ArticlePDF Available

Dynamics of the Urban Lightscape

Authors:

Abstract and Figures

The manifest importance of cities and the advent of novel data about them are stimulating interest in both basic and applied “urban science”. A central task in this emerging field is to document and understand the “pulse of the city” in its diverse manifestations (e.g., in mobility, energy use, communications, economics) both to define the normal state against which anomalies can be judged and to understand how macroscopic city observables emerge from the aggregate behavior of many individuals. Here we quantify the dynamics of an urban lightscape through the novel modality of persistent synoptic observations from an urban vantage point. Established astronomical techniques are applied to visible light images captured at 0.1 Hz to extract and analyze the light curves of 4147 sources in an urban scene over a period of 3 weeks. We find that both residential and commercial sources in our scene exhibit recurring aggregate patterns, while the individual sources decorrelate by an average of one hour after only one night. These highly granular, stand-off observations of aggregate human behavior (which do not require surveys, in situ monitors, or other intrusive methodologies) have a direct relationship to average and dynamic energy usage, lighting technology, and the impacts of light pollution. They may also be used indirectly to address questions in urban operations as well as behavioral and health science. Our methodology can be extended to other remote sensing modalities and, when combined with correlative data, can yield new insights into cities and their inhabitants.
Content may be subject to copyright.
Dynamics of the urban lightscape
Gregory Dobler
a,
n
, Masoud Ghandehari
a
, Steven E. Koonin
a
, Rouzbeh Nazari
a,b
,
Aristides Patrinos
a
, Mohit S. Sharma
a
, Arya Tafvizi
a
, Huy T. Vo
a
,
Jonathan S. Wurtele
c
a
Center for Urban Science and Progress, New York University, New York, NY 11201, United States
b
Department of Civil & Environmental Engineering, Rowan University, Glasboro, NJ 08028, United States
c
Department of Physics, University of California at Berkeley and Lawrence Berkeley Laboratory, Berkeley, CA 94720, United States
article info
Article history:
Received 30 November 2014
Received in revised form
13 May 2015
Accepted 3 June 2015
Keywords:
Urban imaging
Visible light observations
Time series
Image processing
Pattern recognition
abstract
The manifest importance of cities and the advent of novel data abou t them are stimulating
interest in both basic and applied urban science (Bettencourt et al., 2007 [4];
Bettencourt, 2013 [3]). A central task in this emerging field is to document and understand
the pulse of the city in its diverse manifestations (e.g., in mobility, energy use,
communications, economics) both to define the normal state against which anomalies
can be judged and to understand how macroscopic city observables emerge from the
aggregate behavior of many individuals (Louail, 2013 [9]; Ferreira et al., 2013 [6]). Here we
quantify the dynamics of an urban lightscape through the novel modality of persistent
synoptic observations from an urban vantage point. Established astronomical techniques
are applied to visible light images captured at 0.1 Hz to extract and analyze the light
curves of 4147 sources in an urban scene over a period of 3 weeks. We find that both
residential and commercial sources in our scene exhibit recurring aggregate patterns,
while the individual sources decorrelate by an average of one hour after only one night.
These highly granular, stand-off observations of aggregate human behavior wh ich do not
require surveys, in situ monitors, or other intrusive methodologies have a direct
relationship to average and dynamic energy usage, lighting technology, and the impacts
of light pollution. They may also be used indirectly to address questions in urban
operations as well as behavioral and health science. Our methodology can be extended
to other remote sensing modalities and, when combined with correlative data, can yield
new insights into cities and their inhabitants.
& 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
A nighttime image of the Manhattan skyline (see Fig. 1a)
contains thousands of artificial light sources, including
windows, buildings, streets, vehicles, and billboards. While
virtually all of these sources are stationary in position, many
are dynamic in time, changing in color and intensity
throughout the night. The origins and timescales of varia-
bility are diverse, ranging from the 30 Hz flickers of a
display screen to the few-Hz flickering induced by starting
motors to discontinuities as shades are drawn or lights are
turned on and off. Changing atmospheric conditions (for
example the motions of clouds across the sky) also con-
tribute variability. At low resolution, window lights in an
urban night scene are analogous to variable stars on the
night sky and so the techniques of observational astronomy
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/infosys
Information Systems
http://dx.doi.org/10.1016/j.is.2015.06.002
0306-4379/& 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
n
Corresponding author.
E-mail address: greg.dobler@nyu.edu (G. Dobler).
Information Systems ] (]]]]) ]]]]]]
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
can be applied to analyze the urban lightscape. In this
paper, we demonstrate the utility of that approach in
providing a new measure of urban activity and in revealing
the dynamics of individual light sources.
While satellite observations of urban lights have been
used to study city morphology, development, land use
[12], energy consumption [14], and night lights [7], they
are necessarily episodic and cannot probe dynamics on
timescales shorter than a week. Observations from urban
vantage points offer persistent coverage and an unchan-
ging perspective, together with easy and low cost opera-
tions. Such images have been acquired for aesthetic
purposes [13] but have not been analyzed for the scientific
study of cities, with impacts from energy use and effi-
ciency [8] to sleep patterns (which are a significant public
health concern; see for example, [1,11]). in situ measure-
ments with a comparable coverage, while perhaps more
accurate, would be intrusive and would further entail the
cost and operational difficulties of a large-scale sensor
deployment.
In this work, visible light images were acquired from a
rooftop in downtown Brooklyn, the first site of the Urban
Observatory facility created by New York University's
Center for Urban Science þ Progress (CUSP). The northern
view across the East River covers the east side of lower and
midtown Manhattan and offers a diversity of features,
including the tops of the Empire State Building (at a
distance of 6.1 km) and the Chrysler Building, major and
minor building lights, and street and river lights. There are
roughly 20,000 residential and commercial windows in
the scene, and we estimate that some 100,000 people
reside in the 4.4 km
2
covered by our images.
In Section 2 we describe our data acquisition and
analysis pipeline which draws heavily from astronomical
image processing and time series procedures, while in
Section 3 we identify patterns in the light variability and
the implications for aggregate versus individual behavior.
We conclude in Section 4.
2. Methods
The images analyzed in this paper were acquired with a
Point Grey Flea 3 8.8 megapixel camera (equipped with a
25 mm lens) every 10 s between 19:00 and 05:00 h on
each of the 22 nights between October 26 and November
16, 2013. Daylight Saving Time ended during this period on
November 3rd. Upon acquisition, each image ( 25 MB in
three-color raw format) was timestamped, encrypted, and
Fig. 1. The night scene viewed from the CUSP Urban Observatory. (a) This vantage point in downtown Brooklyn faces northward towards Manhattan. A
time-lapse of the scene is available online at https://www.youtube.com/watch?v=D3UKcoh6gig. (b) A typical light curve for one of the sources in the scene.
Vertical green (red) lines indicate on (off) transitions. The off transition corresponding to the largest change in average intensity, as measured before and
after the transition (the t
off
time), is shown in orange. (For interpretation of the references to color in this figure caption, the reader is referred to the web
version of this paper.)
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]]2
stored for analysis. A total of 70,560 images were acquired
(equipment failure terminated observations at 00:00 on
five nights), yielding a total data volume of 4.5 TB.
As we describe in detail below, images were processed
in a two-stage pipeline: data reduction and data analysis.
Our data reduction pipeline to extract light curves for each
source begins by registering each image to a reference
image to account for camera vibration and pointing drift.
With good conditions frame-to-frame offsets were a few
pixels (vertical drifts were of order 10 pixels over the
22 days). Registration failed if the shift exceeded 20 pixels
in either direction and the correlation was not sufficiently
close to unity, which could be caused by poor visibility due
to weather or excessive vibration. Some 4147 rectangular
window apertures were defined manually on a stacked
image, comprising 20% of the total number visible
(40% of the near scene and 20% of the far scene). The
intensity of each source within each frame was calculated
by averaging the intensity of the pixels within the corre-
sponding aperture in three colors (RGB). We refer to the
time history of a source's intensity as its light curve. The
light curves were assembled with the time stamps of the
individual images. Their accuracy is significantly higher
than the 10 s cadence of our observations making that a
negligible source of uncertainty.
The data analysis stage identifies on/off transitions in
each light curve for each night and then from those,
determines t
on=off
, the on/off transition with the largest
change in average intensity as measured before and after.
For each light curve and for each night our algorithm first
convolves the light intensity with a Gaussian filter of
standard deviation 5 min, thereby reducing small-scale
noise. On/off transitions are identified as peaks in the
derivative of the convolved light curve. The situation is
analogous to identifying edges in noisy images, so we use a
variant of the well-known Canny edge detector [5]. This
definition of on/off transitions minimizes effects due to
drift and further cross checks were performed to guard
against spurious on/off detections.
Protecting the privacy of those in our field of view has
been paramount in our image acquisition and analysis and
strict protocols have been observed toward that end. No
more than a few pixels cover the closest sources in our
scene, and atmospheric effects significantly blur the images
further (see Fig. 2), ensuring that no personal detail is ever
captured. In addition, all analyses have been performed at
the aggregate level and any human inspection of an
individual light curve has been done in ignorance of the
precise location of that source within the scene.
2.1. Data reduction: image registration
Wind-driven vibrations make the camera pointing time
dependent, so that each frame must be registered to a
reference image before extracting source intensities. The
reference image was chosen by visual inspection to have
many distinct features and be clear and in sharp focus. This
image registration was performed by correlating a (mean
subtracted and variance normalized) 800 800 pixel sub-
patch in each image with the corresponding patch in the
reference image for various offsets in pixel row and
Fig. 2. A zoom in of two of the closest window sources in our scene. Our
privacy protections which include limiting the maximum number of
pixels per source are enhanced by atmospheric effects which tend to
blur the sources into amorphous blobs of light (see Fig. 1).
Fig. 3. The apertures defining the 4147 light sources in our scene. The apertures designated commercial and residential in our analysis (see Section 3.1)
are shown in blue and orange respectively. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of
this paper.)
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]] 3
Fig. 4. The pulse of city lights curves over 22 nights. The number of transitions per 30 min is plotted for four transition types: all off transitions (light
orange), t
off
s (dark orange), all on transitions (light blue), and t
on
s (dark blue). Pink regions denote missing data due to equipment malfunction and the
solid curves are averages over the final 16 nights (M/T/W/T weekdays and F/S/S weekends). (For interpretation of the references to color in this figure
caption, the reader is referred to the web version of this paper.)
Fig. 5. An illustration of our robustness checks for the on/off transition detection algorithms using two example mock light curves. (a) A clean light curve
with sharp, well-separated transitions. The only noise is the photometric noise of the observations. (b) A light curve with spurious transitions which can
arise due to excessive camera vibrations or rapidly varying cloud conditions. (c) and (d) The derivative of the smoothed intensity for each light curve. In the
absence of noise, peaks in this quantity represent locations of on/off transitions (for on/off spacing greater than the 5 min smoothing scale). Photometric
noise introduces spurious, low-level peaks which we eliminate by performing 10 iterations of 2σ outlier rejection. The final 2σ band is shown in dark
orange. Furthermore, we then only consider peaks which are 10σ outliers from that clipped distribution (i.e., we only consider peaks outside the light
orange band). (e) and (f) The light curve intensity distributions for the 5 minute interval before (red) and after (blue) the transitions circled in green in (c)
and (d) (these time intervals are shown as shaded regions in (a) and (b)). For non-spurious transitions these distributions are well separated, while for
spurious transitions there is significant overlap. We only consider transitions for which the separation is 4 2σ. (For interpretation of the references to color
in this figure caption, the reader is referred to the web version of this paper.)
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]]4
column. Specifically, if I
i
is the image to be registered, I
r
is
the reference image, and we take S
i
and S
r
to be the
subpatch of the image and reference respectively, we find
ð
Δx; ΔyÞ¼ð400x; 400 yÞ where (x,y) are the column
and the row which maximize the 2D convolution matrix
C ¼ S
0
i
n
S
0
r
; ð1Þ
where, for example,
S
0
r
¼
S
r
S
r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
S
2
r
o S
r
4
2
q
ð2Þ
which sets the bounds of C to be ½1; 1.Thus,theappro-
priate registration offset is determined by the pixel shifts that
maximize correlation. With good conditions, frame-to-frame
offsets were a few pixels (vertical drifts were 10 pixels over
the 22 days). However, registration is deemed to fail (and the
frame is not registered) if the shift exceeds 20 pixels in either
direction and the correlation is not sufficiently close to unity .
This might be caused by poor visibility due to weather or
ex cessiv e vibrations. A total of 72,280 frames were success-
fully registered and there were only 79 failures. The final
image size after registration is 4056 2 120 pixels , with a pixel
depth of 8 bits in each of the RGB bands.
2.2. Data reduction: source selection
All registered night time images were stacked and, from
the stacked image, apertures wer e placed around 414 7
windows (of the 20,000 in our scene). Those rectangular
apertures were assigned manually by selecting two diagonally
opposed corners of the sources (upper left and lower right)
which enclosed the light from that source. Aperture definition
using the stacked image rather than a single image av oided
biasing our sample toward lights that are only on at a specific
time. The aperture selection included representation from all
buildings, but, owing to low resolution at large distances as
well as non-optimal viewing angles and the close spacing of
some sources, the selection was sparser at larger distances.
W e estimate that 40% of sources in the near scene and
20% of sources in the far scene were identified for this study ,
which provides a sample size that is more than sufficient to
yield statistically robust (see Section 3). The apertures are
shown in Fig. 3.
Fig. 6. Left: the intensity as a function of time for primarily residential light curves for two nights and for different sortings. Orange dots denote t
off
times of
1706 individual sources. (a) In the top panel light curves from Monday are sorted according to t
off
times. (b) Tuesday light curves are plotted sorting
according to Monday's t
off
times and the pattern in (a) becomes strongly randomized. (c) Tuesday light curves are instead plotted sorting according to
Tuesday t
off
times and a curve similar to that observed on Monday re-emerges. Right: the same but comparing one Monday with the following Monday for
the residential sample. Again, as with the day-to-day comparison, the week-to-week comparison shows that there is a repeatable pattern in aggregate, but
that the individual sources do not exhibit strictly repeating behavior. (For interpretation of the references to color in this figure caption, the reader is
referred to the web version of this paper.)
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]] 5
2.3. Data reduction: source brightness and light curves
Brightnesses for each source were calculated in three
bands by averaging RGB values of the pixels within that
source's aperture. It is important to note that our bright-
nesses are not absolutely calibrated. However, for the
purposes of this paper, relative brightnesses of sources
are sufficient as our results are based exclusively on
feature identification in the light curves, which is inde-
pendent of the absolute scale. The light curves were
reduced to a single intensity IðtÞ¼ðRðtÞþGðtÞþBðtÞÞ=3 (i.
e., the mean of the three bands).
2.4. Data analysis: identifying on/off transitions
The time dependent behavior in our light curves can
take several forms including noise fluctuations, slow zero
point drifts, and sharp on/off transitions. Because the goal
of our work is to assess patterns of activity, we have
developed an algorithm (a modified version of a Canny
edge detector in 1D) for identifying on/off transitions
while suppressing both noise and drifts.
For each light curve I(t) (and for each night), our
algorithm first convolves I(t) with a Gaussian filter of
width 5 min (30 frames)
I
G
¼ G
5
n
I; ð3Þ
thereby reducing small scale noise. Edges (i.e., on/off
transitions) are then identified as peaks in the derivative
of this quantity so that
I
G
t ¼ t
k;on=off

¼
2
I
G
t
2
¼ 0; ð4Þ
where k indexes the transitions in a given light curve (it is
typically less than 5). By concentrating on peaks in the
derivative of I
G
, this definition of t
k;on=off
minimizes effects
due to drift.
We have two tests to protect against spurious transi-
tions due to noise. These are illustrated in Fig. 5. In our first
test, we restrict our transitions in Eq. (4) to only those
peaks in
_
I
G
which are 10
σ
outliers by performing recursive
Fig. 7. An overlay of the recurring t
off
pattern in the residential sample. (a) The t
off
curves for a given Monday with Monday sorting and Tuesday with
Tuesday sorting for our residential sample. The blue points show the significant dispersion away from the pattern when the Monday sorting is used for the
Tuesday points. (b) The t
off
curve overlaid for 8 weekdays in our sample, showing the clearly repetitive nature of the aggregate pattern. (For interpretation
of the references to color in this figure caption, the reader is referred to the web version of this paper.)
Fig. 8. The day-to-day variation in the change of t
off
. The residential
sample is shown in the histogram (binned in 10 min intervals) of the
fraction of sources with a given change Δt
off
. Comparison between 1 day
(orange), 2 days (red), and 3 days (blue) are plotted. The gray line denotes
a correlation histogram generated by pairs of independent draws from
the distribution. The red (gray) shaded regions with boundaries located
at 7 1.0 ( 7 2.0) hours enclose 50% of the data (independent draw)
distribution. (For interpretation of the references to color in this figure
caption, the reader is referred to the web version of this paper.)
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]]6
2
σ
outlier rejection for 10 iterations. Then, we look for 10
σ
0
outliers in
_
I
G
where σ
0
is the standard deviation of the
data points after outlier rejection. For our second test, we
check that, for each transition, the mean of the light curve
for 5 min before the transition is 2
σ
away from the mean
of the light curve for 5 min after the transition where
σ
here is the maximum noise in the light curve during those
two time intervals. This test is particularly effective in poor
weather conditions where very noisy light curves can
induce many spurious edge detections separated closely
in time.
Lastly, we formally define the big off time for each
light curve as
t
off
¼ arg max
k
δðt
k;off
Þ

; ð5Þ
where
δðt
k;off
Þ¼Iðt o t
k;off
ÞIðt Z ðt
k;off
Þ: ð6Þ
Fig. 9. The same as Fig. 8 but with a binning resolution of 1 min. The sources which have t
off
times within a minute from one day to the next (likely lights
on timers) comprise roughly 2.5% of our sample.
Fig. 10. A comparison of the individual t
off
behavior with a independent and identically distributed random draw. Left: the average t
off
distribution from
our residential observational sample (red) and an IID draw of 1100 sources for 8 days (orange) from that parent distribution. Right: for a given source, the
variance in t
off
is calculated for both the observations (light blue) and IID draw (dark blue). (For interpretation of the references to color in this figure
caption, the reader is referred to the web version of this paper.)
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]] 7
That is, t
off
is the time of that transition having the largest
difference between the mean value of the intensity before
and after. We define t
on
analogously.
3. Results
A typical light curve is shown in Fig. 1b. Small high-
freq uency noise fluctuations are interrupt ed by clear disconti-
nuities (transitions) that signal activi ty . While we have not
attempted to analyze the causes of an y particular transition,
we note that large discontinuities could be caused by turning
a light on/off or by opening/closing a shade or drape, while
small discontinuities might be associated with a small light (e.
g., desk lamp) or a translucent curtain or an interior room
light. As described in Section 2, transitions where the intensity
decreases (increases) are referred to as off (on) transitions,
andwedenotebyt
off
thetimeofthatoff each night hav ing
the largest decrease in average intensity before and after (the
orange vertical line in Fig. 1b; similarly the largest on occurs
at t
on
). This transition detection drastically reduces the data
volume from several TB of images t o a few transition times
pernightforeachsourceobserved.Theaggregateofthese
transitions over many nights is one dimension of the pulse of
the city . This is evident in Fig. 4 which shows four approxi-
mately repeating patterns obtained from the on and off
transitions for our complete sample. Given the very different
use cases for commercial versus residential buildings in urban
environme nts and the anticipated difference in weekda ys
versus weekends, residential/commer cial and weekday/week-
end subdivisions of our sample are discussed below .
3.1. Light patterns from residential sources on weekdays
The lower half of our scene contains largely residential
buildings, while those in the upper half are largely
commercial buildings in midtown Manhattan. As we
expect residential and commercial buildings to have dif-
ferent activity patterns, we have isolated a primarily
residential sample of 1706 sources comprising only those
in the lower half of the scene. Those residential light
curves are plotted in grey scale through a Monday -
Tuesday night in Fig. 6a; the sources have been sorted
according to their t
off
times (orange points). The resulting
curve is the cumulative distribution of t
off
, which shows an
inflection point near 23:00 corresponding to a peak rate of
t
off
events. In Fig. 6b the residential light curves for the
following night (Tuesday - Wednesday) are shown in the
same sorting as that of Fig. 6a (i.e., Monday's sorting). If
each source were to turn off at roughly the same time each
night, Fig. 6b would show roughly the same cumulative
distribution; instead, the t
off
times become largely disor-
dered. However, Fig. 6c shows that if Tuesday's light curves
are instead sorted by their own t
off
values, a pattern very
similar to Monday's re-emerges.
It is reasonable to suspect that, while we have shown
that individual light activity does not strictly repeat day-
to-day, there may be a weekly coherence of individual
sources. In the right hand panels of Fig. 6 we show the
behavior of Monday - Monday þ 1 week. As with the
day-to-day comparison, the aggregate pattern recurs from
one Monday to the next, but individual behavior is
strongly decorrelated. In fact (as shown in Fig. 7) the
macro pattern persists for all full weekday nights in our
sample despite the more random micro behavior. In other
words, while individual sources do not habitually turn off
at the same time each night, the aggregate activity is quite
repeatable.
Some insight into the single-source behavior under-
lying these macroscopic regularities can be obtained by
asking By how much does t
off
vary for individual sources
Fig. 11. The same as Fig. 8 but for the commercial sample of lights on weekdays. As in the residential sample, there is a repeated pattern night-to-night and
week-to-week with more random individual activity. The most notable difference between the residential and commercial samples is the lack of inflection
in the t
off
curves, indicating a more random distribution of t
off
prior to midnight, by which time most commercial sources have experienced their t
off
.
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]]8
between nights separated by one, two, or three days?
Fig. 8 shows a histogram of such variations over all of the
residential sources. The overlap of the distributions for
separations of 1, 2, and 3 days indicates randomization
after only 1 day. Nevertheless, a comparison of these
distributions with a random draw from the parent popula-
tion of t
off
s (gray line) shows stronger night-to-night
correlation than random. However, a full 50% of our
residential sample has j
Δt
off
j4 1:0 h, again indicating that
a substantial fraction of our sample do not have regular t
off
times from one night to the next. We also note that 2.5%
of our residential sample repeats nightly with a variation
of less than 1 min as shown in Fig. 9, suggesting the
presence of timers.
Further evidence of non-trivial single-source dynamics
can be seen by comparing (for each source) the variance in
t
off
among the 8 weeknights in our sample for which we
had a full 10 h of observations (see Fig. 4) with that of a
random draw of octuples from the parent t
off
distribution.
Fig. 10 shows the probability distribution function for t
off
averaged over 8 weekdays of our residential sample and an
independent and identically distributed (IID) random
variable draw of 1100 samples of 8 points (i.e., 8 days)
from that parent distribution. The distribution of variances
in the t
off
s over the 8 nights for the residential sources in
our observational sample is markedly different than for
the IID draw. The broader distribution in the data indicates
the difference in source dynamics compared to a random
draw from the aggregate parent population. Specifically,
there are some sources which have more repeatable t
off
s
(within about 1 h) while others have much more random
behavior than what would be expected from a purely
random draw of t
off
s from the average weekday
distribution.
Fig. 12. Top (bottom): the same as Fig. 7 but for the residential (commercial) sample of lights on weekends.
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]] 9
3.2. Light patterns from commercial sources on weekdays
The overlay of the aggregate t
off
pattern for the pri-
marily commercial scene is plotted in Fig. 11. It also
exhibits repeating, albeit distinct, aggregate behaviors.
However, before about 1:0 0 am we do not observe the
significant curvature seen in the residential sample, indi-
cating that there is not necessarily a characteristic off
time for commercial lights, and their transition activity
before midnight is more random. There are however
multiple lights (including lights on timers, whole floors,
splash from spotlights, etc.) that turn off on the hour in the
commercial sample.
3.3. Light patterns on weekends
An overlay of the recurring t
off
pattern on weekends for
both residential and commercial activity is plotted in
Fig. 12. Each exhibits a pattern distinct from that seen on
weekdays. Again, as for weekdays the aggregate pattern
emerges, but there is more scatter on the weekends when
comparing successive days (note, the flat behavior on one
Saturday between about 8:30 pm and 9:15 pm is due to
missing data). Interestingly, the Saturday with Friday
sorting points are less tightly clustered around the Friday
with Friday ordering curve than the Monday - Tuesday
comparison shown in Fig. 7, indicating that the weekend
behavior is in fact more random than the weekday
behavior.
3.4. Robustness to sub-sampling: residential and commercial
To test the robustness of our results (namely repeatable
aggregate behavior with more random individual beha-
vior) we perform subsamplings of our data over various
dimensions in the analysis. First, the t
off
curve for various
random subsamplings of the residential sample on a
Monday night is shown in Fig. 13. In each case, we
randomly subsample some fraction of the total number
of residential light sources and reorder according to t
off
for
that subsample. The fact that the shape of the t
off
curve is
robust to this subsampling (even down to 12% of the
sample) indicates that our parent sample is sufficiently
large. Fig. 14 shows the t
off
pattern for only large ampli-
tude transitions. Specifically, the sample in panel a (b)
represents sources which the brightness after transition is
r 70%ð50%Þ of the brightness before the transitions. Thus,
our main results (repeatable pattern in aggregate but less
repeatable on an individual basis) are robust to restricting
ourselves to only the largest t
off
transitions. Lastly, in
Fig. 13. The t
off
pattern for random subsamplings of the data.
Fig. 14. The same as Fig. 7a but for subsamples of t
off
that are restricted to different amplitude thresholds.
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]]10
Fig. 15. The same as Fig. 7a but for different spatial subdivisions of sources in our scene: ((a) and (b), left and right half of the whole sample; (c) and (d), left
and right half of our residential sample; (e) and (f), different definitions of the dividing line of residential vs. commercial. In each case, our results hold for
different spatial subsamplings.
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]] 11
Fig. 15 we show that various spatial subsamplings also
yield similar macro/micro behavior in the t
off
distributions.
4. Conclusions
We have described a fundamentally new methodology
for quantifying the pulse of a city via time-dependent
changes in the urban lightscape. Our use of astronomical
techniques to analyze the dynamics of city lights applies
physics to the new realm of urban science and human
behavior, complementing the use of statistical physics for
scaling and modeling of urban growth [10,2]. The latter
studies use existing data sources and examine slow
phenomena (e.g., urban sprawl) while our work focuses
on rapidly varying urban signals.
Furthermore, we have demonstrated that an
astronomy-inspired analysis of persistent synoptic ima-
gery of an urban scene leads to insights into aggregate
light activity. In particular, we have discovered a repeata-
ble pattern of behavior characterized by the aggregate on/
off transition times of lights in the scene which recurs
night-to-night and week-to-week. Despite the strong
repeatability of that pattern, the individual sources which
comprise it do not strictly repeat night-to-night or week-
to-week. Specifically, a full 50% of our sample decorrelates
by more than an hour after only one night. This decorrela-
tion does not significantly increase when comparing
nights separated by 2, 3, or 7 days. The same qualitative
behavior is seen for both residential and commercial light
sources and for both weekdays and weekends.
Interestingly, however, the individual source dynamics
of our urban lights sample is more complex than a simple
random draw from the aggregate parent distribution of off
transitions which repeats night-to-night. We find that the
distribution of nightly correlations in our sample of resi-
dential sources is broader than that expected from an
independent and identically distributed draw from the
repeating pattern.
Our results represent the regularity and variability of one
component of the pulse of the city. A much more random, but
temporally correlated, individual behavior has been found t o
underlie that regularity. The method we have described in this
paper re pres ents a firs t ste p in a new appr oach to studying
urban functioning. Correlations of the kind of data we have
present ed with such v ariables as demographics , energy use,
season, income, and met eorological conditions are of clear
interest and will be the focus of our future work. Disturbances
in these patterns by light pollution and noise pollution will
give insights into the public health affects on the circadian
rh ythms of city dwellers while building level aggregation will
inform measurement s of occupancy ( a strong correlate for
energy consumption). Further, beyond basic urban science,
applications include emergency response, envir onmental
monitoring, and urban operations, and the CUSP Urban
Observat ory will push this methodology to other observa-
tional modalities including broadband infrared and hyper-
spectral observations.
Acknowledgments
This work was partially supported by a grant from the
Alfred P. Sloan Foundation grant # 2013-10-37.
References
[1] A. Abbott, Restless nights, listless days, Nature 425 (2003) 896.
[2] C. Andersson, K. Lindgren, S. Rasmussen, R. White, Urban growth
simulation from first principles, Phys. Rev. E 66 (August (2)) (2002)
026204, http://dx.doi.org/10.1103/PhysRevE.66.026204.
[3] L. Bettencourt, The origin of scaling in cities, Science 340 (2013)
1438.
[4] L.M.A. Bettencourt, J. Lobo, D. Helbing, C. Kühnert, G.B. West.
Growth, innovation, scaling, and the pace of life in cities. Proc. Natl.
Acad. Sci. 104 (17) (2007), 73017306, http://www.pnas.org/con
tent/104/17/7301.abstract.
[5] J. Canny. A computational approach to edge detection. IEEE Trans.
Pattern Anal. Mach. Intell. (1986), http://dx.doi.org/10.1109/TPAMI.
1986.4767851.
[6] N. Ferreira, J. Poco, H.T. Vo, J. Freire, C.T. Silva, Visual exploration of
big spatio-temporal urban data: a study of New York city taxi trips,
IEEE Trans. Vis. Comput. Graph. 19 (12) (2013) 214921 58.
[7] J. Hale, G. Davies, A.J. Fairbrass, T.J. Matthews, C.D.F. Rogers, et al.,
Mapping lightscapes: spatial patterning of artificial lighting in an
urban landscape, PLoS ONE 8 (5) (2013) e61460.
[8] C.E. Kontokosta. Local Law 84 Energy Benchmarking Data. Report to
the New York City Mayor's Office of Long-Term Planning and
Sustainability, 2012.
[9] T. Louail. From Mobile Phone Data to the Spatial Structure of Cities,
arXiv:1401.4540,2013.
[10] H.A. Makse, J. Andrade, M. Batty, S. Havlin, H.E. Stanley, Modeling
urban growth patterns with correlated percolation, Phys. Rev. E 58
(December (6)) (1998) 70547062, http://dx.doi.org/10.1103/
PhysRevE.58.7054.
[11] T. Roenneberg, K.V. Allebrandt, M. Merrow, C. Vetter, Social jetlag
and obesity, Curr. Biol. 22 (10) (2012) 939943. URL http://www.
sciencedirect.com/science/article/pii/S0960982212003259.
[12] P. Sutton, A scale-adjusted measure of urban sprawl using night-
time satellite imagery, Remote Sens. Environ. 86 (2003) 353369.
[13] A. Warhol. Empire, 1964.
[14] J. Wu, Z. Wang, W. Li, J. Peng, Exploring factors affecting the
relationship between light consumption and GDP based on dmsp/
ols nighttime satellite imagery, Remote Sens. Environ. 134 (0) (2013)
111119. URL http://www.sciencedirect.com/science/article/pii/
S0034425713000734.
Please cite this article as: G. Dobler, et al., Dynamics of the urban lightscape, Information Systems (2015), http://dx.doi.
org/10.1016/j.is.2015.06.002i
G. Dobler et al. / Information Systems ] (]]]]) ]]]]]]12
... Complementing these studies, in this paper, we follow the operational mode of the "Urban Observatory" (UO) facility [3] by deploying a wide-area IR camera mounted in a tall building at stand-off distance from a target building and acquiring imaging at sub-minute temporal resolution over roughly two months. Analogous to the studies on patterns of lighting variability in [4], we extract time-dependent brightnesses of the "sources" shown in Figure 1 and use characteristics of their variability to quantify patterns of operations and usage. This methodology is complimentary to conventional methods of estimates of air-conditioner end use via surveys or dense sensor deployments in that it is empirically measurement based with the potential to scale up to many buildings with a single sensor. ...
... Our goal in this work is to demonstrate the ability to measure timedependent IR brightness from "sources" in the scene that vary over time, similarly to the lightscape analysis from [4], but the sources in our case being the exterior vents of the air-conditioner (AC) units. Source detection in thermal images is a challenging task due to several factors including resolution, noise floor, and weather, and so in this proof-of-concept work, we select sources via handselection (the use of unsupervised and supervised machine learning methods for source detection will be the subject of future work; see Section 4). ...
... Finally, the aggregate behavioral properties of AC use can yield important information for characterizing energy end-use, particularly if AC IR brightness correlates with energy consumption for a given unit. Relatedly, while the spatial resolution of our IR imaging is certainly sufficiently low to preclude the possibility of collecting personally identifiable information (see also [3,4]), the patterns of AC use may be characteristic of an individual user and so while we have presented completely anonymized and aggregated results here, it is an open question at what level of disaggregation behavioral patterns become unique to a certain AC unit. Each of these use cases and considerations will be the focus of future work. ...
... The Hyperspectral Imaging (HSI) data used in this work were obtained by the "Urban Observatory" (UO) facility in New York City (NYC) [71][72][73][74]. The UO has deployed broadband visible and infrared imaging cameras, as well as Visible and Near-Infrared (VNIR) and Long Wave Infrared (LWIR) HSI cameras that continuously image landscapes in NYC with high persistence and temporal resolution on the order of minutes. ...
... Using a Visible and Near-Infrared (VNIR) single slit, scanning spectrograph, deployed by the Urban Observatory (UO) [71][72][73][74] in New York City, that captures 848 spectral channels in the 0.4-1.0 µm wavelength range, we obtained side-facing scans of an urban scene in 15 min intervals between 08h00 and 18h00 for 30 days between 3 May and 6 June 2016. ...
Article
Full-text available
We examine the impact of changes in ozone (O3), particulate matter (PM2.5), temperature, and humidity on the health of vegetation in dense urban environments, using a very high-resolution, ground-based Visible and Near-Infrared (VNIR, 0.4–1.0 μm with a spectral resolution of 0.75 nm) hyperspectral camera deployed by the Urban Observatory (UO) in New York City. Images were captured at 15 min intervals from 08h00 to 18h00 for 30 days between 3 May and 6 June 2016 with each image containing a mix of dense built structures, sky, and vegetation. Vegetation pixels were identified using unsupervised k-means clustering of the pixel spectra and the time dependence of the reflection spectrum of a patch of vegetation at roughly 1 km from the sensor that was measured across the study period. To avoid illumination and atmospheric variability, we introduce a method that measures the ratio of vegetation pixel spectra to the spectrum of a nearby building surface at each time step relative to that ratio at a fixed time. This “Compound Ratio” exploits the (assumed) static nature of the building reflectance to isolate the variability of vegetation reflectance. Two approaches are used to quantify the health of vegetation at each time step: (a) a solar-induced fluorescence indicator (SIFi) calculated as the simple ratio of the amplitude of the Compound Ratio at 0.75 μm and 0.9 μm, and (b) Principal Component Analysis (PCA) decomposition designed to capture more global spectral features. The time dependence of these vegetation health indicators is compared to that of O3, PM2.5, temperature, and humidity values from a distributed and publicly available in situ air quality sensor network. Assuming a linear relationship between vegetation health indicators and air quality indicators, we find that changes in both SIF indicator values and PC amplitudes show a strong correlation (r2 value of 40% and 47%, respectively) with changes in air quality, especially in comparison with nearby buildings used as controls (r2 value of 1% and 4%, respectively, and with all molecular correlations consistent with zero to within 3σ uncertainty). Using the SIF indicator, O3 and temperature exhibit a positive correlation with changes in photosynthetic rate in vegetation, while PM2.5 and humidity exhibit a negative correlation. We estimate full covariant uncertainties on the coefficients using a Markov Chain Monte Carlo (MCMC) approach and demonstrate that these correlations remain statistically significant even when controlling for the effects of diurnal sun-sensor geometry and temperature variability. This work highlights the importance of quantifying the effects of various air quality parameters on vegetation health in urban environments in order to uncover the complexity, covariance, and interdependence of the numerous factors involved.
... In addition, while the physics of light propagation in the atmosphere is identical during day and night, the light sources are not. Instead of the (comparatively) simple angular distribution of reflected sunlight, the hundreds of millions to billions (Zissis et al., 2021;Zissis & Kitsinelis, 2009) of artificial lights of Earth that emit some or all of their light outdoors have unique angular emission distribution functions, each of which vary over time (Dobler et al., 2015;Meier, 2018;Li et al., 2020). While this complication presents a challenge for working with night lights, it also provides an opportunity: night lights imagery acquired at multiple angles contains information that could potentially be extracted via remote sensing. ...
Article
Full-text available
The spatial and angular emission patterns of artificial and natural light emitted, scattered, and reflected from the Earth at night are far more complex than those for scattered and reflected solar radiation during daytime. In this commentary, we use examples to show that there is additional information contained in the angular distribution of emitted light. We argue that this information could be used to improve existing remote sensing retrievals based on night lights, and in some cases could make entirely new remote sensing analyses possible. This work will be challenging, so we hope this article will encourage researchers and funding agencies to pursue further study of how multi-angle views can be analyzed or acquired.
... The road network data from OpenStreetMap (http://www.openstreetmap.org (accessed on 12 June 2021)) was selected to reflect the road network density, since streetlights and traffic lights are the main sources of NTL in urban areas [51,52]. Figure 2c illustrates the road network adopted by Chongqing, including motorway, trunk road, primary road, secondary road, and tertiary road, covering most areas of the city. ...
Article
Full-text available
Nighttime light (NTL) intensity is highly associated with the unique footprint of human activities, reflecting the development of socioeconomic and urbanization. Therefore, better understanding of the relationship between NTL intensity and human activities can help extend the applications of NTL remote sensing data. Different from the global effect of human activities on NTL intensity discussed in previous studies, we focused more attention to the local effect caused by the spatial heterogeneity of human activities with the support of the multiscale geographically weighted regression (MGWR) model in this study. In particular, the Suomi National Polar Orbiting Partnership/Visible Infrared Imaging Radiometer Suite (NPP/VIIRS) NTL data within Chongqing, China were taken as example, and the point of interest (POI) data and road network data were adopted to characterize the intensity of human activity type. Our results show that there is significant spatial variation in the effect of human activities to the NTL intensity, since the accuracy of fitted MGWR (adj.R2: 0.86 and 0.87 in 2018 and 2020, respectively; AICc: 4844.63 and 4623.27 in 2018 and 2020, respectively) is better than that of both the traditional ordinary least squares (OLS) model and the geographically weighted regression (GWR) model. Moreover, we found that almost all human activity features show strong spatial heterogeneity and their contribution to NTL intensity varies widely across different regions. For instance, the contribution of road network density is more homogeneous, while residential areas have an obviously heterogeneous distribution which is associated with house vacancy. In addition, the contributions of the commercial event and business also have a significant spatial heterogeneity distribution, but show a distinct decrement when facing the COVID-19 pandemic. Our study successfully explores the relationship between NTL intensity and human activity features considering the spatial heterogeneity, which aims to provide further insights into the future applications of NTL data.
... In addition, while the physics of light propagation in the atmosphere is identical during day and night, the light sources are not. Instead of the (comparatively) simple angular distribution of reflected sunlight, the hundreds of millions to billions (Zissis et al., 2021;Zissis & Kitsinelis, 2009) of artificial lights of Earth that emit some or all of their light outdoors have unique angular emission distribution functions, each of which vary over time (Dobler et al., 2015;Meier, 2018;Li et al., 2020). While this complication presents a challenge for working with night lights, it also provides an opportunity: night lights imagery acquired at multiple angles contains information that could potentially be extracted via remote sensing. ...
Article
Full-text available
The spatial and angular emission patterns of artificial and natural light emitted, scattered, and reflected from the Earth at night are far more complex than those for scattered and reflected solar radiation during daytime. In this commentary, we use examples to show that there is additional information contained in the angular distribution of emitted light. We argue that this information could be used to improve existing remote sensing retrievals based on night lights, and in some cases could make entirely new remote sensing analyses possible. This work will be challenging, so we hope this article will encourage researchers and funding agencies to pursue further study of how multi-angle views can be analyzed or acquired.
... The different overpass time may also increase the uncertainty of NTL intensity. As city lights change over time (Dobler et al. 2015), the night-time lights recorded by two satellites may be different due to the 50 min time interval. Besides, the NTL intensity is also influenced by aerosol and cloud conditions (Wang, Mu, and Yan 2020). ...
Article
Night-Time light (NTL) data have been widely used for monitoring the dynamics of human activities and socioeconomics. As a new-generation satellite for acquiring NTL data, the National Oceanic and Atmospheric Administration-20 (NOAA-20) was successfully launched in November 2017. To support broad-scale environmental applications, it is necessary to generate monthly NTL composites of NOAA-20. Taking China as the study area, we produced NOAA-20 monthly NTL composites from April 2019 to December 2019. First, we performed a series of de-noising steps to eliminate NOAA-20 NTL pixels affected by sunlight, moonlight, high scan angles, clouds and further eliminated NTL outliers in time series by using a modified z-score method. Then, we aggregated daily NOAA-20 NTL data to produce monthly NTL composites to improve data coverage and stability. Subsequently, we examined the consistency of monthly composites between NOAA-20 and S-NPP at multiple scales, including provincial, prefectural, and pixel levels. Our results show that the monthly NTL composites of NOAA-20 are in good agreement with that of S-NPP with an R² ranging from 0.81 to 0.99. Besides, we found that the spatial variation trends of the NOAA-20 monthly NTL composites are similar to that of S-NPP monthly NTL composites. We believe that the monthly NTL composites of NOAA-20 are very close to that of S-NPP and will open up more opportunities for relevant NTL studies.
Article
We present a comparative analysis of urban magnetic fields between two American cities: Berkeley (California) and Brooklyn Borough of New York City (New York). Our analysis uses data taken over a four-week period during which magnetic field data were continuously recorded using a fluxgate magnetometer with 70 pT/[Formula: see text] noise. We identified significant differences in the magnetic signatures. In particular, we noticed that Berkeley reaches a near-zero magnetic field activity at night, whereas magnetic activity in Brooklyn continues during nighttime. We also present auxiliary measurements acquired using magnetoresistive vector magnetometers (VMRs), with the noise of 300 pT/[Formula: see text], and demonstrate how cross correlation, and frequency-domain analysis, combined with data filtering can be used to extract urban-magnetometry signals and study local anthropogenic activities. Finally, we discuss the potential of using magnetometer networks to characterize the global magnetic field of cities and give directions for future development.
Article
Indoor light pollution threatens human health. This study examined how built environment characteristics affect nighttime light levels in the bedrooms of urban areas. Indoor illuminances were monitored in the bedrooms of 28 participants in the city of Daejeon, Korea, between 2 a.m. and 4:30 a.m. over five days. Five built environment variables surrounding the bedrooms under study were assessed. Results illustrate that residential floor area on floors of influence (±3 floors from the bedroom under study) increased indoor illuminance, but residential floor area on all floors, including those vertically distant from the bedroom, did not. Tree canopy percentage reduced indoor illuminance, suggesting that tree canopies block artificial light at night emitted at ground level. Non-residential floor area and length of road lanes did not have significant effects, further suggesting that tree canopies protect urban residents on upper floors from light directed upwards from streets and commercial facilities at ground level. These findings indicate that vertical separation of residential areas is an effective measure to mitigate light pollution, especially in high-density urban areas where horizontal separation is not feasible. These findings also imply that policy efforts should focus on reducing light trespass from neighbors at similar heights.
Article
As one of the new and crucial fields of economic development, the digital economy refers to the economic activities associated with digital technologies and information. The traditional measurement mainly relies on the statistical data, which are regionally specific and labour cost. As an appropriate proxy for socio-economic activities, nighttime light (NTL) remote sensing data was used in this study to explore its potential in estimating the digital economy. Then, the Zipf's law was used to evaluate the growth of the digital economy at the city level. The results show that the total NTL intensity has a logarithmic relationship with the digital economy index and can be estimated for each cities' digital economy growth in China from 2017 to 2020 (R² ≈ 0.7). An unbalanced distribution and a decentralized polycentric structure of digital economy are found among all cities in China. But the top 100 cities have a relative harmonious development with a better goodness of Zipf's law. Finally, four main incentives behind the digital economy growth were concluded for three stages of the digital economy growth. This study could enrich the understanding of digital economy and have valuable implications for its future growth in China.
Article
Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS) and Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (SNPP-VIIRS) data are valuable records of nighttime lights in analyzing socioeconomic development. However, inconsistencies between these data have severely restricted long time-series analyses. Published time-series nighttime light (NTL) data sets are not widely available or accurate because the DMSP-OLS calibration is inadequate and some missing data in the SNPP-VIIRS data are seldom considered for patching. To address these issues, we calibrated DMSP-OLS data (1992–2013) by using a quadratic model based on a “pseudo-invariant pixel” method. Thereafter, an exponential smoothing model was used to predict and patch missing data in the monthly SNPP-VIIRS data (2013-2019). Outliers and noise were also removed from the annual data. In addition, a sigmoid model was employed to generate improved simulated DMSP-OLS (SDMSP-OLS) data (2013–2019), which were appended with the calibrated DMSP-OLS data (1992–2013) to develop improved DMSP-OLS-like data (1992–2019) in China. Lastly, we qualitatively and quantitatively compared these data with published NTL data to examine data availability. Results showed that choosing invariant pixels to calibrate DMSP-OLS data can minimize discontinuity. The correlation between the SNPP-VIIRS data synthesized by the patched monthly SNPP-VIIRS data and the official annual SNPP-VIIRS data in 2015 (R2 = 0.931) and 2016 (R2 = 0.930) was higher than those of the two existing correction methods with R2 values below 0.90. Spatial patterns of pixels in the improved SDMSP-OLS data in 2013 were more similar with the DMSP-OLS data than those in the published data. Strong correlations likewise existed between the total (average) pixel values of the improved SDMSP-OLS data (2013-2019) and the DMSP-OLS data in 2012. We also found that the improved DMSP-OLS-like data held strong linear correlations with the different statistics, the average R2 values of which were 0.931 and 0.654 at the national and provincial levels, respectively. Meanwhile, the average regression R2 values between the two published data sets and statistics were 0.858/0.506 and 0.911/0.611, respectively. Our study has proven that the improved DMSP-OLS-like data (1992–2019) have immense potential to effectively evaluate socioeconomic development and anthropic activities.
Article
Full-text available
Despite the increasing importance of cities in human societies, our ability to understand them scientifically and manage them in practice has remained limited. The greatest difficulties to any scientific approach to cities have resulted from their many interdependent facets, as social, economic, infrastructural, and spatial complex systems that exist in similar but changing forms over a huge range of scales. Here, I show how all cities may evolve according to a small set of basic principles that operate locally. A theoretical framework was developed to predict the average social, spatial, and infrastructural properties of cities as a set of scaling relations that apply to all urban systems. Confirmation of these predictions was observed for thousands of cities worldwide, from many urban systems at different levels of development. Measures of urban efficiency, capturing the balance between socioeconomic outputs and infrastructural costs, were shown to be independent of city size and might be a useful means to evaluate urban planning strategies.
Article
Full-text available
Artificial lighting is strongly associated with urbanisation and is increasing in its extent, brightness and spectral range. Changes in urban lighting have both positive and negative effects on city performance, yet little is known about how its character and magnitude vary across the urban landscape. A major barrier to related research, planning and governance has been the lack of lighting data at the city extent, particularly at a fine spatial resolution. Our aims were therefore to capture such data using aerial night photography and to undertake a case study of urban lighting. We present the finest scale multi-spectral lighting dataset available for an entire city and explore how lighting metrics vary with built density and land-use. We found positive relationships between artificial lighting indicators and built density at coarse spatial scales, whilst at a local level lighting varied with land-use. Manufacturing and housing are the primary land-use zones responsible for the city's brightly lit areas, yet manufacturing sites are relatively rare within the city. Our data suggests that efforts to address light pollution should broaden their focus from residential street lighting to include security lighting within manufacturing areas.
Article
Full-text available
This paper describes a computational approach to edge detection. The success of the approach depends on the definition of a comprehensive set of goals for the computation of edge points. These goals must be precise enough to delimit the desired behavior of the detector while making minimal assumptions about the form of the solution. We define detection and localization criteria for a class of edges, and present mathematical forms for these criteria as functionals on the operator impulse response. A third criterion is then added to ensure that the detector has only one response to a single edge. We use the criteria in numerical optimization to derive detectors for several common image features, including step edges. On specializing the analysis to step edges, we find that there is a natural uncertainty principle between detection and localization performance, which are the two main goals. With this principle we derive a single operator shape which is optimal at any scale. The optimal detector has a simple approximate implementation in which edges are marked at maxima in gradient magnitude of a Gaussian-smoothed image. We extend this simple detector using operators of several widths to cope with different signal-to-noise ratios in the image. We present a general method, called feature synthesis, for the fine-to-coarse integration of information from operators at different scales. Finally we show that step edge detector performance improves considerably as the operator point spread function is extended along the edge.
Article
Full-text available
“Urban Sprawl” is a growing concern of citizens, environmental organizations, and governments. Negative impacts often attributed to urban sprawl are traffic congestion, loss of open space, and increased pollutant runoff into natural waterways. Definitions of “Urban Sprawl” range from local patterns of land use and development to aggregate measures of per capita land consumption for given contiguous urban areas (UA). This research creates a measure of per capita land use consumption as an aggregate index for the spatially contiguous urban areas of the conterminous United States with population of 50,000 or greater. Nighttime satellite imagery obtained by the Defense Meteorological Satellite Program's Operational Linescan System (DMSP OLS) is used as a proxy measure of urban extent. The corresponding population of these urban areas is derived from a grid of the block group level data from the 1990 U.S. Census. These numbers are used to develop a regression equation between Ln(Urban Area) and Ln(Urban Population). The ‘scale-adjustment’ mentioned in the title characterizes the “Urban Sprawl” of each of the urban areas by how far above or below they are on the “Sprawl Line” determined by this regression. This “Sprawl Line” allows for a more fair comparison of “Urban Sprawl” between larger and smaller metropolitan areas because a simple measure of per capita land consumption or population density does not account for the natural increase in aggregate population density that occurs as cities grow in population. Cities that have more “Urban Sprawl” by this measure tended to be inland and Midwestern cities such as Minneapolis–St. Paul, Atlanta, Dallas–Ft. Worth, St. Louis, and Kansas City. Surprisingly, west coast cities including Los Angeles had some of the lowest levels of “Urban Sprawl” by this measure. There were many low light levels seen in the nighttime imagery around these major urban areas that were not included in either of the two definitions of urban extent used in this study. These areas may represent a growing commuter-shed of urban workers who do not live in the urban core but nonetheless contribute to many of the impacts typically attributed to “Urban Sprawl”. “Urban Sprawl” is difficult to define precisely partly because public perception of sprawl is likely derived from local land use planning decisions, spatio-demographic change in growing urban areas, and changing values and social mores resulting from differential rates of international migration to the urban areas of the United States. Nonetheless, the aggregate measures derived here are somewhat different than similar previously used measures in that they are ‘scale-adjusted’; also, the spatial patterns of “Urban Sprawl” shown here shed some insight and raise interesting questions about how the dynamics of “Urban Sprawl” are changing.
Article
Full-text available
General and mathematically transparent models of urban growth have so far suffered from a lack in microscopic realism. Physical models that have been used for this purpose, i.e., diffusion-limited aggregation, dielectric breakdown models, and correlated percolation all have microscopic dynamics for which analogies with urban growth appear stretched. Based on a Markov random field formulation we have developed a model that is capable of reproducing a variety of important characteristic urban morphologies and that has realistic microscopic dynamics. The results presented in this paper are particularly important in relation to "urban sprawl," an important aspect of which is aggressively spreading low-density land uses. This type of growth is increasingly causing environmental, social, and economical problems around the world. The microdynamics of our model, or its "first principles," can be mapped to human decisions and motivations and thus potentially also to policies and regulations. We measure statistical properties of macrostates generated by the urban growth mechanism that we propose, and we compare these to empirical measurements as well as to results from other models. To showcase the open-endedness of the model and to thereby relate our work to applied urban planning we have also included a simulated city consisting of a large number of land use classes in which also topographical data have been used.
Article
We consider night light as a type of consumer goods and propose a model for factors affecting the relationship between night lights and GDP. It is then decomposed into agricultural and non-agricultural productions. Further, the model is modified to determine how the factors affect residents' propensity to consume lights. Models are tested with time-fixed regression on a set of 15-year panel data of 169 countries globally and regionally. We find that light consumption propensity is affected by GDP per capita, latitude, spatial distribution of human activities and gross saving rate, and that light consumption per capita has an inverted-U relationship with GDP per capita.
Article
As increasing volumes of urban data are captured and become available, new opportunities arise for data-driven analysis that can lead to improvements in the lives of citizens through evidence-based decision making and policies. In this paper, we focus on a particularly important urban data set: taxi trips. Taxis are valuable sensors and information associated with taxi trips can provide unprecedented insight into many different aspects of city life, from economic activity and human behavior to mobility patterns. But analyzing these data presents many challenges. The data are complex, containing geographical and temporal components in addition to multiple variables associated with each trip. Consequently, it is hard to specify exploratory queries and to perform comparative analyses (e.g., compare different regions over time). This problem is compounded due to the size of the data-there are on average 500,000 taxi trips each day in NYC. We propose a new model that allows users to visually query taxi trips. Besides standard analytics queries, the model supports origin-destination queries that enable the study of mobility across the city. We show that this model is able to express a wide range of spatio-temporal queries, and it is also flexible in that not only can queries be composed but also different aggregations and visual representations can be applied, allowing users to explore and compare results. We have built a scalable system that implements this model which supports interactive response times; makes use of an adaptive level-of-detail rendering strategy to generate clutter-free visualization for large results; and shows hidden details to the users in a summary through the use of overlay heat maps. We present a series of case studies motivated by traffic engineers and economists that show how our model and system enable domain experts to perform tasks that were previously unattainable for them.
Article
Obesity has reached crisis proportions in industrialized societies. Many factors converge to yield increased body mass index (BMI). Among these is sleep duration. The circadian clock controls sleep timing through the process of entrainment. Chronotype describes individual differences in sleep timing, and it is determined by genetic background, age, sex, and environment (e.g., light exposure). Social jetlag quantifies the discrepancy that often arises between circadian and social clocks, which results in chronic sleep loss. The circadian clock also regulates energy homeostasis, and its disruption-as with social jetlag-may contribute to weight-related pathologies. Here, we report the results from a large-scale epidemiological study, showing that, beyond sleep duration, social jetlag is associated with increased BMI. Our results demonstrate that living "against the clock" may be a factor contributing to the epidemic of obesity. This is of key importance in pending discussions on the implementation of Daylight Saving Time and on work or school times, which all contribute to the amount of social jetlag accrued by an individual. Our data suggest that improving the correspondence between biological and social clocks will contribute to the management of obesity.
Article
More and more people's working and social lives are blighted by skewed sleep patterns. Is it time for the medical mainstream to take notice of what neuroscientists are learning about the body clock? Alison Abbott reports.