Abstract:
Often central to a district or an Upazila (subunit of a district), the Paurashavas (municipalities) are regarded as the small urban areas in Bangladesh. Rapid and uncontrolled urbanization in these areas has resulted in undesired transformation of the spatial and temporal dimensions in recent years. Study findings have reported a significant loss of agricultural land in Bangladesh due to the transformation of undesired Land Use Land Cover (LULC). To confront the impending challenges of rapid rural-urban transformation, concerned authorities of the Government have begun taking initiatives recently. One such initiative is the preparation of development plans for Madaripur and Rajoir Upazila, which is a comprehensive effort to regulate LULC transformations in the rural-urban continuum. In this background, this research focuses on devising an assessment procedure of development plans by analyzing satellite data, historical growth trends and future growth projection of the selected study areas by land cover change models.
To assess the development plans, predicted land cover maps of the corresponding years have been simulated by the land cover change model. The methodology of this research is primarily based on the time series raster data collected from satellite images of the years 1995, 2005 and 2015. A supervised Multi-Layer Perceptron Neural Network classification is used to classify the satellite images into five major land covers. To simulate the land covers, the classified land cover maps are applied into three selected models: St_Markov (Stochastic Markov), CA_Markov (Cellular Automata Markov) and MLP_Markov (Multi-Layer Perceptron Markov). The best fit model for the study areas has been determined from the three models by model validation techniques. Land cover data simulated by the MLP_Markov model has shown substantial agreement in kappa statistics with values of 0.77 and 0.72 for Madaripur and Rajoir Upazila, respectively. As these values are higher than the other two models, the MLP_Markov model has been identified as the best fit model to predict land cover maps of 2035 for both study areas. The final predicted maps have been created based on two independent simulations: unregulated simulation and regulated simulation.
