On Mon, Dec 17, 2012 at 12:23 AM, Toby Murray <toby.mur...@gmail.com> wrote: > As mentioned previously on talk-us, I made contact with the Riley > county GIS department and got access to their data. Since then I have > been analyzing and working with the data. I believe I am nearing a > point at which I can proceed with some imports based on this data. > There are a few different data sets, each of which will be handled > individually. The biggest one is an address import of about 20,000 > points. I don't have many fellow mappers around here but I have > brought it up with a few occasional users and they do not have a > problem with this. As stated on the wiki page, I am hoping to use the > church data to involve some of them in the process. > > I have documented the different data sets, the conversion files and > posted .osm files for review on the following wiki page: > http://wiki.openstreetmap.org/wiki/Kansas/Riley_County > > As noted on that page, the address import still needs more processing. > > In particular there are two special cases that I'm not quite sure what > to do with For now I have marked them with FIXME tags in the .osm > file. > > One is apartments. Each apartment has its own point in the county > data. They are spaced apart from each other somewhat arbitrarily. My > current thought is to discard the apartment number and merge the nodes > down to a single node per unique addr:housenumber value. I would place > this node either at the building center or at a stairwell if it can be > identified. This is how I mapped my own apartment complex some time > ago as can be seen here: http://osm.org/go/T59wjkyXy-- > > The other special case is trailer parks. Two of the biggest ones have > one full address point in the middle of the park. Then on each unit > they have a point with nothing but the lot number. I'm not actually > sure how residents in these parks write their mailing address so I'm > not quite sure how to handle it in this data. Some of the smaller > parks are marked similarly to apartments with duplicated housenumbers > and then a lot number added on. I may just exclude the trailer parks > for a first import and come back to them later. > > I plan on doing the county, state and park boundaries in the next few > days. The addresses will likely take a few more days to fix up so that > one might wait until after Christmas. > > Comments welcome. > > Toby
I put some more time into this and have made progress. I ended up completely redoing the address file. The building points shapefile did not have a city name in it so I joined the building points with the parcel polygons which do have a city in the address field. After several regexes and a few hours of manual merging, correcting and checking of data I think I have a pretty good data set. I do however have some questions where I would like some advice. Apartments. As discussed previously, I left off all individual apartment numbers and consolidated down to one node per building. In most cases this was straightforward since each building has its own unique addr:housenumber. However a couple of apartment complexes share a single street number with several buildings, each with a building letter. In the county data this was seen as "Apt G101" and "Apt H101" and so forth. I still consolidated down to buildings but wasn't sure what to do with the building letter. I ended up putting them in addr:housename tags. So there are some nodes that share duplicate addr:housenumber values but have addr:housename tags to differentiate them. Is this appropriate? If you open the second address file linked on the wiki page[1] about this import you can do a search for "addr:housename" and find them. There does seem to be a little fuzziness around zip codes. Sometimes different fields in the shapefile conflicted with each other. I have fixed a few by hand that were obviously wrong but there are still a few I'm not sure about. This does seem to be the least accurate part of the data set. But even then, it is going to be a few tens of zip codes out of 20,000 addresses that are off so it is still batting a pretty high average. I did leave mobile home parks out of the data for now. I intend to do some more research to figure out how best to handle them. There are 17 nodes with FIXME=* tags. These are new properties that are located on street corners and have this as the address in the county shapefile: "<address 1> OR <address 2>". I made a best guess based on the parcel information but added the FIXME tags to make note that they need to be manually surveyed. I actually know someone who lives in that part of town and has shown some interest in OSM but hasn't done any editing yet. I'm hoping maybe I can convince him to go survey them :) I believe the only thing left to do on the addresses is to merge them with addresses that have already been mapped. I will try to use Paul Norman's addressmerge tool[2]. My first attempt ended in failure but that was because I didn't have addr:city tags which it counts on to ensure uniqueness. Now that I have these tags merged in from the parcel file it will hopefully go much better. [1] http://wiki.openstreetmap.org/wiki/Kansas/Riley_County#Imports [2] https://github.com/pnorman/addressmerge Toby _______________________________________________ Talk-us mailing list Talk-us@openstreetmap.org http://lists.openstreetmap.org/listinfo/talk-us
