| چکیده انگلیسی مقاله |
ABSTRACT Khuzestan Province, as one of the main dust storm hotspots in Iran, is consistently confronted with serious challenges to public health and environmental sustainability due to its unique geographical characteristics. This study introduces a hybrid SVR-ACOR-Holt-Winters model to simulate dust storms in Khuzestan over a 50-year statistical period (1971–2020). The results reveal that the SVR-ACOR-Holt-Winters model outperforms the individual and dual hybrid models evaluated, achieving superior performance metrics, including a Root Mean Square Error (RMSE) between 0.293 and 0.317, a correlation coefficient (R) ranging from 0.849 to 0.873, and a Mean Absolute Error (MAE) between 0.275 and 0.293. Among all evaluated stations, Abadan demonstrated the highest prediction accuracy, which can be attributed to the influence of climatic variables such as extreme temperature and wind speed on increasing the frequency index of dust storm days in Khuzestan. These findings suggest that dust storm modeling yields higher accuracy in critical regions with greater dust event intensity. The outcomes of this research can be effectively utilized for dust storm modeling and the design of early warning systems, contributing to the mitigation of damages caused by such hazardous environmental phenomen. Extended Abstract Introduction This study examined the performance of the triple-hybrid SVR-ACOR-HoltWinters metamodel for predicting the FDSD index in Khuzestan Province. The results were compared with those of the GEP-WOA hybrid model. All tested models showed the highest accuracy and efficiency during the first and second seasonal combinations across seven stations. Using data from one or two prior seasons improved the accuracy and performance of the predictions. Among the models, the proposed triple-hybrid metamodel was the most accurate and efficient, making it the best method for predicting dust storms in Khuzestan Province. On the other hand, the SVR-HoltWinters hybrid model had the lowest performance, with the smallest NS value and the highest MAE. This indicates that creating dual-hybrid models does not necessarily improve the accuracy of FDSD index predictions. The results of this study can help in better modeling of dust storms and support managerial decisions to reduce the damages caused by this phenomenon. Methodology The goal is to predict the frequency of dust storm days at seven meteorological stations in Khuzestan Province (Dezful, Safiabad, Masjed Soleyman, Bostan, Ahvaz, Bandar Mahshahr, and Abadan) over a 50-year period (1971–2020). As defined by the World Meteorological Organization, a dust storm day is when at least one of the eight synoptic observations includes a dust-related code (06, 07, 08, 09, 30, 31, 32, 33, 34, 35, or 98) in the present weather report. Additionally, the horizontal visibility associated with the reported code must be less than 1000 meters. In this study, hourly data on horizontal visibility and standardized WMO codes indicating visibility below 1000 meters were used to identify dust storms for all relevant codes. Results and Discussion The simulation results of dust storms in Khuzestan Province using the SVR algorithm show that the model performs more accurately in combinations one and two compared to other combinations. At the Bostan, Masjed Soleyman, and Dezful stations, the RMSE values decreased from 0.476, 0.481, and 0.491 days in combination four to 0.473, 0.477, and 0.488 days, respectively, after applying data from the previous season to predict dust storms in future seasons. Therefore, it can be concluded that simple models using the Support Vector Regression algorithm provide more accurate predictions of the FDSD index in Khuzestan Province. The results of the Holt-Winters model for predicting the FDSD index during the training and testing phases indicate better performance in all stations when using the first and second prediction scenarios compared to other scenarios. For instance, at the Masjed Soleyman station, the Nash-Sutcliffe efficiency coefficient decreased from 0.436 and 0.435 in combinations 2 and 1 to 0.438 and 0.437 in combinations 4 and 3. A comparison of the results from modeling the frequency of dust storm days using the standalone Support Vector Regression model and the dual-hybrid SVR-ACOR model shows improved efficiency and accuracy in the developed hybrid model. Therefore, when combined with the continuous Ant Colony Optimization catalyst, this model shows improved adaptability to variable and non-stationary data, resulting in higher accuracy in predicting the FDSD index. For instance, at the Bandar Mahshahr station, the RMSE and MAE values decreased from 0.439 and 0.422, respectively, in the Holt-Winters model to 0.406 and 0.398 after applying the dual-hybrid Holt-Winters-ACOR model. The triple-hybrid SVR-ACOR-HoltWinters model is capable of modeling both nonlinear and seasonal patterns of dust storm trends. In this regard, the developed hybrid model showed its best performance in the first and second prediction stages. Thus, it can be concluded that using past seasons to model future trends does not provide optimal results. A comparison of the dust storm prediction results using the SVR-HoltWinters model at the Safiabad station in the first seasonal combination shows a significant reduction in the RMSE value, from 0.501 to 0.307 days, in the optimal combination 1 of the triple-hybrid model. Conclusion Dust storms are one of the prominent environmental and climatic challenges in arid and semi-arid regions of Iran and the world, particularly in Khuzestan Province. This phenomenon has widespread impacts on human health, agriculture, infrastructure, quality of life, and the balance of natural ecosystems. Addressing this challenge requires the adoption of comprehensive and multidimensional approaches. The aim of this study was to evaluate the performance of the triple-hybrid model SVR-ACOR-HoltWinters and compare it with the individual models SVR, HoltWinters, and dual-hybrid models SVR-ACOR, HoltWinters-ACOR, and SVR-HoltWinters, as well as to analyze and compare the results with the gene expression programming model optimized by the whale optimization algorithm for predicting the frequency of dust storm days in seven meteorological stations across Khuzestan province.The results showed that the triple-hybrid model SVR-ACOR-HoltWinters demonstrated the best performance among all the models tested in predicting the frequency of dust storm days at all stations. The findings indicated an improvement in the performance of all models when using one or two prior seasons for dust storm prediction. It was thus clear that using older seasons did not improve the models' efficiency or accuracy. In this regard, simple models played a significant role in improving the evaluation metrics. Abadan station had the highest number of dust storm days in the seasonal scale, while Dezful station had the lowest frequency. It was evident that as the FDSD index increased at the stations, the performance of the triple-hybrid and individual and dual-hybrid models also improved. The results of the dual-hybrid models revealed an increase in the performance of the SVR and HoltWinters models after combining them with the continuous Ant Colony Optimization catalyst. Based on the results from the goodness-of-fit metrics, the dual-hybrid HoltWinters-ACOR model ranked second as the best dust storm prediction model in Khuzestan Province. It is also important to note that the gene expression programming model, after being integrated with the whale optimization algorithm, provided relatively better results compared to other dual-hybrid models, securing the third position. Furthermore, the results from the dual-hybrid SVR-HoltWinters model showed a reduction in performance and accuracy when predicting the frequency of dust storm days in Khuzestan compared to the other individual and hybrid models studied in this research. Funding There is no funding support. Authors’ Contribution All of the authors approved the content of the manuscript and agreed on all aspects of the work. Conflict of Interest Authors declared no conflict of interest. Acknowledgments We are grateful to all the scientific consultants of this paper. |