Dust storms in Sanandaj area in the western region of Iran mainly during spring and summer has become an environmental crisis. Prediction of dust storm occurrences lets alarm the residents to helps manage its detrimental effects. However, no study has been conducted to determine the accuracy of Adaptive Neuro-Fuzzy Inference System (ANFIS) model in predicting dust storm occurrence. For that purpose, prediction accuracy of ANFIS model was compared with that of two conventional models used to dust storm prediction including the Artificial Neural Networks (ANN), and Multiple Regression (MLR). Daily mean meteorological variables of Damascus (Syria) and PM10 concentration, measured in a ground station in Sanandaj, Iran from 2009 to 2012, were selected as independent and dependent variables, respectively. After data normalization between zero and one, the data from 2009 to 2011 were used for the simulation while the data of 2012 were utilized for testing the models. The performance of the ANFIS model in simulating dust storm occurrences was acquired higher compared with that of MLR and ANN. In the simulation results, among the three models the highest Pearson correlation coefficient between the observed and the estimated dust storm occurrences was obtained for the ANFIS model. The prediction tests showed that the accuracy of ANFIS model was higher compared with ANN and MLR. From the results of this study it can be concluded that the ANFIS model has the potential of forecasting dust storm occurrence in west of Iran by using meteorological variables of dust storm creation zone in Syrian deserts.
