Defining City Boundaries: How Data Choices Shape Our Perception of Cities

Introduction: The "Uncertainty" Problem of Urban Boundaries

In today's era of accelerated global urbanization, we face a paradox: the more precise urban data we need, the harder it becomes to define where a city's boundaries lie. Census tracts, administrative jurisdictions, sprawling built-up areas, commuting zones, nighttime light contours—each delineation method produces a different portrait of the city. A 2026 perspective article published in Nature Cities (Van Migerode et al., 2026) systematically reviews proxy datasets used to identify urban boundaries and points out that choosing one proxy over another is, in essence, a declarative statement about what constitutes a "city."

This issue is by no means an academic exercise confined to the ivory tower. The United Nations World Urbanization Prospects, global city competitiveness rankings, infrastructure investment decisions, and climate change adaptation planning—all rely on a unified or comparable demarcation of urban boundaries. When different datasets classify the same area as "urban" in one case and "non-urban" in another, cracks appear in the foundation of comparative research.

Three Mainstream Proxies: Population, Built-Up Area, and Nighttime Lights

The most commonly used proxies for urban boundaries currently fall into three categories. The first is based on population density and population distribution grids (e.g., WorldPop, GPW), premised on the assumption that "cities are spaces of high population concentration." The second leverages remote sensing data on built-up areas (e.g., MODIS, GlobeLand30), assuming that "cities are areas continuously covered by buildings." The third relies on nighttime light intensity (e.g., DMSP/OLS, NPP-VIIRS), assuming that "cities are spaces illuminated by concentrated economic activity and infrastructure."

Each proxy has unique strengths and blind spots. Population data can reflect human distribution but is easily affected by administrative boundaries and census timeliness; built-up area data are physically intuitive but struggle to distinguish industrial warehouses from residential communities; nighttime lights capture economic activity but suffer from saturation effects and "light spillover," underestimating the true urbanization patterns in low-density regions such as Africa. As shown in research, different definitions of urban extent in assessing the urban heat island effect across 892 cities can significantly alter estimates of land surface temperature (Yang et al., 2023).

From Methodological Choice to Conceptual Framework

The core contribution of Van Migerode et al. is their call for researchers to link proxy selection with the "conceptualization of the city." In other words, before deciding which data to use, one must first clarify: which dimension of the city are we actually measuring?For example, if the research subject is the impact of urban air quality on health, then "city" might be defined as an area with continuous high population density; if the study is on the global urban economic network, nighttime light or commuting data could better reflect functional urban areas (FUA). When measuring rural-to-urban transformation, built-up area expansion may be more closely related to land-use change. No single proxy is universally applicable to all scenarios—this is precisely the "conceptual drift" that urban science has long faced.

This perspective elevates the definition of urban boundaries from a technical operation to an epistemological issue. As geographer Louis Wirth stated in his classic 1938 essay, a city is not only a physical form but also a way of life (Wirth, 1938). The choice of proxy datasets ultimately reflects our implicit judgment of this "urbanism."

The Data Gap in Global City Comparisons

Differences in proxy selection are particularly pronounced between the Global South and the Global North. In developed countries, urbanization patterns are typically compact, high-density, and covered by an extensive power grid, so nighttime light and population density data align well. However, in sub-Saharan Africa, South Asia, and other regions, a large number of urban areas exhibit low-density sprawl, inadequate infrastructure, and widespread informal settlements. Defining urban boundaries solely by nighttime lights would systematically underestimate urbanization levels in these regions.

The OECD and SWAC (Sahel and West Africa Club) report "Africa's Urbanisation Dynamics 2020" directly addresses this issue by employing a more refined "Africapolis" methodology, which redefines African cities based on built-up areas and high-resolution population data. The results show that Africa's urbanization rate is much higher than traditional estimates. This not only corrects data bias but also reshapes our understanding of global urbanization patterns—urban growth in the Global South is far more dynamic than we previously thought.

Toward Purpose-Driven Data Selection

The "purpose-driven" framework proposed by Van Migerode et al. represents a pragmatic and responsible response. Researchers need to explicitly state: Why was a particular proxy chosen? How does it correspond to the core concept of the study? How do its limitations affect the conclusions? At the same time, new methods such as cross-validation with multiple data sources and machine learning fusion edge detection (e.g., Arribas-Bel et al., 2021 using XGBoost to identify urban boundaries) offer possibilities to bridge the gaps between different proxies.

However, technological advances cannot fully resolve conceptual disagreements. The city itself is a multifaceted spatial entity—simultaneously an administrative unit, an economic basin, and a container of social life. Attempting to lock the city within a single unified boundary is like asking a map to accurately display elevation, rainfall, and political divisions all at once. A better approach might be to acknowledge that every map serves a specific purpose and to make its cartographic rules transparent.

Conclusion: Boundaries as PerspectivesThe debate over delineating city boundaries is fundamentally a contest of urban perceptions. With the emergence of novel data sources such as satellite remote sensing, mobile phone signaling, and social media check-ins, we now have richer "measurement" tools than ever before—but we must also be wary of the unconscious tyranny of data tools. A proxy dataset may be widely used not because it is the most conceptually appropriate, but because it is the most accessible or the most classic.

Future urban research should no longer avoid the assumptions behind methodologies. Advocating for transparency and conceptual consistency is not merely a requirement of academic rigor, but also a necessary prerequisite to ensure that urban policies are grounded in the real world. Because when we draw the line of a city boundary, we are also defining who is included, who is excluded, and how we understand the most complex artificial system on this planet.

*This article is based on the perspective paper "Proxy datasets to identify city boundaries" by Céline Van Migerode, Ate Poorthuis & Ben Derudder, published in Nature Cities (2026).*References

• Van Migerode, C., Poorthuis, A. & Derudder, B. Proxy datasets to identify city boundaries. Nat. Cities (2026).
• Yang, Q. et al. Influence of urban extent discrepancy on the estimation of surface urban heat island intensity. J. Clean. Prod. 426, 139032 (2023).
• Wirth, L. Urbanism as a way of life. Am. J. Sociol. 44, 1–24 (1938).
• OECD/SWAC. Africa’s Urbanisation Dynamics 2020: Africapolis, Mapping a New Urban Geography (OECD, 2020).
• Arribas-Bel, D., Garcia-López, M.-À. & Viladecans-Marsal, E. Building(s and) cities: delineating urban areas with a machine learning algorithm. J. Urban Econ. 125, 103217 (2021).