High spatio-temporal resolution evapotranspiration estimates within large agricultural fields by fusing eddy covariance and Landsat based data

Accurate estimates of evapotranspiration (ET) are difficult to quantify at varying spatial and temporal scales. Eddy covariance (EC) methods estimate ET at high temporal resolutions (30 min), but with little knowledge regarding its spatial variation. In contrast, remote sensing-based methods using Landsat (7 and 8) provide high spatial resolution (30 m) ET with low temporal resolution (8 to16-days for 2 concurrent Landsat platforms). In this study, we developed a new algorithm to generate high spatio-temporal (daily 30 m resolution) ET (ETFUSE) within large agricultural fields by fusing eddy covariance and Landsat ET data. ETFUSE was compared with standardized Penman-Monteith ET (ETPM) and spline interpolated alfalfa reference fraction ET (ETRF) at six sites in the Continental United States. Spatial cross-correlation (using NDVI as a covariable), EC flux footprint modeling, and a source weighted scaling relationship between EC footprints and Landsat ET were used in fusion algorithm to generate ETFUSE. Using Ameriflux and Texas Water Observatory sites as testbeds, ET dynamics were found statistically similar for ETFUSE, ETRF, and ETPM for various land covers and growing seasons. Correlation coefficients for ETFUSE compared with ETPM were 0.77-0.96 at the study sites, during the growing seasons from 2016 to 2019. RMSEs and MAEs ranged between 0.35-0.96 mm d-1 and 0.28-0.51 mm d-1, respectively, for ETFUSE compared to ETPM. The ETFUSE algorithm is limited for use up to 81 km2 extents centered around EC towers. ETFUSE provides spatially variable ET, reflecting areas with low and high ET, useful in variable-rate irrigation systems for precision agriculture water management or in hydrologic models.