Land use classification from combined use of remote sensing and social sensing data

Large amounts of data can be sensed and analyzed to discover patterns of human behavior in cities for the benefit of urban authorities and citizens, especially in the areas of traffic forecasting, urban planning, and social science. In New York, USA, social sensing, remote sensing, and urban land use information support the discovery of patterns of human behavior. This research uses two types of openly accessible data, namely, social sensing data and remote sensing data. Bike and taxi data are examples of social sensing data, whereas sentinel remote sensed imagery is an example of remote sensing data. This research aims to sense and analyze the patterns of human behavior and to classify land use from the combination of remote sensing data and social sensing data. A decision tree is used for land use classification. Bike and taxi density maps are generated to show the locations of people around the city during the two peak times. On the basis of a geographic information system, the maps also reflect the residential and office areas in the city. The overall accuracy of land use classification after the consideration of social sensing data is 85.3%. The accuracy assessment shows that the combination of remote sensing data and social sensing data facilitates accurate urban land use classification.