Deforestation modeling using artificial neural network and GIS (Case study: forests of Khorramabad environs)

In this research, occurred changes in the forests around Khorramabad between 1986 and 2018 using TM and OLI Landsat images were investigated. For this purpose, after making the necessary atmospheric and geometric corrections, the images were classified by the maximum likelihood algorithm in five classes with a total accuracy of 95% and a kappa coefficient of 0.94. By overlaying the images, the amount of lost forest (34 km2) was determined and as a dependent variable was imported into the multilayer perceptron (MLP) model. In the GIS environment, were prepared the effective factors in the process of deforestation (independent variables); then by using MLP, the deforestation process in the years under review was determined. It was also land use changes was extracted that the results show the highest changes belonged to the forest to barren land changes and finally the deforestation forecast for 10, 20 and 30 years displays a decrease of 4.6% for the year 1407, 7.5% for the year 1417 and 9.3 for 1427. The results of the network training involving all variables with mean squared error (RMS) of 0.13 indicate that the MLP-based modeling is accurate and also, using Receiver Operating Characteristic (ROC) index, the real amount of deforestation was compared to the result of the MLP model; which showed the high accuracy of the MLP model with 0.88 of the ROC.