This map shows for every street in Berlin how rough it is on a bike, separately for each riding direction.
It works wherever rides are recorded with a phone on the bike: its accelerometer measures every bump, and GPS places it on the street. For now the rides come from the SimRa project, where cyclists record their rides with a phone app. It builds on OpenStreetMap to match the rides to the streets, and to carry what the rides show over to the surface and smoothness tags where there are not enough rides yet.
Green is smooth, red is rough, faint lines rest on few rides. Click or tap a street for its values; the buttons top left switch between the views.
A demo: all values are estimates and may contain errors or inaccuracies (wrongly matched rides, few rides, biases at low speed and on short segments). They are not a recommendation to change OSM tags. Feedback is welcome.
Code: open source on GitHub
(MIT License).
Roughness: calculated from the SimRa rides
(CC BY-NC 4.0: no commercial use).
Streets and tags: © OpenStreetMap contributors
(ODbL). Base map: OpenFreeMap,
© OpenMapTiles; drawn with
MapLibre and PMTiles
(BSD licenses).
Impressum · Privacy
Every street is shown per riding direction (each on its own side). Roughness is "× as rough as good asphalt": ×1.5 shakes a bike 1.5 times as much as a street tagged surface=asphalt, smoothness=good at the same speed. A value ± x is its uncertainty (one standard error: the true value lies within it about 2 times in 3).
From SimRa rides. Each second of a ride gives the vertical shaking, corrected for speed and for the rider's phone and bike (the ride's median is taken out), averaged over 60 m of the ride: a short segment shows the street around it, not just a kerb at its end. The seconds are matched to the street network; the seconds of one ride on one street segment and direction form a pass. The estimate is the weighted mean of its passes.
The passes, each counted by its weight: age (a pass counts half at 3 years old and ×0.1 at 5), length (a 20 m pass counts 1, a longer one a little more) and the matcher's certainty that the pass is on this street and not on a parallel one (0.5 for a cycle track exactly as close as the road). 1 = one new, clearly matched 20 m pass. Estimated and Difference fade out where it is low.
What OpenStreetMap's surface and smoothness tags predict (surface:bicycle and smoothness:bicycle where a road has its own surface for bikes): the typical measured roughness of the streets with the same pair of tags (if at least 20 are well measured: information ≥ 3), else of all streets with the same surface. Without a surface tag nothing is predictable (grey). Its ± is how much these streets really differ.
OSM and the measurement merged, each weighted by how sure it is (1 / uncertainty²). Without rides it is the OSM prediction, without an OSM prediction the measurement.
Measured minus OSM, in units of their combined uncertainty (z). Beyond ±2 the street is clearly smoother (blue) or rougher (red) than its tags say.
Details and the checks behind every number: how it works.
Wrong values, ideas or questions: please write to .