Estimating half-hourly solar radiation over the Continental United States using GOES-16 data with iterative random forest

Abstract

To reduce carbon emissions, using more solar energy is a feasible solution. Many meteorological-based models can estimate global downward solar radiation (DSR), but they are with limited applications due to the point-based estimation and low temporal resolution. Satellite remote sensing-based models can estimate DSR with better spatial coverage. However, most previous models are restricted to estimate clear-sky or monthly scale DSR at several sites, limiting the solar energy monitoring of nationwide scale. In this study, using high spatiotemporal resolution Geostationary Operational Environmental Satellites (GOES)-16 satellite data, an iterative random forest (RF) model was developed to estimate and map half-hourly DSR at 1-km spatial resolution over the Continental United States (CONUS). The results show that the iterative RF model performed better than multiple linear regression (MLR) and traditional RF models. The accuracy of estimating half-hourly DSR is that R2 = 0.95, root-mean-square-error (RMSE) = 66.92 W/m2, and mean-bias-error (MBE) = 0.06 W/m2. Half-hourly and daily DSR with spatial resolution 1-km over the CONUS were mapped. The GOES-16 estimated DSR showed the similar spatial patterns with the results from the Clouds and the Earth's Radiant Energy System (CERES) DSR product. This study demonstrated the potential of GOES-16 data for mapping DSR over the CONUS, and hence can be further used in solar energy related applications.