LAUSR

Abstract Solar radiation zones serve as guidelines for the assessment, investment, and policymaking for solar energy utilization. Most zoning methods of cluster analysis depend on discrete measurements at solar and meteorological stations (usually sparsely and unevenly distributed); thus, the zone boundaries are empirically determined by artificial delineation, leading to misclassifications at the junction of adjacent zones. In this study, we propose a solar zoning algorithm that combines the station measurements of monthly global solar radiation (GSR) with spatially continuous satellite estimates to identify the definite solar radiation zones in China. To avoid the overfitting effect resulting from a fixed cluster number, this algorithm uses the Gaussian mixture model with the Dirichlet process to automatically infer the appropriate cluster number from the training data in a fully Bayesian manner. Unlike the traditional five-zone opinion, our experiments indicate that the number of reasonable solar radiation zones in China can range from five to ten, thus forming a hierarchical system. The five primary zones reflect the disparity of the annual total GSR caused by dominant climates, while the ten fine zones reveal the differences in the seasonal GSR variations associated with the local topography, altitude, and latitude. The comparative results demonstrate that the identified zone boundaries are accurate and reliable and are in agreement with the natural geographical dividing lines. The statistical analysis of the Chinese-installed photovoltaic capacity suggests that the proposed hierarchical system conforms to the multi-level needs of diverse applications.