Nomination the most suitable of input combination of artificial neural networks method to purpose nomination the Wind parameters on the prospect of dust storms phenomenon (case study: yazd province)

The aim of this study was to determine some factors affecting dust storms phenomenon using different methods. In order to determine the best-input combination, variable reduction techniques such as factor analysis (maximum likelihood, principal component analysis), Gama test, and multivariate forward regression analysis were used. Each of these methods presented different combinations used by feedforward neural network model, with Levenberg–Marquardt algorithm and multivariate forward regression with R²=0.87 and RMSE=0.04 was selected as the best suitable combination of neural network model. In addition, monthly and seasonal data were applied by neural network using the best-input combination, and the simulation of dust storm phenomenon was done in summer and spring during the months of April, May, June, July, August and September with a higher correlation coefficient and lower mean square error, due to the good distribution of the dust storm data. The results showed that based on these methods used in this study, dominant wind speed, horizontal visibility, continuity and average of wind speed were the most important factors affecting dust storm phenomenon in Yazd province.