The data is aggregated into a quad tree based on number of GPS points in each pixel. Tiles are then served on the fly using Go and C using CGO.
Cloudfront tackles most of the load, but the load balanced i2.xlarge instances can do about 300 tiles a second.
I noticed there are quite a few paths through Golden Gate Park in SF (as just one of many examples) that are pretty clear running/biking paths that are not in any of the typical basemap providers' maps and not currently in OSM. If you reached out to the OSM community and gave the go ahead, this would be a great way to improve the trail data in OSM.
I've also built another tool, Slide, to help use this data for faster editing, see http://labs.strava.com/slide/
Unlike a traced layer, the heatmap is self-updating and objective, so I think has a value of its own.
Also how accurate is the bike vs run stuff? I'm seeing a a surprising amount of people biking activity in wilderness areas.
- Where's the source data from?
- You should think about producing an overlay with bike and run in different colours to see where these differ.
An overlay would be nice, but there is a version for both running and riding.
I couldn't get the styles I wanted with mapbox so I just defaulted to google.
We're not considering opening the data. But we are working with cities to help improve their bike infrastructure
http://bikeportland.org/2014/05/01/odot-embarks-on-big-data-...
Maybe unnecessarily complicated, but it does provide really good local normalization. You also get cool looking shading like in this location http://labs.strava.com/heatmap/#16/-121.76487/38.49134/gray/...
(edit: Your color scales are not really intuitive, I have no idea what red or blue means or red vs green.)
Is this now the complete dataset from strava Run/Bike activities?
It's not updated on every upload, but pretty much the complete dataset.