[Talk-GB] [Spam] Looking for places to map?

[Peter Miller]

Great stuff. I assume you are aware of woeids from Yahoo which have  
some interesting boundaries derived from crowd sourced data in Flickr  
and other places. Details here.
http://developer.yahoo.com/geo/geoplanet/guide/concepts.html#woeids

It would be a great project to match up boundaries from yahoo woeids  
with Wikipedia articles for UK places,  OSM place names and Traveline  
NatGaz database and also I guess the Cycle Streets DB. Note that we  
still haven't imported the NatGaz DB (with 50,000 place names in a  
hierarchy) into OSM and are probably missing many places on the map in  
more obscure parts of the country as a result.


Regards,


Peter


On 23 Feb 2010, at 14:18, Peter Reed wrote:

> As the weather improves we are all going to be out and about tracing  
> roads for OSM. So there are various discussions, and work under way  
> to help find, prioritise and then fill the most important gaps in  
> the map. Larger towns and bigger areas that need attention are  
> fairly well known, but experience suggests that it can be more  
> difficult to find smaller and more local places where a short  
> session could make a difference quickly.
>
> So this is an attempt to help, at a bit more of a micro-level –http://www.reedhome.org.uk/Documents/osmembed.html?kml=KML/osmcategory.kml
>
> (to view the same thing on Google maps use this http://maps.google.co.uk/maps?f=q&source=s_q&hl=en&geocode=&q=http:%2F%2Fwww.reedhome.org.uk%2FDocuments%2FKML%2Fosmcategory.kml&sll=51.525483,-0.746802&sspn=0.01287,0.033023&ie=UTF8&z=6 
>  )
>
> Broadly – green areas look as though they are already fairly well  
> covered, red areas look as though they would benefit from some  
> attention, and the rest are grey.
>
> WARNINGS
>
> This is very much a first cut, pre-beta, etc. We know there are  
> places missing, some gaps in the data, and we are aware that it  
> doesn’t always get things right  (we think it does often enough to  
> be useful).
> It really cannot tell how well an area is mapped from the data. It  
> can only try to find areas that look a bit thin - and some areas  
> look thin when they are not.
> At the moment we are concentrating on roads, not the full list of  
> features in areas that are most thoroughly mapped.
>
> As a result this sometimes flags up an area as “poor” when in  
> reality it has been perfectly well mapped. For example, there are  
> two areas near here that a lot of work has gone into, where the  
> classification is not right. So if your favourite areas shows up  
> wrongly, please don’t take it personally.
>
> Still, it seems to point in the right direction quite a lot of the  
> time, and it’s offered up on that basis. In my own area (which is  
> already fairly well covered) it has flagged up some towns that I  
> already knew were a bit thin, and another half dozen options that I  
> can reach fairly easily, but didn’t know about.
> If it looks as though it will prove useful then future plans will  
> address some of the current limitations by refining the borders,  
> filling gaps, correcting errors in the underlying measures, and  
> tweaking the arithmetic.
>
> However, some mis-classification will always be inevitable. Read on  
> to see why.
>
> THIS IS HOW IT WORKS:
>
> We start with a list of about 1,600 UK settlements, and a figure for  
> the population that lives there. Baring a few errors and omissions,  
> the settlements are the same ones that Cyclestreets uses for local  
> areas -http://www.cyclestreets.net/area/
>
> We then try to find a boundary for each of these settlements. This  
> is based either on the local authority admin area where there is  
> one, the naptan pay-scale area if there is one, or if all else  
> fails, a guesstimate of how big the settlement must be based on  
> population density.
> (FWIW we already realise that some of the pre-defined areas are too  
> big, some of guesstimates are off-centre, or the wrong size, and the  
> guesstimates don’t work well on the coastline. But it’s a start, and  
> most settlements don’t look too far out.)
>
> Where we know the actual length of roads in a settlement from  
> Department for Transport data we use that to classify the area, but  
> this only works for bigger towns and cities, so for the rest we are  
> trying to figure out how well they are covered without knowing the  
> true length of roads on the ground.
>
> At the moment we do this using various ratios. Within each  
> settlement boundary we measure the length of roads in the OSM  
> database, and from that we calculate three measures: the length of  
> roads per sq km (the road density within the settlement), the length  
> of road per head of population ,and the proportion of roads that are  
> major (primary, secondary and motorway).
>
> The underlying hypothesis is that a thoroughly mapped area should  
> have a relatively high road density, plenty of road per head of  
> population, and a relatively low proportion of major roads (because  
> it’s the unclassified and residential roads that tend to be missing,  
> not the major roads).
> However, there are anomalies - some areas are thinly populated, some  
> are at the intersection of a lot of major roads, some have more  
> tightly packed houses (so a high road density), while others have  
> big gardens (and hence a low road density). So inevitably our  
> hypothesis sometimes breaks down.
> To avoid tripping over some of the more extreme cases, we therefore  
> highlight as “good” only those areas that fall into the top quartile  
> on at least two of the ratios, and to be classified as “poor” we  
> pick only those areas that fall into the bottom quartile on at least  
> two ratios. Everywhere else is in the middle.
>
> That covers complete settlements. On top of that we have plotted all  
> the bigger residential areas (landuse = residential) where there  
> don’t seem to be many roads. We can’t know the population of these  
> areas so we just highlight those with a low road density (road  
> length / area). If the road density in a residential area is above  
> the threshold, or if it is a very small residential area (where the  
> measures can be unreliable), then we just ignore it. We have only  
> plotted the bigger, low density residential areas.
>
> One final caveat – the base data on which this is all based was  
> extracted several weeks ago, so it’s a little bit out of date. An  
> updated version will follow when time permits.
>
> IN CONCLUSION
>
> We have several ideas on how this can be improved, based on earlier  
> suggestions. We welcome more comments and ideas.
> Even more importantly, we welcome more roads on the map.
> So most of all we hope this helps members of the community find some  
> handy places where they can quickly make a difference by plugging a  
> few gaps. Remember that it’s all a bit hit-and-miss though. It’s  
> probably a good idea to check some other sources as well before  
> rushing out with the GPS.
> _______________________________________________
> Talk-GB mailing list
> Talk-GB at openstreetmap.org
> http://lists.openstreetmap.org/listinfo/talk-gb

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk-gb/attachments/20100224/68cc2ec6/attachment-0001.html>