Analyzing Space-Time Variation of Urban Human Stay Using Kernel Density Estimation by Considering Spatial Distribution of Mobile Phone Towers

In recent years, the availability of mobile phone location data provides an opportunity and challenge for studying human stay. Therefore, we only can extract human stay based on base stations from the dataset, it need estimate to produce a continuous population distribution. Kernel density estimate (KDE) could generate a continuous surface and has been widely used to estimate population distribution, but the traditional KDE assumes that the sample data points are homogeneous and use fixed bandwidth to estimate for all data points, however, the service area of base stations in the city varies with the distribution of population distribution, so fixed bandwidth will bring error. In order to eliminate the errors, this paper introduces a search bandwidth controlling parameter to make the bandwidth to vary with the spatial distribution of mobile phone towers. Least-squares cross validation (LSCV) and log-probability methods were used to test the proposed approach, and the result of experiment demonstrates that this improvement can make the estimation better than fixed bandwidth. Taking mobile location data of Shenzhen as an example, we extract urban human stay for five typical time intervals, and the improved KDE was used to analyze the distribution difference of the five time intervals, which make us have a deep understanding of condition of urban different areas are used by human and how it vary with time going.