From OpenStreetMap Wiki
Jump to navigation Jump to search

OpenSolarMap aims to add roof orientation data into OpenStreetMap, so that this useful data can be reused by as many people as possible.

It is very simple (and highly addictive) way to contribute to OpenStreetMap... and possible to discover OpenStreetMap !

It is also an experiment mixing crowdsourcing and neural networks based automated image classification.

How does it work ?

Just go to, there is no registration, sign up or account needed

Buildings are automatically selected based on their shape and orientation.

Only the buildings closely-orientated with cardinal points (N/S/E/W) are shown:


A dashed line shows the building footprint we have in OpenStreetMap, in the above example, the roof is facing North/South.

You can simply click on the corresponding icon:

  1. roof facing north/south
  2. roof facing east/west
  3. flat roof
  4. (or 0) other cases: like complex roofs, missing building on the imagery, impossible to see (trees, etc) or simply you can't tell !

Keyboard shortcuts 1,2,3,4+0 allow you to quickly go to the next building.

Don't worry if you make a mistake, OpenSolarMap works like captchas, at least 3 identical answers are needed to classify a building.

This tool is a quick and dirty hack created during the Climate Change Challenge hackathon that took place in Paris from 6th to 8th November 2015.

The code is available on github and your pull-requests are welcome !

Contributing to OpenStreetMap

The roof orientation details are added to existing OpenStreetMap buildings.

Changeset are organized to limit uploads to one municipality at a time.

The tags that are updated are: roof:orientation=along/across and roof:shape=0.0 for flat roofs.

See the OpenSolarMap account edits for details.

Neural network project

Based on the manual contributions, a neural network has been trained to classify buildings automatically.

This work has been done by Michel Blancard from Etalab ( The code is available on github.

A first batch of 60000 buildings has been classified using the neural network and injected as "contributions". As the neural network is returning a confidence value, these contributions are considered like 1 or 2 manual contributions, meaning that only 2 or 1 additional contributions are needed to have a final classification to get the new tags back in OSM.

A second batch of 1.4 million buildings has been classified and not yet injected.


Project initiated by User:Cquest