Urban landuse monitoring using neural network classification

The pressure on land in the city of Mumbai (Bombay), India has given rise to an alternative satellite city called Navi Mumbai (New Bombay). The area is being developed in a planned manner by delineating specific areas for various purposes. The strategy permits private land holding and development within the designated areas. This appears to have resulted in construction according to individual needs and priorities and somewhat unsystematic development and changes in landuse. This paper describes the application of neural network techniques to assess the landuse landcover changes using remotely sensed data. This helps in arresting changes, if any, which disturb the ecological balance. Remotely sensed data of Mumbai area of IRS-1A LISS II has been used. An area of 240 sq. km has been classified using supervised classification technique and backpropagation and Kohonen neural network techniques. The performance of the networks have been optimized with appropriate values of parameters such as number of hidden nodes, error tolerance and learning rate in order to achieve a high degree of accuracy. The results of classification by the three methods were judiciously interpreted. In all, there were seven classes - starting with water body (creek), water body (lake) and adjoining marshy land, the landuse changed gradually to vegetation, urban areas, dense urban areas and, lastly, the hilly areas of the Western Ghats. The study showed depletion of vegetation and hilly areas due to tree felling and quarrying; reclamation of marshy lands for urban development resulting in removal of mangrove vegetation; urban areas developing essentially near transportation nodes resulting in congestion.