Main content area

Population bias in nighttime lights imagery

Zhao, Naizhuo, Samson, Eric L., Liu, Ying
Remote sensing letters 2019 v.10 no.9 pp. 913-921
African Americans, Asians, algorithms, demographic statistics, elderly, per-capita income, remote sensing, United States
Brightness of nighttime lights (NTL) derived from remote sensing imagery has been extensively used to estimate population. However, population bias in NTL images has yet to be investigated. In this study, we explore the associations of race and age demographics with brightness of NTL in the United States (US) for 2015 and 2016. We find that if a county has more African Americans or young (<18 years old) people, then the county is more likely to have brighter lights at night. In contrast, a county with more Asian or elderly (≥65 years) people is prone to have smaller brightness of NTL. We further find that a county with more Asian people usually has a larger personal income per capita while one with a larger proportion of African Americans tends to have a smaller personal income per capita. Thus, the brighter/dimmer NTL is not derived from higher/lower economic levels. Finally, based on percentage of increased mean squared error (%IncMSE) of the random forest model, we discover that proportions of elderly and adult Asian population subsets are the most uncertain variables than the other race-and-age subsets (e.g., adult African American and elderly White) associating with estimation of total population in the US.