[Estimation of PM_2.5 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_2.5 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_2.5 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 R², MAE, and RMSE were 0.96, 3.43 μg·m^-3, and 5.52 μg·m^-3, respectively; and the verification accuracy indexes R², MAE, and RMSE were 0.9, 4.98 μg·m^-3, and 7.92 μg·m^-3, respectively. ② The model had a high goodness of fit (R² of the whole dataset was 0.96, and R² of different times ranged from 0.91 to 0.98) when applied to the estimation of PM_2.5 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 PM_2.5hourly 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^-3 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.