DEVELOPMENT OF DATA-DRIVEN MODEL USING BAYESIAN STATISTICS, AND REMOTE SENSING TECHNIQUES FOR CONSTRUING NONLINEARITY IN LULC TRANSFORMATIONS

摘要

Machine learning techniques help in construing the complexities of combinatorial analyses. Application of these techniques helps in assessing the significance of causal factors on spatial events. This work attempts to identify the hotspots of possible randomness in terms of land use / cover changes using machine learning and remote sensing techniques. Transformations in an urban landscape pattern are a consequence of congregation of different spatial and aspatial factors. Furthermore, there is a huge possibility that the characteristics of these factors may vary in spatio-temporal domain. Hence, it is difficult to investigate an urban event using a unidimensional approach. Non-linearity of urban events can be tackled using techniques which are effective in considering and representing the possible transitions of causal factors from one state to another with probabilistic / possibilistic values. Hence, Bayesian model is employed in this study using historical and current data sets. Firstly, the study area is segregated into different grids. Spatiotemporal assessment of land use / cover changes and transition of LULC class from one to another is performed for the years between 1992 and 2014 for each grid. Then, current landscape pattern is quantified using a proposed landscape indices termed as fuzzy-Shannon's heterogeneity index for different grids. It is developed by modifying conventional Shannon's heterogeneity index. The results obtained fromthe application of fuzzy-Shannon's heterogeneity indexand spatiotemporal assessment of LULC changes and transition of LULC class from one to another are fed into the Bayesian model to determine the land use/ cover changes hotspots. Results of the Bayesian model also help in identifying the factors which are the most significant actors in inducing randomness in LULC transformation. Hotspots obtained from the proposed model are witnessing unprecedented changes in LULC. Therefore, it can be inferred that results are in sync with the actual scenario.