GIS/EM4 - Progress in Empirical Measurement of the Urban Environment: An exploration of the theoretical and empirical advantages of using Nighttime Satellite imagery in Urban Studies

Progress in Empirical Measurement of the Urban Environment: An exploration of the theoretical and empirical advantages of using Nighttime Satellite imagery in Urban Studies

GIS/EM4

Paul C. Sutton

Abstract

Nighttime satellite imagery provided by the Defense Meteorological Satellite Program's Operational Linescan System (DMSP OLS) is evaluated as a novel empirical proxy measure of ambient population density in urban areas. Included is a discussion of historical theoretical models of urban population density, issues of temporal and spatial scale as they pertain to conceptions of population density, needs for and applications of finer resolution global and regional population datasets, and a brief review of state-of-the-art global representations of the human population.

Keywords

Empirical measurement, nighttime satellite imagery, Defense Meteorological Satellite Program's Operational Linescan System (DMSP OLS), population density, Global Population of the World (GPW).

Introduction

Understanding the human dimensions of global change and other important and interesting human-environment interactions is difficult if not impossible without knowledge of the spatial distribution of the human population and its diverse behaviors. A major and increasing proportion of the world's population is living in urban areas. Nonetheless the diverse means of collecting social and demographic information (particularly in urban areas) present many problems for incorporating the data into studies of human-environment interactions. Despite the fact that a great deal of important data is being collected, the incommensurate spatial units, reporting methods, and spatial and temporal scales of the information leave many analyses unperformable. In fact, several organizations and institutions have determined that socio-economic data at spatial and temporal scales commensurate with existing environmental data is the most significant need for investigating the human dimensions of global change (Clark and Rhind 1992). Data gathered from satellites will never be the sole source of important and relevant information about the urban environment. However, remotely sensed information can provide important information to both government and private industry that cannot be provided via in situ assessments of the urban environment (Jensen and Cowen 1999). The research described here explores how to produce representations of the urban environment derived from nighttime satellite imagery that can make a substantial contribution to many studies of human-environment interactions.

Background

Initial investigations of nighttime satellite imagery provided by the Defense Meteorological Satellite Program's Operational Linescan System (DMSP OLS) determined that the imagery is a reasonable proxy measure of urban land cover (Imhoff, Lawrence et al. 1997). Subsequent analyses drew upon existing urban geographic theories of allometry and showed that the areal extent of isolated urban clusters can predict the corresponding aggregate population of those areas (Sutton, Roberts et al. 1997). The research described here predicts intra-urban population densities of these urban areas based on light intensity and provides an explanation for the relationship between light and population density. In addition, an argument will be made for the appropriate spatial scale of analysis with respect to a static temporally averaged conception of population density. The development of this research demonstrates how empirical remotely sensed observations of the urban environment can both inform and be informed by theory. The dominant existing theoretical models of urban population density within the urban environment require two measurements: 1) the location of the Central Business District, and 2) distance from the central business district (Clark 1951). Here it is proposed that the intensity of light emitted from the earth at night is a better indicator of population density than those measures derived from exponentially decaying population density as a function of distance from the city center. Both theories require spatially contextual information such as how large the urban area is and what major economic region the urban area exists in. The DMSP OLS measure of light intensity can provide both areal extent and light intensity; and, can easily be geo-referenced allowing for inclusion of available aggregate national statistics. The changing nature of urban form and structure raises many questions as to the generality, utility, and validity of the urban population density decay theories for describing population density within the urban environment. (Clearly, its power at the aggregate level is profound and may even be considered a law of geography). Nonetheless, conurbation of adjacent cities, airports, the phenomenon of edge cities, shopping malls in the suburbs, and other developments have undoubtedly changed urban form in ways that make the exponentially decaying population density function difficult if not impossible to implement in any precise way. As Burgess, Harris, Hoyt, and Ullman have pointed out, cities have multiple nuclei and sectors that make implementing the population density decay function problematic (Burgess 1924; Hoyt 1939; Harris and Ullman 1959). However, the argument made here is that nighttime light intensity will capture sectors, multiple nuclei, and their corresponding population densities.

Methods

A rigorous definition of population density is easy to produce as long as the spatial and temporal scales are defined. At any given moment in time population density is simply the number of people in a given areal unit. Temporally averaged population density is simply a time weighted (by person) measure of the number of people in a given area. For example, consider a given square kilometer that has 100 people in it for the first eight hours of a day, 500 people in it for the second eight hours, and zero people in it for the third eight hours. A one day temporally averaged measure of the population density of that area is 200 persons per square kilometer. Temporally averaged measures of population density are a proxy for many important phenomena related to the intensity of human activity. However, temporally averaged measures of population density are very uncommon. Nighttime imagery shows a strong but imperfect correlation with residence based measures of population density in the Los Angeles Metropolitan area (Figure 1).

However, the imagery correlates significantly with both residence and employment based measures of population density. In addition when residence and employment-based measures of population density are averaged, the correlation with the nighttime imagery is stronger than with either measure of population density alone. Some of the advantages a nighttime imagery based measure of population density are: 1) Uniform spatial units that meet needs for studies of the human dimensions of global change; 2) Objective global coverage on a more frequent repeat cycle (~ 1 year); and 3) A temporally averaged diurnal measure of population density based on human behavior. Figure 2 is a graphical and formal representation of the simple linear model used to predict intra-urban population density based on light intensity in the nighttime imagery.

The comparison of this model to the Los Angeles area's residential population density produces an interesting pattern of error (Figure 3). The model underestimates the high population density of urban centers and over-estimates the population density of the large airports in the area. Residentially based measures of population density clearly miss the tens of thousands of people that work at and travel through major international airports. Nighttime imagery based measures of population density actually capture this kind of population density.

Discussion

These empirical observations of correlation between nighttime imagery and various measures of population density can be explained with very simple theory about the purpose of nocturnal lighting and the spatial scale at which it occurs in actuality and is aggregated to in the imagery. The fundamental geographic measure of distance remains an important variable in questions about the relationship between human presence and light. Let us investigate this relationship as it pertains to the light that can be seen by the nighttime satellite as it observes the earth at midnight. The platform orbits the earth from an average height of 865 kilometers. The swath width of the image is about 3,000 kilometers. Consequently much of the imagery is obtained from an oblique look angle. In many respects the light seen by the sensor is the same light seen by an airplane passenger looking out the window at nighttime city lights (at a reduced spatial and spectral resolution). Nocturnal lighting is a primary method for enabling human activity, notably for the facilitation of transportation; although, it also serves purposes of domestic security, advertising, and aesthetics. Outdoor lighting is used extensively worldwide in residential, commercial, industrial, public facilities and roadways (Elvidge, Baugh et al. 1998). Let us examine the transportation elements of this phenomenon. Modes of transportation can be roughly categorized into three basic categories: 1) Walking, 2) Ground transportation (car, bus, train, etc.), and 3) Air. The average distances traveled via these three modes of transportation are dramatically different, as is the nature of the lighting used to facilitate these modes of transportation. The intensity of the lighting used to facilitate these various modes of transportation is directly proportional to average distance traveled; whereas the spatial density of the lighting is inversely proportional to the average distance traveled. Airports provide very bright lights to facilitate the navigation of airplanes. However, these concentrations of light are scattered as widely in space as are airports. Highways and streets provide linear and relatively sparsely separated lights to facilitate driving. Residential, Industrial, and commercial areas provide high spatial density, lower intensity lighting to facilitate walking on foot. All of this lighting is visible from space at night by a sensitive satellite. The convolution of the intensity and spatial density of these nocturnal emissions is clearly related to human activity in space. The actual population density that nighttime imagery is attempting to capture is a temporally averaged (over 3-6 months) measure of population density at a spatial resolution of 1 km2. Actual measurements of population density defined in this manner would be very difficult to make. A rough approximation could be obtained by estimating the location, and duration of presence, of the population throughout the course of the day. A rough estimate of the temporal breakdown of a typical adult's day might look as follows: 40% of an average adult's time is spent at their residence, 35% is spent at their place of employment, 10% is spent walking, driving, riding or flying (in transportation), and the remaining 15% is spent at various other activities such as shopping, entertainment, vacationing etc. The questions regarding the nighttime imagery with respect to this hypothesized definition of population density are: 1) 'Does the nature of nocturnal lighting match this temporal behavior (e.g. Residence, Employment, Transportation, and Shopping etc.)?' and, 2) 'Is the 1 km2 spatial resolution appropriate to capture this lighting in a way that corresponds with population density?'

DMSP OLS observations and Residential Population Density

Nocturnally visible residential lighting serving the purposes of domestic security, and facilitating walking and ground transportation does correspond in spatial density with population density. Middle income suburbs tend to have more street lighting than higher income, larger lot size, lower population density areas. Suburbs also tend to have less nocturnal lighting than lower income areas with apartment buildings mixed in with some commercial enterprises. High population density areas near the central business districts have the highest measured light intensity. In fact, as has been shown, there is a significant correlation between residence-based measures of population density and light intensity observed by the nighttime satellite. The major spatial discrepancies between light intensity and population density tend to occur in places like airports; which represent both the employment and transportation elements of any temporally averaged measure of population density.

DMSP OLS Observations and Employment Based Measures of Population Density

Commercial and Industrial locations are two significant employment locations. Nocturnal lighting of these facilities is often spatially concentrated because the lighting is used to facilitate walking. Large international airports are very bright in the nighttime imagery and they represent major employment locations. Again, there is a significant correlation between employment-based measures of population density and nocturnal light intensity. Neither the employment based nor the residence based measure of population density account for population density that results from where people are when they are shopping, driving, going to school, or attending a play. Nor do they account for the fact that all people are not adults with a job. However, the fact that a strong correlation exists between nighttime imagery based measures of light intensity and both residential and employment based measures of population density suggest that the nighttime imagery may actually capture these other elements of a temporally averaged measure of population density.

DMSP OLS Observations And Other Measures of Population Density

Some other temporally specific measures of population density that have not been addressed yet are those related to transportation, shopping, and recreation. Nocturnal lighting facilitates human activity in many places other than the residence and the place of employment. The spatial density and intensity of lighting used are functions of the human activity being facilitated. A small suburban shopping center is better lit than the residential neighborhood around it to facilitate the great deal of driving and walking that is taking place there (it probably also provides 'commercial' security). All modes of transportation (walking, ground transport, and airplane) are facilitated by different kinds of lighting. Airplanes land at airports and consequently airports are a very spatially concentrated transportation element of the urban environment. The nighttime imagery reflects this concentration. Automobiles and other ground transportation produce their own light via headlights and are supported with street lights. Roads are not as spatially concentrated as airports and this is reflected in the lighting that supports ground transportation. The nighttime imagery reflects this also. Areas that have a large number of people on foot tend to have the highest density of nocturnal light to support the walking. Yet this is not limited strictly to places of employment, it includes commercial areas such as shopping malls, entertainment venues such as football stadiums, and numerous other locations.

Is the DMSP OLS 1 km2 scale of measurement appropriate?

Nocturnal lighting supports various human activities many of them associated with different modes of transportation. At the 1 km2 resolution of the DMSP OLS imagery the spatial intensity of the lighting is converted to a measurement of light intensity. The major modes of transportation (walking, driving, flying) are related both to an average distance traveled and the amount of time spent in the location of the mode of travel. For example, when people are walking they don't travel as far, and are often spending more time at the location where they are walking (e.g. at work, home, or shopping). Clearly there is a scale that is too fine to generalize the relationship between nocturnal lighting and the population density that results from the various modes of transportation that the lighting supports. Residential street lighting provides an example of this. A suburban subdivision has an average population density. The nature of the relationship between population density and nocturnal light emissions does not suggest that any estimates of the population density of a suburban subdivision could be any better than the average population density of the subdivision. Clearly, some houses in the suburb may have larger families in them and an accurate finer resolution measure of population density would capture this. However, a finer resolution nighttime image would merely capture the bright streetlights in a salt and pepper pattern that would not reflect actual variation in population density. The 1 km2 scale allows for the aggregation of most kinds of nocturnal lighting to a spatial scale somewhat coarser than the lighting exists at. This aggregation has the statistical power of the mean to generalize fine scale variation in the actual lighting to a scale more representative of the phenomenon related to the lighting (e.g. population density). Figure 4 is an aerial photograph of an approximately one kilometer region of Denver, Colorado. The image includes an elementary school, part of the Unviversity of Denver Campus, a shopping center anchored by a Safeway grocery store, a busy commercial street (Evans Avenue), and about 500 residences. This figure provides a means of visualizing what aggregation to one kilometer pixels means. Clearly aggregation to one kilometer does not capture the young age structure of the population of the elementary school nor the diurnal variability of population density at the school. Also the school has a much lower population density on weekends that is partially compensated for the higher population density of the Safeway across the street on weekends. The whole pixel has an annually fluctuating population density due to variation in the students populations at both the university, the nearby homes, and the elementary school in the summer. It may be disappointing to many that the detail in Figure 4 is aggregated to a single number in a one kilometer resolution representation of population density. However, it is important to keep in mind that there are no global representations of human population density at even one kilometer resolution.

State-of-the-art Global Representaions of the Human Population

Two state of the art representations of the global human population are the Global Population of the World (GPW) (http://sedac.ciesin.org/plue/gpw) and LandScan (Dobson, Bright et al. 2000). GPW is derived from second level administrative boundaries (e.g. U.S. counties) and disaggregated to 2.5 minute grids. GPW is considered an unbiased single variable grid in that it is derived only from officially reported national census data. LandScan estimates population density to a 30 second by 30 second resolution using a model that incorporates many variables including roads, slope, coastlines, populated places, and nighttime satellite imagery. LandScan is unbiased at the province level (1st level administrative boundary) and provides a finer spatial resolution representation of population density than GPW. In any case it is interesting to note that both of these state of the art representations of population density are based on administrative boundary data that are difficult, expensive, and/or impossible to obtain for many parts of the world. Many questions remain as to what spatial and temporal resolution of population will be needed for understanding, adapting to, and possibly mitigating the myriad problems and challenges associated with global change. Undoubtedly these datasets would be greatly enriched and more useful if they not only described where we are when but who we are spatially (e.g. age and sex structure, income, employment etc.) Nontheless there is demand for data of this nature and even at the crude level at which global representations exist today these datasets are being used in many important studies.

Conclusion

Global representations of the human population are particularly useful in that they lend themselves to integrated analyses that with global perspectives. A global perspective was essential to recognize climate change and human activity has been recognized as a leading cause of global climate change. Integration of accurate, fine spatial and temporal resolution data concerning the location and behavior of the human species must be a first priority for studies of global change. Global and regional representations of the human population have proved useful in many important studies related to urbanization, agricultural transformation, threats to biodiversity, and generation of environmental sustainability indexes (Vitousek, Ehrlich et al. 1986; Berry 1990; Smil 1997; Cincotta, Wisnewski et al. 2000; Samuel -Johnson 2000). More studies of this nature will undoubtedly occur but their impact on policy will be neglible unless the data they are derived from is accurate at scales appropriate to the respective analyses. Nighttime satellite imagery can clearly improve estimates of the spatial distribution of people throughout the world and may also have potential for enhancing estimates of other socio-economic indicators as well (income, CO2 non-point source emissions, etc).

Acknowledgements

Funding to support this research was provided by a National Science Foundation grant(NSF grant # CMS-9817761). This support is gratefully acknowledged.

References used

Berry, B. J. L. (1990). Urbanization. The Earth as Transformed by Human Action. B. L. Turner, W. C. Clark, R. W. Kateset al. Cambridge`, Cambridge University Press: 103-119.

Burgess, E. W. (1924). "The growth of the city: an introduction to a research project." Publications of the American Sociological Society 18: 85-97.

Cincotta, R. P., J. Wisnewski, et al. (2000). "Human Population in the Biodiversity Hotspots." Nature 404(April 27): 990-992. Clark, C. (1951). "Urban Population Densities." Journal of the Royal Statistical Society Vol. CXIV Part IV(Series A): 490-496.

Clark, J. and D. Rhind (1992). Population Data and Global Environmental Changes, International Social Science Council, Programme on Human Dimensions of Global Environmental Change, UNESCO, Paris.

Dobson, J. E., E. A. Bright, et al. (2000). "A Global Population Database for Estimating Populations at Risk." Photogrammetric Engineering & Remote Sensing 66(In Press).

Elvidge, C. D., K. E. Baugh, et al. (1998). "Radiance Calibration of DMSP-OLS low-light imaging data of human settlements." Remote Sensing of Environment 68: 77-88.

Harris, C. D. and E. L. Ullman (1959). The nature of cities. Readings in Urban Geography. H. M. Mayer and C. F. Kohn. Chicago, Chicago University Press: 277-860.

Hoyt, H. (1939). The structure and growth of residential neighborhoods in American cities. Washington D.C., Federal Housing Administration.

Imhoff, M. L., W. T. Lawrence, et al. (1997). "A Technique for Using Composite DMSP/OLS "City Lights" Satellite Data to Accurately Map Urban Areas." Remote Sensing of Environment 61: 361-370.

Jensen, J. R. and D. C. Cowen (1999). "Remote Sensing of Urban/Suburban Infrastructure and Socio-Economic Attributes." Photogrammetric Engineering and Remote Sensing 65(5): 611-622. Samuel -Johnson, K. (2000). Pilot Enviornmental Sustainability Index. Davos, Switzerland, World Economic Forum.

Smil, V. (1997). Global Population and the Nitrogen Cycle. Scientific American. 277: 70-75.

Sutton, P., D. Roberts, et al. (1997). "A Comparison of Nighttime Satellite Imagery and Population Density for the Continental United States." Photogrammetric Engineering and Remote Sensing 63 (11)(November): 1303-1313.

Vitousek, P., P. R. Ehrlich, et al. (1986). "Human appropriation of the products of photosynthesis." BioScience 36: 368-373

Authors

Paul C Sutton, Department of Geography University of Denver, Denver, Colorado 80208 United States. Email:psutton@du.edu, Tel: (303) 871-2399, Fax: (303) 871-2201.