|
The spatial and temporal distribution of atmospheric aerosols closely affects climate change, air quality, environmental pollution and human health. Exploring and predicting the spatial and temporal characteristics of regional atmospheric aerosols is beneficial to the monitoring and assessment of regional atmospheric environmental quality. Taking the Qinghai-Tibet Plateau and its surrounding areas as an example, this study considers the spatial and temporal non- stationarity of aerosol optical thickness (AOD) and its multiple driving factors, and proposes a geographically and temporally weighted regression method based on kernel principal component analysis (KPCA-GTWR) is proposed. The method eliminates the multicollinearity among the driving factors after the multicollinearity test, extracts the principal components with a cumulative contribution rate greater than 95% as the input of GTWR, and improves the prediction accuracy of GTWR. Finally, the method compared with the prediction results of the conventional VIF-GTWR method and PCA-GTWR method. The results found that (1) there are correlations between AOD and its multiple drivers, as well as linear and nonlinear correlations between the drivers. (2) In comparison, the KPCA-GTWR method has the highest prediction accuracy. Compared with the conventional VIF-GTWR and PCA-GTWR methods, the predicted AOD with MERRA-2 AOD 10-fold cross validated R 2 improved from 0.764, 0.861 to 0.914 , RMSE decreased from 0.059, 0.05 to 0.044 , and MAE decreased from 0.043, 0.037 to 0.033, respectively. (3) Comparing the results in June, July, August in 2020, the spatial distribution of AOD and MERRA-2 AOD predicted using this method in and around the Tibetan Plateau is consistent and shows large spatial differences. The low values of both predicted AOD and MERRA-2 AOD are located in the main part of the Tibetan Plateau, around 0.25 or less, while the high values are found in the Tarim Basin, the Ganges Basin of India and the Sichuan Basin, up to 0.75 or more.
|
