Prediction of Dust Storms in Khuzestan Province Using Artificial Neural Networks

. In this study to predict dust storms, hourly dust data and monthly data maximum, minimum, average temperature, maximum wind speed, and total precipitation in three synoptic stations of Abadan, Ahvaz, and Bostan with statistics period for 25 years (1990-2014) were collected. To investigate the impact of dust storms from climatic fluctuations, in addition to the mentioned variables, the Standardized Precipitation Evapotranspiration Index (SPEI) was also calculated in the seasonal time window. Predicting the number of days with seasonal dust storms using four artificial intelligence methods including MLP, ANFIS, RBF, and GRNN was performed. These were evaluated in the form of three experiments including the effect of adding auxiliary features on the prediction, the effect of the number of previous seasons on the prediction, and the best technique among the models used. The results showed that in all stations, the use of all features has improved dust prediction and the value of the Mean Absolute Error (MAE) for Abadan, Ahvaz, and Bostan stations is equal to 1.15, 1.66, and 0.66, respectively were obtained, all of which were related to the autumn season. In conclusion, it can be said that in Bostan station, if all the features and data of the last four seasons are used, the ANFIS model as input causes the prediction error to be reduced and a better result to be obtained. In the Abadan station, using the MLP model gives such a result.