Application of Normalized Spectral Mixture Analysis (NSMA) to extract urban built-up areas and utilize it in artificial neural network (MLP) to predict the future growth of the city

Using satellite images with a medium spatial resolution to detect, monitoring and prediction of urban built-up areas has a great development in recent decades. The most important step in predicting of the urban areas growth is extracting the urban features with a high precision but also the greatest challenge in this way is the complexity of urban components and the issue of mixed pixels. The purpose of this research is using sub-pixel analysis to extract the surface features of Rasht city to predict the future growth of the city’s built-up areas changes. To achieve this purpose we used three Landsat images related to; 1990 (Landsat Sensor TM), 2002 (Landsat Sensor ETM ) and 2015 (sensor OLI / TIRS) years and Normalized Spectral Mixture Analysis (NSMA). In order to classifying the images, the fraction layers as input layers, endmembers as training samples and maximum likelihood as classifier algorithm were used, As a result, the overall accuracy and kappa coefficient for three period of study were calculated up to 99% and 0.98 respectively. In this study, however, in order to predict urban growth by ANN model, Multilayer Perceptron (MLP) with Back-Propagation learning algorithm (BP) were used. The results of comparison between model’s output and 2015 classification map showed a 92% kappa coefficient, a 89% standard Kappa and a 93% classification Kappa (for classes) respectively. The used model in this research has been successful in predicting the growth of urban boundaries, but also less accurate in predicting the individual built-up areas around the urban areas.