[Estimation of PM Hourly Concentration in Sichuan Province Based on GTWR-XGBoost Model].

Aerosol optical depths of satellites and meteorological factors have been widely used to estimate concentrations of surface particulate matter with an aerodynamic diameter ≤ 2.5 μm. Research on a high time resolution and high-precision PM concentration estimation method is of great significance for timely and accurate air quality prediction and air pollution prevention and mitigation. Himawari-8 AOD hour product and ERA5 meteorological reanalysis data were used as estimation variables, and a GTWR-XGBoost combined model was proposed to estimate hourly PM concentration in Sichuan Province. The results showed that:① the performance of the proposed combination model was better than that of the KNN, RF, AdaBoost, GTWR, GTWR-KNN, GTWR-RF, and GTWR-AdaBoost models in the full dataset; the fitting accuracy indexes , MAE, and RMSE were 0.96, 3.43 μg·m, and 5.52 μg·m, respectively; and the verification accuracy indexes , MAE, and RMSE were 0.9, 4.98 μg·m, and 7.92 μg·m, respectively. ② The model had a high goodness of fit ( of the whole dataset was 0.96, and of different times ranged from 0.91 to 0.98) when applied to the estimation of PM concentration hour. It showed that the model had good time stability for hourly estimation and could provide accurate estimation information for regional air quality assessment. ③ In terms of time, the annual average PMhourly concentration estimation showed an inverted U-shaped trend. It began to increase gradually at 09:00 am to a peak of 44.56 μg·m at 11:00 and then gradually decreased. Moreover, the seasonal variation was very obvious, with winter>spring>autumn>summer. ④ In terms of spatial distribution, it showed the characteristics of high in the east and low in the west and a high degree of local pollution.