After 16 days of publishing Grass&Green, a specific tool for QA of data classification, we got the following contributions.
Thanks for all contributors, however, we still have 1000’s of entities needed to be checked. I could describe simply how does it work?
We have the following assumptions:
Similar entities should be classified consistently, at least within the same country. For example, In Germany, what people know about “park”, as a place for recreation and amusement, doing sport, grilling, pinking, ..etc. Thus, the one can not called a small piece of grass entity in front or backyard of his home as “park”. It is inconsistency.
OSM in lots of cities (e.g., urban cities) has a good and acceptable quality.
Hence, we extract the characteristics that describe specific grass-related classes like: forest, meadow, park, garden, and grass. Afterwards, we develop a recommendation system (Grass&Green) for that classes.
The aim of the tool is to improve/enrich the data classification quality of these features. We just have a start and test on Germany DataSet, however, the entire world would be provided in the next phases. So far, contributors confirm our recommendations with 90% full/partial agreement. That’s sounds good.
Please, don’t hesitate to contact us for further details or feedback. Follow our news also on Twitter and Facebook
Ahmed Loai Ali
Comment from seawolff on 26 September 2015 at 08:52
Your questions will produce a very strong response bias by
* sorting answers
* preselecting answers
* recommending answers
* hiding other answers
* restricting answers to different types of green
* showing green area on OSM map
The example on the initial page would probably be classified as “landuse=residential” by most mappers.
You should distinguish between “surface=grass” and “landuse=grass” and offer both.
On a “green” car park “surface=grass” is correct while “landuse=grass” is wrong.
Comment from Sanderd17 on 28 September 2015 at 09:01
If you plan to include Belgium, you’ll see very strange stuff. We’ve had someone mapping lots of landuse, but mostly just aiming to reproduce the colour on mapnik.
He has drawn forest areas around single treetops (resulting in 300+ forests in a town of only 80000 inhabitants), and also making really bad mistakes like this one: http://www.openstreetmap.org/way/89503452
But mostly just wrongly tagging meadows and parks as landuse=grass. Like http://www.openstreetmap.org/way/298526045
I tried fixing it in the past, but just ended up frustrated.