A new strategy for fine-scale identification of urban PM2.5 spatial patterns and air quality assessment: a case study of the Urumqi-Changji-Shihezi Region
摘要
To thoroughly understand the spatiotemporal evolution characteristics of PM2.5 pollution at the urban scale and to support the development of precise control strategies based on regional differences, this study utilized multi-source data from the Urumqi-Changji-Shihezi region (“U-C-S” region) spanning 2015 to 2020 to construct a dataset comprising 12 key features. The study compared the PM2.5 concentration estimation performance of five machine learning models. The results indicated that GBDT demonstrated the best performance in seasonal modeling (R2 = 0.72) and maintained high accuracy in full-period modeling (R2 = 0.62), achieving outstanding results in external validation in Karamay City with R²=0.97 and RMSE = 5.06 µg/m3. Based on SHAP value analysis, boundary layer height (BLH) and aerosol optical depth (AOD) were identified as the most significant influencing factors, contributing 25.2% and 18.0%, respectively. To address spatial heterogeneity within cities, this study innovatively proposed three statistical boundary extraction strategies: administrative boundaries, land use, and nighttime lights. The comparison revealed that the land use boundary had the highest annual average PM2.5 concentration (51.19 µg/m3), followed by the nighttime light boundary (50.71 µg/m3), both significantly higher than the administrative boundary (47.15 µg/m3). Notably, during winter, the nighttime light boundary concentration (119.89 µg/m3) was almost identical to the land use boundary concentration (120.30 µg/m3), providing a more precise depiction of core exposure areas. Trend analysis showed that approximately 89.8% of grid cells in the study area exhibited a decreasing trend in PM2.5 concentration from 2015 to 2020, with 73.5% showing a significant decline, indicating an overall improvement in regional air quality; however, about 10.2% of grids in the urban core still showed an upward trend. This study, through the integration of multiple boundary extraction methods and model fusion, not only enhances the precision of PM2.5 spatial distribution assessment but also provides scientific evidence for differentiated governance and monitoring in underdeveloped urban agglomerations where administrative boundaries far exceed actual built-up areas.
Graphical Abstract