A new urban landscape in East–Southeast Asia, 2000–2010

East–Southeast Asia is currently one of the fastest urbanizing regions in the world, with countries such as China climbing from 20 to 50% urbanized in just a few decades. By 2050, these countries are projected to add 1 billion people, with 90% of that growth occurring in cities. This population shift parallels an equally astounding amount of built-up land expansion. However, spatially-and temporally-detailed information on regional-scale changes in urban land or population distribution do not exist; previous efforts have been either sample-based, focused on one country, or drawn conclusions from datasets with substantial temporal/spatial mismatch and variability in urban definitions. Using consistent methodology, satellite imagery and census data for >1000 agglomerations in the East–Southeast Asian region, we show that urban land increased >22% between 2000 and 2010 (from 155 000 to 189 000 km2), an amount equivalent to the area of Taiwan, while urban populations climbed >31% (from 738 to 969 million). Although urban land expanded at unprecedented rates, urban populations grew more rapidly, resulting in increasing densities for the majority of urban agglomerations, including those in both more developed (Japan, South Korea) and industrializing nations (China, Vietnam, Indonesia). This result contrasts previous sample-based studies, which conclude that cities are universally declining in density. The patterns and rates of change uncovered by these datasets provide a unique record of the massive urban transition currently underway in East–Southeast Asia that is impacting local-regional climate, pollution levels, water quality/availability, arable land, as well as the livelihoods and vulnerability of populations in the region.


Introduction
We have entered the urban era: cities now form the basis of the human experience for the majority of the Earth's population (UN 2012). Cities today must meet the needs of growing populations and expanding economies, while at the same time minimizing their environmental impacts (Grimm et al 2008, Montgomery 2008. Expansion of built-up land is often the most direct environmental impact associated with urban growth, with far-reaching implications for climate, hydrology, and biogeochemical cycles that extend beyond municipal boundaries (Seto et al 2010). While remote sensing has proven especially useful for characterizing broad-scale land changes, detailed monitoring of urban land use change remains costly and challenging due to the highly heterogeneous nature of cities, the spectral similarity between new urban land and other land cover types, and the lack of cloud-free data in locations where estimates are most needed (e.g. tropics, Mertes et al 2015). As a result, there has been little information on the building boom that is accompanying population growth in many developing countries (China, India, etc) other than case-study analysis of individual cities Woodcock 2008, Angel et al 2011), or country-level assessments (Liu et al 2005, Wang et al 2012. Comparing urban populations has also been notoriously difficult due to differences in census timing, data availability/quality, and most critically, the considerable variability in how cities are defined, whether by population threshold, functional area, or administrative boundaries (Cohen 2004). One of the few studies reporting transnational urban land and population trends concluded that cities are universally spreading out and declining in density (Angel et al 2005). While there is evidence to contradict this in East Asia (Murakami et al 2005, Bagan andYamagata 2012), there has been no systematic way to compare trends across cities, nations, or regions.
To describe urban trajectories across East and Southeast Asia 10 systematically, we characterize urban extent and urban expansion 2000-2010 using Moderate Resolution Imaging Spectroradiometer (MODIS) satellite observations (Mertes et al 2015). In these maps, urban land refers to places dominated by the 'built environment', which includes all non-vegetative, human-constructed elements (e.g. roads, buildings) with >50% coverage of a landscape unit (here, a 250 m pixel). We synthesize this information with population density maps developed using demographic data at the finest administrative unit available and empirically-tested population-land cover relationship-based methods (Tatem et al 2007). To address the issue of comparability, we conduct our analysis of regional urbanization trends using the urban agglomeration as the unit of analysis. We perform a comparative analysis to understand withinnation and across-nation trends in East-Southeast Asia (figure 1, A1) recognizing that such a regional approach cannot account for each city's circumstances or individual drivers/impacts. Our results likely produce a conservative estimate of urban change in the region, and may differ from 'official' statistics (World Bank 2015) as a result of necessary choices regarding definitions, spatial scale, and data sources. Our aim is not to replace national estimates, but to offer a consistent approach for regional comparability of all cities >100 000 in the region.

Background
Great strides have been made to map population distribution using consistent data and methods (Balk et al 2006, Tatem et al 2007, but they depict population as measured at one point in time, and at best adjust only for changing population growth rates at the country level. Similarly, urban maps from remote sensing data have been limited to either static global maps that sacrifice detail to provide areal coverage , or local maps of metropolitan growth that forego coverage in favor of spatial detail (Seto et al 2010). Moreover, of the many global urban maps now available (Elvidge et al 2007, none characterize changes in urban land.
Recent comparative work on urbanization has found a middle ground by focusing on local maps (typically 30 m resolution) for a sample of cities (Seto et al 2011, Taubenbock et al 2012, Schneider and Mertes 2014. These studies point to several key findings: (1) cities in developing countries are consistently smaller or more compact than those in developed nations (Huang et al 2007, Schneider andWoodcock 2008); (2) developing country cities typically undergo some decrease in population density in the core during their development trajectory (Murakami et al 2005); and (3) cities are declining in density (Angel et al 2011. Although they represent advances in our empirical understanding of urbanization, these studies (and their conclusions) suffer from several limitations. Many rely on a limited very city sample (Murakami et al 2005, Taubenbock et al 2012; most exclude small-and medium-sized cities (100 000-1 million) where the majority of new growth is taking place (UN 2012). Definitional issues also jeopardize comparability: nearly all utilize municipal boundaries to clip the city extent, making it nearly impossible to compare trends across places. In addition, administrative divisions often under-bound the built-up extent, so new growth in fringe areas is not captured. Finally, many studies rely only on population or remote sensing, failing to connect the two to provide a complete picture of urban trends meaningful for environmental assessment, land use planning, and regional policy implementation.

Satellite-based maps of urban expansion
To establish potential locations of urban land, the study extent was first established by synthesizing all contemporary city point data (table A1) with a c. 2000 map of urban extent developed from MODIS 500 m data (Schneider et al , 2010. The MODIS 500 m urban extent map has been shown to have the highest locational accuracy of available maps and a zero omission rate for cities globally . Where city points did not align with the MODIS map or vice versa, the locations were manually checked against Google Earth data and adjusted. The final study extent was created by categorizing the identified urban patches into small, medium, and large classes according to their spatial extent and population, and buffering by 5, 25, and 100 km respectively to include potential areas of urban and peri-urban growth (Webster 2002).
Urban expansion 2000-2010 was mapped in two steps, beginning with delineation of the c. 2010 urban extent. A probability surface of urban land was developed from three years of 500 m MODIS imagery (table A2) and training samples for urban and non-urban areas photo-interpreted using very high resolution (VHR) Google Earth imagery (1-4 m resolution). A separate probability surface based on vegetation characteristics of urban and non-urban areas was produced from 250 m MODIS enhanced vegetation index (EVI) data (Tan et al 2011) and integrated with the 500 m probabilities according to Bayes' Rule (Mertes et al 2015).
To detect change, we assume all urban expansion 2000-2010 to be unidirectional and occur within the mapped 2010 urban extent. We again exploit EVI data in a multi-date composite technique (annual maximum for each year, 2001-2010) by stacking all images for classification with a boosted decision tree (Quinlan 1993) to map (a) stable urban areas; and (b) areas that were developed 2000-2010. This approach relies on the assumption that any conversion from a non-urban land cover to developed land is detectable through changes in vegetation content (Schneider et al 2010, Mertes et al 2015. The final maps were assessed for accuracy using a two-tiered approach. The 2010 urban map was first assessed using a stratified random sample of 6528 sites 0.132 km 2 in size, and the maps of urban expansion were assessed using a separate random sample of 2086 sites (0.06 km 2 , to align with the 250 m resolution). Test sites were assessed within Google Earth against VHR data in a double-blind assessment by a team of photo-interpretation analysts, and labeled as urban/ non-urban land (tier one), or urban land/urban expansion 2000-2010 (tier two) according to the >50% built-up threshold (note that the 50% threshold is used throughout to maintain consistency with previous urban remote sensing efforts). Overall accuracy measures for the maps were calculated by comparing the maps against the test sites. The results indicate that map accuracies for urban extent (tier one) range between a maximum of 93% to a minimum of 79% for each country, and for urban expansion (tier two), between 91% and 70%, confirming their suitability for this analysis (Mertes et al 2015).

Population density maps
Human population census data and corresponding administrative boundaries at the finest level available were obtained from multiple recent censuses in each nation (table A3). If they did not align with the c. 2000 and 2010 time points, the population data were adjusted forward or backward using inter-censal growth rates and linear estimation methods. High resolution census data were then used to establish population densities for each time point (2000,2010) on a biome-by-biome basis for each land cover type in the region, following previous WorldPop (www. worldpop.org.uk) mapping approaches (Tatem et al 2007, Gaughan et al 2013. These population densities were then used as weights to distribute the population across the raster cells, an approach that has been shown to produce more accurate disaggregations than previous approaches that rely on disaggregation to very coarse data (nighttime lights data) or areal weighting alone , Gaughan et al 2013. After synthesizing all population data with land cover information and built-up extent to map population density, we count only the population cells fully contained within the built-up area. With this approach, we avoid the problems common to urban population data, including the lack of data at disaggregated scales, country-dependent definitions and delineations of urban versus rural that lead to drastically different population estimates, and changes in census geographies that require adjustment so measures reflect true population growth/decline rather than differences due to changes in administrative boundaries (Cohen 2004).

Analysis
We defined an urban agglomeration as the extended area comprising the built-up area of a central place (i.e. a city) and any suburbs or small cities linked by continuous urban land (UN 2012). To delineate agglomerations for this analysis, we collected the most detailed administrative boundary data available (typically county or finer) for c. 2010, to reflect the most recent units used for governing. We then assigned any administrative unit containing part of the contiguous built-up area of the city to its agglomeration, so that the agglomeration boundary is made up of one or more official administrative units. This was repeated for each city >100 000 persons, resulting in 1036 agglomerations across 17 countries (figure A1). For each agglomeration, we estimated the built-up extent for 2000 and 2010 from the satellite-based maps, as well as the 2000 and 2010 urban population within the built-up extent from the population density maps. The 1036 agglomerations were then stratified into five categories (UN 2012) based on their 2010 agglomeration population: >10 million; 5-10 million; 1-5 million; 500 000-1 million; and 100 000-500 000. To understand regional urbanization trends within the 30 largest agglomerations, we also conducted a separate analysis measuring urban expansion for all established cities within the administrative core, within the urban agglomeration defined by the builtup extent, and directly adjacent or near the urban agglomeration boundary (within 120 km of each agglomeration's center, following distance recommendations from the peri-urbanization literature, Webster et al 2002). For this analysis, we followed convention in urban geography, and standardized the study extent for each small city using adaptive radial zones (5, 10, 15 km) based on 2010 population size (Dietzel et al 2005, Seto andFragkias 2005).

Results
4.1. Regional and country-level results Across the region, the total net increase in urban land area was >34 000 km 2 from 2000 to 2010, expanding from 155 000 to 189 000 km 2 . While urban land area increased >22%, urban populations climbed >31%, adding 231 million persons in just ten years (from 738 to 969 million). The rapid pace of population change is clear in the average rates of change for each country (table A4): cities in the region grew annually at 2.8%, with Malaysia, Vietnam, Cambodia, and Laos all observing rates well above this average (4.0-7.8%). In contrast, the rates of change for urban land average 2.0% annually, with only China, the Philippines, Cambodia, and Laos having rates above this level (2.2-3.2%).
The results suggest that urban population growth has outpaced land expansion, a trend we measure explicitly using urban density. Here we estimate persons per square kilometer of built-up land since the conventional measure, persons per square kilometer within an administrative region, does not account for the vastly different sizes of municipal boundaries. While the results show a great degree of variability (figure 2), there are two common trends across nations: (a) urban densities are high (mean 2010, 5850 persons/km 2 ), and (b) urban densities increase 15-30% from 2000 to 2010, adding between 270 and 2020 persons/km 2 in ten years. Although on average, urban densities are decreasing in China (from 6150 to 5290 persons/km 2 across 677 cities), there is considerable variability here as well: roughly half of Chinese agglomerations are decreasing in density, while the remaining half witnessed no change or an increase in urban density, similar to other agglomerations in the region. The growth of these 'mega-agglomerations' is not the whole story, however. The region has an additional 101 large agglomerations, each with populations between 1 and 5 million persons, totaling >207 million. Although rates of expansion in these areas are on par with the 30 large agglomerations (>3%), the average rates of population increase surpass those of the top 30, at >3.4%. These trends are also apparent in agglomerations 100 000-1 million: small cities in Myanmar, Indonesia, Vietnam and the Philippines, especially, have added population without much expansion ( figure 4). Nearly all trajectories are headed in the same general direction, with an average increase of 970 persons/km 2 for the 2000-2010 period.

City-level results: governance and policy perspectives
While the agglomeration provides a consistent way to compare metropolitan areas since they are defined by built-up extent, many agglomerations are comprised of a large number of independently-governed cities. For instance, the Manila agglomeration has 17 cities in its administrative core (where resources and planning are concentrated) and another 15 cities on the outskirts. Alternatively, many large agglomerations have a small core area governed as one unit, with expansion that spills into the jurisdiction of nearby county-or city-level governments (e.g. Shanghai, Seoul, Hanoi). To understand how cities within an agglomeration view and govern themselves, we measure urban expansion for all established cities within 120 km of the city core for the top 30 agglomerations. Here we delineate each core according to its 2010 municipal area, and standardize the size of each small city extent using adaptive radial zones corresponding to each city's 2010 population.
On average, >60% of 2010 urban land and >71% of new development 2000-2010 are located outside the core administrative area, but within the urban agglomeration defined by this study (figure 5). The results also highlight three distinct urban typologies for large agglomerations: (1) a core surrounded by rapidly growing cities, with expansion rates that decline with distance (e.g. Hangzhou, Guangzhou, Chengdu, Jakarta); (2) a core with numerous nearby cities, but with limited growth due to geophysical factors (e.g. Manila, Bangkok, Kuala Lumpur); and (3) a core with few nearby cities (Hanoi, Bangkok). Some of these latter areas are witnessing peri-urbanization (expansion up to 100 km from the core, Kontgis et al 2014), but this trend may not be fully captured in satellite-based estimates or census data due to its small, patchy nature.

Discussion and conclusions
This research presents new evidence that East-Southeast Asia is undergoing unprecedented urbanization and urban expansion, coincident with well-established trends of rapid industrialization, economic growth, and globalization. These results were generated using directly comparable, spatially-detailed datasets derived from multiple sources of remote sensing and disaggregated census data, with close attention to how urban land, urban expansion, urban population and agglomeration boundaries were defined and operationalized. When the factors limiting comparative urban analysis are addressed, the results reveal that urban agglomerations across East Asia are experiencing increasing urban densities. While these trends are not surprising for some scholars and local experts, they do contradict established empirical work that showswith similar attention to consistency in data and definitions but with results modeled using static c. 2000 urban maps-that cities are universally declining in density (Angel et al 2011, 2012).
The trend toward increasing urban densities is clear in nearly all countries, and at multiple scales. At the country level, Japan and South Korea lead the region with highly urbanized populations (80-90%) spread across multiple large urban agglomerations covering 3-5% of each country's land area. Although growth has tapered off in these countries, their aggregate urban densities are still climbing. On average, population growth rates for large, middle-income countries (China, Indonesia, the Philippines, Thailand) are high (3.5%) relative to their average rates of urban expansion (2.6%). Cities of all sizes are growing in these countries, with higher rates of population growth for small cities than for large agglomerations during the last decade. China is clearly a unique case, however. At the country level, Chinese cities appear to be decreasing in density, a result that is expected given the central government's planning and policy initiatives focused on small cities outside major metropolitan areas (Lin 1999). Results at the city level reveal that half of all Chinese cities have urban densities that increase or remain unchanged.
East-Southeast Asia is also home to several low or low-middle income countries with 30% of their total populations living in urban areas, including Vietnam, Myanmar, Laos, and Cambodia. During the last decades, these countries have witnessed major shifts from predominantly subsistence agrarian economies to increasingly commercialized agriculture, leading to rapid urbanization of rural populations (Hall Finally, this work also examines how differences in administrative boundaries and urban definitions impact how we characterize, monitor, and understand urban change. We defined 30 large agglomerations by contiguous urban land, but evaluated rates/amounts of growth within the core administrative area and for the individual cities comprising these agglomerations (figure 5). Most administrative cores contain multiple cities on average, while an additional 2-21 cities exist within the built-up area of the agglomeration, but outside the jurisdiction of the core. It is in these outer cities where the majority of urban growth is concentrated. From these results, we therefore conclude that cities as they are experienced on the ground (i.e. contiguous built-up regions) are often not the same as how they are governed. Given rising urban densities, continued expansion, and a lack of coordinated governance, the question for governments and planners becomes whether adequate services, infrastructure, housing, and employment are available or can be provided to incoming populations.
There are several potential sources of uncertainty in this study that should be noted. With respect to the remote sensing data, the 250 m pixel size combined with the population threshold of 100 000 makes it difficult to capture all small settlements. In China, Indonesia, and Vietnam, villages are spectrally distinct and sufficiently large (>1 km 2 ), and disaggregated population data are available. Accordingly, they are wellmapped with our methods (figure 1). In Laos, Cambodia, and North Korea, villages are small and comprised of local materials that are spectrally similar to surrounding land cover types. These countries have no reliable village population estimates, and consequently, the results may under-report urban land or growth. On average, the total land area and population of these settlements is a fraction of the urban extent and urban population in each country, and should therefore have a limited effect on interpreting the results of this study. Finally, the urban extent does not include low-density settlements (e.g. 30-40% builtup), although these areas may function as urban space. If we relax the 50% threshold, higher rates and amounts of urban land would be likely.
One additional area of uncertainty is related to the availability of population data. Locations with lessthan-ideal data include Malaysia, Thailand, Laos, Myanmar, and North Korea (table A3); results for these countries should be considered in light of this bias. In addition, population estimates have greater uncertainty when the administrative unit is large relative to urban extent, and rural populations within the unit are dense (Hay et al 2005). In these areas (e.g. Indonesia), population densities may be overestimated. Finally, the approach here does not capture growth within existent urban areas, including redevelopment or vertical growth. The lack of within-city monitoring remains a critical limitation of both population data sources and remote sensing for land use planning. New datasets (crowd sourcing, social media, etc) and advances in radar/lidar have the potential to significantly change how we monitor urban change (Frolking et al 2013, Tsou et al 2013. Urban growth has increased in scope, scale, and complexity in recent decades, and has become one of the most important challenges of the 21st century. The urban expansion and urban growth datasets 11 presented here provide a valuable, practical, and consistent way to monitor a broad range of issues, including impacts to local-regional climate (Kaufmann et al 2007), pollution levels (Grimm et al 2008), water quality/availability (McDonald et al 2011), arable land (Lambin and Meyfroidt 2011) as well as the livelihoods and vulnerability of populations in the region (Solecki et al 2011). These datasets are unique in that they represent the first comprehensive mapping of urban expansion and growth for all cities >100 000 in East-Southeast Asia, and they also form the basis of ongoing work to examine land and population trends globally for all cities and agglomerations. While uncertainties may always be present no matter the data source, spatially-and temporally-detailed maps of urban expansion and population growth based on the best available data are nevertheless critical for researchers, urban planners, land managers, and government officials interested in a sustainable urban future.