Introduction: Suspended particles with a diameter of less than 2.5 microns have a remarkable contribution to air pollution and human health. These particulate matters are among the most critical air pollutants with severe impacts on human health. Due to the importance of these particles, this study aimed to model the increasing trend of PM2.5 concentrations in Isfahan, as one of Iran's metropolises, using machine learning models, and to determine the contribution of environmental and human-demographic factors influencing it over 24 years (2001-2024).
Materials and Methods: Initially, using ground-based data of PM2.5 from 12 air pollution monitoring stations and satellite-derived Aerosol Optical Depth (AOD) images (MCD19A2 product from MODIS), monthly PM2.5 images were estimated for the entire city of Isfahan through linear regression modeling. The Mann-Kendall trend test was then applied in TerrSet software to determine the monthly trend of PM2.5 changes over the 24 years. Based on the results, six months, including April, May, June, August, October, and March, which showed the highest percentage of area with an increasing trend, were selected as the study months. For modeling the increasing trend of PM2.5 using machine learning models including the Random Forest (RF), Support Vector Machine (SVM), and Boosted Regression Trees (BRT) in RStudio. Validation was performed using the Receiver Operating Characteristic (ROC) curve and Area Under the Curve (AUC), while factor prioritization was conducted using Jackknife sensitivity analysis.
Results and Discussion: Trend analysis revealed that June (81.16% of the area), April (78.07%), August (71.41%), May (52.91%), October (45.72%), and March (40.68%) exhibited the largest spatial extent of increasing PM2.5 trends, respectively. This indicates that while air pollution has increased across all seasons, the intensity is more pronounced during the warm months and early spring, potentially due to higher temperatures, intensified photochemical reactions, and dust phenomena. The highest modeling accuracy, with an AUC value of 0.9, was for the SVM model in August, which indicates the high ability of this model to identify complex and nonlinear patterns governing air pollution in this month. Also, according to the Jackknife sensitivity analysis, temperature emerged as the most dominant variable in five out of six months (except October). This confirms temperature's critical role in shaping air pollution through its influence on atmospheric stability, photochemical reactions, and secondary particle formation.
Conclusion: This study, integrating ground-based PM2.5 data with satellite AOD imagery and employing machine learning models coupled with Jackknife sensitivity analysis, successfully identified spatial-temporal pollution patterns and their driving factors in Isfahan. The results demonstrate that while environmental variables, particularly temperature, are fundamental in increasing the trend of PM2.5, human-demographic factors such as industry, traffic, agriculture, and population concentration constitute the primary source of emissions and play a decisive, sometimes dominant, role in increasing PM2.5 concentrations. Based on these findings, it is recommended that new industries be established with appropriate buffer zones, preferably downwind of the prevailing wind. Strict monitoring of existing industrial stacks and mandatory adoption of emission reduction technologies are essential. Given the significant role of roads, especially in June and October, developing clean public transportation, expanding low-traffic zones, and enforcing stringent vehicle emission inspections should be prioritized. Prohibiting agricultural residue burning and promoting conservation agriculture are crucial measures. Increasing green spaces with native, drought-resistant species in high-population-density areas and near industrial zones is suggested as a strategy for particulate matter absorption. Finally, any urban development policies must incorporate environmental considerations aimed at reducing population concentration in critical, high-pollution areas.