The sustainable management direction of PM2.5 concentrations in the Yangtze River Delta region remains unclear due to regional spatial effects. This study combined the random forest model, spatial econometric model, and multi-scale geographically weighted regression model (MGWR) to explore the multi-scale spatial response of PM2.5 concentration to land use/cover conversion. The results show that:① PM2.5 concentrations in the Yangtze River Delta region from 2000 to 2018 showed four types of spatial-temporal patterns of spatially continuous aggregation, with strong regional synchronous changes. ② The relative influence of land conversion on PM2.5 concentrations showed a complex performance, and the source-sink effect of cultivated land and forest land was obvious. Neighborhood analysis indicated that the effect of surrounding aggregated land use conversion was generally more significant than that of single cells on PM2.5 concentration change, and the spatial effect was obvious. ③ PM2.5 concentration changes were mostly significantly negatively correlated with forest land and grassland conversion types and significantly positively correlated with conversion types between cropland, construction land, and water bodies. The importance ranking of the random forest model and correlation coefficient intensity indicated that the conversion between cropland-cropland (29.65%; 0.650), forest land-forest land (26.98%; 0.726), construction land-cropland (22.57%; 0.519), cropland-forestland (17.84%; 0.602), and cropland-construction land (16.34%; 0.424) contributed more to the variation in PM2.5 concentration. The spatial Durbin model revealed a significant spatial dependence of the change in PM2.5 concentration and a strong spatial spillover effect. ④ The MGWR model revealed the scale effects and non-stationary characteristics of the spatial relationships between different land use conversions acting on PM2.5 concentration change, and its spatial relationship showed strong differences in transfer types. However, the multi-models revealed that different land conversions drove the PM2.5 concentration change in different ways, so it is necessary to formulate targeted joint management strategies in a categorical and hierarchical manner.
Keywords: K-means clustering; PM2.5; land use/cover conversion; multi-scale geographically weighted regression model (MGWR); random forest; spatial econometric model.