Using suburb boundaries to align imagery & data in Burnie, Australia
Posted by bdhurkett on 18 November 2021 in English.I’m undertaking a bit of a project in Burnie, a city in the state of Tasmania in Australia, and thought that I should leave a more public note about what I’m doing and the reasoning behind it. Don’t expect anything particularly entertaining from this entry. The short version: a recent import of official suburb boundary data can be used to accurately offset aerial imagery, which can in turn be used to consistently align existing mapped items across the city.
While there are other mappers in the area, and I was most active in OSM mapping several years back, I’m responsible for a decent amount of data in the Burnie area. My process was pretty straightforward: drive to a new area, survey on foot, record GPS traces as I went. Back home I would upload the address data and anything else that needed updating and trace buildings from imagery. The best-quality imagery at the time was Bing, which was fairly average resolution and not always well-aligned, so I’d align the map to my GPS traces or publicly uploaded ones, depending on subjective recording quality. While this was in good faith, it meant that each little area I’d surveyed tended to have slightly different offsets, with no way for me to tell if any were correct. It’s minor, but many of the buildings I traced are also poorly outlined, though the older imagery probably limited how well I could do at the time.
Having resumed OSM mapping recently (after moving house - what better motivation than having a new neighbourhood to explore?) nobody had done much to improve this issue. Not surprising; it’s not the most populated area, things were good enough for routing purposes, and there was plenty that still wasn’t mapped at all. But it bothered me, and I noticed a new feature that made it much easier to resolve.
Burnie suburbs often split along fencelines and other structures visible from imagery, and Bing has improved their aerial imagery which now seems very well georectified. (Maxar imagery is slightly newer, but the local offset varies noticeably over small distances, especially if there are changes in elevation.) That means it’s easy to align the imagery to suburb boundaries, check that it matches well across the entire suburb/city (it does), and then align items to that imagery offset. Bing almost doesn’t need to be offset, to be honest.
I’ve thus been spending my time aligning roads and buildings to this offset across the city, rather than the hodge-podge of inconsistent alignments that existed thanks to my past efforts. I’m also doing the same with landuses, and unglueing and splitting them where necessary (which is “often”, again thanks to my past efforts). Because the city is now too detailed to download in a single request, I’m now attempting to get public roads consistent first and hope to go back and adjust all the other buildings & landuses and other data as necessary in a more piecemeal effort, probably with more reliance on the newer Maxar imagery. However, I’m not adjusting anything in the CBD as I mapped this to a relatively high level of detail in the recent past - it’s probably not perfectly aligned, but it is at least consistent within itself, and is much better than most of Burnie.
There is one obvious question - whether the suburb boundary data itself is trustworthy. In this case they were added as part of an Australia-wide import from official data, with documentation on the wiki at https://wiki.openstreetmap.org/wiki/Import/Catalogue/PSMA_Admin_Boundaries - some may have existed previously but were presumably checked for accuracy during the import. I don’t think there’s any need to doubt the original data, produced by a government with access to much better surveying equipment than I have. For safety I downloaded a copy of the imported data from https://github.com/FrakGart/psma-admin-bdy-2020-08 and compared in JOSM to the current boundaries, and couldn’t see any significant changes around Burnie. So the suburb boundaries (or at least identifiable shapes along them) can be considered a high-quality data source, reducing the need for GPS traces to align items visible on aerial imagery. And if I’m wrong, at least it will be easier to offset everything correctly in future.