Developing a 5-min 2-km solar radiation product from GOES-R geostationary satellite towards hourly 10-m ET estimations in real time for the Continental United States

Abstract:

Solar radiation incident on land surface is the largest component of surface energy budget and the primary driver of evapotranspiration (ET). While there is a urgent demand for high resolution ET datasets for real time applications, a key limitation is the reanalysis solar radiation products as ET model inputs, with coarse spatial resolution (e.g., 0.25°) and large time legacy (e.g., days to weeks). Characterizing spatial and temporal variations of solar radiation in real time is therefore key to fill in the big gap in the resolution mismatch. Here we generated a hyper-spatiotemporal resolution (5 min, 2-km) solar radiation product (HyperRad) over the Continental United States from 2018 to present, by fully taking advantages of a range of cloud and atmosphere data observed by the GOES-R satellite. Specifically, we combined an atmospheric radiative transfer model with several machine learning approaches to estimate direct/diffuse visible/near-infrared/shortwave-infrared solar radiation components. We evaluated our satellite estimations against high-quality field observations from the Surface Radiation Budget (SURRAD) network, and the overall accuracy is about 0.9 for our 5-min 2-km radiation estimates. In addition, our estimates agree well with field observations with regards to diurnal cycle, seasonal cycle, annual sum and spatial distribution. We further used our operational HyperRad product as inputs of a satellite-driven water–carbon–energy coupled bio-physical model BESS-STAIR (version 2) to generate hourly 10-m ET at two counties in the U.S. Corn Belt. We evaluated the ET performance using eddy covariance observations at multiple sites, and the overall accuracy is about 0.85 for our hourly 10-m ET estimates. This work is promising for advancing the estimation of ET and the evaluation of surface energy budget for real time applications.