The City That Left: How Urban Populations Respond When Heat Arrives

On the hottest days of summer 2024, something unexpected happened across Germany's ten largest cities: people disappeared. Not dramatically, not visibly—but measurably, in the way a citywide body count can be measured. When temperatures climbed and stayed high for three or more consecutive days, urban presence dropped by 1.5 percentage points below what the calendar would have predicted on an ordinary week. No major sporting event, no public holiday, no school vacation could explain it. The heat itself was sending people elsewhere—outward to lakes and forests, inward to air-conditioned homes, or simply away from streets that had become inhospitable.

But here's what makes the pattern worth studying: the people who remained weren't distributed randomly. They clustered around swimming pools, beer gardens, and lakeside cafés—places where heat becomes tolerable, even enjoyable. Meanwhile, activity around schools, clinics, and government offices either held steady or declined. The city wasn't just getting quieter during heat waves; it was reshaping itself around what heat makes bearable.

This is the finding at the center of new research from Dominique Geissler, Felix Creutzig, Ramit Debnath, and Stefan Feuerriegel, published on arXiv in July 2026. Using mobile phone location data from more than 100 million anonymized observations across Berlin, Hamburg, Munich, and seven other German cities, they tracked where people were—hour by hour, block by block—during a summer that ranged from unremarkably warm to genuinely oppressive. Their resolution was extraordinary: grid cells of 100 meters by 155 meters, covering approximately 160,000 distinct locations. This isn't aggregate city data or survey estimates. This is the actual, granular fingerprint of human presence across an entire urban system.

The implications extend well beyond Germany. As climate change intensifies, cities everywhere will face more frequent and severe heat. Understanding how urban populations respond—not just in health outcomes but in daily behavior—has become essential for anyone designing cities that can function under new thermal regimes. The difference between a resilient city and a dangerous one might not be just how hot it gets, but where people go when it does.

The Science

Measuring the Invisible

For decades, researchers studying how people respond to heat relied on surveys, transportation records, or health data. Surveys capture what people say they do, not what they actually do. Bus and subway ridership tells you about one slice of urban mobility but misses everyone walking, cycling, or staying put. Hospital admissions reveal consequences but not behavior.

Mobile phone location data has emerged as a powerful alternative—not a perfect one, but far richer than anything available before. When your phone communicates with cell towers, it leaves a timestamped breadcrumb. Aggregated across millions of devices and processed with privacy protections, these breadcrumbs become a detailed map of where populations concentrate, how they move, and when they appear in different places. The resolution has improved dramatically over the past decade. In this study, the researchers worked with hourly counts at the level of individual city blocks—small enough to capture the difference between a park and the street beside it, but large enough to protect individual privacy.

The data came from a major telecommunications provider in Germany, which anonymized and aggregated the information before transferring it to the research team. No individual identifiers, no movement trajectories, no personal information—just hourly counts of detected devices within each grid cell. The entire pipeline complied with GDPR requirements, and the data covered the period from May 13 to July 12, 2024. That window captured everything from crisp spring days to a summer heat that pushed temperatures above 33°C in several cities.

The researchers paired this activity data with two other datasets. First, hourly weather observations from the Deutscher Wetterdienst (DWD), Germany's meteorological service, which provided temperature readings for each city. Second, approximately 126,000 points of interest (POIs) from OpenStreetMap—restaurants, parks, schools, swimming pools, museums, and dozens of other location types—categorized into six broad groups: Food & Drink, Retail & Shopping, Public Service, Green Spaces, Sport Facility, and Tourism & Culture. By linking activity counts to POI locations, they could ask not just whether people were in the city, but where specifically they were congregating.

The ten cities chosen for the study represent a meaningful cross-section of German urbanism. Berlin, with 3.76 million residents, anchors the sample. Hamburg and Munich each exceed 1.5 million. Cologne, Frankfurt, and Stuttgart fall between 600,000 and 1.1 million. The remaining four—Düsseldorf, Leipzig, Dortmund, and Gelsenkirchen—are smaller but capture different urban forms and regional contexts. Together, these cities account for 11.76 million people, roughly a seventh of Germany's entire population.

Defining Heat

The study uses a definition of "hot day" that is both technically precise and intuitively meaningful: any day when the maximum temperature reaches or exceeds 25°C. This is the German Meteorological Service's official definition of a summer day. "Sustained hot periods" are defined as three or more consecutive summer days—a threshold that matters because consecutive heat creates cumulative physiological stress. Single hot days are uncomfortable; a week of them becomes dangerous.

During the observation window, there were 109 hot-day observations across the ten cities and 19 distinct sustained hot periods, spanning 89 days in total. Hamburg and Frankfurt experienced the most intense sustained heat, with mean daily maximum temperatures reaching 29.6°C. Berlin's sustained periods were milder on average, topping out around 25.7°C. These are not extreme events by the standards of future climate projections, but they are representative of the kind of heat that European cities increasingly experience each summer.

The Analytical Approach

The researchers used several complementary methods. For the overall relationship between temperature and city-wide activity, they examined how daily totals changed across temperature bins: below 20°C, 20-25°C, 25-30°C, and above 30°C. They tested whether differences between bins were statistically significant using standard t-tests. For the effect of sustained heat periods specifically, they built a predictive model of expected presence based on calendar controls—workday status, public holidays, school holidays, and a dummy variable for the UEFA Euro 2024 championship, which took place in Germany from June 14 to July 14 and measurably boosted urban presence in host cities. They then compared observed presence to model predictions, creating an "activity index" where 100 represents exactly expected presence, 102 represents 2% above expected, and so on.

For the POI analysis, they ran fixed-effects regressions at the grid-cell-hour level, relating activity counts to temperature bins, POI categories, and their interactions. The interactions are the key: they capture whether grid cells near a particular type of place (say, a swimming pool) show relatively more or less activity as temperatures rise, compared to otherwise similar grid cells without that amenity. The models controlled for regular temporal patterns, city-wide activity levels, and the same calendar factors used in the activity index analysis.

What They Found

The Curvilinear Curve

The most fundamental finding is one that contradicts simple intuitions about heat and behavior. Most people assume that hotter weather means more outdoor activity—more time in the park, more walking, more general city life. The data reveal a more complicated, nonlinear relationship.

Across all ten cities combined, urban activity increased as temperatures rose from cool to warm, peaked somewhere between 25°C and 30°C, and then plateaued or declined above 30°C. Days in the 25-30°C range showed significantly higher activity than cooler days. But beyond that threshold, the gains stopped. The difference between a 30°C day and a 33°C day, in terms of observed urban presence, was negligible—suggesting that very high temperatures don't draw more people outside; they drive people inside or out of the city entirely.

The statistical evidence for this pattern is robust. In comparisons of mean activity per grid cell, the difference between days below 20°C and days in the 25-30°C range was statistically significant at conventional levels (t = -2.579, p ≤ 0.05). The same was true for the difference between 20-25°C days and 25-30°C days (t = -2.302, p ≤ 0.05). For total city activity, the patterns were even stronger, with significance at p ≤ 0.001 for the coolest versus warmest non-extreme category.

Below is a chart showing how city-wide activity relates to temperature across all ten cities. The pattern reveals that moderate warmth increases presence, but extreme heat does not continue to draw people outward—instead, something appears to switch.

City-wide activity increases with moderate warmth but plateaus at extreme heat

Label	Value
<20°C	96
20-25°C	98
25-30°C	100
>30°C	100

Source: Geissler et al., 2026. Activity index measures observed divided by expected presence times 100. Values above 100 indicate above-expected presence.

Workdays Versus Weekends

The relationship between temperature and activity depended heavily on whether it was a workday.

On weekdays, when commuting, school schedules, and fixed work routines structure behavior, temperature had a weak and inconsistent effect. The pooled estimate across all cities was positive but not statistically distinguishable from zero. In several cities—Berlin, Düsseldorf, Gelsenkirchen, and Leipzig—a 1°C increase in midday temperature was actually associated with a small decrease in urban activity, ranging from 0.6% to 1.0% per degree. The interpretation is straightforward: when you're obligated to be somewhere (office, school, factory), you go there regardless of whether it's 24°C or 30°C outside. Thermal discomfort reduces discretionary movement but cannot easily cancel non-discretionary movement.

On non-workdays, the story changed dramatically. The pooled estimate showed a 6.1% increase in city-wide activity per 1°C rise in midday temperature—roughly ten times the weekday effect and statistically significant. This makes intuitive sense: on a Saturday or Sunday, if the weather is pleasant, you go out. If it's oppressive, you stay home. The temperature is no longer background context but a primary driver of the decision itself.

Leipzig was a striking exception to this pattern. On non-workdays, Leipzig showed a significant negative relationship between temperature and activity—more heat meant fewer people in the city, even on weekends. The researchers note this heterogeneity but do not fully explain it. Leipzig has a large student population, and university calendars may have influenced behavior. It also has extensive green spaces nearby, and residents may flee the city for lakes and forests more readily than residents of other German cities.

Weekend behavior responds to heat; weekday behavior holds steady

Label	Value
Berlin	-0.6 %
Düsseldorf	-0.7 %
Gelsenkirchen	-1 %
Leipzig	-1.1 %
Hamburg	-0.01 %
Munich	0.3 %

Source: Geissler et al., 2026. Coefficients from OLS regressions of logged city-wide activity on midday temperature. Values represent percentage change in activity per 1°C increase.

Sustained Heat Changes Everything

The sharpest finding in the paper concerns sustained hot periods—the three-day-or-more stretches that the meteorological literature identifies as the threshold for elevated health risk.

During these sustained periods, observed city-wide presence fell below expected levels by an average of 1.5 percentage points. The deviation was statistically significant at p < 0.001, and it held across the majority of hot-period days: 60 out of 89 days showed below-expected presence after adjusting for workdays, public holidays, school holidays, and the Euro 2024 effect. The activity index, which measures observed divided by expected presence times 100, averaged 98.45 during sustained hot periods compared to 100.23 on milder days.

This is not a trivial finding. When a city experiences sustained heat, people are not simply staying home or altering their routes—they are leaving the city entirely, or reducing their outdoor presence so substantially that mobile networks register meaningfully fewer devices in urban grid cells. For city managers, this represents a measurable behavioral signal that corresponds to the physiological reality of heat stress. Consecutive days of high temperature create cumulative strain. The body cannot recover overnight if nighttime temperatures remain elevated. People respond to this reality, even when they cannot articulate it: they minimize exposure.

Where People Go

But the city doesn't empty completely. The POI analysis reveals where the remaining activity concentrates.

The clearest pattern: during hot weather, people gravitate toward leisure and culture-oriented amenities. Grid cells characterized by tourism and culture POIs—museums, galleries, stadiums, attractions—showed up to 3.86% higher activity on days above 30°C compared to cooler days, relative to otherwise comparable grid cells without those POIs. Food and drink venues (cafés, restaurants, bars) showed modest but consistent increases. Retail and shopping showed similar patterns, with increases of up to 1.41%.

The researchers describe these as "third places"—locations beyond home and work where people spend discretionary time. The logic is intuitive: when the choice is "what do I do today?" and the weather is hot, you're more likely to seek out a lakeside beer garden than to wander aimlessly through a commercial district. The city becomes a menu of climate-controlled or climate-adapted options.

By contrast, activity around public service environments—schools, clinics, hospitals, government offices—showed weaker or negative shifts during heat. These are places people go for essential functions, not leisure. They cannot easily be avoided when needed, but they also aren't destinations that people actively seek during hot weather. The pattern suggests that public service environments are thermal environments people endure rather than seek out during heat.

Green spaces and sport facilities showed heterogeneous patterns that defied simple summary. In Munich, grid cells near green spaces showed roughly 5% higher activity in the hottest temperature bin compared to cooler conditions—suggesting that Munich's parks and gardens serve as genuine heat refuges. In Hamburg, the same analysis showed near-zero or negative associations, suggesting that Hamburg's green spaces don't function as heat refuges in the same way, perhaps because of local accessibility, design, shading, or the specific types of green space captured in the OpenStreetMap data.

Heterogeneity Across Cities

The city-by-city variation in these patterns is one of the study's most important contributions—and one of its most underappreciated challenges.

Berlin, Germany's largest city, showed the most consistent temperature responses. Weekday activity declined with rising temperature (about 0.6% per degree), while non-workday activity increased (about 0.57% per degree). Munich and Stuttgart showed positive weekday responses, suggesting that work routines in those cities allow for more temperature-driven flexibility—perhaps because outdoor work is more common, or because air-conditioned office environments make heat less of a deterrent. Leipzig's unusual monotonic decline in activity with temperature—more heat, fewer people, on all days—suggests either a different urban morphology, a different population demography, or a different relationship to nearby green space and recreational destinations.

These differences matter enormously for policy. A heat adaptation strategy that works in Munich may not translate to Leipzig. The same green space that draws people in Frankfurt may be irrelevant in Hamburg. The paper's value lies not just in identifying aggregate patterns but in revealing the granularity beneath them.

Why This Changes Things

Beyond Temperature Maps

Urban heat adaptation has historically relied on temperature maps—spatial visualizations of where cities are hottest. These are valuable tools. They identify the urban heat island effect, where dense building coverage traps and radiates heat; they highlight neighborhoods with limited tree canopy; they reveal which blocks lack reflective surfaces or ventilation corridors. Cities including Barcelona, Paris, and Rotterdam have used such maps to guide tree-planting programs, cool roof mandates, and public space redesign.

But temperature maps capture only half the problem. They tell you where the heat is. They do not tell you where the people are when the heat arrives. A park that appears to offer cooling potential on a GIS overlay may be inaccessible to the residents who need it most. A clinic that sits in a heat island may see reduced attendance during heat waves if patients defer non-urgent visits. A commercial street that seems ideally positioned for pedestrian activity may empty entirely on very hot days, undermining the business case for shading investments.

This study demonstrates that mobile phone location data can fill that gap. It provides a real-time, fine-grained picture of where populations actually are during heat events—not where residential censuses say they live, not where pedestrian counts were conducted on mild days, but where devices are actually detected as temperatures fluctuate.

For urban planners, this is a fundamentally new kind of evidence. It allows them to ask: during last summer's heat wave, where were the crowds? Which neighborhoods retained activity, and which emptied? Which types of amenities drew people in, and which became thermal liabilities? The data to answer these questions now exist, at least in cities with sufficient mobile network coverage and data partnerships.

The Third Place Imperative

The finding that leisure and culture amenities attract activity during heat has implications for urban design that extend well beyond climate adaptation.

"Third places" is a term coined by sociologist Ray Oldenburg to describe the informal gathering spaces—cafés, pubs, parks, barbershops—that sustain community life outside of home and work. These spaces are not destinations of necessity but of desire. People choose to be there, and that choice is sensitive to comfort. During heat, comfort becomes a scarce resource, and the spaces that provide it become more attractive.

The researchers' finding that tourism, culture, food, and drink venues show increased activity during heat is consistent with this framework. These are precisely the third places that people seek when heat makes home uncomfortable and work obligations are absent. They are also, critically, the spaces where climate adaptation infrastructure—shading, misting systems, water features, ventilation—would have the highest behavioral return. Installing a shade canopy over a café terrace during a heat wave isn't just making a business more comfortable; it's concentrating an existing behavior that people are already motivated to perform.

The corollary is equally important: public service environments—schools, clinics, government offices—show weaker heat-related attraction. These are places where people have limited choice about being present. They are often thermally poor—older buildings with limited air conditioning, rooms facing west with afternoon sun exposure, waiting areas designed for throughput rather than comfort. The finding that these environments do not draw additional activity during heat is not surprising. The implication—that targeted cooling, shading, and hydration infrastructure in these spaces could improve attendance and reduce heat-related disruptions—is significant.

The Behavioral Cliff at 30°C

The curvilinear relationship between temperature and activity—one of the paper's most striking findings—deserves particular attention.

Most heat-related policy and research assumes a monotonic relationship: hotter weather means more outdoor activity until some physiological limit is reached. This study suggests a more threshold-driven dynamic. Up to a point—roughly 25-30°C—warmer weather is associated with more urban activity. Above that threshold, the relationship flattens or reverses. People are not drawn further outside as temperatures climb from comfortable to dangerous; they withdraw.

This has implications for how cities should interpret heat warning systems. Current heat-health warning frameworks, including those used in Germany, typically trigger public communications when temperatures exceed certain thresholds. The assumption is that people respond to warnings by altering behavior. This study suggests that behavior may already be altering itself before warnings are issued—that populations are making independent judgments about thermal comfort that roughly track the same thresholds that health authorities use.

The challenge is that the withdrawal is partial, uneven, and potentially inequitable. Not everyone can leave the city on hot days. Not everyone has air conditioning. Not everyone can reschedule discretionary activities. The people who remain in urban centers during sustained heat may be precisely those with the fewest alternatives—the elderly, the low-income, the socially isolated, the essential workers whose jobs cannot be done remotely. This is the population most at risk of heat-related illness and death. Understanding their location during heat events—not as a matter of survey response but as a matter of observed device presence—is essential for targeted outreach and resource allocation.

Leipzig as Warning

The case of Leipzig deserves special consideration because it breaks the study's primary narrative.

In every other city in the sample, non-workday activity increased with temperature up to the 25-30°C range. In Leipzig, it declined monotonically. More heat meant fewer people in the city, on weekends as well as weekdays. This is the pattern you'd expect if residents of Leipzig have easy, attractive alternatives outside the city—a lake, a forest, a regional green space—and use them aggressively when conditions permit.

If that's the correct interpretation, Leipzig's pattern may represent a kind of adaptive capacity that other cities lack. When heat arrives, Leipzig's residents vote with their feet. They don't endure the city; they leave it. This reduces heat exposure at the individual level but may simply redistribute the problem to surrounding areas, creating crowding and pressure at green spaces and waterfront destinations.

Alternatively, Leipzig's pattern may reflect cultural or demographic factors that the data cannot capture. A younger, more mobile population may be more likely to flee heat. A city with a weaker "third place" culture may have fewer attractions that keep people in town despite discomfort. The researchers do not definitively resolve this question, and it's one that deserves further study.

What's Next

Extending the Window

The study covers just two months—May 13 to July 12, 2024. This is a meaningful window for German summer conditions, capturing both the tail end of spring and the peak of early summer heat. But it leaves open questions about later summer months, when heat is typically more severe and sustained, and about inter-annual variation. The summer of 2024 included both mild periods and significant heat events, but a single season cannot establish the full range of conditions that German cities will face as climate change progresses.

Extending this analysis across multiple years and across different climate regions would strengthen the findings considerably. Do the same behavioral patterns hold in hotter summers? In cooler ones? Do cities with different urban forms—more green space, more waterfront access, different building densities—show systematically different responses? The methodology demonstrated in this paper could be applied to answer these questions, given sufficient data access and institutional partnerships.

The POI Data Question

The study's use of OpenStreetMap POI data is simultaneously one of its greatest strengths and one of its most significant limitations.

POIs capture what exists in the urban landscape—the park, the clinic, the restaurant. They provide a rich typology of urban environment that can be linked to observed behavior. But POIs are not equally distributed across all urban contexts. They may underrepresent informal or temporary uses of space: street vendors, pop-up events, informal gatherings under shade trees. They may be more complete in some cities than others, depending on community editing activity and local mapping culture. And they capture spatial location but not thermal quality—the POI for a park doesn't tell you whether it has mature trees, water features, or open lawn with no shade.

The researchers acknowledge this limitation and note that green spaces showed highly heterogeneous associations with heat-related activity. Some cities (Munich) showed strong positive associations; others (Hamburg) did not. This variation may reflect real differences in how green spaces function as heat refuges, or it may reflect differences in how OpenStreetMap maps green space in different cities. Disentangling these explanations would require more detailed data on the physical characteristics of green spaces—their canopy coverage, water features, and thermal measurements—combined with the activity data used here.

Privacy and Scale

The study's methodology depends on access to mobile phone location data from a major telecommunications provider. This is a significant logistical and ethical undertaking. The data used here were anonymized and aggregated before transfer, following GDPR requirements, and the research team accessed only hourly counts at the grid-cell level without individual identifiers. These are appropriate safeguards.

But the approach raises questions about scalability and equity. Not all cities have telecommunications providers willing to share data with researchers. Not all cities have the legal frameworks, data infrastructure, or research partnerships to enable this kind of analysis. The result may be that the most data-rich, evidence-based heat adaptation planning is concentrated in cities that already have advantages—large populations, strong research institutions, provider relationships—while smaller cities and lower-income regions remain dependent on coarser data and simpler analyses.

Addressing this gap will require deliberate effort. Open data standards for urban activity monitoring, anonymization pipelines that allow cross-city comparison, and partnerships between providers, researchers, and municipalities could broaden access. The researchers' demonstration that fine-grained behavioral data can inform heat adaptation planning is a case for investing in these infrastructure elements.

The Health Translation Problem

The study measures activity, not health outcomes. It tells us where people are during heat, not how heat affects their bodies. Connecting behavioral change to health consequences is the next frontier.

The logic is straightforward in outline: when people leave cities during heat, they reduce exposure to urban heat islands but may also lose access to air conditioning, medical care, or social support networks that cities provide. When people concentrate around certain amenities, they may increase crowding and competition for cooling resources. When public service activity declines, appointments are missed and preventive care is deferred. Each of these shifts has health implications, but tracing them requires linking behavioral data to health records—an undertaking that raises privacy concerns of its own.

Some researchers are already exploring this direction. Studies in the United States have linked mobile phone mobility data to ZIP-code-level health outcomes during heat events. Similar approaches in Germany would face different legal constraints but could yield valuable insights into how behavioral adaptation mediates the relationship between heat exposure and health risk.

From Description to Prescription

The study is careful to describe rather than prescribe. It documents how urban activity changes during heat; it does not argue for specific interventions. This is methodologically appropriate but practically incomplete. The gap between "we now know where people are during heat" and "here is the policy that should follow" requires an additional layer of analysis that this study does not provide.

But the direction is clear. If leisure and culture venues draw activity during heat, then shading and cooling investments in those locations are likely to have high utilization. If public service environments show weak heat attraction, then improving their thermal quality may improve attendance and reduce heat-related disruptions to essential services. If sustained heat periods are associated with measurable departures from the city, then emergency planning should account for reduced urban presence and the implications that has for service delivery and communication.

Each of these implications could be the subject of further study. The research team has demonstrated a methodology; the policy community now has an invitation to build on it.

The Bigger Picture

Heat is not a future problem. For cities across Europe, South Asia, the Middle East, and beyond, the summers of the 2020s already include conditions that were projected for mid-century under earlier climate scenarios. Adaptation is not a theoretical exercise; it is an operational necessity.

The finding that cities measurably empty during sustained heat—that populations vote with their feet when conditions become uncomfortable—has implications that extend beyond climate adaptation. It speaks to the relationship between cities and the natural environments that surround them. The green spaces, forests, lakes, and coastal areas that draw people out of cities during heat are not separate from urban infrastructure; they are part of it, even when they fall outside municipal boundaries. Regional coordination, not just city-level planning, may be essential for managing heat exposure.

It also speaks to the unevenness of heat's impact. The people who leave cities during heat are exercising a form of thermal privilege—the ability to relocate, to defer activities, to avoid exposure. The people who remain are often those without that privilege: essential workers, the elderly, the poor, the socially isolated. Their exposure is not a behavioral choice but a structural condition. Understanding where they are during heat events, as this study demonstrates, is the first step toward protecting them.

Cities have always been shaped by climate. The heat island effect is not new; urban form has evolved over centuries in response to wind patterns, precipitation, and seasonal temperature. What is new is the pace and intensity of change. Climate adaptation is not about returning to a previous thermal equilibrium; it is about building cities that can function under conditions that have no modern precedent in most locations.

The behavioral data analyzed in this study offers a map—not of temperature, but of response. It shows how populations already behave under heat stress, where they cluster, where they retreat. This is not a complete adaptation plan, but it is the evidence base that adaptation planning has been missing. The gap between knowing where the heat is and knowing where the people are during the heat is finally, for at least ten German cities, closed. The next step is to use that knowledge.