Using Decision Tree Method for Dust Detection from MODIS Satellite Image

Iran has always been exposed to dust storms because of its climate، geographical location and proximity with the neighbor’s desert such as Iraq، Syria and the Saudi. Hence، detection of dust phenomenon is a critical issue facing our world. In this research، the dust storms which occurred in Eilam and Khuzestan provinces during 2005 to 2012 were detected using multispectral technique and applied criteria in global models were customized for the study area. For this purpose، a decision tree algorithm is utilized for distinguishing cloud and dust and then the dust from ground surface is separated using the same reflectance behavior. First، appropriate training data for three classes of cloud، dust over dark and bright surfaces and clear sky is selected. Secondly، the reflectance behavior of pixels in the mentioned classes is analyzed. In the next step the best bands for the detection of dust pixels are chosen and the improved decision tree is recommended for the study area. Finally، the accuracy of the proposed algorithm is evaluated and compared with the previous algorithms using criteria such as visibility and weather codes from the meteorological data of the study area. The results show if the improved method is used the accuracy would increase. Eventually، if the Normalized Difference Dust Index (NDDI) Indicators and Ln (b1) are used for dust detection over bright surface، the accuracy will be 58 percent. Moreover، for dark surfaces the accuracy of 53 percent is achieved using the NDDI and BTD (BT20-BT31).