Estimating actual evapotranspiration across China by improving the PML algorithm with a shortwave infrared-based surface water stress constraint

Accurate estimation of evapotranspiration (ET) is essential for the precise quantification of energy and water budgets under climate change. Remote sensing ET models provide an effective way to map ET across different spatial and temporal scales. However, conductance-based ET models such as PML_V2 are associated with limited or no water stress constraints on soil evaporation and canopy transpiration that could cause significant bias for sparse vegetation in arid and semi-arid areas. To meet this challenge for using conductance-based ET models, we proposed to use shortwave infrared information to serve as a water stress constraint to vegetation transpiration and soil evaporation, and an improved ET model (PML_SWIR) was proposed. The PML_SWIR model was calibrated with ET measurements from 21 eddy covariance flux towers distributed across China, and showed good performance for estimating ET (R² = 0.70 and RMSE = 0.72 mm/day) for the cross-validation dataset. PML_SWIR outperformed PML_V2 in estimating ET for arid and semi-arid areas, indicated by RMSE being 7.86 and 25.93 mm/year lower and bias being 4.74 and 16.63 % less compared with PML_V2(China) and PML_V2(Global) for ET estimation over Xinjiang Province. In addition, PML_SWIR was noticeably better than PML_V2 for depicting the ET patterns for these seasonal rivers in the arid areas. The ET values estimated by PML_SWIR were further compared with other ET products. The results indicated that PML_SWIR well characterized the ET pattern in arid and semi-arid areas, and the estimated ET values showed good agreement with the water balance-based ET (R² = 0.87, RMSE = 91.37 mm/year) in major river basins of China. The PML_SWIR ET estimates indicated that 20.2 % of the area of China increased significantly in ET over the study period, mainly due to vegetation greening caused by cropland expansion and the large-scale afforestation program. Overall, our results demonstrated that the incorporation of SWIR-based water stress constraints into the conductance-based ET model was a very promising way for accurately mapping ET in arid and semi-arid areas, and that the PML_SWIR model was highly applicable to regional high spatiotemporal ET mapping.