Strava just published a new map of all their data. 3 trillion GPS points turned into a heatmap. It’s lovely.
They wrote a fantastic writeup of how they made the map. All custom software. It’s not really a “heat map” in the sense I use the term, in that it lacks heat diffusion that leads to blobbiness of a true heat map. It’s more of a spatial histogram. That’s just a quibble, but it helps me keep this visualization straight from the new Mapbox JS GL heatmaps I’m playing with.
The smartest thing they do is adaptive normalization of the colormap to spatially local data. They describe it in detail with the CDFs they calculate. Basically bright yellow indicates the highest number for a local region near where you’re looking now, not an absolute number for the highest value in the whole world. This allows seldom-visited areas to still have compelling and readable visualizations. You can see this in effect if you zoom in to Alcatraz which has relatively few visits; the brightness changes radically from z=15 to z=16. A trick worth stealing.
I was also struck at how all the tracks look nicely snapped to roads. Mobile phone locations are never that accurate. And they deliberately fuzz all points by 2 meters, why are the roads so sharply defined? I think they simply have enough points that the average that comes through visually really is the actual road. Neat! You can see this in a hyper-zoom on the Golden Gate bridge where you can see faint traces of tracks off the bridge, but the main bridge path is highlighted. Note also the little bump, that’s where you have to walk about 2m to the outside to avoid the tower. (I believe the two tone-line is because this is walking data, pedestrians tend to walk the east side.)