Detection and Prediction of Sundarban Reserve Forest using the CA-Markov Chain Model and Remote Sensing Data

Abstract

The present paper highlights the classification, change detection and future prediction of Sundarban reserve forest using multi-spectral satellite data over last 43 years (1975–2018). The remote sensing data are classified using back-propagation neural network algorithm which are healthy vegetation, unhealthy vegetation, wet land, and water bodies. The classification result demonstrates that the net forest areas were gradually declined by around 6.83% during 1975–2018, while it was not uniform over the whole period. Besides the other features also correspondingly changes. The change detection results revealed that some of the forest areas were converted into wet land and part of the wet land also flooded by water bodies due to rising sea level. To validate the forest cover classification on the images, the overall accuracy and Kappa coefficient were used. The resulting overall accuracy were 91.8%, 94.1%, 87.5%, 88.1% and 90.1% and Kappa coefficients were 0.8903, 0.9201, 0.8292, 0.8413 and 0.8680 for 1975, 1990, 2000, 2010, and 2018 respectively. Future predictions were obtained through CA-Markov Chain model which is based on the probabilistic modeling methods. The CA–Markov model shows that constant changes in forest cover. Changes in the extent of forest cover of the study area were further projected until 2034, representing that the area of net forest will be continuously reduced to 12.89%. The outcomes of this study may be offer quantitative information, which signify the base for measurement of forest ecosystem and for taking actions to reduce their degradation.