Authors: Shrobona Karkun*, Temple University, Victor Hugo Gutierrez-Vélez, Temple University
Topics: Land Use, Urban Geography, Asia
Keywords: urban remote sensing, Indian cities, urban spatial research, machine learning, supervised classification, urban transportation
Session Type: Paper
Presentation File: No File Uploaded
India’s metropolitan regions have rapidly spread to colossal proportions in the past two decades. In this era of rapid urbanization, it is becoming increasingly vital to understand the patterns of their land use. The availability of high quality, medium resolution satellite products, and advancements in spatial research methodologies are enabling researchers to inquire about the current state of land use in urban areas. However, supervised classification studies using Sentinel products are relatively new. Here, we utilize linear spectral unmixing and data fusion techniques with Random Forest to classify densely built, heterogeneous urban areas in Delhi’s metropolitan region. We also analyze the classification results in proximity to transportation systems. SAR Data from Sentinel-I, optical reflectance data from Sentinel-II, and the locations of Metro Rail infrastructure were used to complete the analysis. The linear unmixing involved the calculation of pixel values as a function of three pure end members, namely vegetation, concrete, and water. The findings demonstrate that the fusion of optical and SAR data from Sentinel satellites has better accuracy than classification models that utilize data from only one source. However, classification models that use linear unmixing data instead of optical bands produce more realistic classification maps. Thus, we also show that there is a tradeoff between classification accuracy and the spatial representation of urban features. Finally, this study demonstrates there is a wide variety of land use in the study area, which needs to be further understood in the context of its urban planning and transportation development.