Forecasting areas vulnerable to forest conversion using artificial neural network and GIS (case study: northern Ilam forests, Ilam province, Iran)

Forest conversion due to illegal logging and agricultural expansion is a major problem that is hampering biodiversity conservation efforts in the Zagros region. Yet, areas vulnerable to forest conversion are unknown. This study aims to predict the spatial distribution of deforestation in western Iran. Landsat images dated 1988, 2001, and 2007 are classified in order to generate digital deforestation maps which locate deforestation and forest persistence areas. Meanwhile, in order to examine deforestation factors' investigation, deforestation maps with physiographic and human spatial variables are entered into the model. Areas vulnerable to forest changes in the Zagros forest region are predicted by a multilayer perceptron neural network (MLPNN) with a Markov chain model. The results show that about 19,294 ha forest areas are deforested in the last 19 years. The predictive performance of the model appears successful, which is validated using the actual land cover map of the same year from Landsat data. The validated map is found to be 94% accurate. The validation is also tested using the relative operating characteristic approach which yielded a value of 0.96. The model is then further extended to predict forest cover losses for 2020. The MLPNN approach was found to have a great potential to predict land use/land cover changes because it permits developing complex, nonlinear models.