APAN_conference_QING_2011

Slide 1:

Relationships between Nighttime Imagery and Population Density for Hong Kong Qing Liu Paul C. Sutton Christopher D. Elvidge Asia Pacific Advanced Network Hong Kong, China February, 2011

Introduction:

Introduction Previous analyses have revealed a striking correlation between city-lights and human population density Some researches suggested correlation between measures of population or population density and nighttime imagery is low and not statistically significant. DMSP-OLS imagery saturated in most of the areas of high population density. ISS photograph provide more information of nighttime lights with finer spatial resolution. Will ISS photograph produce better model to estimate population for areas with high population density than DMSP-OLS imagery?

Data:

Data Nighttime Imagery Radiance calibrated DMSP Nighttime Lights 2006 Hong Kong at Night - Astronaut Photography from ISS 2003 Population Representations 2008 LandScan population grid 2006 Hong Kong population by administrative district

Radiance Calibrated DMSP Nighttime Lights of Hong Kong 2006 (F16):

Radiance Calibrated DMSP Nighttime Lights of Hong Kong 2006 (F16) Cloud-free composite derived DMSP-OLS data collected at low, medium and high gain settings. 30 arc-second grid or approximately 1 km 2 at the equator

Hong Kong at Night - Astronaut Photography from ISS 2003.03.10:

Hong Kong at Night - Astronaut Photography from ISS 2003.03.10 Spatial Resolution: estimated to be 6m per pixel

LandScan Population Density of Hong Kong 2008:

LandScan Population Density of Hong Kong 2008 Representing ambient population count per cell 30 arc-second grid or approximately 1 km 2 at the equator US Laboratory Department of Energy, Oak Ridge National Laboratory

First-Principal Component of ISS photograph:

First-Principal Component of ISS photograph

Hypotheses:

Hypotheses Nighttime light intensity will positively correlate with the population density. Integration of nighttime lights within a district polygon will positively correlate with the total population of this polygon. ISS imagery will produce better model than existing models using DMSP-OLS imagery for estimation of population in areas with high population density.

Methods:

Methods Pixel-based Mean aggregation of ISS bands to 1km 2 grid Multi-variate Analysis LandScan, DMSP-OLS, ISS red, ISS green, ISS blue, ISS PC1 Correlation and scatterplot matrix Polygon-based Zonal statistics: sum of lights in each district polygon Ordinary least squares regression Independent variable: sum of lights within district polygon Dependent variable: total population for each district

Correlations:

ISS PC1 ISS R ISS G ISS B DMSP-OLS LandScan ISS PC1 1.0000 0.9988 0.9989 0.9851 0.7706 0.6118 ISS R 0.9988 1.0000 0.9976 0.9768 0.7677 0.6116 ISS G 0.9989 0.9976 1.0000 0.9796 0.7709 0.6127 ISS B 0.9851 0.9768 0.9796 1.0000 0.7618 0.5976 DMSP-OLS 0.7706 0.7677 0.7709 0.7618 1.0000 0.6239 LandScan 0.6118 0.6116 0.6127 0.5976 0.6239 1.0000 Correlations

Scatterplot Matrix:

Scatterplot Matrix

Parameter Estimation:

Parameter Estimation R 2 = 0.15 p-value = 0.1072 Weak linear relationship between sum of lights from DMSP-OLS imagery and total district population R 2 = 0.03 p-value = 0.4381 No linear relationship between sum of lights from ISS PC1 imagery and total district population

Slide 13:

Hong Kong International Airport Central & West Wan Chai Eastern Islands North Tuen Mun Kwun Tong Southern Hong Kong Island New Territories Kowloon Yuen Long Kowloon City Sham Shui Po Sai Kung

Slide 14:

Hong Kong International Airport Central & West Islands North Tuen Mun Southern Wan Chai Eastern Hong Kong Island New Territories Kwun Tong Kowloon Yuen Long Kowloon City Sham Shui Po Sai Kung

Discussion:

Discussion Overestimation for districts: including airport, commercial areas, residential areas in new town; with residents of high income and education. Underestimation for districts: including large residential areas and industrial areas; with residents of low income and education. “Reasonable” Estimation for districts: including relatively low density residential areas.

Conclusions:

Conclusions ISS photograph is no better than the DMSP-OLS imagery at predicting population density for Hong Kong at the spatial resolution of LandScan. The finer spatial and spectral resolution of the ISS photographs does not increase correlation between nighttime lights and measures of population or population density. Residence-based population at district level of Hong Kong provided by census data are not well modeled by nighttime imagery alone. Areas of high ambient population density that are not residential (airports, commercial areas, etc.) tend to emit more light than residential areas.