High-spatiotemporal-resolution estimation of solar energy component in the United States using a new satellite-based model.

Diffuse solar radiation (Rd), known as an important component of global solar radiation (Rg), is a key parameter for solar energy related applications and ecosystem photosynthesis. Some meteorological models have been developed to estimate Rd with acceptable accuracy, but their spatial scales are often small due to the limited meteorological station number. Satellite-based models provide accurate and large-scale Rg estimates. However, remote sensing estimations of Rd are often with low spatial resolutions and large uncertainties, because their methods were based on inaccurate surface and atmospheric parameters. To address these challenges, the high-spatiotemporal-resolution (half-hourly and 1-km) Rd in the United States was estimated using the top-of-atmosphere (TOA) data from the new-generation geostationary satellite Geostationary Operational Environmental Satellites (GOES-16) and the method iterative random forest (RF). The results showed that the iterative RF model had higher accuracy than the simple RF and artificial neural network (ANN) models, and using TIR (thermal infrared) bands can improve models' accuracy. The best model can estimate half-hourly Rd with the accuracy R ²  = 0.88, RMSE = 37.81 W/m ² , and MBE = 0.01 W/m ² . Compared with the previous 0.01-degree (∼11-km) Rd product Earth Polychromatic Imaging Camera (EPIC), the GOES-16 estimated 1-km Rd had similar spatial patterns. Moreover, based on the GOES-16 estimated half-hourly and 1-km Rd in the United States, the spatiotemporal heterogeneity of Rd was quantitatively observed. The proposed approach can be used to produce more high-spatiotemporal-resolution Rd products, and these products are very helpful for many solar-related research topics and industrial applications.
(Copyright © 2021 Elsevier Ltd. All rights reserved.)