Accurate measurement of surface visibility is vital for transportation. Satellite remote sensing technology offers the advantage of monitoring vast areas over extended time periods, providing a novel approach to visibility monitoring. However, it currently lacks the capability to directly measure visibility, necessitating further theoretical exploration of visibility remote sensing methods. In this study, we introduced an innovative model for remote retrieval of global surface visibility. It leverages the Extreme Random Forest (ERF) model to optimize the residual components of the physical method. The proposed model comprehensively explores the intricate relationship among surface visibility, aerosol optical depth (AOD), atmospheric extinction coefficient, and aerosol heights. Utilizing surface visibility data collected from global ground-based meteorological stations in 2018, we assessed the performance of the proposed model through 10-fold cross-validation, yielding correlation coefficient (R) of 0.85, root mean square error (RMSE) of 2.7 km, and mean absolute error (MAE) of 1.6 km. Results from temporal and site cross-validation further confirm the model’s strong performance in both time and spatial domains. Additionally, this study has developed a global visibility dataset at a daily resolution of 0.1° × 0.1°. Temporal analysis of this dataset reveals that visibility values are higher from June to October, while values are lower in other months. Regarding the spatial distribution of visibility, oceanic areas exhibit markedly superior visibility compared to land areas. Additionally, Oceania, South Africa, and South America boast notably better visibility than other regions.
Read full abstract