Using nighttime DMSP/OLS images of Citylights to Estimate District-level Population Distribution in Developing Countries.
Takahiro Konami, Roysouke Shibasaki, and Guoxin Tan
Center for Spatial information Science, University of Tokyo
7-22-1, Roppongi, Minato-Ku Tokyo 106, Japan
Tel: (81)-3-3402-6231 Fax(81)- 3-3479-2763
E-mail: konami@skl.iis.u-tokyo.ac.jp
Abstract
With the world having a population of more than 6 million . it is important for us to forecast future populating of the earth, especially in the developing countries which face a populating explosion. To forecast future population , it is necessary to grasp the fluctuation of district-level changes in population frequently and globally. In this research, a model was developed which yields district-level population distribution based on an input of DMSP/OLS images of night-time city lights. the existing detailed population distribution data of china was used to estimate the parameters of the model.
1. Introduction
Under the situation of exploding population on the earth, it is necessary to recognized the present situation of population distribution of the world in order to be able to estimate the future situation correctly.
For this purpose one of the most important issues is to develop a method t grasp in detail the distribution of population , especially in the developing countries.
To see detailed transition of population distribution, It is desirable to have population distribution data by grid cells, but in developing counties such detailed statistics of population is not yet available.
As a consequence of these situations, the final purpose of this research is to develop a method to estimate detailed population distribution from data that can be obtained remotely, by observations from the space. To achieve this, the purpose of this paper is to explore a relationship between stable night light data generated by National Oceanic and Atmospheric Administration's National Administration's National Geophysical Data Center (NOAA/NGDC), and population distribution, which can be obtained form governments. The study area for this research is the People's Republic of china, as detailed population statistics is available in China, which is the largest one among developing countries.
2. Overview of the research
2.1. Source of the data
Stable night light data was obtained as a prototype " city light" data set form NOVAA/NGDC using the Defense Meteorological Satellite Program's Operational Line-scan system (DMSP/OLS). This data shows a cumulative percentage of lighted area for each pixel, of size 2.7 X km X 2.7 km. (Fig. 1-1).
Fig.1-1 Distribution of Night-Light Around China
When using DMSP/OLS data, thresholding has to be applied because of the very high in the OLS nighttime photo-multiplier configuration . Unless thresholding is done, OLS night-time light data may contain fisher's lights or other lights which have no relation to the inhabitants in that place. Thresholding cab eliminate such lights, which has no population.
As for the population data in the analysis, the author used " China Country -Level Data on Population (CENSUS) and Agriculture " which includes 1990 censes and agricultural economic variables at he county level for the People's Republic of China ( Fig.1.2). and the authors used 1:1 M GIS map as the boundary data of the counties in china . this data was developed by china in Time and Space (CITAS) Project Funded by Center for International Earth Science Information Network .
2.2 Overview of the analysis
As we mentioned above , the purpose of this research is to propose a method to estimate population distribution of China using DMSP/OLS night-light images .
In this research , the authors have tried to construct amodel, which can estimate county-level population from the night-light images.
To make such model, firstly the authors have assumed some models between population and night-light rate of each cell. After that , the authors have evaluated the accuracy of the assumed models by checking the output of them with the aggregated county-level population data . then, the best-fit model cab be found by comparing correlation values between output of the models and the actual population data .
Parameter that should be estimated are
(i)parameter of the initial model conditions that explain the relations between night-light and population of each grid cell , and (ii) appropriate thresholding rate for the night-light data to eliminate non-populated lights .
Fig.1-2 Distribution of Population of China from the Census
In summary, the procedure of the analysis is as follows.
-
make models with unknown parameter between night light rate and population in each grid cell.
- set thresholding rate under which the night-light rat is ignored.
- Estimate parameters of the model by minimizing square of he errors between the estimated populations and the actual ones .
- Find the best threshold rate, which can be maximize the correlation between estimated and actual population .
- Compare the models to find the best model.
As the initial models, the authors proposed the following two equations to explain the relationships between night-lights rates and populations.
(a) liner models
P=
aX+
b
(b) exponential model
P=
aX
b
In this research , these models were applied for (1) whole of China, and for (2) every province of China. The following sections will explain the results of estimations of based on these models.
