Urban Heat Islands

As cities grow, they create their own climate. Urban areas become islands of heat, significantly warmer than the surrounding countryside. This is the story of how satellite data reveals this hidden phenomenon and what it means for millions of people.

01

What Are Urban Heat Islands?

Picture a city from above, viewed through a thermal camera. What you'd see is striking: the urban core glows like a hot ember, while the surrounding countryside appears cool and dark. This is an Urban Heat Island—a metropolitan area that experiences significantly warmer temperatures than its rural surroundings.

The temperature difference is most dramatic at night, when urban areas can be 1-3°C warmer on average, and up to 12°C warmer in extreme cases. This isn't just a curiosity—it affects energy consumption, public health, and the daily lives of millions of urban residents.

How Do They Form?

Urban Heat Islands emerge from a simple principle: what cities are made of. Sidewalks, roads, parking lots, and buildings—constructed from asphalt, brick, and concrete—share a critical characteristic. These materials absorb sunlight during the day and release stored heat slowly, particularly at night, creating a persistent thermal blanket over urban areas.

While the Sun shines equally on cities and countryside, urban surfaces behave fundamentally differently. Many hard, dark city surfaces have low albedo—they reflect less sunlight and absorb more energy. Crucially, because these surfaces hold virtually no water, they cannot cool through evaporation the way soil, grass, or trees can. Instead, absorbed energy converts directly to heat, spreading through the air via infrared radiation, convection, and conduction.

The result: a warmer "bubble" over urban areas—sometimes several degrees hotter than nearby suburbs or countryside. The contrast between rocky, impervious city surfaces and natural, water-rich rural landscapes creates one of the most significant human modifications to local climate. Understanding this thermal signature is the first step toward designing cities that remain livable as global temperatures rise.

Genoa Urban Heat Island: Satellite Thermal Imagery

Loading Genoa UHI data from satellite imagery...

Figure 1: Downscaled Land Surface Temperature (LST) map of Genoa from satellite thermal data. Uses Inferno color scale (colorblind-friendly). Warmer areas (yellow/orange) are urban centers; cooler areas (purple/black) are rural/vegetated. Temperature range uses 1-99 percentiles. For methodology, check the Methodology section.

The data tells a clear story: cities are not just collections of buildings and people—they are heat-generating systems that fundamentally alter local climate. As we'll see, this has profound consequences for everything from energy bills to public health.

02

Why Do Urban Heat Islands Happen?

Urban Heat Islands (UHIs) are not accidental. They emerge from a combination of physical, material, and structural factors that systematically alter how cities absorb, store, and release heat. While these factors interact, some play a much stronger role than others.

To understand UHIs, it helps to first outline the main drivers, and then examine the most influential ones in detail.

Main Drivers of Urban Heat Islands

Urban heat islands are primarily caused by:

  • Lack of vegetation
  • Heat-retaining urban materials and surfaces
  • Urban structure and geometry
  • Anthropogenic (waste) heat

Among these, land cover and surface materials play a dominant role in shaping urban temperature patterns.

1. Lack of Vegetation

Vegetation provides cooling through:

  • Shade, which reduces surface heating
  • Evapotranspiration, which dissipates heat into the atmosphere

Urban areas typically contain 20–40% less vegetation than surrounding rural regions, removing this natural cooling mechanism. As a result, surfaces heat up faster during the day and cool down more slowly at night.

2. Urban Materials and Surfaces (Primary Driver)

Built environments are dominated by materials such as concrete, asphalt, and brick. These materials:

  • Have low albedo, absorbing large amounts of solar radiation
  • Store heat efficiently and release it slowly after sunset
  • Remain warm during the night, intensifying thermal discomfort

Land Cover → Heat: Direct Evidence

Figure 2a directly compares land cover and land surface temperature in Genoa. Using the same temperature data and applying different land-cover masks shows that:

  • Vegetated areas consistently exhibit lower temperatures
  • Built-up surfaces are significantly hotter
  • Bare soil and sparse vegetation fall in between

This confirms that surface type, not location or weather alone, is a primary determinant of urban heat.

Loading Genoa land-cover and temperature data...

Figure 2a: Side-by-side comparison of Genoa's land cover and temperature patterns. Land Cover Only: WorldCover classification showing spatial context. Vegetation Heat: Temperature for vegetated areas only. Urban Heat: Temperature for built-up/urban areas only. The same temperature data is used; only the mask changes. This visualization clearly demonstrates that built-up areas are significantly hotter than vegetated areas.

Which Surfaces Contribute Most to Extreme Heat?

Figure 2 further quantifies this relationship using a Sankey diagram linking land cover classes to temperature ranges.

The visualization shows that:

  • Built-up surfaces disproportionately contribute to the hottest temperature class
  • Tree-covered areas are overwhelmingly associated with cooler temperatures

This makes clear that urban materials are not just warmer on average—they dominate extreme heat conditions.

Loading land cover and temperature data...

Figure 2: Sankey diagram showing the relationship between land cover types and temperature classes in Genoa. Flow width represents the number of pixels (area) associated with each land cover–temperature class combination. Built-up surfaces dominate the hottest temperature categories, while tree cover is predominantly linked to cooler conditions.

3. Additional Amplifying Factors

Urban Structure and Geometry

Dense building arrangements create urban canyons that:

  • Trap heat between surfaces
  • Reduce wind flow
  • Limit nighttime cooling

Multiple heated surfaces re-radiate energy toward each other, intensifying warming.

Anthropogenic (Waste) Heat

Cities generate additional heat through:

  • Air conditioning
  • Vehicles
  • Industrial activity

This waste heat directly raises ambient temperatures and reinforces existing heat accumulation.

A Self-Reinforcing Cycle

These factors do not act independently. As cities warm:

  • Cooling demand increases
  • More waste heat is released
  • Nighttime temperatures rise further

This feedback loop explains why UHIs are persistent and difficult to mitigate.

Understanding which factors dominate—particularly land cover and surface materials—is essential for designing effective urban cooling strategies.

Now we understand why cities are hotter. But what does this actually mean for the people who live in them? Let's translate satellite data into human experience.

03

Why does this matter?

The answer is simple: more people live in cities than ever before. In 1960, just over 1 billion people lived in urban areas. Today, that number has grown to nearly 4.7 billion—representing more than half of the world's population. As cities expand and densify, the Urban Heat Island effect intensifies, putting more lives at risk.

Figure 3: Proportional square comparison of global urban population in 1960 vs 2025. The area of each square is directly proportional to the population size, making the dramatic 4.5× growth visually immediate. What housed just over a billion urban dwellers in 1960 must now accommodate nearly 5 billion people. Data: World Bank World Development Indicators.

With billions of people now concentrated in urban environments, the impacts of heat islands are no longer a niche concern—they're a global crisis affecting energy consumption, public health, and quality of life for the majority of the world's population.

What does 10°C actually feel like?

Imagine walking from a tree-lined park into a concrete plaza on a summer afternoon. That sudden wave of heat you feel? That's the Urban Heat Island effect. A 10°C surface temperature difference translates to roughly 3-5°C higher air temperatures—enough to turn a manageable 30°C into a dangerous 35°C. For vulnerable populations—the elderly, children, outdoor workers—this difference isn't just uncomfortable. It can be deadly.

The impacts of Urban Heat Islands extend far beyond uncomfortable summer days. They create cascading effects that touch nearly every aspect of urban life, from energy bills to public health to economic productivity.

The data reveals a stark reality: built-up areas are approximately 10°C hotter than vegetated areas. This temperature difference isn't just a number—it's a force that shapes how we live, how we consume energy, and how we survive extreme heat.

And here's the cruel irony: the hottest places coincide with where people are—areas of highest population density—and where shade is scarce—places with the least green cover. The people who need cooling the most often have the least access to it.

"When temperatures rise, our response is measured not just in kilowatts, but in lives. Every degree of warming triggers a cascade: higher energy demand, more waste heat, increased health risks. The data doesn't lie—summer peaks in electricity consumption directly track summer peaks in temperature. And for those who can't afford air conditioning, those peaks can be fatal."

At a city scale, Urban Heat Islands amplify two major side effects. First, they push up energy consumption as households and businesses rely more on cooling during hot periods. Second, they increase health risks by intensifying heat exposure—especially for vulnerable populations—raising the likelihood of heat stress and heat-related illness.

The numbers tell the story: Higher temperatures drive up air conditioning demand by 20-25%, creating a vicious cycle. More cooling generates more waste heat, which requires more cooling. Let's see exactly when and how this happens.

Italy Monthly Electricity Consumption

Loading Italy electricity consumption and temperature data...

Figure 4a: Electricity consumption and temperature patterns in Italy (2021-2025). The top panel shows monthly electricity consumption (TWh); the bottom panel shows average temperature (°C). Summer shading (beige) and shared time axis enable direct vertical comparison. Key insight: Peak electricity consumption directly coincides with peak summer temperatures, revealing how extreme heat drives energy demand through air conditioning use. Annotations and visual connectors highlight this correlation. The visualization animates on scroll, with lines growing left-to-right. Data: Terna (electricity), ERA5 Copernicus (temperature).

From monthly patterns to yearly trends

Over decades, the pattern becomes undeniable. As temperatures climb, our need for cooling grows, while our need for heating diminishes. The trend is clear: we're moving toward a future where cooling dominates our energy consumption, especially during the hottest months.

Cooling and Heating demand in Italy (1979-2024)

To quantify heating and cooling demand, climate scientists use two key metrics:

  • Heating Degree Days (HDD): Measures how much energy is needed to heat buildings during cold weather. Higher HDD values indicate colder years requiring more heating.
  • Cooling Degree Days (CDD): Measures how much energy is needed to cool buildings during hot weather. Higher CDD values indicate hotter years requiring more air conditioning.

Both metrics are calculated relative to a baseline temperature (typically 18°C for HDD and 24°C for CDD) and summed over the year to capture total heating or cooling demand.

Loading HDD and CDD data...

Figure 4c: Annual average Heating Degree Days (HDD) and Cooling Degree Days (CDD) in Italy (1979-2024). HDD (warm/orange, above 0) declines over time (milder winters), while CDD (blue, below 0) rises (hotter summers and growing cooling needs). The seasonal strips summarize when heating vs cooling is concentrated across the year. Data source: Eurostat (nrg_chdd_m).

But energy consumption is only part of the story. When heat becomes extreme, when temperatures soar beyond what our bodies can handle, the consequences are measured not in kilowatts, but in lives. Heat-related illnesses become more common, especially among vulnerable populations. During heat waves amplified by UHI effects, emergency room visits spike, and mortality rates increase. The elderly, children, and those with pre-existing conditions are most at risk.

Heat as a global public health stressor

Beyond cities and beyond energy demand, rising temperatures act as a widespread stress test for human health. The indicators below capture a progression of impacts—from early disruption of daily recovery to acute health crises, economic constraints, and unequal exposure across vulnerable groups.

Sleep disruption is often one of the earliest and most widespread impacts of heat exposure, affecting populations even before severe health outcomes emerge.

Global Sleep Loss Due to Heat

Loading sleep loss data...

Figure 4b: Percentage of sleep lost globally due to heat exposure (2015-2024), obtained by collecting data from 68 different countries. Rising heat disrupts sleep patterns worldwide, with a sharp increase in recent years reaching 8.7% sleep loss in 2024. Data source: Lancet Countdown 2025.

When heat exposure becomes extreme, its effects turn lethal, producing sharp spikes in mortality during major heatwave years rather than a smooth, gradual trend.

Global Heat-Related Mortality

Loading mortality data...

Figure 4c: Annual heat-related deaths globally (1990-2021). Heat mortality shows significant variability with extreme peaks during major heatwave years. The 2010 peak (721K deaths) corresponds to severe heatwaves across multiple continents. Data source: Lancet Countdown 2025.

Heat also constrains economic activity, particularly outdoor and manual labour, reducing the number of hours people can safely work each year.

Global Potential Work Hours Lost to Heat

Loading work hours lost data...

Figure 4d: Global potential work hours lost per person per year due to heat exposure (1990-2024). Rising temperatures increasingly impact outdoor work capacity, with 178 hours lost per person in 2024 - a 34% increase since 1990. Data source: Lancet Countdown 2025.

These impacts are not evenly distributed: age, health, and social vulnerability strongly shape who is most exposed to dangerous heat.

Heatwave Exposure of Vulnerable Populations Worldwide

Loading vulnerable population data...

Figure 4e: Global heatwave exposure events for vulnerable age groups (1980-2024). Adults aged 65+ face dramatically higher exposure than infants, with both groups showing increasing vulnerability over time. Exposure measured as person-events exceeding heat thresholds, aggregated across all countries. Data source: Lancet Countdown 2025.

Taken together, these indicators show how rising temperatures translate into layered human impacts. Heat first disrupts daily comfort and recovery through sleep loss, then escalates into health emergencies and mortality during extreme events. At the same time, it erodes economic productivity by limiting safe working hours and disproportionately exposes vulnerable populations, particularly older adults. While energy demand metrics such as HDD and CDD reveal how societies adapt to warming through infrastructure, these global indicators highlight the limits of adaptation when heat exceeds what technology, physiology, and social systems can absorb.

04

What Can Be Done?

The good news is that Urban Heat Islands are not inevitable. Cities around the world are implementing strategies to mitigate the effect, and the data shows these interventions can make a real difference.

Green Infrastructure

Increasing vegetation through parks, green roofs, and street trees can reduce temperatures by 1-3°C. Green spaces provide shade, cooling through evapotranspiration, and improved air quality.

Cool Materials

Using high-albedo materials for roofs and pavements can lower surface temperatures by 5-7°C. Cool roofs and cool pavements reflect more solar energy, reducing heat absorption.

Urban Planning

Thoughtful design can mitigate heat: preserving natural landscapes, creating green corridors, optimizing building orientation for ventilation, and maintaining open spaces for air circulation.

Energy Efficiency

Reducing waste heat through better building insulation, efficient HVAC systems, and promoting public transportation can help break the heat-energy feedback loop.

The most effective approach combines multiple strategies. Cities that integrate green infrastructure, cool materials, and smart planning see the greatest reductions in heat island intensity. The data shows that these interventions are not just environmentally beneficial—they're economically sound investments that pay dividends in reduced energy costs and improved public health.

As our cities continue to grow, understanding and addressing Urban Heat Islands becomes increasingly urgent. The satellite data tells us not just what's happening, but what's possible when we design cities with climate in mind.

05

How We Know: Methodology

Data Sources

This analysis integrates multiple publicly available datasets from Earth observation, climate, energy, and health sources.

Satellite Imagery & Land Cover:

  • Landsat 8-9 TIRS - Thermal infrared imagery for Land Surface Temperature (LST) mapping. Downloaded programmatically using custom Python implementation of USGS Machine-to-Machine (M2M) API. See footer for API documentation links. Resolution: 100m (thermal band ST_B10), Date: July 9, 2025 (Genoa region). Used as target variable for XGBoost-based spatial downscaling to 10m resolution.
  • Sentinel-2 - Multispectral imagery (10m resolution) for vegetation, built-up, and water analysis. Downloaded using Copernicus Data Space Ecosystem (CDSE) API client (custom implementation). See footer for CDSE documentation links. Bands used: B02 (Blue), B03 (Green), B04 (Red), B08 (NIR), B11 (SWIR1), B12 (SWIR2). Used for calculating spectral indices (NDVI, NDBI, NDWI) and as predictor variables for LST downscaling.
  • Digital Elevation Model (DEM) - 10m resolution elevation data for Genoa region. Used to derive topographic features (slope, aspect) as predictors in LST downscaling model. Topography influences local temperature patterns through shading, exposure, and cold air drainage.
  • ESA WorldCover 2021 - Global land cover classification at 10m resolution. Accessed via ESA WorldCover. Used for categorizing land cover types (built-up, tree cover, grassland, etc.) in Genoa.

Climate Data:

  • ERA5 Reanalysis - Monthly average 2-meter air temperature for Italy (2021-2025). Accessed via Copernicus Climate Data Store (CDS). Dataset: "ERA5 monthly averaged data on single levels from 1940 to present". Spatial resolution: 0.25° × 0.25°, aggregated to country-level monthly averages.
  • Eurostat CHDD - Heating Degree Days (HDD) and Cooling Degree Days (CDD) for Italy (1979-2024). Dataset code: nrg_chdd_m. Accessed via Eurostat Data Browser or Eurostat REST API. HDD measures heating demand (colder years = higher HDD), CDD measures cooling demand (hotter years = higher CDD).

Energy Data:

  • Terna S.p.A. - Monthly electricity consumption in Italy (2021-2025), measured in TWh. Terna is the Italian transmission system operator. Data downloaded from official Terna reports and processed from Excel files. Files located in data/electricity_consumption_italy/.
  • Eurostat Energy Statistics - European energy balance data. Dataset codes: nrg_cb_pem (energy balances), nrg_10m (monthly energy statistics). Accessed via Eurostat Data Browser or REST API. Used as supplementary/validation data for energy consumption patterns.

Heat & Health Data:

  • Lancet Countdown (2025) - Heat and health indicators used in this project (global heat-related mortality, sleep lost due to heat, potential work hours lost, and vulnerable-population exposure metrics). Source: Lancet Countdown on Health and Climate Change. Processed into web-ready JSON in data/heat_and_health/ and exported to data/json/.

Data Location: Processed data files are in data/processed/ and data/json/. Raw data (if stored) is in data/raw/. Preprocessing scripts are in data/ directory.

Data Cleaning and Imputation

Raw data underwent systematic preprocessing to ensure quality and consistency:

Satellite Imagery Processing:

  • Cloud Cover Removal: Cloud-contaminated pixels identified using quality assessment bands and masked out
  • Gap Filling: Sensor gaps and cloud shadows filled using spatial interpolation (nearest neighbor or bilinear interpolation)
  • Atmospheric Correction: Thermal bands corrected for atmospheric effects using standard algorithms (e.g., split-window method for Landsat)
  • Coordinate System: All raster data transformed to consistent CRS (EPSG:4326 or UTM) for spatial alignment

Temperature Data Processing:

  • Outlier Detection: Extreme values outside plausible ranges (-50°C to 60°C for LST, -30°C to 50°C for air temperature) flagged and excluded
  • Missing Values: ERA5 monthly data gaps (rare) handled via temporal interpolation or excluded from analysis
  • Validation: Cross-checked against meteorological station data where available

Energy Data Processing:

  • Unit Conversion: Terna data converted from MWh to TWh for consistency
  • Missing Months: No missing data in Terna dataset (complete 2021-2025 coverage)
  • Validation: Compared against Eurostat energy statistics for consistency checks

Eurostat CHDD Processing:

  • Data Parsing: Complex Eurostat TSV format parsed and reshaped into structured JSON
  • Value Cleaning: Removed flags, whitespace, and non-numeric characters from cell values
  • Aggregation: Monthly data aggregated to annual averages for trend analysis

Preprocessing Scripts: All preprocessing code is available in the repository: data/download_and_preprocess_heat_data.py, data/eurostat_chdd/preprocessing_estat_chdd.ipynb, data/average_monthly_temperature_ERA5/avg_monthly_temperature_italy_era5.ipynb, and data/electricity_consumption_italy/electricity_italy_2021_2025.ipynb.

Data Processing and Analysis Pipeline

The complete pipeline transforms raw data from multiple sources into interactive visualizations. Below is a visual overview of the workflow:

01 | Ingestion Satellite + API Data 02 | Preprocessing Clean & Transform 03 | Features LST, NDVI, Metrics 3.5 | LST Downscaling XGBoost ML: 100m → 10m (Genoa case study) 04 | Aggregation Export to JSON 05 | Visualization Interactive Charts

Parallel workflow: satellite imagery flows through feature extraction with optional ML downscaling (orange path), while tabular data flows through aggregation. Both converge at visualization.

Stage 1: Data Ingestion

  • Download satellite imagery via custom APIs
  • Fetch climate and energy data from public sources

Stage 2: Preprocessing

  • Cloud masking, coordinate transforms, quality filters
  • Handle missing values and remove outliers

Stage 3: Feature Extraction

  • Calculate LST, NDVI, UHI intensity, HDD/CDD

Stage 4-5: Aggregation & Visualization

  • Group by time/space, export to JSON
  • Create interactive D3.js charts

Scripts: data/*.py, data/**/*.ipynb — see repository for details

Stage 3.5: LST Downscaling (Genoa)

Challenge: Landsat thermal data is 100m resolution. We need 10m to see detailed urban heat patterns.

Solution: XGBoost machine learning model trained to predict temperature at 10m using:

  • Landsat-9: Accurate temperature at 100m
  • Sentinel-2: Surface details at 10m (vegetation, buildings, water)
  • DEM: Topography at 10m (elevation, slope, aspect)

Process:

  1. Compute spectral indices (NDVI, NDBI, NDWI) and topographic features
  2. Create training dataset: aggregate 10m features to 100m, pair with Landsat LST
  3. Train XGBoost with hyperparameter tuning (80/20 split, cross-validation)
  4. Predict temperature at every 10m pixel

Result: 10× higher resolution temperature maps revealing micro-scale UHI effects.

Code: LST_downscaling/01_data_processing.ipynb, 03_train_xgboost.ipynb

Limitations & Uncertainty

Key constraints to consider:

  • Spatial Resolution: Satellite resolution limits street-level detail (100m Landsat, 10m Sentinel-2)
  • Temporal Coverage: Single snapshot (July 9, 2025). UHI varies by season and weather
  • Cloud Gaps: Incomplete coverage due to cloud cover
  • Measurement Uncertainty: LST accuracy ±2-3°C; health data has confidence intervals
  • Correlation ≠ Causation: We show patterns, not causal proof

Note: Relative comparisons (urban vs. rural) are more reliable than absolute values. Spatial patterns are more reliable than point measurements.

Technical Details

Tools & Libraries: Python (rasterio, xarray, geopandas, pandas, XGBoost), D3.js v7, HTML/CSS/JavaScript

Performance Optimizations: Lazy loading, Canvas rendering, asynchronous data loading, pre-aggregated JSON

Interactions: Tooltips, zoom/pan, responsive design, mobile-optimized

Team

Stefano Infusini

AI Student, University of Genoa

Roles: Data visualization design and implementation, D3.js development, data preprocessing and analysis pipeline, web development, project coordination.