ABSTRACT
The rapid expansion of cities and the impacts of urbanization on local and global environmental factors such as biodiversity and climate change are of great concern. Reliable rapid approaches for mapping the expansion of cities are of increasing importance today. In this paper, we explore the use of Google Earth Engine to classify land cover in Indian cities from Landsat imagery, using a Random Forest approach, a robust per-pixel approach to supervised classification which generates classification trees based on the band values of the desired classes. Cities were classified into four classes – urban, vegetation, waterbody, and fallow land. We developed global and individual random forest models and used them to classify India’s 10 largest cities. Our results show that the global model produces accuracies greater to individual models, with an overall classification accuracy greater than 80% for each city. This research provides an empirically grounded method to map cities.
Read more (subscription required)