Code from this video — shared by request.
Turns a Strava data export into an interactive heatmap. No API needed - just the zip file Strava lets you download.
The output is a single HTML file with six layers you can switch between:
| Layer | Colour | Shows |
|---|---|---|
| Frequency (linear) | Orange | How often you've run each path |
| Frequency (log) | Orange | Same, log scale - better when a few paths dominate |
| Pace (average) | Blue | Average pace - brighter = faster |
| Heart rate (average) | Red | Average HR - brighter = higher |
| Gradient (absolute) | White | Steepness - brighter = steeper |
| Gradient (change) | Green / purple | Direction - green = descending, purple = ascending |
pip install -r requirements.txt
- Request your data from Strava: Settings → My Account → Download or Delete Your Account → Download Request
- Unzip the export and place the folder next to
heatmap.ipynb - Update the config cell:
ACTIVITIES_DIR = "your_export_folder" # name of the unzipped folder
ACTIVITY_TYPES = ["Run"] # Run, Ride, Hike, Walk, ...
DATE_FROM = "2024-01-01" # or None for no lower limit
DATE_TO = "2024-12-31" # or None for today- Run all cells. Map is saved to
outputs/heatmap.html.
Home is auto-detected from the most common activity start point in the date range, then only activities within RADIUS_KM of that point are included. It's a heuristic — if you started more runs from somewhere else (work, a club) than home in that period, that location wins. Override it with HOME_LAT / HOME_LON if needed.
Parsing .fit.gz files is slow so GPS data is cached after the first run. Changing the date range or config won't re-parse files you've already loaded.
GPS records at ~1 Hz, so the frequency layers count GPS samples per pixel rather than distinct activities. A slower run deposits more points on the same path than a faster one. In practice this means the map shows something closer to time spent on each road than how many times you've run it - which is arguably more useful, but worth knowing.
The log scale version exists because a few favourite routes tend to dominate completely on a linear scale, washing out everything else.
Each pixel is the mean across every activity that ever crossed it. A route you used to run slowly but now run fast will show somewhere in the middle. Narrow the date range if you want a specific period.
GPS altitude is much noisier than horizontal position - typically ±10–20 m vertically versus ±3–5 m horizontally. The gradient layers are reliable on hilly terrain but can look noisy on flat routes where the signal-to-noise is poor.
The raster grid is built in Web Mercator (EPSG:3857) so it aligns directly to the map tile basemap without any reprojection. But Web Mercator distorts distances at higher latitudes, so anything involving real-world metres - the clip radius around home, and the rise/run calculation for gradient - uses a local UTM projection instead. The visual output is unaffected, it just means the underlying measurements are accurate.
