=Paper=
{{Paper
|id=Vol-3006/65_short_paper
|storemode=property
|title=Using remote sensing data in population density estimation
|pdfUrl=https://ceur-ws.org/Vol-3006/65_short_paper.pdf
|volume=Vol-3006
|authors=Evgeniy G. Shvetsov,Nadezhda M. Tchebakova,Elena I. Parfenova
}}
==Using remote sensing data in population density estimation==
https://ceur-ws.org/Vol-3006/65_short_paper.pdf
Using remote sensing data in population density
estimation
Evgeniy G. Shvetsov1,2 , Nadezhda M. Tchebakova3 and Elena I. Parfenova4
1
V.N. Sukachev Institute of Forest, Federal Research Center “Krasnoyarsk Science Center SB RAS”, Krasnoyarsk, Russia
2
Siberian Federal University, Krasnoyarsk, Russia
Abstract
In recent decades, remote sensing methods have often been used to estimate population density, especially
using data on nighttime illumination. Information about the spatial distribution of the population is
important for understanding the dynamics of cities and analyzing various socio-economic, environmental
and political factors. In this work, we have formed layers of the nighttime light index, surface temperature
and vegetation index according to the SNPP/VIIRS satellite system for the territory of the central and
southern regions of the Krasnoyarsk krai. Using these data, we have calculated VTLPI (vegetation
temperature light population index) for the year 2013. The obtained values of the VTLPI calculated for a
number of settlements of the Krasnoyarsk krai were compared with the results of the population census
conducted in 2010. In total, we used census data for 40 settlements. Analysis of the data showed that the
relationship between the value of the VTLPI index and the population density in the Krasnoyarsk krai
can be adequately fitted (𝑅2 = 0.65) using a linear function. In this case, the value of the root-mean-
square error was 345, and the relative error was 0.09. Using the obtained model equation and the spatial
distribution of the VTLPI index using GIS tools, the distribution of the population over the study area
was estimated with a spatial resolution of 500 meters. According to the obtained model and the VTLPI
index, the average urban population density in the study area exceeded 500 people/km2 . Comparison of
the obtained data on the total population in the study area showed that the estimate based on the VTLPI
index is about 21% higher than the actual census data.
Keywords
Population density, VIIRS, nighttime lights, VTLPI, Krasnoyarsk krai.
1. Introduction
Detailed information on the spatial distribution of the population is important for understanding
the dynamics of cities and analyzing various socio-economic, environmental and political
factors [1, 2]. The most readily available source of population data is usually census data. This
method, however, is expensive and time consuming, and therefore, population censuses are
rarely conducted [3].
In recent decades, remote sensing methods have often been used to estimate population
density, especially using nighttime light (NTL) data. In addition, in recent years, to improve the
accuracy of the population density estimates, nighttime lights data have been used in conjunction
with other data sources (vegetation indices, data on types of vegetation cover) [4]. We have
used several sets of remote sensing data to assess the spatial distribution of the population
SDM-2021: All-Russian conference, August 24–27, 2021, Novosibirsk, Russia
" eugeneshvetsov11@yandex.ru (E. G. Shvetsov)
© 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073 CEUR Workshop Proceedings (CEUR-WS.org)
550
Evgeniy G. Shvetsov et al. CEUR Workshop Proceedings 550–556
density in the southern and central regions of the Krasnoyarsk krai. The day night band (DNB)
of the VIIRS radiometer located on the Suomi NPP satellite was used as a source of nighttime
lights data. In addition, data on the Earth’s surface temperature and the NDVI vegetation index
were used to improve the accuracy of the population density estimates. These datasets were
also obtained using measurements made by the SNPP/VIIRS radiometer. The VTLPI (vegetation
temperature light population index) index [5] was used as a composite index to estimate the
population density.
The main purpose of this work was the estimation of the population density in the central
and southern regions of the Krasnoyarsk krai using remote sensing data. This included the
following objectives: (1) creation of the composite index, including data on nighttime lights in
the study area, surface temperature and vegetation index; (2) estimation of population density
using a model derived from the 2010 census.
2. Calculation of the Vegetation Temperature Light Population
Index from satellite data
We have generated the GIS layers containing nighttime lights; surface temperature and vegeta-
tion index using data from the SNPP/VIIRS satellite system.
The study area included the central and southern regions of the Krasnoyarsk krai, covering
an area from 52∘ N to 58∘ N and from 88∘ E to 99∘ E. (Figure 1). Its total area was about
250 thousand km2 , including three forest zones: taiga, forest-steppe, and the South Siberian
mountain zone [6]. According to the vegetation map [7], the area of forested lands is about 50%.
We have used satellite data obtained using the VIIRS radiometer of the SNPP satellite. As a
source of nighttime lights data, we have used a product generated by the National Geophysical
Data Center (NOAA, USA) and available at (https://www.ngdc.noaa.gov/eog/viirs/download_
dnb_composites.html). This product contains information on the light intensity from areas of
anthropogenic activity, as well as certain other objects or phenomena with a spatial resolution
of 15 arc seconds. We used data on nighttime lights for the territory of the Krasnoyarsk krai for
2013, which were averaged over 12 months, so each image pixel represented the mean annual
illumination value. After the averaging procedure, the final layer was re-projected into the
Albers Equal-Area Conical Projection with a spatial resolution of 500 meters.
To calculate the vegetation indices, we used the VNP13A1 product, containing vegetation
indices, such as NDVI, EVI, as well as reflectance values in the visible and near infrared wave
ranges. During data preprocessing, we also have excluded areas containing clouds or cloud
shadows. Further, a composite image was created using the maximum NDVI values obtained
between the beginning of June and the end of August of the year 2013.
Additionally, to calculate the composite VTLPI index, we also used data on the earth’s
surface temperature. The VNP21A1 product created using the VIIRS radiometer data was
used as a source of surface temperature data. Using product quality flags we have excluded
pixels containing clouds, cloud shadows, or low quality data. The final temperature layer was
calculated as the mean annual temperature. Further, the GIS layers containing the NDVI values
and surface temperature were also re-projected into the Albers equal-area conical projection
with a spatial resolution of 500 meters using the nearest neighbor method.
551
Evgeniy G. Shvetsov et al. CEUR Workshop Proceedings 550–556
Figure 1: GIS layer containing VTLPI spatial distribution for the territory of the central and southern
regions of the Krasnoyarsk krai, obtained using VIIRS radiometer data.
To exclude areas occupied by water bodies, a map of land surface types was used [7]. To
assess the population density, we used the VTLPI index [5], calculated as follows:
√
DNBnorm
VTLPI = · LSTnorm ,
NDVImax
where DNBnorm is the normalized DNB value, NDVImax is the maximum NDVI value, and
LSTnorm is the normalized surface temperature value. The normalized DNBnorm and LSTnorm
values were calculated as follows:
LST − LSTmin
LSTnorm = .
LSTmax − LSTmin
Here LSTnorm is the normalized surface temperature, ranging from 0 to 1. LSTmax and LSTmin
are the maximum and minimum temperature values within the study area, respectively, and
LST is the temperature value of the analyzed pixel.
√ √
DNB − DNBmin
DNBnorm = √
√︀
√ .
DNBmax − DNBmin
552
Evgeniy G. Shvetsov et al. CEUR Workshop Proceedings 550–556
Here DNBnorm is the normalized value of the nighttime light for the area, ranging from 0 to 1.
DNBmax and DNBmin are the maximum and minimum nighttime light values within the study
area, respectively, and DNB is the value of the analyzed pixel.
Unlike [5], we did not correct the VTLPI value for the height of the terrain above sea level,
since there were no mountain regions within our study area.
2.1. Modeling population density using remote sensing data
To use the VTLPI to assess the population density of the Krasnoyarsk krai, a model relationship
relating these two data sets was obtained. The materials of the year 2010 census were used as a
source of data on the actual population density. Census materials are publicly available on the
Rosstat website (https://rosstat.gov.ru/free_doc/new_site/perepis2010/croc/perepis_itogi1612.
htm).
We compared the averaged VTLPI values calculated for a number of settlements within
the Krasnoyarsk krai with the population values obtained from the census data. To calculate
the mean VTLPI value, we used geoinformation layer of the settlements boundaries. Using
these boundaries, we have extracted fragments from the VTLPI raster layer corresponding to
settlements for which several statistical parameters were calculated. In total, we used data for
40 settlements in the central and southern regions of the Krasnoyarsk krai.
According to the census data, the number of people in the analyzed settlements ranged
from 756 to 973826. At the same time, the mean VTLPI value for the considered settlements
ranged from 0.7 to 17. Figure 2 shows a diagram of the VTLPI values and population density
in the analyzed settlements according to the census data. Analysis of the data showed that
the relationship between these data series can be fitted using a linear function, such as 𝑃 =
136.5 * VTLPI − 105, which reasonable well describes the dependence between the population
density and the value of the composite VTLPI index (𝑅2 = 0.65). Here 𝑃 is the population
density (people/km2 ). In this case, the value of the root-mean-square error was 345, and the
median relative error was 0.09. The relative error was calculated as the difference between the
census data and the model data referred to the census data.
Figure 2: The relationship between the VTLPI index and population density according to the 2010
census. Each point corresponds to one settlement. The solid line represents the linear fit.
553
Evgeniy G. Shvetsov et al. CEUR Workshop Proceedings 550–556
Figure 3: Spatial distribution of population density in the study area.
Using the obtained model equation and the spatial distribution of the VTLPI index, we have
estimated the distribution of the population over the territory of the study area with a spatial
resolution of 500 m (Figure 3). According to the resulting model and the VTLPI index, the highest
population density was observed in cities, which corresponds to the actual census data. The
average density of the urban population in the considered territory exceeded 500 people/km2 .
Comparing our results with the census data for the study area we found that the VTLPI-based
estimate is about 21% higher than the actual census data. Thus, according to the model estimate,
the total population was 2.9 million, while according to the census data, this was approximately
2.4 million.
There are several factors that can influence the accuracy of estimates. First, the VTLPI index
does not allow excluding the influence of industrial objects, which are characterized by rather
high values of the nighttime light measurements, temperature, as well as low values of the NDVI
index. For example, on the territory of the Achinsk refinery, the population density, according
to the model estimate, reached 4–5 thousand people/km2 , which is obviously associated with
significant emissions of thermal and light radiation. Also, errors in determining the boundaries
of settlements used to create a sample in the process of the model parameterizing can have an
impact on the accuracy of estimates. In addition, the time difference between census data (2010)
and the time of satellite data acquisition (2013) is likely to have influenced the results.
554
Evgeniy G. Shvetsov et al. CEUR Workshop Proceedings 550–556
3. Conclusions
Using the archive of thematic products, generated using the data of the VIIRS radiometer,
we have created the GIS layers of the nighttime lights, the vegetation index NDVI and the
land surface temperature for the central and southern regions of the Krasnoyarsk krai for the
year 2013. Using these layers, we have calculated the VTLPI index related to population density.
According to the year 2010 census data obtained for 40 settlements of the Krasnoyarsk krai, a
training sample was created, which was used to parameterize a linear model relating the VTLPI
index and the actual population density.
Using the obtained model equation and the spatial distribution of the VTLPI index, the
population density for the study area was estimated with a spatial resolution of 500 m. At the
same time, comparison with the census data showed that the estimate based on the VTLPI index
is about 21% higher than the actual data.
Acknowledgments
This work was financially supported by the Russian Foundation for Basic Research, the Krasno-
yarsk Regional Fund for the Support of Scientific and Scientific and Technical Activities, project
No. 19-45-240004-r_a “Predictions of the ecological-economic potential for possible “climatic”
migrations in the Angara-Yenisei macroregion in a changing climate of the 21st century”.
References
[1] Ahola T., Virrantaus K., Krisp J.M., Hunter G.J. A spatio-temporal population model to
support risk assessment and damage analysis for decision-making // International Journal
of Geographical Information Science. 2007. Vol. 21. P. 935–953.
[2] Aubrecht C., Özceylan D., Steinnocher K., Freire S. Multi-level geospatial modeling of
human exposure patterns and vulnerability indicators // Natural Hazards. 2013. Vol. 68.
P. 147–163.
[3] Yang X., Huang Y., Dong P., Jiang D., Liu H. An updating system for the gridded population
database of China based on remote sensing, GIS and spatial database technologies //
Sensors. 2009. Vol. 9. P. 1128–1140.
[4] Roy Chowdhury P.K., Maithani S., Dadhwal V.K. Estimation of urban population in Indo-
Gangetic Plains using night-time OLS data // International Journal of Remote Sensing.
2012. Vol. 33. Is. 8. P. 2498–2515.
[5] Luo P., Zhang X., Cheng J., Sun Q. Modeling population density using a new index derived
from multi-sensor image data // Remote sensing. 2019. Vol. 11. P. 2620.
[6] Decree of the Ministry of Natural Resources of Russia “On approval of the list of forest
zones of the Russian Federation and the list of forest regions of the Russian Federation”
No. 367 dated 18.08.2014 (as amended on 19.02.2019). Moscow, 2014. (In Russ.)
[7] Bartalev S.A., Belward A., Erchov D.V., Isaev A.S. A new SPOT4-VEGETATION derived
land cover map of Northern Eurasia // International Journal of Remote Sensing. 2003.
Vol. 24. Is. 9. P. 1977–1982.
555