Modeling Land Use Change Process by Integrating the MLP Neural Network Model in the Central Desert Regions of Iran

Abstract

To understand and manage the natural and human-made ecosystems and develop long-term planning, it is necessary to model Land Use Change (LUC) and predict future changes. Therefore, we used Landsat satellite imagery, Multilayer Perceptron neural network (MLP) and Markov Chain model (MCA) to monitor the regional changes over 30 years in the central arid regions of Iran. In the present research, the stratified maps derived from the object-oriented algorithm were used to detect and map the changes of land use classes from 1986 to 2016. Furthermore, the land use in 2030 was predicted using Land use Change Modeler (LCM). Slop, contour elevation lines, distance from river, road, afforestation, agricultural lands/gardens, barren lands, poor rangelands, residential lands, rocky land, and sand dunes were considered as factors influencing the changes in the ANN. The Cramer's V coefficient was employed to select appropriate parameters with the highest significant correlation. Our results showed that the sub-models performed well (75-85%). Besides, the highest and lowest accuracy of sub-models were related to the distance from barren lands and distance from residential areas (75.23 and 85.91%, respectively). The results of land use change monitoring from 2016 to 2030 revealed that land use such as forest, residential lands, gardens, and sand dunes would be increased by about 0.11, 1.53, 2.36 and 0.56 %, respectively, by 2030 compared to 2016. On the other, the area of barren land and poor rangeland would be reduced by 2.88 and 1.68 %, respectively. Our results can be used in land change evaluations, environmental studies, and integrated planning and management regarding appropriate and logical use of natural resources and reducing resource degradation.