Comparison of geospatial-temporal modeling approaches in air pollution estimations

Recent advancements in the geographic information systems and remote sensing technology have supported the development of geospatial-temporal modeling approaches for air pollution. Particulate matter (PM10) and ozone (O3) are two pollutants of great concern in all pollutants. Previous studies estimated the spatial-temporal variability of PM10 and O3 using a single model, but only a few studies considered exposure assessment using multiple models and compared model performance. In this study, PM10 and O3 data during 2015 to 2018 were collected from specific industrial monitoring stations provided by the Taiwan Environmental Protection Agency. Three geospatial-temporal modeling approaches including land-use regression (LUR), geographically weighted regression (GWR), and geographically and temporally weighted regression (GTWR) were used to predict PM10 and O3 exposure. Furthermore, the kriging-based hybrid model was integrated with these three geospatial-temporal models, and totally performs six models for each pollutant for our comparison. The results showed that integrating the GTWR and kriging-based hybrid models have the greatest performances compared to LUR, GWR, and the combination of both with kriging-based hybrid models. R² obtained from the GTWR coupled with kriging-based hybrid models for PM10 and O3 was 0.96 and 0.92, respectively. Of all variables used, wind speed, pure residential area, manufacturing, park; rice field, orchard; and forest land were important predictors for PM10. Whereas, wind direction, industrial area, dry farming, and orchard were variables selected to predict O3.