA2) to improve the accuracy of evapotranspiration (ET) over East Asia. Here, we calculate the uncertainty of E and T separately through triple collocation (TC), which then used as a weighting factor to produce merged estimates of E and T. The combination of the two was quantitatively evaluated by comparing it with ground-based measurements. Before assessing the merged ET estimates, the error standard deviation (sigma mean) of TC was analyzed. ERA5-Land showed the smallest sigma mean for E for most of the study region, although MERRA2 achieved superior performance for T compared with ERA5-Land. This behavior further resulted in the highest weighting factors of ERA5-Land (59%) and MERRA2 (49%) for E and T, respectively. An evaluation of merged ET estimates by comparison with flux tower measurements revealed the close agreement of the two datasets in crop and forest regions. For croplands, the merged ET estimates showed better statistics compared with the two worst original products. However, the merged ET showed limitation in reflecting crop rotation due to the uncertainty of ERA5-Land E, which yielded the largest weighting factor over cropland. At forest sites, merged ET estimates yielded better statistical performance than the three individual ET products as well as the ET estimated through TC without separating E and T. Detailed analysis of the ratio of T and ET revealed that the merged ET was similar in magnitude to ground-based measurements. Spatial distribution of annual and seasonal ET indicated that the overall improvement of merged ET resulted from superior estimates of T during the spring and summer. These results confirm that separate error characterization of E and T through TC can enhance the accuracy of terrestrial ET estimates.