|
Why do we need global maps of urban areas?
|
|
|
City smog is just one example of environ- mental changes caused by expanding urban areas. |
|
|
|
2
|
|
While growth and expansion of urban areas has long been studied at the local scale, the cumulative impact of urban expansion in global change processes is virtually unknown (Lambin et al. 2001). To advance our understanding of the possible aggregate influence of urban areas, it is important to have a means to map and monitor cities over large areas.
As cities grow, the degree and complexity of environmental change caused by human settlements increase. Cities are known to produce microclimatic changes through the urban heat island effect (Stephenson 1991). The size and rate of current urban expansion also have the potential to produce regional climate and biogeochemical changes (Lakshmanan 1997), and linkages exist between local air pollution and global processes involving energy transfer, hydrology and climate interactions (Calbo et al. 1998, Decker et al. 2000).
Moreover, the growth of cities can wreak environmental damage through the conversion of natural ecosystems, the removal of flora and fauna, the compaction of soils, the pollution of water, and the alienation of ground for landfill and waste treatment (Douglas 1994).
In addition, social scientists have studied urban areas as a global network or system articulated by global economic activity for more than two decades (Hall 1984, Knox and Taylor 1995). The acceleration and intensification of globalization and integration of national economies has reinforced the idea that cities are important nodes of production and consumption, in essence, command centers for contemporary global capitalism (Smith 2001). Researchers in this arena require more synoptic and spatial views of population growth, migration, and distribution, settlement patterns and urban morphology, as well as broader views of city network linkages, megalopolises and conurbations.
|
|
How have global maps of cities been made in the past?
|
|
|
Maps produced by the Digital Chart of the World have been the only source of global information on cities for many years. |
|
|
|
|
|
The Nighttime Lights data are often used to depict cities, but this is a representation of light, not what is on the ground. |
|
|
|
|
|
To understand many global processes, maps of land cover have become an important tool for circulation, hydrology and carbon models. These maps have continued to rely on urban boundaries available from the Digital Chart of the World populated places digital data (Danko 1992), an out-of-date map source which has proven inconsistent globally and poorly georeferenced. Frustration with DCW data as well as the conception of urban land cover as "small enough to be ignored" may explain, in part, why researchers have so deliberately not included urban information in global land surface parameterization.
More recently, data from the Defense Meteorological Satellite Program's "low light" sensor have been used to depict lights at night (Elvidge et al. 1999). While the data have not been validated for accuracy, the images provide a striking depiction of human activity on the planet. It is important to recognize that maps derived from DMSP data provide a representation of light, but do not necessarily represent the built environment or settlement patterns. In particular, brightly lit agricultural areas and non-urban light sources such as gas flares and fires are captured in these data sets. Despite these problems, the data have received much attention from both researchers and the media.
|
|
Is there a better way to map urban land cover?
|
|
|
A global view of the map of urban areas produced in this research using the fusion of MODIS data, Nighttime Lights data and Gridded Population data. Click on the image for a larger view (12mb) with city snapshots, or see below for a regional comparison. |
|
|
|
|
|
The primary goal of the first year's research was to assess the utility of 1 km MODIS data for mapping and monitoring urban areas at continental to global scales. Two major tasks were involved in this study. First, a supervised decision tree classification method was developed, where map errors were minimized by exploiting data from two ancillary data sources: the Nighttime Lights data and Gridded Population density data. The second task was to establish the best means for evaluating the accuracy of maps produced over large regions, an issue that is especially problematic when the class of interest is a small fraction of the area mapped. (Schneider et al. 2003)
The method involved three main steps. In the first step, the Nighttime Lights data and Gridded Population density data were combined in a logistic regression model to produce a probability surface for urban areas. In the second step, a decision tree algorithm was trained using a global set of training sites for 17 land cover classes (including urban) defined by the International Geosphere-Biosphere Program (IGBP), and the trained tree was applied to the MODIS data. The output from this first stage provided a map of per-pixel probabilities for each of 17 classes. The class probabilities and the probability surface were then used as input to the third step, where Bayes' Rule was applied at every pixel. To do this, the probabilities of urban areas derived from the logistic regression were used as prior probabilities, and the final pixel label was assigned based on the maximum likelihood derived from the posterior probabilities. In this way, information from all three data sources was fused to create a final map of urban areas.
|
|
A comparison of global urban maps...
|
|
|
|
|
|
|
Results show that MODIS data alone is not sufficient to reliably map urban areas. However, by fusing MODIS data with ancillary sources, it was possible to map urban areas on a continental scale for the first time. A regional view of San Francisco, U.S., provides a useful basis for comparison: (from upper left) (a) fine resolution Landsat TM imagery (urban areas appear purple, vegetation green); (b) the urban data from the Digital Chart of the World (note misregistration); (c) the continuous values of the Nighttime Lights data; (d) 1 km resolution MODIS data (urban areas also appear purple, vegetation green); and (e) the map created in this research using the fusion of MODIS data, Nighttime Lights data, and Gridded Population data.
Understanding the methodological and validation requirements for mapping urban areas from coarse resolution data will provide a foundation for urban and urban growth mapping in the future, and will serve as a basis for developing methods for use with 250 m and 500 m MODIS data. In addition, the Fused Data maps will be compared against local maps of urban areas currently being developed from Landsat data. This analysis will provide an assessment of the utility of coarse resolution data products.
|
|
Availability of the data
Click here for more information on the data.
|
