[Estimation of Hourly PM_2.5 Mass Concentration in Guanzhong Based on Spatio-temporal XGBoost Model]

To improve model estimation accuracy and obtain hourly-scale PM_2.5 concentration data， a spatiotemporal XGBoost （STXGBoost） model incorporating data heterogeneity was developed， integrating multi-source datasets including Himawari-8 hourly apparent reflectance， ground-level PM_2.5 measurements， meteorological variables， population density， DEM， and forest coverage. The model was applied to estimate hourly PM_2.5 concentrations across the Guanzhong region in 2022 and analyze their spatiotemporal characteristics. The results showed that： ① The STXGBoost model demonstrated superior estimation accuracy and generalization capability compared to the SVM， RF， and XGBoost models. Ten-fold cross-validation on the full dataset achieved an R² of 0.96， with RMSE and MAE values of 7.17 μg·m^-3 and 3.82 μg·m^-3， respectively. The model exhibited robust hourly-scale estimation precision and stability， confirming its applicability for PM_2.5 monitoring in the Guanzhong region. ② Annual PM_2.5 concentrations from 08：00 to 17：00 local time displayed an "M-shaped" diurnal curve. Seasonal variations were pronounced， with winter exhibiting the highest pollution levels （characterized by a "morning-peak， evening-trough" diurnal pattern） and the most severe air quality degradation. ③ Spatially， elevated PM_2.5 concentrations clustered predominantly in the central Guanzhong Plain， while lower values prevailed in the western， northern， and southern mountainous areas.