remote sensing

Article
Quantitative Responses of Satellite-Derived
Nighttime Lighting Signals to Anthropogenic
Land-Use and Land-Cover Changes across China
Ting Ma 1,2,3
1 State Key Laboratory of Resources and Environmental Information System, Institute of Geographical
Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China;
[email protected]; Tel.: +86-10-64889769
2 University of Chinese Academy of Sciences, Beijing 100049, China
3 Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and
Application, Nanjing 210023, China




Received: 4 August 2018; Accepted: 7 September 2018; Published: 11 September 2018


Abstract: Remotely sensed artificial lighting radiances at night can provide spatially explicit
proxy measures of the magnitude of human activity. Satellite-derived nighttime light images,
mainly provided by the Defense Meteorological Satellite Program (DMSP) and the Visible Infrared
Imaging Radiometer Suite (VIIRS) day/night band (DNB), have been increasingly used to study
demographic and socioeconomic activities for a wide range of issues—for instance, human population
dynamics, economic growth, and urbanization process—at multiple scales. In practice, the lack of
texture information regarding man-made surfaces would usually lead to substantial difficulty in
delineating the spatial dynamics in human settlements due to the diverse distributions of artificial
nocturnal lighting sources, which are closely related to the predominant land-use/land-cover
(LULC) types and their evolutions. An understanding of how nighttime lighting signals respond to
synchronous anthropogenic LULC changes, therefore, is crucially important for the spatiotemporal
investigations of human settlement dynamics. In this study, we used DMSP-derived nighttime light
(NTL) data and Landsat-derived LULC maps to quantitatively estimate the pixel-level responses of
NTL signals to different types of human-induced LULC conversions between 1995 and 2010 across
China. Our results suggest that the majority (>70%) of pixel-level LULC conversions into artificial
lands (including urban, rural, and built-up lands) might show a statistically significant increase in
nighttime brightness with an average >20 (in digital number, DN) step change in nighttime lights
(dNTL), both of which are distinctly higher than that in the LULC conversions into non-man-made
surfaces on the whole. A receiver operating characteristic (ROC) curve-based analysis implies
that we might have an average chance of ~90% to identify the nationwide LULC conversions into
man-made surfaces from all types of conversions through the observed changes in artificial nocturnal
luminosity signals. Moreover, ROC curve-based analyses also yield two nation-level optimal dNTL
thresholds of 4.8 and 7.8 DN for recognizing newly emerged three types of artificial lands and urban
lands between 1995 and 2010 across the entire country, respectively. In short, our findings reveal
fundamental insights into the quantitative connections between the anthropogenic LULC changes and
the corresponding responses of synchronous nightlight signals at the pixel-level, which are generally
essential for further applications of satellite-derived nocturnal luminosity data in the spatiotemporal
investigations of human settlement dynamics.

Keywords: nighttime light; land-use/land-cover change; step change; artificial lands;

optimal threshold

Remote Sens. 2018, 10, 1447; doi:10.3390/rs10091447 www.mdpi.com/journal/remotesensing

Remote Sens. 2018, 10, 1447 2 of 15

1. Introduction
Satellite-based observations of artificial nocturnal brightness have been extensively used for a
range of issues related to human activity because of the well-documented significant relationships
between nighttime lighting (NTL) signals and several urbanization, demographic, and socioeconomic
variables over time and space [1–10]. The outstanding characteristic of remotely sensed nighttime light
datasets is that they can provide us with timely and consistent proxy measures of the degrees of various
human activities in a spatially explicit manner [11–15] closely related to land features [16]. The widely
used nighttime light images were previously provided by the Defense Meteorological Satellite
Program (DMSP) and are currently being derived from the Visible Infrared Imaging Radiometer Suite
(VIIRS) day/night band (DNB) onboard the Suomi National Polar-Orbiting Partnership (Suomi-NPP)
satellite [17–21]. Both satellite-derived observations show significant quantitative responses of
sensed nighttime light radiances to gross domestic product, human population size, urbanized areas,
and electric power consumption at local, regional, and national scales [22–33].
Compared with visible or infrared remote sensing images representing the spatial patterns of
natural and man-made land features, satellite-derived nighttime light data are primarily determined
by the spatial distributions of artificial lighting sources at night. Thus, the lack of detailed texture
information regarding landscape features poses a major challenge to characterizing the spatial and
temporal patterns in human settlements [34]. Moreover, the saturation (for DMSP data due to the
lack of on-board radiance calibration) and over-glow (i.e., pixels can be lighted by adjacent lighting
sources, for both DMSP and VIIRS DNB data) effects of nightlight signals can markedly affect the
spatial identifications of artificial land features [25,31,34]. In practice, satellite-derived land-use and
land-cover maps and vegetation index data have typically been applied in nighttime light data-based
studies to diminish the impacts of the abovementioned effects and improve the recognition accuracy
of man-made surfaces across human settlements [35–37].
Although the combination of nighttime light data and other sources of remote sensing data can
be helpful for studying human activity in terms of the magnitude and its spatial patterns, the spatial
partition of human settlements is still challenging, largely because of both the diversity of human
activity and the spatial heterogeneity of artificial lighting sources at night, even though we have
more detailed looks as provided by the current VIIRS DNB images. Hence, an understanding of the
connection between nighttime lights and corresponding land features and how nightlight signals
respond to anthropogenic land-use and land-cover (LULC) changes at the pixel level is crucial for
further applications of remotely sensed nocturnal luminosity data. Unfortunately, despite their
fundamental importance, these two issues are less well understood.
Using VIIRS DNB nighttime light data and Landsat-derived LULC maps, our previous study [16]
has represented a nationwide investigation with respect to the pixel-level relationship between NTL
and various types of LULC across China. In the present study, we aim for the quantitative responses
of nighttime lighting signals to corresponding anthropogenic LULC changes between 1995 and 2010 in
China. DMSP-derived NTL data and Landsat-derived LULC data are used to perform a nationwide
comparative analysis of the pixel-level alterations in nighttime lights between the two observational
periods among different types of human-induced LULC conversions. Welch’s t-test is employed to
examine whether there exists a statistically significant step change in the pixel-level NTL signals in
response to synchronous LULC conversions. This study, together in association with our previous
investigation [16], represents an attempt to provide a basic understanding of the connections between
two satellite-based observations of nocturnal light radiances and the LULC types and their synchronous
alterations for further applications of nighttime light data in studies of human settlement dynamics.
Remote Sens. 2018, 10, 1447 3 of 15

2. Materials and
Remote Sens. Methods
2018, 10, x FOR PEER REVIEW 3 of 15

2.1. Land-Use andand

2. Materials Land-Cover
MethodsData
Two LULC maps with a spatial resolution of 100 m under the Albers equal-area projection
2.1. Land-Use and Land-Cover Data
across China’s mainland in 1995 and 2010 were provided by the Institute of Geographical Sciences
and Natural TwoResource
LULC maps with a spatial
Research resolution
(IGSNRR) of 100
[38]. m under
These LULC thedatasets
Albers equal-area projection
were derived fromacross
Landsat
China’s mainland in 1995 and 2010 were provided by the Institute of Geographical Sciences and
Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (EMT+) images and produced by using
Natural Resource Research (IGSNRR) [38]. These LULC datasets were derived from Landsat
a supervised classification method in cooperation with ground truth verification. Detailed information
Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (EMT+) images and produced by using
regarding the LULC
a supervised data can bemethod
classification found inin[38,39]. In thiswith
cooperation study, the LULC
ground truthtypes initially Detailed
verification. presented in
the IGSNRR maps were re-categorized into seven different categories: cropland (including
information regarding the LULC data can be found in [38,39]. In this study, the LULC types initially rain-fed
presented in the IGSNRR maps were re-categorized into seven different categories: croplandlakes,
and irrigated), vegetation (including grassland and forestland), water (consisting of rivers,
(including
reservoirs, andrain-fed and irrigated),
bottomland), vegetation
urban land (includingrural
(settlements), grassland
landand forestland),
(residential waterincluding
areas, (consistingsmall
towns, of villages,
rivers, lakes,
andreservoirs,
hamlets),and bottomland),
built-up urban landof(settlements),
land (consisting rural land
industrial areas, (residential
factories, mines,areas,
oil fields,
including
salt fields, andsmall towns, villages,
transportation), andand hamlets),
unused land built-up land (consisting
(consisting of deserts,ofGobi,
industrial areas, and
swamps, factories,
bare soil).
mines, oil fields,
As exemplified by thesalt
Pearlfields, and
River transportation),
Delta, and unused
there are visible LULC land (consisting
conversions intoofhuman-dominated
deserts, Gobi,
swamps, and bare soil). As exemplified by the Pearl River Delta, there are visible LULC conversions
lands between 1995 and 2010 (see Figure 1a,b), mainly driven by the underlying rapid economic
into human-dominated lands between 1995 and 2010 (see Figure 1a, b), mainly driven by the
growth and urbanization process in this region and across the entire country as well. A total of
underlying rapid economic growth and urbanization process in this region and across the entire
12,210,817
countrypixels (covering
as well. A total of~12.2 × 104pixels
12,210,817 km2 (covering
of land surface) appear
~12.2 × 10 4 km2 oftoland
have LULCappear
surface) changes between
to have
1995 and 2010 across China and are thus involved in subsequent analyses.
LULC changes between 1995 and 2010 across China and are thus involved in subsequent analyses.

Figure
Figure 1. Illustrations
1. Illustrations of satellite-derivedobservations
of satellite-derived observations of
ofanthropogenic
anthropogenicland-use/land-cover (LULC)
land-use/land-cover (LULC)
changes and synchronous dynamics in nighttime lighting (NTL) signals over the Pearl River Delta
changes and synchronous dynamics in nighttime lighting (NTL) signals over the Pearl River Delta
region in China. (a) Landsat-derived human-dominated land uses in 1995. (b) Landsat-derived
region in China. (a) Landsat-derived human-dominated land uses in 1995. (b) Landsat-derived
human-dominated land uses in 2010. (c) The Defense Meteorological Satellite Program (DMSP)-
human-dominated land uses in 2010. (c) The Defense Meteorological Satellite Program (DMSP)-derived
derived average NTL changes between 1992–1995 and 2010–2013.
average NTL changes between 1992–1995 and 2010–2013.
2.2. Nighttime Light Brightness Data
2.2. Nighttime Light Brightness Data
Time-series data of annual cloud-free composited stable DMSP nighttime light products (version
Time-series data of annual cloud-free composited stable DMSP nighttime light products
4.0) for the period 1992–1995 and the period 2010–2013 were used in this study. DMSP NTL images
(version
with4.0) for the
a spatial period 1992–1995
resolution and the
of 30 arc-seconds period
under the 2010–2013 were
WGS84 spatial used insystem
reference this study.
(~1 kmDMSP
at the NTL
equator and ~0.8 km at 40°N) were provided by the National Oceanic and Atmospheric
images with a spatial resolution of 30 arc-seconds under the WGS84 spatial reference system (~1 km
at the equator and ~0.8 km at 40◦ N) were provided by the National Oceanic and Atmospheric
Administration (NOAA) National Centers for Environmental Information (NCEI) (we downloaded
Remote Sens. 2018, 10, 1447 4 of 15

Remote
the data Sens. 2018,
from the 10, x FOR PEER
website REVIEW
at https://ngdc.noaa.gov/eog/dmsp.html). DMSP NTL images 4 of 15 were
re-projected into the Albers equal-area projection by a nearest neighbor resampling method to match the
Administration (NOAA) National Centers for Environmental Information (NCEI) (we downloaded
LULC datasets. Original DMSP NTL images are composed of the annual average visible band digital
the data from the website at https://ngdc.noaa.gov/eog/dmsp.html). DMSP NTL images were re-
numbers (DN, hereafter) ranging from 0 to 63. To diminish the yearly variations and differences among
projected into the Albers equal-area projection by a nearest neighbor resampling method to match
sensors,
the the
LULC second-order regression
datasets. Original DMSP model
NTL proposed
images areby [40] wasofused
composed to empirically
the annual average intercalibrate
visible band the
DMSPdigital
NTL products.
numbers (DN, Moreover, the annual
hereafter) rangingcomposites
from 0 to of 63.DMSP NTL images
To diminish that show
the yearly the best
variations andfitting
differences among sensors, the second-order regression model proposed by [40] was used to used
regression for years with two separate sensor composites were used in this study (here, F10 was
empiricallyF12
for 1992–1994, intercalibrate
for 1995 andthe F18
DMSPforNTL products.As
2010–2013). Moreover, the annual
demonstrated composites
in Figure of DMSP
1c, the NTLmean
four-year
images
of DMSP NTL thatsignals
show the best fitting
shows marked regression
changesfor years with two the
accompanying separate
LULC sensor composites
conversions were
(see used1a,b),
Figure
in thisover
prevalent studythe(here,
PearlF10 was used
River Delta,forbetween
1992–1994,
theF12
twoforobservational
1995 and F18 for 2010–2013).
periods. As demonstrated
Hence, we would be able
to estimate the pixel-level responses of DMSP NTL signals to corresponding LULC changes. the
in Figure 1c, the four-year mean of DMSP NTL signals shows marked changes accompanying
LULC conversions (see Figure 1a, b), prevalent over the Pearl River Delta, between the two
observational
2.3. Step Change inperiods. Hence,
Nighttime Lightwe would be able to estimate the pixel-level responses of DMSP NTL
Signals
signals to corresponding LULC changes.
Theoretically, anthropogenic LULC conversions usually accompany alterations in artificial
2.3. Step
nighttime Change sources,
lighting in Nighttime Light Signals
especially with respect to man-made surfaces. Hence, such changes
can be captured by satellite-derived measurements
Theoretically, anthropogenic LULC conversions of nocturnal brightness.
usually accompany Figure 2a
alterations inillustrates
artificial the
nighttime
observed changes lighting sources,
in DMSP NTL especially with respectand
during 1992–1995 to man-made
2010–2013surfaces. Hence,
over a given such pixel,
LULC changes can was
which
be captured
recognized as croplandby satellite-derived
in 1995 and measurements
converted intoofurban nocturnal
landbrightness.
in 2010. For Figure 2a illustrates
quantifying the
the response
observed signals
of nightlight changesto in corresponding
DMSP NTL during LULC1992–1995 and 2010–2013
changes, we defined over
thea given LULC pixel,
step change which NTL
in DMSP
wasdNTL
(termed recognized as cropland
throughout, in 1995
in units of DN)and as
converted into urban
the difference land in
between the2010. For quantifying
four-year arithmeticthe means
response of nightlight signals to corresponding LULC changes, we defined the step change in DMSP
of 2010–2013 and 1992–1995. We use here the four-year mean instead of that of a single year in an
NTL (termed dNTL throughout, in units of DN) as the difference between the four-year arithmetic
attempt to reduce the effect of inter-annual variations and uncertainty in DMSP NTL signals and
means of 2010–2013 and 1992–1995. We use here the four-year mean instead of that of a single year
further to allow us to perform the statistical test on the mean change in NTL signals responding to
in an attempt to reduce the effect of inter-annual variations and uncertainty in DMSP NTL signals
the LULC conversions.
and further to allow Inus the presentthe
to perform study, the two-sided
statistical Welch’s
test on the mean t-testin(also
change knownresponding
NTL signals as the unequal
to thet-test)
variances LULC wasconversions.
used to test In the
the statistical
present study, the two-sided
significance of dNTL Welch's
basedt-test (also
on the known
null as the (H0)
hypothesis
unequal
that two variances
observed timet-test)
serieswas used toNTL
of DMSP test the
datastatistical
have equal significance of dNTL
means (i.e., dNTL based
= 0).onInthe null
subsequent
hypothesis
analyses, we mainly (H0) use
that the
twoaverage
observeddNTLtime series of DMSP
and the NTL data
proportion have equalpixels
of converted meansshowing
(i.e., dNTL = 0).
statistically
In subsequent
significant dNTL (P-valueanalyses,<we mainly
0.05) at theuse the average
pixel-level to dNTL
estimate andthe
theresponse
proportion ofofDMSP
converted
NTLpixels
signals to
different types LULC conversions. Moreover, we use a receiver operating characteristic (ROC)ofcurve
showing statistically significant dNTL (P-value < 0.05) at the pixel-level to estimate the response
DMSP NTL signals to different types LULC conversions. Moreover, we use a receiver operating
to obtain optimal dNTL thresholds for identifying LULC conversions into artificial lands. The flow
characteristic (ROC) curve to obtain optimal dNTL thresholds for identifying LULC conversions into
chart for the study is presented in Figure 2b. Data processing and analyses were performed in Esri’s
artificial lands. The flow chart for the study is presented in Figure 2b. Data processing and analyses
ArcGIS and
were the R software
performed in Esri'senvironments.
ArcGIS and the R software environments.

Formatted: MDPI_5

Figure
Figure 2. (a)2.An
(a) illustrative
An illustrative exampleofofobserved
example observed changes
changes in
innighttime
nighttimelight (NTL)
light overover
(NTL) a given pixel pixel
a given
converted from cropland (1995) into urban land (2010) between 1992–1995 and 2010–2013. Welch’s t-
converted from cropland (1995) into urban land (2010) between 1992–1995 and 2010–2013. Welch’s t-test
test indicates that there is a statistically significant step change in nighttime lighting signals (dNTL)
indicates that there is a statistically significant step change in nighttime lighting signals (dNTL) between
between the two observational periods. (b) Flow chart of the study.
the two observational periods. (b) Flow chart of the study.
Remote Sens. 2018, 10, 1447 5 of 15
Remote Sens. 2018, 10, x FOR PEER REVIEW 5 of 15

3. Results
3. Results and
and Discussion
Discussion

3.1. LULC
3.1. LULC Changes
Changes and
and Nightlight
Nightlight Dynamics
Dynamics
Driven by
Driven by the
thefast-growing
fast-growingeconomyeconomy andandrapidrapid urbanization
urbanization process,
process, China China appears
appears to have to
have undergone remarkable changes in the nationwide LULC over the
undergone remarkable changes in the nationwide LULC over the past several decades [39]. As past several decades [39].
As represented
represented in Figure
in Figure 3a, human
3a, human activity activity contributes
contributes the majority
the majority of theconversions
of the LULC LULC conversions
between
between 1995 and 2010. Most notably, sizeable pixels (~75.9% of the total
1995 and 2010. Most notably, sizeable pixels (~75.9% of the total 10,452,217, ~7.9 × 10 km 10,452,217, ~7.94 × 102,4 after
km2 ,
after excluding
excluding minorminor
types oftypes
LULC of LULC conversions,
conversions, hereafter)hereafter)
were foundweretofound to be converted
be converted into human- into
human-dominated
dominated lands (including
lands (including croplands, croplands,
urban lands, urban lands,
rural lands,rural
and lands,
built-upand built-up
lands) fromlands) from
vegetation,
vegetation, water, and unused lands. Approximately 88.7% of LULC conversions
water, and unused lands. Approximately 88.7% of LULC conversions were related to croplands (~4.5 were related to
croplands (~4.5 × 10 4 km2 ). ~21.7% of pixels (~2.3 × 104 km2 ) were converted into human settlements
× 104 km2). ~21.7% of pixels (~2.3 × 104 km2) were converted into human settlements (including urban
(including
and urban
rural) and and rural)
built-up and
lands. built-up
It is shouldlands. It is should
be mentioned thatbe mentioned
some types ofthat
LULCsome types of LULC
conversions with
conversions with a pixel number less than 10 4 were excluded from Figure 3a because these minor
a pixel number less than 10 were excluded from Figure 3a because these minor conversions typically
4

conversions
occurred typically
among occurred among
non-man-made surfacesnon-man-made
with dim nighttimesurfaceslighting
with dim nighttime lighting signals.
signals.

Figure 3. (a) Summary of the pixel number for major kinds of LULC conversions between 1995 and
2010 in China.
China. (b)
(b)Average
Averagestep
stepchange
changeinin nighttime
nighttime lighting
lighting signals
signals between
between 1992–1995
1992–1995 and and 2010–
2010–2013
for corresponding
2013 kindskinds
for corresponding of LULC conversions.
of LULC conversions.

As exhibited
As exhibited in in Figure
Figure 3b,
3b, all
all types
types of
of LULC
LULC conversions likely show
conversions likely show increased
increased nighttime
nighttime light
light
radiances on average. In general, the LULC conversions into artificial lands, including
radiances on average. In general, the LULC conversions into artificial lands, including urban, rural urban, rural and
built-up lands (except the unused-to-built-up conversion), show a distinctly higher average
and built-up lands (except the unused-to-built-up conversion), show a distinctly higher average step step change
in DMSPinNTL
change DMSP between
NTL 1992–1995 and 2010–2013
between 1992–1995 than thosethan
and 2010–2013 in other
thosetypes of LULC
in other conversions.
types of LULC
conversions. The most notable step change in DMSP NTL occurred in the LULC conversions with
The most notable step change in DMSP NTL occurred in the LULC conversions into urban lands into
a meanlands
urban of 23.2 ± 12.4
with (standard
a mean deviation,
of 23.2 SD). In particular,
± 12.4 (standard deviation,a relatively large average
SD). In particular, of dNTLlarge
a relatively was
found inofboth
average dNTL thewas
cropland-to-urban
found in both the (24.1 ± 12.4) and unused-to-urban
cropland-to-urban (24.1 ± 12.4) and (27.6 ± 12.4) conversions.
unused-to-urban (27.6 ±
The LULC conversions among non-man-made surfaces—including croplands,
12.4) conversions. The LULC conversions among non-man-made surfaces—including croplands, vegetation, water,
and unusedwater,
vegetation, lands—commonly show a slightly increase
and unused lands—commonly in NTL,
show a slightly whichinisNTL,
increase generally
whichless than 6.0
is generally
(estimated for the cropland-to-water conversion). Overall, these results clearly indicate
less than 6.0 (estimated for the cropland-to-water conversion). Overall, these results clearly indicate measurable
alterations inalterations
measurable nighttimein light data in light
nighttime response
datatointhe human-induced
response LULC changes,
to the human-induced especially
LULC for
changes,
the LULC for
especially conversions
the LULCinto artificial into
conversions lands. In the lands.
artificial following sections,
In the followingwe sections,
mainly paywe attention
mainly pay to
detailed statistical comparisons of dNTL for three major categories of anthropogenic
attention to detailed statistical comparisons of dNTL for three major categories of anthropogenic LULC changes,
including
LULC the conversions
changes, including from croplands, vegetation,
the conversions and rural
from croplands, lands intoand
vegetation, other types
rural of LULC.
lands into other
types of LULC.
3.2. Statistical Comparisons of dNTL for the LULC Conversions from Croplands
3.2. Statistical
A total ofComparisons
~4.5 millionofcropland
dNTL forpixels
the LULC × 104 km2from
(~4.5Conversions Croplands
) have been converted into other types of
LULC (~1.6 × 10 4 km2 ) and 49.3%
A between 1995million
total of ~4.5 and 2010 acrosspixels
cropland the entire
(~ 4.5country, of2)which
× 104 km 35.2%converted
have been into other types of
LULC between 1995 and 2010 across the entire country, of which 35.2% (~1.6 × 104 km2) and 49.3%
(~2.2 × 104 km2) were changed into vegetation and man-made lands, respectively (see Figure 4a). The
Remote Sens. 2018, 10, 1447 6 of 15

(~2.2 × 104 km2 ) were changed into vegetation and man-made lands, respectively (see Figure 4a).
Remote Sens. 2018,
The former 10, x conversion
major FOR PEER REVIEW
from 6 of 15
croplands likely resulted from the reforestation of the cultivated
lands, and the latter was driven by the epic urbanization process since the end of 1990’s in China [39].
former major conversion from croplands likely resulted from the reforestation of the cultivated lands,
Figure 4b displays the density distribution of dNTL for different types of LULC conversions from
and the latter was driven by the epic urbanization process since the end of 1990’s in China [39]. Figure
croplands. Most obviously, the overall response of NTL signals to the LULC conversions from
4b displays the density distribution of dNTL for different types of LULC conversions from croplands.
croplands into three types of artificial lands (median = 24.7 with 1st–3rd quartile = 14.8–33.5 for
Most obviously, the overall response of NTL signals to the LULC conversions from croplands into
urban lands, 14.3 with 4.4–30.8 for rural lands, and 21.4 with 8.0–33.5 for built-up lands) are
three types of artificial lands (median = 24.7 with 1st–3rd quartile = 14.8–33.5 for urban lands, 14.3
generally much higher than those in other types of the LULC conversions. 88.8% (~0.9 × 104 km2 ),
with 4.4–30.8 for4rural2 lands, and 21.4 with 8.0–33.5 for built-up lands) are generally much higher
76.7% (~0.5 × 10 km ), and 86.7% (~0.4 × 104 km2 ) of pixels show statistically significant dNTL
than those in other types of the LULC conversions. 88.8% (~0.9 × 104 km2), 76.7% (~0.5 × 104 km2), and
(P-value < 0.05) in the conversions into urban lands, rural lands, and built-up lands, respectively
86.7% (~0.4 × 104 km2) of pixels show statistically significant dNTL (P-value < 0.05) in the conversions
(Figure 4e,f,g). In contrast, a major proportion of pixels in other types of LULC conversions show
into urban lands, rural lands, and built-up lands, respectively (Figure 4e,f,g). In contrast, a major
no statistically significant increases in NTL signals (i.e., P-value > 0.05 or dNTL ≤ 0, here 84.8%
proportion4 of pixels in other types of LULC conversions show no statistically significant increases in
(~1.4 × 10 km2 ) for vegetation, 57.9% (~0.1 × 104 km2 ) for water and 83.3% for unused lands)
NTL signals (i.e., P-value > 0.05 or dNTL ≤ 0, here 84.8% (~1.4 × 104 km2) for vegetation, 57.9% (~0.1 ×
(see Figure 4c,d,h).
104 km2) for water and 83.3% for unused lands) (see Figure 4c,d,h).

Figure
Figure 4.4. Statistical
Statisticalresults
results of pixel-level
of the the pixel-level step change
step change in nighttime
in nighttime light
light signals forsignals for LULC
LULC conversions
conversions
from croplands from croplands
into into of
other types other types
lands of lands
between between
1995 1995(a)and
and 2010. 2010. (a)
Relative Relative proportions
proportions of different
of different
types of LULCtypes of LULC
pixels pixels from
converted converted from croplands.
croplands. (b) Density(b) Density distributions
distributions of dNTL.
of dNTL. (c–h) (c–h)
Analyzed
Analyzed results oft-test
results of Welch’s Welch’s t-test of
for dNTL forLULC
dNTLchanges
of LULC changes
from from converted
croplands croplands into
converted into (c)
(c) vegetation,
vegetation,
(d) water, (e)(d) water,
urban (e) (f)
lands, urban
rurallands,
lands,(f)
(g)rural lands,
built-up (g)and
lands, built-up lands,lands,
(h) unused and (h) unused lands,
respectively.
respectively.
3.3. Statistical Comparisons of dNTL for the LULC Conversions from Vegetation
3.3. Statistical
BetweenComparisons of dNTL
1990 and 2010, ~4.3 for the LULC
million Conversions
pixels (~4.3 × 10from
4 kmVegetation
2 ) were found to be converted from

natural vegetation
Between into2010,
1990 and croplands (accounting
~4.3 million pixels for 87.9%
(~ 4.3 × 10of
4 kmthe
2) total
were converted
found to bevegetation
convertedpixels,
from
~3.8 × 10 4 2
km ) and artificial surfaces (12.1%, ~0.5 4
× 10 of 2
kmthe) (Figure 5a). Overvegetation
the past two decades,
natural vegetation into croplands (accounting for 87.9% total converted pixels, ~3.8
the LULC conversions from forestlands and grasslands into croplands mainly occurred
× 10 km ) and artificial surfaces (12.1%, ~0.5 × 10 km ) (Figure 5a). Over the past two decades, the
4 2 4 2 in Northeast
and Northwest
LULC conversions China due
from to notablyand
forestlands increased agricultural
grasslands activities
into croplands in these
mainly two regions
occurred [39]. Onand
in Northeast the
whole, no visible
Northwest changes
China due in NTLincreased
to notably signals (0.0 with 0.0–1.6)
agricultural wereinfound
activities these in theregions
two LULC [39].
conversions
On the
whole, no visible changes in NTL signals (0.0 with 0.0–1.6) were found in the LULC conversions from
vegetation into croplands (Figure 5b), in which merely 11.5% of pixels (~0.4 × 104 km2) shows a
statistically significant, but very slight, increase in nighttime light brightness between 1990 and 2010
(Figure 5c). The most marked average change in NTL signals appears to occur in the conversions of
vegetation into urban lands (20.6 with 10.7–29.6), in which most pixels (83.7%, ~0.1 × 104 km2) show a
Remote Sens. 2018, 10, 1447 7 of 15

from vegetation into croplands (Figure 5b), in which merely 11.5% of pixels (~0.4 × 104 km2 ) shows a
statistically significant, but very slight, increase in nighttime light brightness between 1990 and 2010
(Figure 5c). The most marked average change in NTL signals appears to occur in the conversions of
Remote Sens. 2018, 10, x FOR PEER REVIEW 7 of 15
vegetation into urban lands (20.6 with 10.7–29.6), in which most pixels (83.7%, ~0.1 × 104 km2 ) show a
significant
significant positive
positive change
change in
in NTL
NTL signals
signals (Figure
(Figure 5c). Two types
5c). Two types of conversions, from
of conversions, from vegetation
vegetation into
into
both
both rural
rural lands
lands and
and built-up
built-up lands,
lands, show
show closely
closely similar
similar dNTL
dNTL (5.5
(5.5 with 0–19.0 vs.
with 0–19.0 vs. 6.6
6.6 with
with 1.3–19.0,
1.3–19.0,
4 4 2
and
and 52.4%
52.4% vs.vs. 61.4%
61.4% of of pixels
pixelsshowing
showingdNTL
dNTL>>00andandP-value
P-value<<0.05, ~0.05××10
0.05,~0.05 104 and 0.1×× 10
and0.1 104 km
km2,,
respectively)
respectively) (see
(see Figure
Figure 5b,e,f).
5b,e,f).

Figure 5.5.Statistical
Figure Statistical results
results of pixel-level
of the the pixel-level step change
step change in nighttime
in nighttime light
light signals for signals for LULC
LULC conversions
conversions from vegetation into other types of lands between 1995 and 2010. (a) Relative
from vegetation into other types of lands between 1995 and 2010. (a) Relative proportions of different proportions
of different
types of LULCtypes of LULC
pixels pixelsfrom
converted converted from croplands.
croplands. (b) Density(b) Density distributions
distributions of dNTL.
of dNTL. (c–f) (c–f)
Analyzed
Analyzed
results results of
of Welch’s Welch’s
t-test t-test of
for dNTL forLULC
dNTLchanges
of LULC changes
from from vegetation
vegetation converted
converted into into (c)
(c) croplands,
croplands,
(d) (d) urban
urban lands, lands,
(e) rural (e) rural
lands, lands,
and (f) and (f)
built-up built-up
lands, lands, respectively.
respectively.

3.4. Statistical Comparisons

3.4. Statistical Comparisons of
of dNTL
dNTL for
for the
the LULC
LULC Conversions
Conversions from
from Rural
Rural Land
Land
Diverse human activities
Diverse human activitiesand andseveral
severalland-use
land-use policies
policies (e.g.,
(e.g., urbanurban expansion,
expansion, ruralrural
exodusexodus
and
and hollowing, and consolidation of rural lands) can lead to the conversions
hollowing, and consolidation of rural lands) can lead to the conversions of rural lands into other types of rural lands into
other
of LULCtypes of LULC in China
in present-day present-day[41,42].China [41,42].
As shown As shown
in Figure in Figure
6a, 76.4% (~0.176a, 76.4%
× 10 4 km2(~0.17 × 104(~0.02
) and 12.2% km2 )
and
× 10412.2%
km2) of (~0.02 104 pixels
initial×rural km2 ) of (a initial
total ofrural pixels (a~total
~0.2 million, 0.2 × of
104~0.2
km2)million, ~0.2 ×urban
have become 104 km 2 ) have
lands and
become urban lands and croplands, respectively, between 1995 and 2010
croplands, respectively, between 1995 and 2010 with the rapid urbanization process and increasingwith the rapid urbanization
4 2
process and increasing
rural depopulation [41].rural depopulation
Meanwhile, [41]. Meanwhile,
the remaining 5.4% (~0.01 the remaining
× 10 4 km2) and 5.4%
6.0% (~0.01××10
(~0.01 104 km
km2))
4 2
and
of initial rural ×
6.0% (~0.01 10 km
lands have ) ofbeen
initialconverted
rural landsintohave been converted
vegetation into vegetation
and built-up lands, and built-up
respectively.
lands, respectively. Conversions into both urban lands and built-up
Conversions into both urban lands and built-up lands commonly show a notable average dNTL lands commonly show a notable
(21.9
average dNTL (21.9
with 11.8–31.3 with with
and 24.0 11.8–31.3 and 24.0
12.2–34.1, with 12.2–34.1,
respectively) respectively)
(Figure (Figure 6b). Correspondingly,
6b). Correspondingly, the statistically
the statistically
significant significant
positive positive
step change step signals
in NTL change was
in NTL signals
found overwas85.8%found
(~0.1over
× 1085.8% × 104(~0.01
(~0.188.9%
4 km2) and km2 )
and 4
× 10 km 2
× 10488.9%
km2) of(~0.01
converted pixels) ofinconverted
these twopixels
types in
of these
LULCtwo types of LULC
conversions (Figure conversions
6e,f). On the (Figure
whole, 6e,f).
no
On the whole,
remarkable no remarkable
changes changes
in NTL signals in NTL
have beensignals have
observed forbeen observed
the LULC for the LULC
conversions fromconversions
rural lands
from ruralcroplands
into both lands intoand both croplands
natural and natural
vegetation vegetation
with respect with
to the respect to of
distribution thedNTL
distribution
(Figureof6b) dNTL
and
(Figure 6b) and the
the proportions proportions
of converted of converted
pixels pixels with
with a statistically a statistically
marked step change marked step change
in NTL signals in NTL
(Figure
signals
6c,d). (Figure 6c,d).
Remote Sens. 2018, 10, 1447 8 of 15
Remote Sens. 2018, 10, x FOR PEER REVIEW 8 of 15

Figure
Figure 6.6.Statistical
Statisticalresults
results of pixel-level
of the the pixel-level step change
step change in nighttime
in nighttime light
light signals for signals for LULC
LULC conversions
conversions from rural lands into other types of lands between 1995 and 2010. (a) Relative
from rural lands into other types of lands between 1995 and 2010. (a) Relative proportions of different proportions
of different
types of LULCtypes of LULC
pixels pixelsfrom
converted converted from croplands.
croplands. (b) Density(b) Density distributions
distributions of dNTL.
of dNTL. (c–f) (c–f)
Analyzed
Analyzed results of
results of Welch’s Welch’s
t-test t-test of
for dNTL forLULC
dNTLchanges
of LULC changes
from rural from
landsrural lands into
converted converted into (c)
(c) croplands,
croplands,
(d) (d) vegetation,
vegetation, (e) urban
(e) urban lands, lands,
and (f) and (f)
built-up built-up
lands, lands, respectively.
respectively.

3.5. Statistical Comparisons

3.5. Statistical Comparisons of
of dNTL
dNTL among
among Different
Different Types
Types of
of Newly
Newly Emerged
Emerged Lands
Lands
The above
The above results
results jointly
jointly gave
gave the descriptive summaries
the descriptive summaries of of quantitative
quantitative responses
responses of of nighttime
nighttime
lighting brightness to corresponding anthropogenic LULC changes
lighting brightness to corresponding anthropogenic LULC changes across China with a focus across China with a focus onon
three major categories of the LULC conversions from croplands, vegetation,
three major categories of the LULC conversions from croplands, vegetation, and rural lands. Next, and rural lands. Next,
we
we pay
pay attention
attention to to statistical
statistical comparisons
comparisons of of the
the step
step change
change in in nighttime
nighttime lightslights among
among different
different
types of newly emerged lands converted from other types between
types of newly emerged lands converted from other types between 1995 and 2010. First, our results1995 and 2010. First, our results
suggest that most of the pixels converted into artificial lands (87.7, 74.1,
suggest that most of the pixels converted into artificial lands (87.7, 74.1, and 70.6% for urban lands, and 70.6% for urban lands,
4 4 4 2
rural
rural lands,
lands, andand built-up
built-up lands, ~1.3×× 10
lands,~1.3 0.6 ×
104,, 0.6 104,, and
× 10 0.7 ×
and 0.7 104 km
× 10 km2, , respectively)
respectively) likelylikely have
have a a
marked P-value × 4 2
marked step change in NTL signals (dNTL > 0 and P-value < 0.05) (Figure 7a). 42.1% (~0.2 × 104 km
step change in NTL signals (dNTL > 0 and < 0.05) (Figure 7a). 42.1% (~0.2 10 km2))
of the
of the pixel-level
pixel-level conversions
conversions into into water
water showshow significant
significant nightlight responses largely
nightlight responses largely because
because ofof
increased human activities locally or in adjacent settlement/built-up
increased human activities locally or in adjacent settlement/built-up areas. Such effects might also areas. Such effects might also
contribute enhanced nighttime
contribute enhanced nighttime light light radiances
radiances for for few
few proportions
proportions of of the
the LULC
LULC conversions
conversions into into
croplands and
croplands and other
other non-human-dominated
non-human-dominated lands. Second, as
lands. Second, as exhibited
exhibited in in Figure
Figure 7b,
7b, the
the statistically
statistically
significant
significant (P-value
(P-value << 0.05)
0.05) positive
positive step
step change
change in in NTL
NTL signals
signals was
was found
found across
across aa total
total of of ~3.7 million
~3.7 million
4 2 4 2
converted pixels
converted 3.7 × ×
pixels(~(~3.7 10410km2km), in ),which
in which the majority
the majority (69.6%,(69.6%,
~2.6 × 10~2.6 4 km× 10 km
2) were ) were
likely likely
converted
converted into man-made surfaces, including urban lands (34.9%, ~1.3 × 104 km2 ), rural lands
into man-made surfaces, including urban lands (34.9%, ~1.3 × 10 km ), rural lands (16.2%, ~0.6 × 104
4 2

(16.2%,
km2), and ~0.6 × 104 km
built-up
2 ), and built-up lands (18.4%, ~0.7 × 104 km2 ). Third, in spite of the statistical
lands (18.4%, ~0.7 × 104 km2). Third, in spite of the statistical significance, there are
significance,
distinct there in
differences arethe distinct differences
distributions of thein the distributions
pixel-level net increases of the pixel-level
in NTL signals net
among increases in
different
NTL signals among different LULC conversions into man-made
LULC conversions into man-made lands and other types of lands (see Figure 7c). The median is lands and other types of lands
(see Figureto7c).
estimated beThe median
larger than is estimated
22.0 commonly to befor
larger than 22.0
the LULC commonly
conversions forthree
into the LULC
typesconversions
of artificial
into three
lands, whiletypes
the of artificial
median and lands, whilequartile
the third the median and the third
are typically quartile
less than are typically
8.1 and less than for
18.3, respectively, 8.1
and 18.3, respectively, for the LULC conversions into other types of non-man-made
the LULC conversions into other types of non-man-made lands. Moreover, there are visible intercity lands. Moreover,
there are visible
differences in dNTLintercity differences
with respect to LULCin dNTL with respect
conversions to LULC
into urban landsconversions
among 10 major into urban lands
cities (Table
among 10 major cities (Table 1), probably
1), probably because of the variations in the city form. because of the variations in the city form.
Remote Sens. 2018, 10, 1447 9 of 15
Remote Sens. 2018, 10, x FOR PEER REVIEW 9 of 15

Figure 7. (a)
Figure 7. (a) LULC
LULC type-specific
type-specific percentages
percentages of
of pixels
pixels with
with significant positive dNTL
significant positive dNTL (P-value < 0.05)
(P-value < 0.05)
converted from other types between 1995 and 2010. (b) Relative proportions of pixels
converted from other types between 1995 and 2010. (b) Relative proportions of pixels showing showing
significant positive dNTL
significant positive dNTL (P-value
(P-value <
< 0.05)
0.05) and
and converted
converted into
into different
different types
types of LULC between
of LULC between 1995
1995
and 2010. (c)
and 2010. Density distributions
(c) Density distributions of the pixel-level
of the pixel-level positive
positive dNTL
dNTL with
with P-value
P-value <
< 0.05
0.05 for different
for different
types of LULC converted from other types between 1995 1995 and
and 2010.
2010.

Table 1.1.Summary
Table Summaryof the pixel-level
of the step change
pixel-level in nighttime
step change light signals
in nighttime light(dNTL) for(dNTL)
signals LULC conversions
for LULC
into urban lands between 1995 and 2010 across 10 major cities in China.
conversions into urban lands between 1995 and 2010 across 10 major cities in China.
dNTLdNTL
City CityC1 C1 C2 C2 C3 C3
C4
C4 C5 C5 C6 C6
C7 C7
Guangzhou Guangzhou
403.3 403.3806.1 806.1 402.8
402.8 13.71
13.71 5.755.75 21.87 21.87
80.22 80.22
Beijing 1017.0
Beijing 1404.6 1404.6
1017.0 387.6
387.6 21.34
21.34 12.9528.28 28.28
12.95 85.12 85.12
Suzhou 393.9
Suzhou 393.9728.1 728.1 334.2
334.2 32.11
32.11 22.5939.64 39.64
22.59 98.20 98.20
Tianjin 547.0
Tianjin 853.6 853.6
547.0 306.6
306.6 19.92
19.92 11.0731.00 31.00
11.07 83.16 83.16
Chengdu Chengdu
323.4 323.4562.1 562.1 238.7
238.7 32.41
32.41 25.61
25.6137.73 37.73
99.54 99.54
Nanjing Nanjing
316.3 316.3
545.4 545.4 229.2
229.2 31.74
31.74 17.61
17.6137.24 37.24
92.50 92.50
Shanghai Shanghai
763.3 763.3987.3 987.3 224.0
224.0 10.75
10.75 4.074.07 19.40 19.40
46.23 46.23
Zhengzhou Zhengzhou
237.9 237.9433.4 433.4 195.5
195.5 26.92
26.92 16.14
16.1435.38 35.38
85.80 85.80
Hangzhou 206.5 397.0 190.5 26.59 16.24 32.61 89.19
Hangzhou 206.5 397.0 190.5 26.59 16.24 32.61 89.19
Xi’an 207.4 316.3 108.8 10.88 5.14 23.45 73.43
Xi’an 207.4 316.3 108.8 10.88 5.14 23.45 73.43
C1—urban area in 1995 (km2 ); C2—urban area in 2010 (km2 ); C3—increased urban area (km2 ); C4—median of
C1—urban areaquartile;
dNTL; C5—first in 1995C6—third
(km2); C2—urban area in 2010of(km
quartile; C7—proportion
2); C3—increased urban area (km2); C4—
pixels with P-value < 0.05 (%).
median of dNTL; C5—first quartile; C6—third quartile; C7—proportion of pixels with P-value < 0.05 (%).
3.6. Threshold Selections of dNTL for Identifying LULC Conversions into Artificial Lands
3.6. Threshold Selections of dNTL for Identifying LULC Conversions into Artificial Lands
On one hand, it is quite reasonable that more salient responses of nighttime lighting signals
On one hand,
are typically founditinisthe
quite reasonable
LULC that more
conversions salient responses
into artificial of nighttime
lands in which lighting
a relatively signals
large are
average
typically found in the LULC conversions into artificial lands in which a relatively large
increase in NTL and a large proportion of converted pixels with a statistically significant step change average
increase in NTL
in NTL were and afound.
usually large proportion of converted
This is mainly because pixels with a statistically
satellite-derived significant
nocturnal stepdata
luminosity change
are
particularly sensitive to land surfaces with high-density artificial lighting sources at night, generallyare
in NTL were usually found. This is mainly because satellite-derived nocturnal luminosity data in
particularly sensitiveand
human settlements to land surfaces
built-up areas.with
Onhigh-density
the other hand,artificial lighting sourcesmarked
the aforementioned at night, generally
differences
in
in human settlements
quantitative andof
responses built-up
NTL toareas. On the
different other
types hand, changes
of LULC the aforementioned
could suggestmarked differences
that most LULC
in quantitative responses of NTL to different types of LULC changes could suggest that most LULC
conversions into artificial lands might be distinguishable through the observed differences in the
Remote Sens. 2018, 10, 1447 10 of 15

conversions into artificial lands might be distinguishable through the observed differences in the
responses of nighttime lighting signals. Then, a crucial question arises of how to determine the
threshold value of changes in NTL for spatial identification of newly emerged man-made surfaces
between two observational periods among various types of LULC conversions. In practice, the answer
to this question plays a pivotal role in guiding nighttime light data-based investigations of human
settlement dynamics, especially for urban expansions. As represented in the above analysis results,
there are always a few pixels converted into non-man-made lands and having marked responses in
NTL signals in terms of both the statistical significance and the magnitude, primarily because of the
diversity and the spatial heterogeneity of human activities, as well as the over-glow effect in nighttime
light images. By contrast, a small quantity of pixels converted into artificial surfaces have no visible
changes in nighttime light brightness. These findings can jointly state that the nightlight signals-based
partition might not completely distinguish the LULC changes into man-made lands and those into
non-man-made lands from each other.
As in our previous study [16], we used a receiver operating characteristic (ROC) curve to evaluate
the performance of dNTL threshold-based identification rules (a binary partition) in detecting the
newly emerged artificial lands among various types of LULC conversions and towards finding an
optimal threshold. The ROC curve is generally created by the sensitivity (known as the true positive
rate) against the specificity (the true negative rate). Here, the sensitivity measures the proportion of
actual LULC conversions from other types into artificial lands that are correctly identified as such.
Correspondingly, the specificity is defined as the proportion of LULC conversions into non-artificial
lands that are correctly excluded from the partition.
Analyses with two ROC curves were carried out for dNTL threshold-based identifications of
newly added artificial lands (including urban, rural, and built-up lands together) and urban lands
between 1995 and 2010, separately as displayed in Figure 8a,b. The area under the ROC curve (AUC)
provides a quantitative measure of the overall performance of dNTL threshold-based partitions.
For this study, the AUC value (with an expected value of 1) can be explained as the probability that
a randomly selected pixel with the condition (i.e., its dNTL value is larger than a given threshold)
from all converted pixels has the result indicating a greater likelihood of being an actual artificial
land in 2010 than that of a randomly chosen pixel without the condition. Here, the AUC value was
estimated to be 0.87 for all types of newly added artificial lands (Figure 8a) and 0.89 for emerged urban
lands (Figure 8b). More generally speaking, we might have an overall nearly 90 percent chance of
distinguishing newly added artificial lands converted from other types of lands between 1995 and
2010 by using the observed step change in nighttime lighting signals.
As illustrated in Figure 8a, a perfect identification is always located on the top-left point with an
expected value of 1 for both measures, the sensitivity and specificity. In our two cases, the optimal
threshold of dNTL can be obtained by a trade-off between the sensitivity and specificity using the
minimum distance between the point on the ROC curve and the perfect classification point. An optimal
dNTL threshold of 4.8 was estimated for the nationwide identification of the LULC conversions into
three types of artificial lands (see Figure 8a), while ~79.3% of target DMSP pixels (totally covering
~3.0 × 104 km2 ) were included and ~84.1% of non-target pixels (~9.2 × 104 km2 ) were excluded.
The majority (~95.5%) of non-target pixels can be precluded as the dNTL threshold is enhanced
to 12. Meanwhile, ~37.4% of target pixels were also omitted. By contrast, ~87.3% of target pixels
can be identified while the dNTL threshold is deduced to a low value of 2. A large proportion
(~31.1%) of non-target pixels, however, can also be contained. As shown in Figure 8b, the estimated
optimal dNTL threshold for identifying the LULC conversions into urban lands is equal to 7.8,
while ~86.6% of nationwide target pixels (~1.5 × 104 km2 ) were recognized and ~83.8% of non-target
pixels (~10.7 × 104 km2 ) were ruled out. When we increased the dNTL threshold to 12, ~87.8% of
non-target pixels and ~21.6% of target-pixels were ruled out. Although the reduction of the dNTL
threshold can apparently result in a marked increase in the sensitivity, a sizeable proportion of
non-target pixels might also be included in this case. Hence, given the capability of nighttime light data
might indicate the overall effectiveness of nighttime lighting signals for detecting the spatial
expansions of man-made-surfaces during the urbanization process. Moreover, we can find that both
the omitted target pixels and included non-target pixels likely occur in the periphery of local hot
spots of urbanizing areas, where the capability of nighttime lighting signals is probably limited by
both the
Remote Sens.complexity
2018, 10, 1447of human activity and the over-glow effect of DMSP NTL images. Similar 11 results
of 15
can also be found in other regions like the Yangtze River Delta urban agglomeration and the Pearl
River Delta urban agglomeration (see Figure A1). Table A1 gives quantitative comparisons of LULC
for identification
conversions of the LULC
into artificial and conversions
urban lands into man-made
observed lands, a trade-off
by Landsat-derived LULCbetween
maps the
andsensitivity
identified
and specificity in the threshold-based partition is crucially
by optimal DMSP dNTL thresholds across 10 major cities in China. important.

Figure 8.8.Receiver
Figure Receiveroperating
operating characteristic
characteristic (ROC)
(ROC) curve-based
curve-based nationwide
nationwide threshold
threshold selections
selections of dNTL of
dNTL
for for identifying
identifying LULC conversions
LULC conversions intolands
into artificial artificial lands
(a) and (a) lands
urban and urban
(b). (c)lands (b).
Spatial (c) Spatial
distributions
distributions
of newly emergedof newly emerged
artificial artificial 1995
lands between landsand
between 1995Beijing
2010 over and 2010
City.over Beijing
(d) The City. (d)
partitioned The
result
partitioned
based on theresult based in
step change onnighttime
the step lights
change in nighttime
according to thelights
optimalaccording to the optimal
dNTL thresholds dNTL
estimated by
thresholds
the estimated
ROC curves by the ROC curves in (a,b).
in (a,b).

As exemplified by Beijing City, Figure 8c,d illustrate the spatial distributions of the LULC
4. Conclusions
conversions into urban, rural and built-up lands between 1995 and 2010 and the optimal dNTL
The utility of satellite-derived artificial lighting radiances at night for investigations of various
threshold-based partitions, respectively. The visible consistency of two distributions over the region
types of human activity at multiple scales has been widely addressed. Apart from the capability of
might indicate the overall effectiveness of nighttime lighting signals for detecting the spatial expansions
sensors, the features of remotely sensed nocturnal luminosity images are mainly determined by the
of man-made-surfaces during the urbanization process. Moreover, we can find that both the omitted
spatial distributions of diverse artificial lighting sources on the ground. In the absence of explicit
target pixels and included non-target pixels likely occur in the periphery of local hot spots of urbanizing
texture information regarding artificial surfaces, it is often very hard to spatially identify diverse
areas, where the capability of nighttime lighting signals is probably limited by both the complexity
human activities and their dynamics at a fine scale merely using the observed magnitude of nighttime
of human activity and the over-glow effect of DMSP NTL images. Similar results can also be found
light brightness over time and space. Hence, the pixel-level quantitative estimates for both the
in other regions like the Yangtze River Delta urban agglomeration and the Pearl River Delta urban
agglomeration (see Figure A1). Table A1 gives quantitative comparisons of LULC conversions into
artificial and urban lands observed by Landsat-derived LULC maps and identified by optimal DMSP
dNTL thresholds across 10 major cities in China.

4. Conclusions
The utility of satellite-derived artificial lighting radiances at night for investigations of various
types of human activity at multiple scales has been widely addressed. Apart from the capability of
Remote Sens. 2018, 10, 1447 12 of 15

sensors, the features of remotely sensed nocturnal luminosity images are mainly determined by the
spatial distributions of diverse artificial lighting sources on the ground. In the absence of explicit
texture information regarding artificial surfaces, it is often very hard to spatially identify diverse human
activities and their dynamics at a fine scale merely using the observed magnitude of nighttime light
brightness over time and space. Hence, the pixel-level quantitative estimates for both the connections
between nighttime light radiances and land-use/land-cover types and the responses of NTL signals to
anthropogenic LULC changes are fundamentally important for further uses of nighttime light data,
especially in studying the patterns and dynamics of diverse human settlements.
In our previous study [16], we investigated the pixel-level connections of the VIIRS-derived
nighttime light image to the LULC dataset (both in 2015) and revealed the notable differences in the
NTL signals between man-made surfaces and non-artificial lands. In this study, we further studied
the pixel-level quantitative responses of DMSP NTL signals to anthropogenic LULC changes between
1995 and 2010 across China. First, our results represent the high prevalence of distinct changes in
NTL signals responding to the LULC conversions into man-made land surfaces. These responses can
be quantitatively measured by a step change between two observational periods with the statistical
significance test. Second, a minor proportion of the LULC conversions into non-man-made surfaces
can also show marked increases in nighttime lights; the magnitude of changes in NTL on average,
however, is relatively low. Third, we thus are able to identify the LULC conversions into man-made
lands using the step change in nighttime lights with a high chance as estimated by the ROC curve.
Finally, an optimal threshold of step change in nighttime light brightness—for detecting newly added
artificial lands during the observational period—can be obtained by weighting both the sensitivity and
the specificity equally on the ROC curve.
In addition, the relatively coarse spatial resolution and visible over-glow effects in DMSP
nighttime light images are likely to cause a large uncertainty with respect to the spatial identifications
of artificial land dynamics, especially in the peripheral areas of human settlements and built-up lands.
By comparison, the current nighttime light images derived from VIIRS DNB with a higher spatial
resolution, a wider radiometric detection range, and a lower over-glow effect can provide us with a
detailed look at land surfaces over human settlements. Whether the quantitative response of VIIRS
DNB NTL signals to anthropogenic LULC changes can be beneficial for identification of artificial
land dynamics should be investigated in future studies, while the long-term archives of VIIRS DBN
nightlight data (especially the annual composite product) become available. Such investigations are
quite important for the increasing uses of satellite-derived nocturnal luminosity data, particularly in the
context of the uncertainty and mis-specification in nighttime light data-related results and applications.

Author Contributions: T.M. conceived and designed the study and methods; T.M. analyzed the data; T.M. wrote
the paper.
Funding: This study was funded by the National Natural Science Foundation of China (Nos. 41590845,
41771418, 41421001), the Key Research Program of Frontier Science, Chinese Academy of Sciences
(No. QYZDY-SSW-DQC007), the National Science and Technology Key Project (No. 2016YFB0502301), and the
National Key Basic Research Program of China (No. 2015CB954101).
Acknowledgments: The author would like to thank Xinliang Xu for providing the land-use/land-cover datasets.
We would also like to extend our thanks to all the reviewers for their helpful comments and suggestions.
Conflicts of Interest: The author declares no conflict of interest.
Remote Sens. 2018, 10, 1447 13 of 15

Remote Sens. 2018, 10, x FOR PEER REVIEW 13 of 15

Appendix A
Appendix A

Figure A1. Spatial

Figure Spatialdistributions of of
distributions newly emerged
newly emergedartificial lands
artificial between
lands 1995 and
between 19952010
andover
2010(a) the
over
Yangtze
(a) River Delta
the Yangtze urbanurban
River Delta agglomeration and (b)
agglomeration and the
(b)Pearl RiverRiver
the Pearl DeltaDelta
urban agglomeration.
urban The
agglomeration.
partitioned
The result
partitioned based
result on on
based thethe
step change
step changeininnighttime
nighttimelights
lightsaccording
accordingtoto the
the optimal dNTL
dNTL
thresholds over
thresholds over (c)
(c)the
theYangtze
YangtzeRiver
RiverDelta urban
Delta urbanagglomeration
agglomeration and and
(d) the
(d)Pearl River River
the Pearl Delta urban
Delta
urban agglomeration.
agglomeration.

Table A1. Summary of optimal dNTL threshold-based identifications of LULC conversions into
Table A1. Summary of optimal dNTL threshold-based identifications of LULC conversions into
artificial and urban land areas (in km22) between 1995 and 2010 across 10 major cities in China.
artificial and urban land areas (in km ) between 1995 and 2010 across 10 major cities in China.

C1 C1 C2 C2 C3 C3 C4C4 C5
C5 C6C6 C7 C7 C8 C8
Guangzhou
Guangzhou 8152.113398.8
8152.1 13,398.85246.7
5246.7 4486.4
4486.4 403.3
403.3 806.1
806.1 402.8402.8271.9 271.9
BeijingBeijing 21937.4
21,937.4 27,837.25899.8
27837.2 5899.8 5293.5
5293.5 1017.0
1017.0 1404.6
1404.6 387.6387.6329.6 329.6
Suzhou Suzhou 8647.6 8647.617044.4
17,044.48396.8
8396.8 8402.6
8402.6 393.9
393.9 728.1
728.1 334.2334.2326.6 326.6
TianjinTianjin 18159.6
18,159.6 22,979.34819.7
22979.3 4819.7 4653.3
4653.3 547.0
547.0 853.6
853.6 306.6306.6255.6 255.6
Chengdu 11,089.1 14,973.2 3884.1 3851.2 323.4 562.1 238.7 238.1
Chengdu 11089.1 14973.2 3884.1 3851.2 323.4 562.1 238.7 238.1
Nanjing 9394.0 13,910.8 4516.8 4109.5 316.3 545.4 229.2 207.3
Nanjing 9394.0 13910.8 4516.8 4109.5 316.3 545.4 229.2 207.3
Shanghai 13,748.6 21,615.8 7867.2 6890.7 763.3 987.3 224.0 134.8
Shanghai
Zhengzhou 13748.6
8067.021615.8
10,702.57867.2
2635.5 6890.7 763.3
2456.7 237.9 987.3
433.4 224.0
195.5 174.2 134.8
Zhengzhou
Hangzhou8067.0 5025.210702.5
8441.6 2635.5
3416.4 2456.7
3243.6 237.9
206.5 433.4 190.5195.5172.6 174.2
397.0
Hangzhou Xi’an 5025.2 7834.18441.69253.4 3416.4
1419.3 3243.6
1063.7 206.5
207.4 397.0 108.8190.567.8 172.6
316.3
Xi’an 7834.1 9253.4 1419.3 1063.7 207.4 316.3C2—artificial
C1—artificial land area (including urban, rural and built-up lands together) in 1995; 108.8 land area
67.8in
C1—artificial
2010; landartificial
C3—increased area (including urban, rural artificial
land area; C4—increased and built-up lands
land area withtogether) inaccording
dNTL > 4.8 1995; C2—artificial
to Figure 8a;
C5—urban area in 1995; C6—urban area in 2010; C7—increased urban area; C8—increased urban area with
land area in 2010; C3—increased artificial land area; C4—increased artificial land area with dNTL >
dNTL > 7.8 according to Figure 8b.
4.8 according to Figure 8a; C5—urban area in 1995; C6—urban area in 2010; C7—increased urban area;
C8—increased urban area with dNTL > 7.8 according to Figure 8b.
References
References
1. Sutton, P.C. A scale-adjusted measure of “urban sprawl” using nighttime satellite imagery.
1. Sutton,Sens.
Remote P.C.Environ. 2003, 86, 353–369.
A scale-adjusted [CrossRef]
measure of “urban sprawl” using nighttime satellite imagery.
2. Doll, C.; Muller, J.; Morley, J. Mapping regional economic activity from night-time light satellite imagery.
Remote Sens. Environ. 2003, 86, 353–369.
Ecol. Econ. 2006, 57, 75–92. [CrossRef]
2. Doll, C.; Muller, J.; Morley, J. Mapping regional economic activity from night-time light satellite
3. Ghosh, T.; Powell, R.L.; Elvidge, C.D.; Baugh, K.E.; Sutton, P.C.; Anderson, S. Shedding light on the global
imagery. Ecol. Econ. 2006, 57, 75–92.
distribution of economic activity. Open Geogr. J. 2010, 3, 148–161.
3. Ghosh, T.; Powell, R.L.; Elvidge, C.D.; Baugh, K.E.; Sutton, P.C.; Anderson, S. Shedding light on
the global distribution of economic activity. Open Geogr. J. 2010, 3, 148–161.
4. Chen, X.; Nordhaus, W.D. Using luminosity data as a proxy for economic statistics. Proc. Natl.
Acad. Sci. USA 2011, 108, 8589–8594.
Remote Sens. 2018, 10, 1447 14 of 15

4. Chen, X.; Nordhaus, W.D. Using luminosity data as a proxy for economic statistics. Proc. Natl. Acad. Sci. USA
2011, 108, 8589–8594. [CrossRef] [PubMed]
5. Zhang, Q.; Seto, K.C. Mapping urbanization dynamics at regional and global scales using multi-temporal
DMSP/OLS nighttime light data. Remote Sens. Environ. 2011, 115, 2320–2329. [CrossRef]
6. Small, C.; Elvidge, C.D. Mapping decadal change in anthropogenic night light. Procedia Environ. Sci. 2011, 7,
353–358. [CrossRef]
7. Li, X.; Ge, L.; Chen, X. Detecting Zimbabwe’s decadal economic decline using nighttime light imagery.
Remote Sens. 2013, 5, 4551–4570. [CrossRef]
8. Kyba, C.C.M.; Kuester, T.; Sánchez de Miguel, A.; Baugh, K.; Jechow, A.; Hölker, F.; Bennie, J.; Elvidge, C.D.;
Gaston, K.J.; Guanter, L. Artificially lit surface of Earth at night increasing in radiance and extent. Sci. Adv.
2017, 3, e1701528. [CrossRef] [PubMed]
9. Bennett, M.M.; Smith, L.C. Advances in using multitemporal night-time lights satellite imagery to detect,
estimate, and monitor socioeconomic dynamics. Remote Sens. Environ. 2017, 192, 176–197. [CrossRef]
10. Small, C.; Sousa, D.; Yetman, G.; Elvidge, C.; MacManus, K. Decades of urban growth and development on
the Asian megadeltas. Glob. Planet. Chang. 2018, 165, 62–89. [CrossRef]
11. Sutton, P.; Roberts, D.; Elvidge, C.; Baugh, K. Census from heaven: An estimate of the global human
population using night-time satellite imagery. Int. J. Remote Sens. 2001, 22, 3061–3076. [CrossRef]
12. Small, C.; Pozzi, F.; Elvidge, C.D. Spatial analysis of global urban extent from DMSP/OLS night lights.
Remote Sens. Environ. 2005, 96, 277–291. [CrossRef]
13. Elvidge, C.; Baugh, K.; Anderson, S.; Sutton, P.; Ghosh, T. The night light development index (NLDI):
A spatially explicit measure of human development from satellite. Soc. Geogr. 2012, 7, 23–35. [CrossRef]
14. Huang, Q.; Yang, X.; Gao, B.; Yang, Y.; Zhao, Y. Application of DMSP/OLS nighttime light images:
A meta-analysis and a systematic literature review. Remote Sens. 2014, 6, 6844–6866. [CrossRef]
15. Li, X.; Li, D.; Xu, H.; Wu, C. Intercalibration between DMSP/OLS and VIIRS night-time light images to
evaluate city light dynamics of Syria’s major human settlement during Syrian Civil War. Int. J. Remote Sens.
2017, 38, 5934–5951. [CrossRef]
16. Ma, T. An estimate of the pixel-Level connection between Visible Infrared Imaging Radiometer Suite
Day/Night Band (VIIRS DNB) nighttime lights and land features across China. Remote Sens. 2018, 10, 723.
[CrossRef]
17. Elvidge, C.D.; Baugh, K.E.; Kihn, E.A.; Kroehl, H.W.; Davis, E.R. Mapping city lights with nighttime data
from the DMSP Operational Linescan System. Photogramm. Eng. Remote Sens. 1997, 63, 727–734.
18. Hillger, D.; Kopp, T.; Lee, T.; Lindsey, D.; Seaman, C.; Miller, S.; Solbrig, J.; Kidder, S.; Bachmeier, S.; Jasmin, T.
First-light imagery from Suomi NPP VIIRS. Bull. Am. Meteorol. Soc. 2013, 94, 1019–1029. [CrossRef]
19. Elvidge, C.D.; Baugh, K.E.; Zhizhin, M.; Hsu, F.-C. Why VIIRS data are superior to DMSP for mapping
nighttime lights. Proc. Asia-Pac. Adv. Netw. 2013, 35, 62–69. [CrossRef]
20. Elvidge, C.D.; Baugh, K.; Zhizhin, M.; Hsu, F.-C.; Ghosh, T. VIIRS night-time lights. Int. J. Remote Sens. 2017,
38, 5860–5879. [CrossRef]
21. Falchi, F.; Cinzano, P.; Duriscoe, D.; Kyba, C.C.; Elvidge, C.D.; Baugh, K.; Portnov, B.A.; Rybnikova, N.A.;
Furgoni, R. The new world atlas of artificial night sky brightness. Sci. Adv. 2016, 2, e1600377. [CrossRef]
[PubMed]
22. Amaral, S.; Câmara, G.; Monteiro, A.M.V.; Quintanilha, J.A.; Elvidge, C.D. Estimating population and energy
consumption in Brazilian Amazonia using DMSP night-time satellite data. Comput. Environ. Urban Syst.
2005, 29, 179–195. [CrossRef]
23. Li, X.; Xu, H.; Chen, X.; Li, C. Potential of NPP-VIIRS nighttime light imagery for modeling the regional
economy of China. Remote Sens. 2013, 5, 3057–3081. [CrossRef]
24. Shi, K.; Yu, B.; Huang, Y.; Hu, Y.; Yin, B.; Chen, Z.; Chen, L.; Wu, J. Evaluating the ability of NPP-VIIRS
nighttime light data to estimate the gross domestic product and the electric power consumption of China at
multiple scales: A comparison with DMSP-OLS data. Remote Sens. 2014, 6, 1705–1724. [CrossRef]
25. Shi, K.; Huang, C.; Yu, B.; Yin, B.; Huang, Y.; Wu, J. Evaluation of NPP-VIIRS night-time light composite data
for extracting built-up urban areas. Remote Sens. Lett. 2014, 5, 358–366. [CrossRef]
26. Ma, T.; Zhou, Y.; Zhou, C.; Haynie, S.; Pei, T.; Xu, T. Night-time light derived estimation of spatio-temporal
characteristics of urbanization dynamics using DMSP/OLS satellite data. Remote Sens. Environ. 2015, 158,
453–464. [CrossRef]
Remote Sens. 2018, 10, 1447 15 of 15

27. Yu, B.; Shi, K.; Hu, Y.; Huang, C.; Chen, Z.; Wu, J. Poverty evaluation using NPP-VIIRS nighttime light
composite data at the county level in China. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2015, 8, 1217–1229.
[CrossRef]
28. Jing, X.; Shao, X.; Cao, C.; Fu, X.; Yan, L. Comparison between the Suomi-NPP Day-Night Band and
DMSP-OLS for correlating socio-economic variables at the provincial level in China. Remote Sens. 2016, 8, 17.
[CrossRef]
29. Sharma, R.C.; Tateishi, R.; Hara, K.; Gharechelou, S.; Iizuka, K. Global mapping of urban built-up areas of
year 2014 by combining MODIS multispectral data with VIIRS nighttime light data. Int. J. Digit. Earth 2016,
9, 1004–1020. [CrossRef]
30. Levin, N.; Zhang, Q. A global analysis of factors controlling VIIRS nighttime light levels from densely
populated areas. Remote Sens. Environ. 2017, 190, 366–382. [CrossRef]
31. Dou, Y.; Liu, Z.; He, C.; Yue, H. Urban land extraction using VIIRS nighttime light data: An Evaluation of
Three Popular Methods. Remote Sens. 2017, 9, 175. [CrossRef]
32. Zhao, N.; Hsu, F.-C.; Cao, G.; Samson, E.L. Improving accuracy of economic estimations with VIIRS DNB
image products. Int. J. Remote Sens. 2017, 38, 5899–5918. [CrossRef]
33. Zheng, Q.; Deng, J.; Jiang, R.; Wang, K.; Xue, X.; Lin, Y.; Huang, Z.; Shen, Z.; Li, J.; Shahtahmassebi, A.R.
Monitoring and assessing “ghost cities” in Northeast China from the view of nighttime light remote sensing
data. Habitat Int. 2017, 70, 34–42. [CrossRef]
34. Ma, T.; Yin, Z.; Zhou, A. Delineating spatial patterns in human settlements using VIIRS nighttime light data:
A watershed-based partition approach. Remote Sens. 2018, 10, 465. [CrossRef]
35. Lu, D.; Tian, H.; Zhou, G.; Ge, H. Regional mapping of human settlements in southeastern China with
multisensor remotely sensed data. Remote Sens. Environ. 2008, 112, 3668–3679. [CrossRef]
36. Zhang, Q.; Schaaf, C.; Seto, K.C. The vegetation adjusted NTL urban index: A new approach to reduce
saturation and increase variation in nighttime luminosity. Remote Sens. Environ. 2013, 129, 32–41. [CrossRef]
37. Ma, T.; Xu, T.; Huang, L.; Zhou, A. A human settlement composite index (HSCI) derived from nighttime
luminosity associated with imperviousness and vegetation indexes. Remote Sens. 2018, 10, 455. [CrossRef]
38. Liu, J.; Liu, M.; Tian, H.; Zhuang, D.; Zhang, Z.; Zhang, W.; Tang, X.; Deng, X. Spatial and temporal patterns
of China’s cropland during 1990–2000: An analysis based on Landsat TM data. Remote Sens. Environ. 2005,
98, 442–456. [CrossRef]
39. Liu, J.; Kuang, W.; Zhang, Z.; Xu, X.; Qin, Y.; Ning, J.; Zhou, W.; Zhang, S.; Li, R.; Yan, C.; et al. Spatiotemporal
characteristics, patterns, and causes of land-use changes in China since the late 1980s. J. Geogr. Sci. 2014, 24,
195–210. [CrossRef]
40. Elvidge, C.D.; Ziskin, D.; Baugh, K.E.; Tuttle, B.T.; Ghosh, T.; Pack, D.W.; Erwin, E.H.; Zhizhin, M. A fifteen
year record of global natural gas flaring derived from satellite data. Energies 2009, 2, 595–622. [CrossRef]
41. Liu, Y.; Liu, Y.; Chen, Y.; Long, H. The process and driving forces of rural hollowing in China under rapid
urbanization. J. Geogr. Sci. 2010, 20, 876–888. [CrossRef]
42. Long, H. Land consolidation: an indispensable way of spatial restructuring in rural China. J. Geogr. Sci. 2014,
24, 211–225. [CrossRef]

© 2018 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).