مشخصات پژوهش

Wind erosion and dust storms are among the main causes of air pollution and represent one of the most significant environmental threats in the arid and semi-arid regions of Iran. This study was conducted to identify the controlling environmental factors and to predict the monthly frequency of dust events in these regions over the period 1990–2019. Relevant environmental predictors were selected using feature selection techniques. Several machine learning models were developed and evaluated using a train–test framework. The contribution of environmental factors to this goal was measured using Shapley Additive Explanations (SHAP), while the interaction between dust events and these factors was assessed using partial dependence plots (PDPs) and accumulated local effects (ALE). The results indicated that Random Forest (RF) performed best, with R-squared values of 0.78 and 0.51 for the arid and semi-arid regions, respectively. SHAP analysis identified wind speed as the most important controlling factor for dust events, with a contribution approximately 20% higher in the arid regions than in the semi-arid regions. In addition, in the arid regions, monthly variations in precipitation (SHAP: 5.6%) along with medium-term drought (SHAP: 3.7%) and the lagged effects of vegetation cover (SHAP: 4%) played important roles in altering dust event activity. In the semi-arid regions, monthly variations in vapor pressure (SHAP:10.3%) along with the lagged effect of vegetation cover (SHAP: 9.9%) and the spring season (SHAP: 8.2%) were identified as key factors. Based on the PDPs and ALE analysis, precipitation and vegetation cover exhibited saturation effects, while wind speed showed threshold-like behavior and vapor pressure displayed a complex nonlinear response, suggesting nonlinear relationships between dust event frequency and key environmental variables in both study regions. Furthermore, the form of these responses differed between the arid and semi-arid regions. These results can be taken as the basis for designing early warning systems and methods to counter the impact of dust events in arid and semi-arid climates