Mapping urban land use by combining multi-source social sensing data and remote sensing images

Abstract

Knowledge of detailed urban land-use patterns is essential in urban management, economic analysis, and policy-making aimed at sustainable urban development. To extract this information, previous studies relied on either the physical features extracted from remote sensing images or human activity patterns analyzed from social sensing data, but seldom on both of them. In this study, we proposed a framework to map the land-use patterns of New York City by combining multiple-source social sensing data and remote sensing images. We started by generating urban land use parcels using the transportation network from the Open street map and grouping them into built-up and non-built-up categories. Then, the random forest method was applied to classify built-up parcels and the National Land Cover Data was used to determine the land use type for non-built-up parcels. Results indicate that a satisfying overall testing accuracy with 77.31% was achieved for the level I classification (residential, commercial, and institutional regions) and 66.53% for level II classification (house, apartment, public service, transportation, office building, health service, education, and retails). Among the Level II classes, the residential land use has achieved the highest accuracy in built-up parcels with the user’s accuracy at 74.19% and producer’s accuracy at 80.99%. In addition, the classified map indicates that most commercial areas are concentrated in the Manhattan, residential land uses are distributed in the boroughs of Staten Island, Bronx, Queens, and Brooklyn, and institutional areas are evenly distributed in Manhattan, Brooklyn, Queens, Bronx, and Staten Island. The classified land use and functional information could further be used in other studies, such as urban planning and urban building energy use modeling.