A New Method for Building-Level Population Estimation by Integrating LiDAR, Nighttime Light, and POI Data

Abstract

Building-level population data are of vital importance in disaster management, homeland security, and public health. Remotely sensed data, especially LiDAR data, which allow measures of three-dimensional morphological information, have been shown to be useful for fine-scale population estimations. However, studies using LiDAR data for population estimation have noted a nonstationary relationship between LiDAR-derived morphological indicators and populations due to the unbalanced characteristic of population distribution. In this article, we proposed a framework to estimate population at the building level by integrating POI data, nighttime light (NTL) data, and LiDAR data. Building objects were first derived using LiDAR data and aerial photographs. Then, three categories of building-level features, including geometric features, nighttime light intensity features, and POI features, were, respectively, extracted from LiDAR data, Luojia1-01 NTL data, and POI data. Finally, a well-trained random forest model was built to estimate the population of each individual building. Huangpu District in Shanghai, China, was chosen to validate the proposed method. A comparison between the estimation result and reference data shows that the proposed method achieved a good accuracy with R2=0.65 at the building level and R2=0.79 at the community level. The NTL radiance intensity was found to have a positive relationship with population in residential areas, while a negative relationship was found in office and commercial areas. Our study has shown that by integrating both the three-dimensional morphological information derived from LiDAR data and the human activity information extracted from POI and NTL data, the accuracy of building-level population estimation can be improved.

Evaluating spatial statistical and machine learning models in urban dynamic population mapping

“…Third, we integrated the covariates most representative of population mapping studies available at 100 m resolution. However, we dropped some commonly used variables due to the coarse spatial resolution or data accessibility issues, such as night-time light remote sensing data (Chen et al, 2021; Li and Zhou, 2018). We admit that incorporating those missing covariates might influence the conclusion we drew.…”

From auxiliary data to research prospects, a review of gridded population mapping

“…Raster DMSP/OLS, NPP/VIIRS, Luojia-1 01 star data, ISSP (need HSL transformation), etc. NTL brightness value (Sorichetta et al, 2015;Stevens et al, 2015), the number/ area of unlit and lit pixels (Dobson et al, 2000;Zeng et al, 2011), slope value of NTL brightness (Chen et al, 2021), the standard deviation of NTL brightness (Chen et al, 2021), etc.…”

Population estimation beyond counts—Inferring demographic characteristics

“…For that reason, we include also traditional population count estimation, as an intermediate step towards our enhanced demographic-aware population estimation. The selection of these datasets follows our hypothesis that amenities and real estate in neighbourhoods have been shaped by the demographics of its residents, a reasoning that has been inspired by recent work using such data for population estimation [ 18 , 43 , 44 ]. Hence our work also contributes to the body of knowledge by uncovering the value of different amenity and real estate data in population count, besides inferring demographic characteristics.…”