An NDVI-based statistical ET downscaling method

Abstract

This study proposes a new method for downscaling ETWatch 1-km actual evapotranspiration (ET) products to a spatial resolution of 30 m using Landsat8 normalized difference vegetation index (NDVI) data. The NDVI is employed as an indicator of land-surface vegetation, which displays periodic spatial patterns on the land surface. A 30-m-resolution ten-day ET dataset is then calculated primarily using the NDVI and the historical ratio of coarse NDVI and ET that considers different land cover types. Good agreement and correlations were obtained between the downscaled data and observations from three flux sites in two study areas. The mean bias (MB) per ten-day period ranges from 4.21 mm in Guantao to 1.55 mm in Huazhaizi, and the coefficient of determination (R2) varies from 0.87 to 0.95. The downscaling results show good consistency with the original ETWatch 1-km data over both temporal and spatial scales for different land cover types, with R2 values ranging from 0.82 to 0.98. In addition, the downscaled results capture the progression of vegetation growth well. This study demonstrates the applicability of the new "de-pixelation" downscaling method in the management of water resources.