Application of artificial neural network and ordinary least squares regression in modeling land use changes

Owing to the vital effects of future land use changes, it is necessary to predict land use growth pattern before any decision making by the authorities and decision makers. Purpose of this research is to model land use change of Kohmare Scorch plain of Shiraz province using Ordinary Least Squares regression (OLS) for pre-processing variables and Modeling using neural networks. To perform this model, the land use maps using Landsat images of the years 1987, 2000 and 2012 were prepared. Next, the validation of classified images and change detection analysis performed. Results of change detection between 1987 and 2000 with accuracy of 83% kappa, shows the greatest increase in rangeland area (4224.24 ha) and the greatest decrease was on forest area (3953.75). Considering these changes, selection of the best combination of explanatory variables, potential land use changes for year 2012 was performed using multi-layer perceptron algorithm of artificial neural network. Next, using Markov chain method the land use map for 2012 was predicted. The error matrix for modeled land use map and that of Landsat image of year 2000 is 75%. Next, the revealed changes for the second period (2000-2012) with Kappa of 88% show greatest increase for rangeland area (1807.02ha). In contrast the greatest decrease was for forest (2132.82). Considering change detection at second period, land use for year 2024 was predicted and result shows that irrigated agriculture would have the greatest change.