Document Type : Research Article
Authors
1
PhD Student of Climatology, Department of Physical Geography, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardabil, Iran
2
Professor of Climatology, Department of Physical Geography, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardabil, Iran.
10.22067/geoeh.2025.91013.1531
Abstract
Introduction
Healthy and pollution-free air is one of the most basic human needs, and among these, dust is considered one of the most important air pollutants. Forecasting the risks of dust phenomena in affected areas and being aware of its temporal and spatial changes is essential for preparing for risks and preventing their occurrence and possible damages. Neural networks can be considered as one of the most widely used methods in forecasting, which are also widely used for modeling dust storms.
Method
The data used in this study for dust storm forecasting include horizontal visibility data of less than 1 km and current weather codes expressing dust, data on average temperature, average relative humidity, average wind speed and direction as seasonal time series for the period 1987-2023 at three synoptic stations of Abadan, Ahvaz and Bushehr. Many methods have been developed for time series modeling and forecasting, and choosing an appropriate method is very important. In this study, a multi-layer perceptron (MLP) neural network model was used to forecast the frequency of dust storms for the future (2024-2040) and its results were compared with the SARIMA model. To evaluate the efficiency of these models in forecasting, the criteria (MSE), (RMSE) and correlation coefficient (R) were used.
Also, in the next step, in order to provide accurate information on the time, location, intensity, and path of these storms, the most important factors affecting the dust concentration and the routing of incoming dust were examined. Because the studied stations were jointly involved in dust storms on July 29, 2009, June 8, 2010, June 3, 2011 and March 11, 2012; therefore, in the continuation of the research, using the Lagrangian HYSPLIT model, the path of dust particles was tracked at three levels of 200, 1000 and 1500 meters above the ground surface at a time interval of 6 hours before the dust entered using the Backward method. Then, its output was compared with the NAAPS and COAMPS models
Results and Discussion
The correlation study between dust frequency and average relative humidity, temperature, wind speed and direction data showed that the effect of these variables on dust frequency at all stations is associated with direct and acceptable correlation coefficients; therefore, the existence of correlation between variables makes it possible to implement modeling and forecasting future values. Based on the evaluation criteria, the MLP model had less error and more correlation than the SARIMA model at all stations; therefore, it was used as the optimal model in forecasting. Comparing the actual and predicted values of dusty days using the MLP neural network model in the testing phase showed consistency between the actual values and the values predicted by the model, indicating the acceptable accuracy of this model in estimating target values in the testing phase. The investigation of dust changes on a seasonal scale for the stations under study during the observation period (1987-2023) and the future period (2024-2040) by both models showed that all three synoptic stations had a higher frequency in spring and a lower frequency in autumn both during the observation period and in the coming decades. The routing of dust events by the HYSPLIT model on common dates showed that the main mechanism was the transport of dust particles in an area between Iraq and parts of Turkey and Syria to the study area, and its results show a good agreement with the optical depth of dust, surface dust density, and the movement of the concentration mass based on the NAAPS and COAMPS models.
Conclusions
From the temporal dimension, dust is a highly variable event, and its location and direction of movement change from hour to hour. From the results of this study, it can be seen that the longer the dust storm is considered for a statistical period, the greater the model's ability to predict and the better the predicted data will perform. On a practical scale, managers and officials can make appropriate and timely decisions based on the highly accurate models and forecasts provided before the onset of an alert situation, and take the necessary precautions at stations with high dust potential so that no damage is caused in the event of a dust event, or at least the damage is not significant.
Keywords
Main Subjects
)
Send comment about this article