Performance Evaluation of Genetic Algorithm and GA-SA Hybrid Method in Forecasting Dust Storms (Case Study: Khuzestan Province)

The increase in dust storms occurrence in recent years in southwestern Iran, especially in Khuzestan province, and consequently the decrease in air quality in these areas, has doubled the importance of forecasting and linking this phenomenon with climate variations. The aim of this study was to investigate the efficiency of hybrid Genetic-Annealing (GA-SA) and Genetic Algorithm (GA) methods for selecting optimal input variables in forecasting the frequency of days with dust storm (FDSD). For this purpose, hourly dust data and meteorological organization codes, as well as climatic data including maximum temperature, minimum temperature, average temperature, total rainfall and maximum wind speed on a seasonal scale with a statistical period of 35 years (1984-2018) in seven synoptic stations in Khuzestan province were used. Then, by having a time series of FDSD index and other climatic variables, it was compared to the efficiency of different modes of input variables, in order to forecasting the frequency of days with dust storm in the next season. The results showed that the hybrid Genetic-Annealing method (GA-SA) had the best performance among all the modes of selecting the input variables; In this case, the evaluation criteria of R, MAE and RMSE varied from 0.91, 0.29, and 0.44 to 0.99, 0.13 and 0.17 in the studied stations, respectively. Also, the average frequency of days with dust storm on a seasonal scale in the studied stations varied from 1.68 to 4.10, respectively, so that with increasing FDSD index in the study station, the predictive accuracy of all modes increased so that in the first forecast state (based solely on the FDSD index), the correlation coefficient between the observational values of the days associated with dust storms and its computational values increased from 0.87 to 0.95. For the second case (forecast based on frequency of days with dust storm and all Auxiliary Characteristics, ie FDSD & AC), the third mode (based on the optimization of the Genetic Algorithm) and the fourth mode (based on the hybrid Genetic-Annealing method) the correlation coefficient also varied from 0.93 to 0.94, 0.91 to 0.97 and 0.94 to 0.99 in the studied stations, respectively. In general, by comparing the modes used, the hybrid Genetic-Annealing method (GA-SA) performed the best, followed by the Genetic Algorithm (GA). The results of this study can be useful in managing the consequences of dust storms and desertification programs in the study areas.