| چکیده انگلیسی مقاله |
PM2.5 particulate matter represents one of the most critical environmental challenges globally, with far-reaching impacts on human health, quality of life, and ecosystem functionality. In this study, the MERRA-2 dataset—covering the period from 1983 to 2023—was utilized to analyze the long-term behavior and spatiotemporal patterns of PM2.5 concentrations during the months of April, May, and June.The research was conducted in three phases: Preprocessing and Preliminary Analysis: Initial steps involved data cleaning, reconstruction, and exploratory analysis to prepare the dataset for spatial modeling. Spatial Distribution and Trend Analysis: The spatial distribution and temporal trends of PM2.5 were examined, along with their correlations with climatic and environmental variables. Modeling with Random Forest Algorithm: PM2.5 concentrations were modeled using the random forest algorithm to assess predictive accuracy and spatial representation. The spatial analysis revealed that the highest concentrations of PM2.5 occurred in the southwestern regions of the country, while the lowest levels were observed in the northwest. Trend analysis showed a significant positive trend in PM2.5 levels in southwestern areas—particularly in parts of Khuzestan Province—with increases exceeding 4 Mg/m² in April and May, and over 5 Mg/m² in June. The random forest model demonstrated strong performance in spatial modeling of PM2.5. The coefficient of determination (R²) values for environmental parameters were 0.87, 0.88, and 0.98 for April, May, and June, respectively. For climatic parameters, the R² values were 0.86, 0.98, and 0.97. These results confirm the robustness and suitability of the random forest algorithm in capturing the spatial and temporal dynamics of PM2.5 concentrations. Extended Abstract 1-Introduction Pollutants refer to any natural or artificial substances that enter the atmosphere in abnormal quantities, and their presence has increased at an alarming rate in recent years. Due to their widespread impact on both global and local climate systems, pollutants pose a growing threat—particularly to developing countries. Among the most significant pollutants affecting air quality is particulate matter, especially dust. Dust particles have numerous health and environmental consequences throughout their atmospheric cycle. While they may carry nutrients and salts that enrich agricultural soils, they also contribute to soil degradation, reduce crop quality, contaminate water resources, obstruct roads and communication routes, impair visibility, and lead to airport closures. Additionally, dust can alter wind patterns and contribute to atmospheric scattering, further complicating climate dynamics. Given the extensive influence of particulate matter and airborne microbes on climatic, human, and environmental variables at both regional and global scales, this study employed landfill-based analytical methods in combination with a national-scale random forest algorithm. The goal was to model pollutant behavior and provide data-driven insights to inform and guide environmental policy-making. 2-Materials and Methods This study utilized the MERRA-2 dataset at a monthly temporal resolution, obtained from the Earth Data portal in NetCDF format, to analyze PM2.5 particulate matter across Iran. The research was conducted in four key phases: Data Acquisition and Preprocessing: PM2.5 data were extracted and prepared using MERRA-2 settings. Initial preprocessing included data cleaning, formatting, and drafting operations to ensure consistency and usability for spatial analysis. Spatial Distribution Analysis: Using geospatial techniques, the average spatial distribution and temporal trends of PM2.5 concentrations were mapped and examined. This step provided insights into regional variations and long-term behavioral patterns of suspended particles. Correlation with Climatic and Environmental Variables: The study investigated the relationship between PM2.5 concentrations and various climatic and environmental parameters across Iran. This included assessing how factors such as temperature, humidity, vegetation cover, and land use influence particulate levels. Modeling with Random Forest Algorithm: Finally, the random forest algorithm was applied to model the spatial distribution of PM2.5. This machine learning approach enabled robust prediction and pattern recognition, offering a reliable framework for understanding the dynamics of suspended particulate matter in relation to environmental and climatic variables. 3- Results and Discussion The spatial distribution map of average column concentrations of PM2.5 over Iran during April, May, and June reveals a relatively consistent pattern, with elevated concentrations primarily observed in the southwest (notably Khuzestan Province) and the northwest (particularly West Azerbaijan). This concentration is largely influenced by a combination of geographical, climatic, and environmental factors, including proximity to major transboundary dust sources such as the deserts of Iraq, Saudi Arabia, and Yemen, reduced soil moisture, and intensified agricultural and industrial activities—all of which contribute to increased pollutant levels. In contrast, the lowest PM2.5 concentrations were recorded in Iran’s mountainous regions, including the Alborz and Zagros ranges and parts of the northwest. These areas benefit from higher altitudes, increased rainfall, cloud cover, penetration of humid air masses, and dense vegetation, which collectively reduce the accumulation of airborne particles. Analysis of variable importance across the spring months identified minimum temperature as the most influential climatic factor in April, May, and June. Among environmental variables, soil moisture was most significant in April and May, while soil temperature played a dominant role in June. Modeling results using the random forest algorithm demonstrated strong predictive performance in estimating PM2.5 concentrations. The model achieved an explanation coefficient (R²) exceeding 85%, with an average absolute error below 6% and a root mean square error under 9%. These metrics confirm the suitability of the random forest approach for capturing the spatial and temporal dynamics of suspended particulate matter using climatic and environmental parameters. 4- Conclusion The evaluation of the random forest algorithm's performance in this study demonstrates its strong capability in modeling PM2.5 concentrations based on climatic and environmental parameters. However, discrepancies were observed in certain regions, notably West Azerbaijan and Khuzestan Provinces, where modeled PM2.5 levels diverged from actual measurements. These differences are primarily attributed to the influx of transboundary dust, which is not fully captured in the MERRA-2 dataset, as well as limitations in input data and geopolitical factors such as the instability of local dust sources and complex, uncontrollable environmental conditions. These limitations highlight the need to enhance model accuracy by integrating high-resolution local datasets and conducting simultaneous analyses of climatic and environmental variables. Additionally, greater attention must be paid to transboundary dust sources, which significantly influence regional air quality. To address these challenges, it is recommended that, alongside local dust mitigation efforts, regional and intercontinental cooperation be strengthened to manage transboundary dust flows. Such collaborative strategies are essential for more effective air pollution control and for informing evidence-based environmental policymaking. |