Mapping Urban Footprint Using Machine Learning and Public Domain Datasets

Abstract

The urban footprint is termed as the physical cover of the urban built-up. In the past several decades urbanization has been accelerated due to rural–urban migration, economic growth, globalization, etc., and it is observed that over half of the world’s population now living in cities. Mostly, the unintentional urbanization causes impermeable surface which triggers several environmental challenges such as trash disposal, groundwater scarcity, heat island effect, and so on which need to be managed to support urban sustainability. The main objective of this study is to map urban footprint of Kolkata metropolitan area using machine learning (ML) algorithms (e.g., SVM, Random Forest) and public domain dataset. Landsat TM satellite data, Night time light data, and census data were employed in this study. Satellite imagery was used for mapping Lulc and reclassifying the built-up area into rural and urban built up using socio-economic data like census data. The robustness of the ML algorithms was tested based on classification accuracy and transferability assessment. In Lulc analysis band and feature stacked images give the high accuracy than the normal image and PCA image in three ML algorithms. SVM-Linear gave the high accuracy comparatively to another ML algorithm. The building footprint of KMA was extracted from top three high accuracy LULC map of different ML algorithm. The built-up area of KMA was validated using the test sample and for the validation of test sample these test sample uses in GHSL images. This finding will aid in the categorization of rural and urban areas and gives the idea of urban extent in the Kolkata metropolitan area. This study will help urban planners, local governments, and policymakers for the urban policy improvement and sustainable urban development planning.