The relationship between teleconnection indices and Aerosol Optical Depth (AOD) at selected stations in Sistan and Baluchistan Province

Extended Abstract
Introduction
Dust storms, which are particularly prevalent in arid and semi-arid regions such as Sistan and Baluchestan Province in Iran, pose significant environmental and health challenges. These storms are influenced by climatic factors and large-scale atmospheric patterns known as teleconnections, which modulate dust activity by affecting wind patterns, precipitation, and temperature. This study investigates the relationship between teleconnection indices and Aerosol Optical Depth (AOD) at two stations, Iranshahr and Zabol, aiming to improve the understanding and prediction of dust storms in the region. By leveraging machine learning models, the research seeks to identify key climatic drivers and develop accurate predictive tools for dust storm management.
Materials and Methods
The study utilized meteorological and climatic data from local weather stations, satellite sources (e.g., MODIS), and global teleconnection indices obtained from NOAA’s Physical Sciences Laboratory. Data preprocessing involved normalization and standardization to enhance model performance. Relationships between teleconnection indices and AOD were examined using Pearson correlation analysis. Feature selection was performed with the Boruta method, followed by the application of five machine learning algorithms Bagged CART, LightGBM, Gradient Boosting, Random Forest, and XGBoost for AOD prediction. Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R²). Furthermore, Shapley values, Sobol sensitivity analysis, and Partial Dependence Plots (PDPs) were employed to assess variable importance and interpret model behavior.
Results and Discussion
Correlation analysis revealed distinct patterns between teleconnection indices and Aerosol Optical Depth (AOD) at the two study stations. At Iranshahr, a strong negative correlation (-0.437) was observed with the Atlantic Meridional Mode (AMM), while the North Atlantic Oscillation (NAO) showed a positive correlation (0.236). In contrast, the most influential indices at Zabol were the Trans-Niño Index (TNI) and the Western Hemisphere Warm Pool (WHWP).Feature selection identified AMM, WHWP, and the Tropical Northern Atlantic index (TNA) as critical drivers for Iranshahr, whereas TNI and WHWP emerged as dominant predictors for Zabol. The applied machine learning models demonstrated strong predictive performance for AOD, with XGBoost and Gradient Boosting achieving the highest accuracy (R²=1 for Iranshahr and R²=0.99  for Zabol). Sensitivity analyses confirmed nonlinear and threshold-dependent relationships between teleconnection indices and AOD. Both Shapley and Sobol analyses highlighted AMM as the dominant factor, particularly at short-term lags, while Partial Dependence Plots (PDPs) further corroborated the threshold-dependent and nonlinear nature of these interactions.
Conclusion
The analysis of results from the Iranshahr and Zabol stations indicates that teleconnection indices significantly influence Aerosol Optical Depth (AOD) variations in these regions. This influence stems from the indices' impact on atmospheric circulation patterns, dust transport pathways, and regional moisture conditions. While similar general patterns have been observed elsewhere, the intensity and direction of these relationships vary due to the unique geographical characteristics of each location. Pacific Ocean indices dominate AOD variations at Zabol, with increasing influence over longer lags, whereas Atlantic indices are the primary drivers at Iranshahr, due to distinct local wind and geographical conditions. From a modeling perspective, boosting-based algorithms (e.g., XGBoost, Gradient Boosting) outperformed bagging models, demonstrating higher efficiency in capturing the nonlinear relationships between climatic indices and AOD. This study advances the understanding of AOD control mechanisms by identifying key teleconnection drivers and developing accurate predictive models. It also accounts for spatial variations in influential factors, which can support the design of region-specific early warning systems for dust storms. The identification of threshold-dependent relationships and critical behavioral thresholds in the indices can significantly improve the accuracy of both short-term and long-term AOD predictions. Furthermore, these results provide a robust scientific foundation for adaptive management planning in sectors such as water resources, agriculture, public health, and transportation. By leveraging this enhanced understanding of regional climatic mechanisms, policymakers and planners can develop more targeted and effective strategies to mitigate the impacts of dust storms.