A novel satellite-based hybrid triple source (HTS) model of terrestrial evapotranspiration

Recent developments in satellite-based evapotranspiration (ET) have generated methodological advances across different facets of ET, yet they remain as independent methods. Here, we combined these disparate methods into an innovative satellite-based hybrid triple-source (HTS) model that couples: i) a thermal infrared (TIR)-based temperature domain-differencing model for estimating daily soil evaporation (〖ET〗_s); ii) a visible and near-infrared (VNIR)-based enhanced Priestley-Taylor (PT) model that incorporates ecophysiological and biometeorological constraints for estimating daily vegetation transpiration (〖ET〗_v); and, iii) a VNIR-based improved interception loss model for estimating daily canopy interception evaporation (〖ET〗_i). Our proposed HTS model directly utilizes the variation in instantaneous land surface temperature for estimating daily 〖ET〗_s, thereby avoiding the need to upscale instantaneous 〖ET〗_s estimates. It introduces a nonlinear synthesis function of the NDVI to characterize vegetation ecophysiological and biometeorological constraints and also considers the influence of atmospheric 〖CO〗_2 levels on 〖ET〗_v to improve transpiration estimation. We validated the HTS model using ground observations from 38 in situ sites (36 eddy covariance sites and 2 sites with sap flow and microlysimeter measurements) across mainland China. The results demonstrate that the HTS model forced by MODIS and site-observed meteorological data, provides reliable daily ET estimates across various plant functional types (PFTs) [coefficient of determination (R^2) of 0.50–0.93 (p<0.01), Bias of -0.33–0.17 mm/day, root mean square error (RMSE) of 0.46–0.80 mm/day, and Kling-Gupta efficiency (KGE) of 0.65–0.91]. The HTS model shows satisfactory accuracy for ET component (〖ET〗_i, 〖ET〗_s and 〖ET〗_v) estimation, with an average KGE greater than 0.50 at both the Tuqiaogou (TQG) and Yangjuangou (YJG) sites. Compared to the RS-PM and the PT-JPL models, the HTS model demonstrates superior performance across most PFTs and incorporates more robust physical processes. In comparison with GLASS, PML_V2, BESS, and MOD16A2 ET datasets, the HTS-derived ET results demonstrated the best consistency with the calculated ET using water balance approach (KGE=0.95) at basin scale. Furthermore, HTS provides more detailed spatial ET information than other satellite datasets, effectively depicting the distribution of irrigated oases and seasonal rivers across Inland River basins. The HTS-derived ET dataset shows a substantial increase in ET across 28.9% of mainland China from 2002 to 2020，primarily associated with variations in precipitation (P) and vegetation greening due to crop expansion and afforestation programs. This HTS model provides an alternative method for generating a reliable terrestrial ET dataset, useful for investigating regional energy, water, and carbon budgets.