This paper focuses on the global environmental issue of light pollution. Using remote sensing data and related side data, the light pollution risk assessment models are established for five latitudes: economy (ECO), health (HEA), environment and sustainability (SAE), climate and topography (CAT), and regional light index (RLI). First, Topsis method is used to construct a light pollution risk assessment model. In order to develop the light pollution risk coefficient system, five firstlevel indicators and 11 second-level indicators are identified based on the five latitudes identified above. Considering the limitation of single weight calculation method, entropy weight method and coefficient of variation method are combined to obtain comprehensive index weight. Secondly, based on the grey prediction model optimized by BP neural network, the light pollution degree of Shanghai, China and Mount Nimba, Guinea in the next 7 years is predicted, and specific optimization strategies are proposed. Finally, the robustness of the model is proved by sensitivity analysis.