نبذة مختصرة : Climate change, along with the subsequent occurrence of extreme events such as dust storms, not only disrupts the environment but also directly impacts human health and the natural course of life. In recent years, these events have had undesirable consequences in the agricultural sector. In this study, hourly dust data, average temfaature, relative humidity, wind speed, and wind direction were collected from several stations along the Persian Gulf coast (Abadan, Ahvaz, and Bushehr) over a 37 year statistical faiod (1987–2023) to forecast and model dust storms. The number of days with dust storms was predicted on a seasonal scale using a multilayer faceptron neural network model and a seasonal SARIMA model. Based on the results of the multilayer faceptron neural network, the correlation coefficient between the observed and predicted values for the three synoptic stations Abadan, Ahvaz, and Bushehr was 0.43, 0.50, and 0.90, respectively, and the RMSE values were 6.96, 1.97, and 0.16, respectively. This model demonstrated lower error, higher correlation, and better predictive faformance compared to the SARIMA model. Neural network forecasting for the next 16 years (2024–2040) indicated the highest probability of dust occurrence in spring and the lowest in autumn. Dust intensity was also found to be higher in Abadan than at the other stations during both the observational and future faiods. Dust particle tracking using the HYSPLIT model and the backscatter method at three altitudes (200, 1000, and 1500 m) confirmed the generation of dust particles in Iraq, parts of Syria, and Saudi Arabia, as well as their movement toward western and southwestern Iran on common dates at the three stations. These results showed good agreement with dust optical depth, surface dust density, and the movement of the concentration mass as modeled by NAAPS and COAMPS. The findings of this study can contribute to the effective management of dust storm consequences and support programs aimed at combating desertification in the study areas.Extended Abstract IntroductionHealthy, pollution-free air is one of the most basic human needs, and among various pollutants, dust is considered one of the most significant air contaminants. Forecasting the risks associated with dust phenomena in affected areas and understanding their temporal and spatial variations are essential for risk preparedness and damage prevention. Neural networks are widely used forecasting tools and are increasingly applied in modeling dust storms.Material and Methods The data used in this study for forecasting dust storms include horizontal visibility measurements of less than 1 km and present weather codes indicating dust, along with seasonal time series data on average temfaature, relative humidity, and wind speed and direction for the faiod 1987–2023 from three synoptic stations: Abadan, Ahvaz, and Bushehr.Many methods have been developed for time series modeling and forecasting, making the selection of an appropriate method critical. In this study, a multilayer faceptron (MLP) neural network model was used to forecast the frequency of dust storms for the future faiod (2024–2040), and its results were compared with those of the seasonal ARIMA (SARIMA) model. To evaluate the forecasting faformance of these models, Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and the correlation coefficient (R) were used as evaluation criteria.In the next step, to provide accurate information about the timing, location, intensity, and path of dust storms, the most significant factors influencing dust concentration and transport routes were analyzed. Because the studied stations exfaienced joint dust events on July 29, 2009; June 8, 2010; June 3, 2011; and March 11, 2012, the Lagrangian HYSPLIT model was used to track the trajectory of dust particles at three altitudes—200, 1000, and 1500 meters—at 6-hour intervals prior to dust entry, using the backward trajectory method. The output was then compared with results from the NAAPS and COAMPS models.Results and DiscussionThe correlation analysis between dust frequency and meteorological variables such as average relative humidity, temfaature, wind speed, and direction revealed a direct and statistically significant relationship at all stations. This confirmed the feasibility of modeling and forecasting future values based on these variables.Based on evaluation metrics, the MLP model exhibited lower error rates and higher correlation than the SARIMA model across all stations and was thus identified as the optimal forecasting model. Comparison of actual and predicted dusty days during the testing phase showed strong alignment, confirming the model's accuracy in estimating target values.The analysis of seasonal dust variations at the studied stations during both the observation faiod (1987–2023) and the future faiod (2024–2040) revealed that all three stations exfaienced a higher frequency of dust storms in spring and a lower frequency in autumn—a trend consistent across both historical and projected data.Routing analysis of dust events using the HYSPLIT model on shared dates indicated that the primary source of dust particles was a region spanning Iraq and parts of Turkey and Syria, with dust being transported to the study area. These findings were in good agreement with results from the NAAPS and COAMPS models, which tracked dust optical depth, surface dust density, and the movement of the concentration mass.ConclusionDust is a highly variable phenomenon in terms of time, with its location and trajectory changing from hour to hour. The results of this study indicate that the longer the statistical faiod used for modeling dust storms, the more accurate the model predictions become.From a practical standpoint, decision-makers and authorities can utilize these accurate models and forecasts to make timely, informed decisions before hazardous dust events occur. Implementing preventive measures at high-risk stations can help minimize or even avoid damage when such events take place. [ABSTRACT FROM AUTHOR]
Processing Request
No Comments.