A study on the urbanization of Bengaluru, India between 2015 and 2020 based on Sentinel-1/2 data

Abstract: Availability of timely information about land-use and land-cover (LULC) changes is essential for understanding and modelling the urbanization pattern, and thereby in planning sustainable urban development of a region accordingly. Remote sensing technology, using Earth observation satellites, has emerged as a reliable tool for timely monitoring of changes anywhere on Earth. This project focusses on studying urbanization and LULC changes in Bengaluru, one of the fastest-growing metropolitan areas in India, between 2015 and 2020 using remotely sensed satellite images from Sentinel missions. LULC image classifications and change detection techniques were used for studying urbanization and LULC changes in the study area. Multispectral Sentinel-2 images of the area, from 2015 and 2020, were used for producing LULC classification maps for the respective years. The classification maps produced were then used for calculating changes in the area under different LULC classes including the urban class between 2015 and 2020. Change detection was performed using three different techniques viz, post-classification change detection, change vector analysis (CVA), and Sentinel-1 SAR based log-ratio method. An overall change map showing changes in LULC classes, and an urban change map showing changes in urban class were generated using each of the change detection methods. Accuracy assessment of the LULC classification maps and the change maps was performed using confusion matrices and kappa values. The study was implemented on Google Earth Engine (GEE) platform. The image classification results for 2015 and 2020 showed good accuracy with kappa values of 0.88 and 0.87 respectively. The results showed a substantial increase in the urban class area and a considerable decrease in area under park and agriculture classes. An increase in the urban class area by 18.70% between 2015 and 2020 was observed. Post-classification change detection and CVA methods performed with good accuracy levels (kappa > 0.8) in detecting LULC class changes using overall change maps. CVA method was found to provide the best results in detecting the LULC class changes. In detecting the urban class changes, post-classification change detection was found to provide the best results. However, the urban change maps produced using all the three methods provided a kappa value below 0.8, showing a moderate level of accuracy in identifying urban class changes. It was observed that the log-ratio method did not provide as good accuracy as the other two methods in detecting LULC changes and urban class changes.