Estimation of population distribution using satellite imagery and GIS data

Tran Thanh Dan*, Manzul Kumar Hazarika, Hiroyuki Miyazaki, Syams Nashrrullah, A. Dahal, Ryosuke Shibasaki

*Corresponding author for this work

Research output: Contribution to conferencePaperpeer-review


Spatial distribution of population map at a finer scale is useful for planning and policy development. A number of population estimation techniques have been developed to disaggregate census data and predict density of population at finer scale. Therefore, this research is one of those attempts to improving high resolution on human population distributions, by presenting a new modeling approach to map the population using census, building footprints, satellite imagery, and ancillary data. The data were processed through four main steps: (1) data collection and pre-processing including: population and building footprints extraction from census data and cadastral map and/or satellite data, respectively; socioeconomic and building information survey using DRM Survey mobile application which developed by Geoinformatics Center, Asian Institute of Technology (GIC-AIT, Thailand); and ancillary data collection, including: topographic, infrastructure, river network, road network, satellite data, and night-time light imagery; (2) covariates preparation for fitting and predicting randomForest models; (3) model adjustment and estimation population at building level; and (4) geospatial population distribution mapping at 30m spatial resolution. Validation of results was made by comparing the estimation with the observation data at building level, which showed a good correlation with R2 = 0.83. We found that RF model performs better than several other commonly used models. An assessment of covariates is important for accurately estimating population. The values of variable importance may fluctuate as the number of covariates is reduced. However, relative ranking is quite stable among top covariates, for example: distance to function area (hospital, school, post office, …), road networks, or night-time light are most important predictors for reducing amount of variability left in log population of training data. An advantage with the approach is that we can aggregated population can be re-distributed to a fine scale, providing quantitative information of planning and policy development.

Original languageEnglish
Number of pages9
Publication statusPublished - Oct 2018
Externally publishedYes
Event39th Asian Conference on Remote Sensing: Remote Sensing Enabling Prosperity, ACRS 2018: Remote Sensing Enabling Prosperity - Kuala Lumpur, Malaysia
Duration: 15 Oct 201819 Oct 2018
Conference number: 39


Conference39th Asian Conference on Remote Sensing: Remote Sensing Enabling Prosperity, ACRS 2018
Abbreviated titleACRS 2018
CityKuala Lumpur
Internet address


  • Building Extraction
  • Machine Learning
  • Mobile Application
  • Population Mapping
  • ITC-CV


Dive into the research topics of 'Estimation of population distribution using satellite imagery and GIS data'. Together they form a unique fingerprint.

Cite this