Assessment of satellite- and reanalysis-based evapotranspiration products with two blending approaches over the complex landscapes and climates of Australia

Numerous evapotranspiration (ET) products with different spatio-temporal resolutions at continental and global scales have been developed by utilizing commonly available satellite imagery and ground-based observations. However, the developed ET products present large uncertainties that limit their operational hydro-meteorological applications, specifically in water limited regions where water consumption is competing. This study presents the uncertainties among four widely available ET (GLDAS, GLEAM, MOD16, and MERRA) products and then assesses the performance of two blending approaches (Maximize R and simple Taylor skill's score; STS)) for generating a fused ET product using combinations of the above ET datasets in a dry continent (Australia) during the period of 2005-2014. The accuracy for all four ET products compared with Australian Water Resource Assessment Landscape (AWRA-L) ET dataset demonstrated large uncertainties across seven different land cover classifications and four different climate zones, with a coefficient of correlation (R) ranging between 0.1-0.85. GLEAM and GLDAS showed better agreement (R ∼ 0.8) over forest and cropland areas, respectively, while better performance of all products was noted in tropical regions compared with other climate zones in the region. Similarly, implementation of the two blending approaches to generate merged ET products revealed an overall higher contribution (>25%) from GLEAM followed by the GLDAS, while that of MOD16 was lowest (<20%) over various climatic zones as well as all of Australia. The accuracy assessment of the two merged ET products exhibited relatively better performance by reducing the Bias toward 0 and Root Mean Square Error (RMSE) of 0.2-0.4 mm/8day with an index of agreement (IOA) >0.8 compared with individual ET products under all climatic and land cover conditions. These statistical indicators explained the relatively large differences among the spatial distributions of fused ET and those of individual products. Among the two blending approaches, the STS method produced more reliable results compared to the accuracy of fused ET generated by the Maximize R method. This reason for this was the discrepancy in the number of datasets used to derive each merging method. The STS method allows combining all given datasets by providing their corresponding weights, whereas the Maximize R method employed the weights only based on the two best of the given products. Overall, the assessments made in the current study provide comprehensive insights on the quality and integration of four globally available ET products to benefit regional water users and the hydro-meteorological community on the basis of major landscape and climatic conditions over long time scales.