A cluster-based method to map urban area from DMSP/OLS nightlights
Introduction
The urban system is complex with various interacting components. Urbanized area, a major feature of the urban system, represents population centers and economic hubs largely characterized by surfaces occupied by buildings, streets, and other infrastructure (Zhang & Seto, 2011). Although urban areas occupy a relatively small fraction of total Earth's surface, urbanization is one of the most important components of human-induced land cover and land use change (LCLUC) and has profound impacts on energy (e.g. urban heat island), water (e.g. flooding), pollution, ecosystems, and carbon cycle from local to regional and even global scales (Brabec, 2002, Foley et al., 2005, McKinney, 2008, Shepherd, 2005, Zhou et al., 2010, Zhou et al., 2013, Zhou et al., 2012). For example, a previous study indicated that 37–86% of direct fuel consumption in buildings and industry and 37–77% of on-road gasoline and diesel consumption in the US occurred in urban areas (Parshall et al., 2010) and the placement of urban infrastructure, while small in area, has a disproportionate impact on potential net primary productivity because of the high native fertility of transformed soils (Imhoff et al., 2004, Nizeyimana et al., 2001).
Remote sensing has been recognized as a major source of consistent and continuous data, and has been used to study urbanization and its change across a variety of temporal and spatial scales (Schneider et al., 2010, Zhang and Seto, 2011, Zhou and Wang, 2007, Zhou and Wang, 2008). Much progress has been made in urbanization research using remote sensing in terms of methodology development and analysis. Urbanization and its related dynamics have been studied not only for individual cities or greater metropolitan areas, but also across selected cities for comparative purposes (Schneider & Woodcock, 2008). Although researchers have started to pay attention to urbanization over large areas, even at global scales (Zhang & Seto, 2011), there are still limited investigations of large scale urban dynamics primarily due to the lack of efficient and timely methods for mapping urban extent over large areas.
Moderate spatial resolution remote sensing data have demonstrated their capability in large scale and global urbanization mapping (Elvidge, Sutton, et al., 2009, Elvidge, Tuttle, et al., 2007, Loveland et al., 2000, Schneider et al., 2010). For example, Schneider et al. (2010) developed a 500 m resolution global urban map using MODIS data from 2000 to 2002. Elvidge, Safran et al. (2007) built a global impervious surface areas (ISA) map using nighttime lights, population counts, and high-resolution ISA data. European Space Agency (ESA) generated a global land cover map using the 300 m Medium Resolution Imaging Spectrometer (MERIS) time series dataset (ESA, 2013). Moreover, with the help of other data and techniques, a number of global urban or population distribution maps have been developed, i.e. the LandScan product (Dobson, Bright, Coleman, Durfee, & Worley, 2000) and the Global Rural–Urban Mapping Project (GRUMP) urban extent (CIESIN, 2011). However, most of these global products have limited temporal coverage, with limited usefulness for dynamic analysis at large scales. Although urban density (fractional urbanization) maps, e.g. global map of ISA by Elvidge, Safran, et al. (2007) and Elvidge, Tuttle, et al. (2007), can provide more information for the study of urbanization, these products require further information such as population or higher resolution supplementary data, which may be difficult to obtain for long time periods over large scales. Moreover, some of these methods require labor-intensive processing of a sufficient number of cloud-free images, and issues of spectral and spatial consistency from different scenes may exist.
The Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) nighttime stable light data (NTL) data are, therefore, a valuable resource for regional and global urban mapping and application to the study of human activities such as population density, economic activity, energy use, and CO2 emissions (Amaral et al., 2005, Cao et al., 2009, Doll et al., 2000, Elvidge, Baugh, Kihn, Kroehl, Davis, et al., 1997, Elvidge, Safran, et al., 2007, Elvidge, Tuttle, et al., 2007, Imhoff et al., 1997, Oda and Maksyutov, 2011, Sutton, 2003, Zhang and Seto, 2011). However, there are several shortcomings in this data, including limited dynamic range, signal saturation in urban centers, contamination from other sources such as gas flares, lack of a well-characterized point spread function (PSF), and lack of a well-characterized field of view (Elvidge, Sutton, et al., 2009). In particular, OLS-derived light features are substantially larger than the lighting sources on the ground, and local economic conditions may have different impacts on the detection and brightness of satellite observed lighting (Elvidge, Sutton, et al., 2009). It was found that the DMSP/OLS NTL data tend to exaggerate the size of urban areas compared to the Landsat analysis, due to several contributing factors, including the reflectance of light from surrounding water and non-urban land areas, georeferencing errors, and warm atmospheric phenomena (Henderson, Yeh, Gong, Elvidge, & Baugh, 2003). For example, the considerable excursion of reflected light onto water bodies causes pixel blooming along the shorelines of large metropolitan areas and the resulting overestimation produces enlarged small towns and expanded boundaries of large cities (Imhoff et al., 1997).
A number of methods have been developed to map urban areas using the DMSP/OLS NTL data (Cao et al., 2009, Elvidge, Tuttle, et al., 2007, Frolking et al., 2013, He et al., 2006, Liu et al., 2012, Lu et al., 2008, Owen, 1998, Small et al., 2005, Sutton et al., 2006). Simple threshold techniques showed potential in generating reasonable urban mapping products at the regional and national scales by using the DMSP/OLS NTL data (Amaral et al., 2005, Henderson et al., 2003, Imhoff et al., 1997, Kasimu et al., 2009). However, the choices of optimal thresholds may vary across regions and countries due to the regional variation in physical environment and socioeconomic development status (Cao et al., 2009, Liu et al., 2012, Small et al., 2005). The determination of appropriate thresholds in delineating urban areas using the DMSP/OLS NTL data is one of the major challenges in urban mapping over large areas (Henderson et al., 2003). Applying a single threshold to the DMSP/OLS data may be problematic, especially across multiple cities or political boundaries (Imhoff et al., 1997). For example, Henderson et al. (2003) found a range in optimal thresholds for urban mapping across different cities with stable light land area lit thresholds of 92% for San Francisco, 97% for Beijing, and 88% for Lhasa, all of which were higher than the thresholds of 82% and 89% for the continental US reported by Imhoff et al. (1997).
Due to the issues in existing global and regional based threshold techniques and their inflexibility, it is necessary, and also a research challenge, to derive optimal thresholds specific to different cities or urban clusters using the DMSP/OLS NTL data in ways that are neither costly nor complex and are globally applicable. In this study, we developed a cluster-based method to estimate the optimal thresholds and delineate the urban extent, and selected the contiguous United States and China, two countries with different urbanization patterns, and also with high quality land-cover data, as experimental areas. This paper focuses on the development of the new threshold method through calibration and validation using a sub-set of regional high-resolution reference data. The remainder of this paper describes the study area and data (Section 2), details of the five major steps of our method (Section 3), a discussion of the results and findings (Section 4), and concluding remarks (Section 5).
Section snippets
Study area and data
In this study, the contiguous US and China were chosen as the experimental areas. These two study areas have different urbanization patterns. In particular, urbanization levels in China vary greatly across space, attributable to the heterogeneous socioeconomic development whereas urbanization is somewhat more uniform in the US. The different urbanization densities and patterns in the US and China provide ideal experimental regions for evaluating the global applicability of the proposed urban
Methods
Threshold techniques have shown potential in generating reasonable urban mapping products at the regional and national scales by using the DMSP/OLS NTL data. In these methods, the pixels with NTL magnitude values larger than some optimal threshold value are classified as urban and all others as nonurban. However, determining optimal thresholds for all cities in a study region is difficult and still remains a challenge. We develop here a cluster-based method to estimate optimal thresholds and
Threshold
The optimal thresholds tend to be larger in larger clusters with higher NTL DN in both countries (Fig. 5), which is an expected finding (Fig. 3, Eq. (2)). In large urban clusters such as Boston and Beijing, the optimal thresholds reach as high as 60 while they are as low as 20 in small urban clusters. Our technique of determining optimal thresholds based on size and NTL DN magnitude can help to reduce under- and over-estimation, which has been a major issue with the use of a single threshold in
Conclusions
In this study, we developed a cluster-based method to estimate optimal thresholds and delineate the urban extent from DMSP/OLS NTL data. In this method the optimal threshold for each potential urban cluster is estimated based on urban cluster size and overall nightlight magnitude in the cluster using a logistic model, resulting in thresholds specific to each urban cluster. The derived optimal thresholds are not highly sensitive to the parameter choices in the logistic model when these
Acknowledgment
We acknowledge funding support from the NASA ROSES Land-Cover/Land-Use Change Program (NNH11ZDA001N-LCLUC) with additional support for Steven J. Smith from the Global Technology Strategy Project. We would like to thank Dr. Benjamin Bond-Lamberty and the anonymous reviewers for their constructive comments and suggestions, and the many colleagues and organizations that shared the data used in this project.
References (54)
- et al.
Estimating population and energy consumption in Brazilian Amazonia using DMSP night-time satellite data
Computers, Environment and Urban Systems
(2005) Object based image analysis for remote sensing
ISPRS Journal of Photogrammetry and Remote Sensing
(2010)- et al.
A SVM-based method to extract urban areas from DMSP-OLS and SPOT VGT data
Remote Sensing of Environment
(2009) - et al.
A comparison of three image-object methods for the multiscale analysis of landscape structure
ISPRS Journal of Photogrammetry and Remote Sensing
(2003) - et al.
The consequences of urban land transformation on net primary productivity in the United States
Remote Sensing of Environment
(2004) - et al.
Extracting the dynamics of urban expansion in China using DMSP-OLS nighttime light data from 1992 to 2008
Landscape and Urban Planning
(2012) - et al.
Regional mapping of human settlements in southeastern China with multisensor remotely sensed data
Remote Sensing of Environment
(2008) - et al.
Strategies for image segmentation combining region and boundary information
Pattern Recognition Letters
(2003) - et al.
Modeling energy consumption and CO2 emissions at the urban scale: Methodological challenges and insights from the United States
Energy Policy
(2010) - et al.
Mapping global urban areas using MODIS 500-m data: New methods and datasets based on ‘urban ecoregions’
Remote Sensing of Environment
(2010)