An Assessment of Urban Area Extraction Using ALOS-2 Data

Urbanization has a dynamic structure especially in megacities and therefore rapid detection of the urban is vital for sustainable management of the city. In this work, we apply a multi-source feature data approach to investigate the urban area of Istanbul, Turkey which is a megacity with an approximate 15 million inhabitant, and under strong both anthropogenic and natural pressures. In order to analyse and compare the spatial pattern of the urban footprint, different techniques are applied. Speckle divergence, backscatter and repeat pass interferometric coherence values are considered for the analysis. To this aim, L-band HH and HV polarized ALOS-2 Synthetic Aperture Radar (SAR) data were acquired from Japan Space Exploration Agency's (JAXA). Pixel based Random Forest Classification method was used for the urban mapping. During the classification, different scenarios have been applied using speckle divergence, backscatter and coherence information. Overall, user and producer accuracies were calculated from the error matrix. While comparing HH and HV polarimetry, in each scenario HH provided much higher accuracies than HV results. Speckle divergence and backscatter values yielded similar accuracies which is around 88% for urban class. However, coherence gave approximately 69% while it is classified individually. The contribution of coherence was extracted while coherence was stacked with speckle divergence, and the result was improved to 91%. The urban areas was extracted with a maximum accuracy of maximum 93% while all information was combined. The preliminary results allow us to obtain a comprehensive image of urban structure, and indicate that the results may reference address for further analysis of multi-temporal SAR data over large and complicated mega cities.