The Lights Are On, All Over the World*
Sandra Lach Arlinghaus
The University of Michigan
Adj. Prof., School of Natural Resources and Environment; College of Architecture and Urban Planning
A few years ago, scientists across the internet were sending the breathtaking image below (Figure 1), showing lights of the Earth at night, to colleagues, friends, and family.  When most scholars I know first looked at it, they noted immediately that the image portrayed mirrored the locations of the cities of the Earth.    On second glance, they wondered how the image was obtained:  the entire surface of the Earth is never dark, simultaneously.  According to a NASA website  (,
"The images were taken by a Defense Meteorological Satellite Programís (DMSP) Operational Linescan System (OLS). This network of satellites was originally designed to pick up on lunar illumination reflecting off of clouds at night in order to aid nighttime aircraft navigation. What the Air Force discovered is that on evenings when there was a new moon, the satellites were sensitive enough to record the illumination from city lights. Over a period of several new moons, the data the satellites retrieved could be pieced together to produce a global image of city lights."

Figure 1.  "Bright Lights, Big City."  The Earth at night--sequence of shots from satellite imagery.  Data courtesy Marc Imhoff of NASA GSFC and Christopher Elvidge of NOAA NGDC. Image by Craig Mayhew and Robert Simmon, NASA GSFC.** 

Marc Imhoff (see links below for references) apparently was the first scholar to note the striking correspondence between the light pattern and the spatial distribution of urban areas.  His interests in developing a spatial view of the global distribution of sprawl led him to work with the city lights map.  He noted that the light pattern overestimates urbaninzation, when taking a more careful look at a larger-than-global scale, and worked with a research team from Goddard Space Center to study, with greater accuracy, the noted correspondence ( ):
"The researchers classified the lights left on the image, after [a] dimming process, as urban area. The previously lit areas on the image that shrank back were classified as peri-urban (low-density suburban areas or farmland). Any areas that had no lights to begin with were labeled as non-urban. They compared these classifications to the boundaries on the actual urban areas of the city and found there was a close match. Imhoff and his team now had a set of numbers (threshold values), which told them to what extent the lights from any portion of the United States should be dimmed to get an accurate and spatially explicit representation of urbanization."
To look at the world's population, in a broad view only, consider using the data in the files of the Digital Chart of the World (DCW).  The data is general, but is designed for viewing broad global pattern.  Thus, Figure 2 shows a sequence of static images representing the world's urbanized areas and populated places.  DCW categorizes both urbanized area and populated places in a hierarchy according to size, represented as 1, 2, and 3, from highest to lowest, in the maps below. There are also a few other categories noted in the legends below (Figure 2a-2j--scroll across to see the entire sequence).  The urbanized areas are colored in tones of yellow from lightest at the high end to darkest at the low end; the populated places are colored in tones of khaki, from lightest at the high end to darkest at the low end.  In addition, there is an overall continuum of shading from lightest yellow to darkest khaki across all the categories.

Figure 2a.  Layers of data: City (polygon), Kampong (polygon), City 1, City 2, City 3, Populated Place 1, Populated Place 2, Populated Place 3, Village.

Figure 2b.  Layers of data:  Kampong (polygon), City 1, City 2, City 3, Populated Place 1, Populated Place 2, Populated Place 3, Village.

Figure 2c.  Layers of data:  City 1, City 2, City 3, Populated Place 1, Populated Place 2, Populated Place 3, Village.

Figure 2d.  Layers of data:  City 2, City 3, Populated Place 1, Populated Place 2, Populated Place 3, Village.

Figure 2e.  Layers of data:  City 3, Populated Place 1, Populated Place 2, Populated Place 3, Village.

Figure 2f.  Layers of data:  Populated Place 1, Populated Place 2, Populated Place 3, Village.

Figure 2g.  Layers of data:  Populated Place 2, Populated Place 3, Village.

Figure 2h.  Layers of data:  Populated Place 3, Village.

Figure 2i.  Layers of data:  Village.

Figure 2i.  Layers of data:  None.
Data Layer information from:  Atlas Data Product Documentation, Digital Chart of the World (DCW), Strategic Mapping, Inc., (File Release  Jan 8, 1993).  Extract from Defense Mapping Agency's (DMA) Online Documentation for the Digital Chart of the World (DCW) in Vector Product Format (VPF) [italics added below]:
Special automation techniques 
Populated place area features were automated from open window peel coat sources.  Positive film separates containing these features were optically scanned, and their outlines vectorized.  Before automation, PP area features were manually extended into inland water body features and across coastlines, intentionally creating overlap zones.  The overlap zones were then eliminated through an automated spatial overlay process in ARC/INFO.
Feature coincidence 
All boundaries between populated places and other area feature classes (including inland water bodies, land cover areas, and coastlines) have identical coordinate representations in the separate coverages.  No attempt was made to make PP area feature boundaries coincident with line features (such as roads) that were depicted as being coincident on the source manuscripts.  Road and railroad line features are split precisely at the positions at which they intersect PP area feature boundaries, except for connecting road and railroad line features compiled from independent sources, which occasionally extend across city outline boundaries without being split.
Database design issues
No features were eliminated on the basis of size, since some type of  size-based representation rule had already been applied to the source manuscripts.  Source map area features classified as having indefinite shape (represented as square, filled objects) were captured as point features in the PP coverage. 
Special automation techniques 
All populated place points were manually digitized.  Annotation text strings were manually positioned and later verified using image background coverages of the scanned sheets for reference. 
Feature coincidence 
PP point data do not have explicit relationships with other DCW feature classes. 
Database design issues 
Font sizes are in accordance with ONC chart specifications.  In instances where the source materials deviated from established specifications, font size assignments were chosen that reflected known relative population characteristics on a regional level.  All population characteristics as expressed through font sizes and symbology on the source materials are maintained within the database as annotation text attributes.  No diacritical marks were captured, with the exception of apostrophes which are included only in conjunction with possessive nouns.  Text features do contain attributes indicating whether they included diacritical marks on the source.  All PP annotation text is stored with a single location point at the lower left corner of the string.  All PP annotation text for a given object is stored as single strings even when represented as multiple, stacked strings on the source manuscripts.  All located objects with a proper name were placed in the PP point feature class."

The sequence of images in Figure 2 shows, successively, the entire hierarchy on the left with one hierarchical layer removed as the reader scrolls toward the right.  One might also wish to see each layer individually rather than looking at the entire stack of layers and subsets of that stack.  Figure 3 shows each of these layers, separately.

When the sequence of images in Figure 3 is animated (Figure 4a), interesting visual comparisons emerge of the global distribution of urbanized areas, populated places, and lights.  The animation begins with all layers showing and successively removes them while applying 5 stages of computer-generated (in Adobe ImageReady) between-smoothing-frames (tweening).  It appears that it is urbanization, the yellow in Figure 4a, rather than mere population, the khaki in Figure 4a, that appears the greatest source of light in Figure 4b.


Figure 4a.  Animation of frames in Figure 3 with "tweening" applied in Adobe ImageReady (trademarked).  Compare to Figure 4b, a copy of Figure 1.
Figure 4b.

Indeed, this comparison offers support for the observation that (
"The brightest areas of the Earth are the most urbanized, but not necessarily the most populated. (Compare western Europe with China and India.) Cities tend to grow along coastlines and transportation networks. Even without the underlying map, the outlines of many continents would still be visible. The United States interstate highway system appears as a lattice connecting the brighter dots of city centers. In Russia, the Trans-Siberian railroad is a thin line stretching from Moscow through the center of Asia to Vladivostok. The Nile River, from the Aswan Dam to the Mediterranean Sea, is another bright thread through an otherwise dark region."
The information from the Digital Chart of the World offers a number of alternate ways to view information about population.  No doubt a whole host of questions might be asked about the global pattern of lights, urbanization, and population.  One question that is often discussed in the contemporary urban municipal setting is that of improving light dispersal in existing cities.  One might imagine looking at the image in Figure 1, together with tables of population, and trying to find places with high population and low light use in relation to a set of population peers.  Clearly, this sort of differencing would not be possible with maps that are as crude as those above.

As a general strategy, though, one might consider an application of Feigenbaum's graphical analysis to find critical values based on increasing urbanization.
  • Assume a functional relationship between lights and urbanization with lights depending on urbanization:
    • lights = f(urbanization)
  • Suppose that the functional relationship is increasing without bound, as for example in an exponential curve:  as population increases, lighting increases without bound (Figure 5a).
  • Suppose that the functional relationship now has an upper bound placed on it, in the form of a lighting ordinance, for example (Figure 5b).

Figure 5a: no upper bound Figure 5b:  purple line is upper bound representing lighting ordinance

The graphical dynamics in Figure 5a are out of control; the initial seed (yellow) fired at the purple exponential bounces back and forth from curve to y=x in a fashion that goes on in ever-increasing steps toward infinity.  When a lighting ordinance is applied as an upper bound, and that ordinance has sufficient strength to dampen the exponential so that it pushes back to the right of the line y=x (quickly), then the graphical dynamics can be made to settle down around the intersection point of the damped curve and y=x (as a fixed point that attracts graphical process).  The latter situation suggests one form of intervention to control graphical dynamics representing real-world situations.  The challenge then becomes implementation of such ideas as well as extension of them into more distant realms.


*Play on a World War II song title:  Vaughn Monroe, 1943, "When the Lights Go On Again, All Over the World"
**With thanks to Robert Simmon for permission to use the city lights image.