Spatialization Of Population Using Nighttime Light Remote Sensing Images And Social Sensing Data

Compared to census data based on administrative boundary, gridded population distribution data better reflects the distribution of residents, which provides important detailed information for city planning and public security management. In previous research models are established based on the relationship between land cover data, indicator of population and the census population data. Multi-variable linear regression or multi-variable weighted matrix are employed to build the population estimation model. Under the circumstance that night-time light images has been used for various research purposes such as the extraction of urban area and estimation of socioeconomic index, night-time light images have been widely utilized to create population model by integrating other types of data such as land cover data. However, the existing studies mainly focus on population spatialization at the national or the city level. Few are conducted at the town level. Moreover, the problem on the accuracy of population estimates based on night-time light images has not been tackled properly. By combining NPP/VIIRS night-time light images, census data and taxi trajectory data, this study proposes a method to redistribute the census data at grid scale in shanghai.,. The main contents of this research are as follows:1. Population spatialization using nighttime light remote sensing images. The correlation between nighttime light radiance and census data was investigated at different scales, ranging from the city to the town level. Taking spatial complexity and heterogeneity into consideration, the study area was divided into two districts to create estimation models. Gridded population distribution was then obtained at the first stage.2. Modification layer derived from social sensing data to modify the previous gridded population data. Origin-destination points relating to the travel place of residents were extracted from millions of taxi trajectory records after filtering the errors of GPS coordinates. Spatiotemporal pattern and semantic information of residents’ travel were analyzed at the town and grid scales, which help recognize functional region and dynamic flow of population in city.3. Correction of estimation residual caused by nighttime light using the modification layer. Based on density of OD pairs in specific time window, previous estimated population were corrected to reduce errors. Gridded population distribution data with the spatial resolution of 500m was obtained and validated for Shanghai city.As a result, NPP/VIIRS nighttime light radiance and census population are correlated at the town scale, which can help estimate the distribution of population at a finer scale. The coefficient of determination between modified estimation and census population at the town scale is 0.713 with relative error equal to 25%, which is better than nighttime light estimation result with R2 0.68 and relative error 28%. Within the random sample gridded area, coefficient of determination at the grid scale is 0.41, better than 0.36 of the nighttime light estimates.This study contributes to the existing studies by providing a new perspective of population spatialization. The gridded population distribution data established reflect detailed information of how residents distribute. Nighttime light images and taxi trajectory data were combined together to implement population spatialization, which is more accurate than only using land use data or night-time light data.