Search

Cloud cover is the main physical factor limiting the downward short-wave (SW) solar radiation flux. In climate and weather forecasting, physical models describing radiative transfer through clouds may be used. However, this option is computationally expensive. Instead, one may use parameterizations, which are simplified schemes for approximating environmental variables including SW radiation flux. In this study, we assessed the capability of machine learning (ML) models in the scenario of statistical approximation of downward SW radiation flux using all-sky optical imagery. We assume that an all-sky photograph contains complete information about the downward SW radiation. We examined several types of ML models that we trained on a dataset of all-sky images each accompanied by a SW radiation flux measurement. The Dataset of All-Sky Imagery over the Ocean (DASIO) containing the imagery and flux values is collected during several oceanic expeditions in Atlantic, Indian and Arctic oceans. We found out that the quality of the best classic ML model is better compared to existing SW radiation parameterizations known from literature. We demonstrate the results of our study regarding classic ML models as well as the results of an end-to-end ML approach involving convolutional neural networks. Our results allow one to acquire downward SW radiation fluxes directly from all-sky imagery with reasonable quality. [ABSTRACT FROM AUTHOR]