I ran into many of the same issues when I started importing Canvec data for the province of Newfoundland and Labrador. But, in my case, the arguments I had were with myself. When I started, most of the province was a blank slate with only the (PGS) coastline in the OSM database. There were a fair number of Canvec tiles which had been imported along the Trans Canada Highway. I found when I tried to update some roads in the imported areas, I ran into considerable difficulties. The main problem was that the error detecting methods I was using (JOSM validator, Keepright, Geofabrik) weren't of much use. When there are a large number of errors in an area when you start modifying something, it's very difficult to detect your own errors and mistakes, since you "can't see the forest for the trees". At the same time I wanted to set up a basic structure in the blank areas so that others could start adding points of interest, correcting highway routes, naming lakes and streams, etc. One of the issues I ran into from the outset was that often, depending on the area, the Canvec data set was not the best source of data for a particular layer. The NL provincial government has produced a database of roads that appears to be based on high quality, current GPS tracks, and includes highway and street names, and provincial reference numbers. Given that this data is available, and easier to import, it doesn't make much sense to import Canvec road data. As for other Canvec highway data (tracks, trails), I have never found this layer to be very reliable in this area. Going back to government issues paper topo maps, I can find trails on the map (and in Canvec) in the Cape Race area that (after extensive hiking in thbe area) I have never been able to find, and that local residents have no knowledge of. On the other hand, there are traditional tracks in the Butterpot Park area that the local residents tell me have "always" been there, and which are quite visible on arial photography, but are not present on the topo maps, or in the Canvec data set. It's probably a lot healthier and more accurate to collect this data by going on a hike with a GPS in your pocket. Another area where the Canvec data causes difficulties is with boundaries, particularly residential areas. All of the data in this area seems to be from about 30 years ago. A more useful source for this data is the municipal/town zoning maps. Of course, the zoning maps can be much more difficult to import, depending on what format they are in. The layers that seem to cause the most difficulty for me are wetlands and forest. For wetland areas, the definition of a wetland seems to be quite variable. There are areas of the province that are marked as wetland near Stephenville on the West coast of the island. I worked in construction in that area for a while, and know that the area is peatland, but solid enough to drive a car over with little difficulty. On the other hand there are areas tagged as wetland near St. John's that I have crossed in a canoe. With such a wide range of conditions all being classified as wetland, the usefulness of incorporating the data comes into questions. I find the same difficulty with forested areas. Some areas tagged as forest consist of widely spaced trees, and you can drive a car (or ATV) easily through the area. Some areas tagged as forest are what's locally known as "tuckamoor", and are difficult to get through with a chainsaw. There doesn't seem to be a lot of agreement between what's in the Canvec data set and what I see on the ground. This is perhaps due to the fact that one of our primary industries is forestry.
If the forest, wetland, and roads are eliminated from the Canvec data set, what's left is mostly hydrology. This layer seems to be pretty accurate, for this region, Canvec seems to be the best source for the data. (In areas with high resolution bing coverage, this may not be the case). The compromise that I arrived at was to filter the Canvec data using osmosis to take out the layers I didn't want (forest, wetland, roads), and to keep the data I did want (hydrology, coastline, place names, etc.). Importing this level of data makes cleaning up the data quite a bit easier, and it is not difficult to bring the data to a near zero error level (from all validators). This in turn allows the production of a "clean" map, but allows reasonably quick installation of of a basic structure (hydrology, roads, coastline, names) in large areas. To get a feel for the maps produced this way, you can look at most of the province of NL. There's a small area on the West coast near Stephenville where another user has added forest and wetland areas for comparison. If you look at the area using keepright, you'll find a LOT of deprecated tag errors. This happened because the decision to deprecate the tag "natural=land" occurred after I had completed these areas. Since Canvec uses the natural=land tag for a number of different features, it's going to take a while to translate all of these tags. The Great Northern Peninsula, and the Avalon Peninsula have been cleaned up, and should be representative of what the province will look like in the near future. I realize that these methods may not work for everyone, but I like the idea of getting a minimal, clean infrastructure in place. This should help people using miminal editors like potlatch, or smartphone applications to contribute to the map. It also makes it easier for me (or someone else) at some later point, to add wetland or forest areas from Canvec. (Although I think I prefer the look of the map without the forested areas) Starting from an error free map makes the validation methods into powerful tools for checking the quality of the import. _______________________________________________ Talk-ca mailing list Talk-ca@openstreetmap.org http://lists.openstreetmap.org/listinfo/talk-ca
