🔬 physics

A global framework to estimate urban spatial cycling patterns based on crowdsourced data

This paper presents a global framework that utilizes population and point-of-interest weighted Strava Global Heatmap data to accurately estimate and validate urban cycling patterns, offering a low-effort solution for planning and large-scale analysis where official count data is scarce.

Original authors: Robert Klein, Elias Willberg, Silviya Korpilo, Tuuli Toivonen

Published 2026-03-03

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Robert Klein, Elias Willberg, Silviya Korpilo, Tuuli Toivonen

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to understand how people move through a city, specifically how many of them are riding bicycles. In the past, city planners had to stand on street corners with clipboards or install expensive sensors to count bikes. This was slow, expensive, and often missed huge chunks of the city.

This paper introduces a clever new way to "see" cycling patterns using a tool most people already know: Strava, the popular fitness app.

Here is the story of the research, explained simply:

1. The Problem: The "Foggy Map"

The researchers started with the Strava Global Heatmap. Think of this heatmap like a glowing map of the world where the brightest spots show where Strava users (mostly fitness enthusiasts) ride their bikes the most.

However, there was a catch. The heatmap is like a foggy photograph. It shows where people are riding, but the brightness is normalized (adjusted) so that the whole world looks balanced. This means a busy street in a small town might look just as bright as a super-busy avenue in a huge city. If you just looked at the raw colors, you couldn't tell which city actually had more cyclists. It was a "relative" map, not an "absolute" one.

2. The Solution: Adding "Context Ingredients"

To clear up the fog, the researchers decided to mix the heatmap with two other ingredients: Population and Points of Interest (POIs).

Population Data: This is like knowing how many people live in a neighborhood. More people usually mean more potential cyclists.
POI Data: These are places like cafes, shops, schools, and parks. Think of these as "magnets" that pull people in. If a street has lots of shops, it's likely to have more cyclists stopping by, not just riding through.

The researchers created a recipe:

Raw Heatmap + (Population × POIs) = A Clearer Picture

They tested this by weighing the heatmap values. Imagine the heatmap is a flat sheet of metal. They placed heavy weights (representing people and shops) on top of it. Where the weights were heaviest, the metal sank, revealing the true "depth" of cycling activity.

3. The Test: The "Taste Test"

To see if their new recipe worked, they compared their "weighted map" against official bike counters in 29 cities around the world (from London to New York to Buenos Aires).

Think of the official counters as the "gold standard" taste test. They asked: "Does our new map predict where the official counters see the most bikes?"

The Results:

The Raw Map (Foggy): Failed the test. It was often wrong about where the busiest streets were.
The Weighted Map (Clear): Passed with flying colors!
- In European cities, the new map matched the official counts almost perfectly (like a 90% match).
- In North American cities, it was also very good, especially on the East Coast.
- The "POI" ingredient (shops and services) turned out to be the secret sauce, working better than just counting people.

4. The Secret Sauce: Buffer Zones

The researchers also had to figure out how far to look for these "magnets" (shops/people).

If you only look at the immediate block (100 meters), you miss the bigger picture.
If you look too far (5 kilometers), you include too much noise.

They found the "Goldilocks Zone":

In dense European cities: Looking about 2.5 to 3 kilometers out worked best.
In spread-out American cities: You needed to look further out, about 3.5 to 4 kilometers, to capture the same effect.

5. Why This Matters

This study is a game-changer for city planners for three reasons:

It's Free and Open: You don't need to buy expensive data or wait for the government to install sensors. The data is already there, waiting to be decoded.
It's Global: You can use the same method to compare cycling in Paris, Toronto, and Sydney instantly. Before, comparing these cities was like comparing apples to oranges because the data didn't exist in the same format.
It Helps Build Better Cities: If a city wants to build a new bike lane, they can use this method to see where people actually want to ride, rather than guessing.

The Bottom Line

The researchers took a "foggy" fitness app map, added some "context weights" (people and places), and turned it into a powerful tool for understanding how cities move. It's like taking a blurry photo and using AI to sharpen it, revealing the true rhythm of the city's cyclists.

In short: If you want to know where the bikes are going in a city, don't just look at the fitness app. Look at the fitness app plus the shops and the people, and you'll see the whole picture.

1. Problem Statement

Urban cycling is a critical component of sustainable mobility, yet effective planning and policy-making are hindered by a lack of spatially accurate, comprehensive, and comparable cycling data.

Data Scarcity: Traditional data collection (manual counts, static sensors) is often sporadic, labor-intensive, and lacks system-wide coverage.
Access Barriers: While the Strava Metro platform offers high-resolution cycling data, access is restricted to specific cities and subject to changing policies, preventing global comparative analysis.
Limitations of Open Alternatives: The Strava Global Heatmap is openly accessible and covers the globe, but it is a normalized, ordinal visualization (0–255 scale) rather than absolute volume data. Previous studies have not systematically validated its utility for estimating relative cycling intensities across diverse global cities or corrected for its internal normalization biases.

2. Methodology

The authors developed a four-step framework to extract, refine, and validate urban cycling patterns using the Strava Global Heatmap.

A. Data Extraction

Source: Strava Global Heatmap (cycling/"ride" activities), updated monthly based on the previous 12 months of user data.
Processing: The heatmap was accessed via a Python script using Strava login credentials. Data was extracted as georeferenced rasters at zoom level 16 (approx. 2.4m resolution) for 29 cities across Europe, North America, South America, and Oceania.
Ground Truth: Official cycle count data (automatic counters, manual counts, or strategic estimates) from 2023–2025 were collected for validation.

B. Refinement (Weighting Strategy)

To address the heatmap's internal normalization (which makes raw values incomparable across large areas) and to model local cycling intensity, the authors applied two weighting variables:

Population Density: Derived from the Global Human Settlement Layer (GHSL) at 100m resolution.
Points of Interest (POI): Derived from OpenStreetMap (OSM). Specific tags were filtered to exclude generic or irrelevant features (e.g., "building," "streetlamp") and retain amenities relevant to daily cycling (e.g., shops, transit stops, parks).

Weighting Algorithm:
The raw heatmap value ( $h_l$ ) for a location $l$ was weighted by the density of the variable (Population or POI) within a specific buffer radius ( $b$ ).

Formula: $h_{l,b,weighted} = h_l \times \frac{n_{l,b}}{\max(n_{b,c})}$ $h_{l, b, w e i g h t e d} = h_{l} \times \frac{n _{l, b}}{m a x ( n _{b, c} )}$
- Where $n_{l,b}$ is the count of the variable in the buffer, and $\max(n_{b,c})$ is the maximum count in that city.
Combined Weighting: A geometric mean of population and POI weights was also tested ( $\sqrt{pop \times POI}$ ).
Buffer Sizes: Tested across 12 buffer radii ranging from 100m to 5000m.

C. Validation

Comparison: Weighted heatmap values were compared against official cycle counts at 29 locations per city (totaling thousands of data points).
Metrics: Pearson and Spearman rank correlations were calculated. Spearman was prioritized as the heatmap represents ordinal intensity rather than absolute counts.
Contextual Analysis: Correlation results were further analyzed against city-level metrics: cycling modal share, population density, and count location density.

3. Key Results

A. Impact of Weighting

Raw Heatmap: Performed poorly in most cities (average Spearman $\rho < 0.5$ ), failing to capture spatial patterns accurately due to global normalization.
Population Weighting: Improved correlations significantly in European cities (often $\rho > 0.7$ ) but showed mixed results in North America.
POI Weighting: Performed best overall.
- Europe: All 13 cities showed strong correlations ( $\rho > 0.7$ ), with 9 cities exceeding $\rho > 0.8$ (e.g., London $\rho=0.89$ , Hamburg $\rho=0.86$ ).
- North America: Cities along the East Coast (e.g., Boston, New York, Washington) showed high correlations ( $\rho > 0.7$ ). Western cities (e.g., Los Angeles, Austin) showed weaker correlations.
- Combined Weighting: Did not significantly outperform POI weighting alone.

B. Optimal Buffer Sizes

Europe: Optimal correlations were found with 2500–3000 m buffers.
North America (East Coast): Optimal correlations were found with 3500–4000 m buffers.
Reasoning: The difference is attributed to the Web Mercator projection used by Strava (distorting distances at higher latitudes) and the generally denser urban morphology of European cities compared to the more dispersed North American cities.

C. Drivers of Accuracy

The study found that the framework's accuracy is positively correlated with:

Cycling Modal Share: Cities with higher cycling adoption rates yielded better correlations.
Count Location Density: Cities with a higher density of official counters provided better validation results, suggesting that sparse ground truth data introduces noise.
Population Density: Denser cities generally performed better.

4. Key Contributions

First Global Validation: This is the first study to systematically validate the Strava Global Heatmap against official count data across a diverse set of 29 global cities.
Methodological Framework: It provides a reproducible, low-effort method to convert ordinal heatmap data into spatially comparable relative cycling intensity estimates using open data (OSM and GHSL).
POI Superiority: It demonstrates that Point of Interest data is a more effective proxy for local cycling intensity than population density alone, likely because POIs better capture the "destination" aspect of daily cycling.
Buffer Optimization: It establishes region-specific optimal buffer sizes for weighting, acknowledging the impact of urban morphology and map projections.

5. Significance and Implications

Planning Support: The framework offers a viable alternative for urban planners in cities lacking official cycling infrastructure data. It allows for the estimation of relative usage patterns to prioritize infrastructure investments.
Global Comparability: By standardizing the extraction and weighting process, the study enables large-scale comparative analyses of cycling behavior across different countries and cultures, which was previously impossible due to data fragmentation.
Data Democratization: It leverages open-source data (Strava Heatmap + OSM) to overcome the limitations of proprietary, restricted-access platforms like Strava Metro.
Limitations & Future Work: The authors note that the method estimates relative intensity, not absolute volumes. Accuracy is lower in cities with low cycling modal shares or sparse ground-truth data. Future research should expand this framework to Asian and African cities as open data availability increases.

In conclusion, the paper proves that POI-weighted Strava Global Heatmaps can accurately represent urban spatial cycling patterns, providing a powerful, scalable tool for advancing sustainable urban mobility planning.