Estimating Small Area Population Density Using Survey Data and Satellite Imagery : An Application to Sri Lanka

Engstrom, Ryan ; Newhouse, David ; Soundararajan, Vidhya

World Bank, Washington, DC

关键词: POPULATION DENSITY;    SATELLITE IMAGERY;    MACHINE LEARNING;    SMALL AREA ESTIMATION;    CENSUS DATA;   

DOI  :  10.1596/1813-9450-8776
RP-ID  :  WPS8776

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

Country-level census data are typicallycollected once every 10 years. However, conflict, migration,urbanization, and natural disasters can cause rapid shiftsin local population patterns. This study uses Sri Lankandata to demonstrate the feasibility of a bottom-up methodthat combines household survey data with contemporaneoussatellite imagery to track frequent changes in localpopulation density. A Poisson regression model based onindicators derived from satellite data, selected using theleast absolute shrinkage and selection operator, accuratelypredicts village-level population density. The model isestimated in villages sampled in the 2012/13 HouseholdIncome and Expenditure Survey to obtain out-of-sampledensity predictions in the nonsurveyed villages. Thepredictions approximate the 2012 census density well and aremore accurate than other bottom-up studies based onlower-resolution satellite data. The predictions are alsomore accurate than most publicly available populationproducts, which rely on areal interpolation of census datato redistribute population at the local level. Theaccuracies are similar when estimated using a random forestmodel, and when density estimates are expressed in terms ofpopulation counts. The collective evidence suggests thatcombining surveys with satellite data is a cost-effectivemethod to track local population changes at more frequent intervals.