Linking ECOSTRESS, MODIS and Eddy-covariance for high-resolution daily evapotranspiration

Abstract:

Launched in June 2018 by the National Aeronautics and Space Administration (NASA), ECOsystem Spaceborne Thermal Radiometer Experiment (ECOSTRESS) provides high resolution evapotranspiration (ET) at 70m with a variable temporal latency across space. NASA also provides 8-day averaged ET using Moderate Resolution Imaging Spectroradiometer (MODIS) at 500m spatial resolution around the globe. While ECOSTRESS and MODIS contain critical spatial information about ET, they have little temporal data at a daily scale. Eddy-covariance, conversely, provide daily temporal ET information but is only representative of a small area. High resolution ET across space and time is critical for improved water-budget estimation and irrigation-scheduling systems; ECOTRESS, MODIS and Eddy-Covariance provide incomplete fragments of spatio-temporal variability of ET. Data-driven fusion algorithms which combine data from these platforms—while accounting for individual strengths and weaknesses—are thus crucial for providing ET estimates across space and time.

Texas Water Observatory (TWO) established in 2015 consists of Eddy-covariance towers installed under diverse land-covers and surface heterogeneity across the Brazos basin in Texas, US. Using the Brazos Basin as our study area, we present a multi-scale spatio-temporal algorithm combining ET data from Eddy-covariance, ECOSTRESS and MODIS while accounting for 1) inherent spatio-temporal dependence of ET, 2) different resolution of the platforms across space and time, 3) bias and errors inherent in remote sensing platforms, and 4) the effect of controls such as vegetation, soil texture and land-cover on ET distribution. We first present an exact likelihood-based algorithm to fuse the three data platforms for a small study area of 36 km² in the Brazos Basin. We quantify the effect of different controls on the ET distribution and validate our ET predictions at multiple scales. We then present a likelihood approximation to extend the algorithm for Big Data and fuse ET for the entire Brazos Basin. We generate high-resolution daily ET maps across the basin and provide 3-7 day ET forecasts with uncertainty metrics. Finally, we discuss the potential application of the fusion algorithm for the entire Contiguous US.