I have some alignment questions that I am sure many people have spoken about in the past but I can’t find such discussions.
When aligning the cadastral overlay with aerial imagery. How big an area do you consider the offset to be fixed for? And when you have some “best alignment” for a few streets do you draw the edge of buildings where you see them or snap them to the cadastral line when it looks like it is built up to the boundary? Or do you just say to yourself we can’t get more accurate that a meter or 2 and try not to spend too long worrying about it?
Have been doing quite a lot of building splitting for terraced houses as well as geometry improvements, so I get the alignment struggles. I tend to consider the aerial imagery offset to be good for a couple of streets, but I think it can be pretty variable - if heights of buildings are changing then that will throw it off. I try to snap to the cadestral lines wherever it looks like structures conform to the land borders, as I believe they are theoretically meant to be aligned perfectly.
When I am unsure about alignment I tend to map with less detail rather than micromapping, as it’s important for the base structures to be aligned correctly before adding too much detail. My experience in trying to realign houses with individually mapped gardens in the ID editor has been very poor.
I generally consider it still aligned until I notice that it isn’t? With the way the mosaics are stitched I don’t think it’s terribly useful to have a fixed distance. Sooner or later you notice that a fence is offset and drag the imagery back into alignment. I think it helps to have multiple imagery layers to cycle through as they drift at different rates.
Thanks. Specifically I was mapping here and was dragging the imagery offset around often but now the buildings don’t all match the imagery at the same time, which looks wrong to me. That would only be correct if there was a lot of stiching in this imagery. Or I should have picked a fixed offset and assumed there is some discrepancy between fences and actual boundaries. In the latter case I would need to make some judgement of the scale over which to do a “global” best offset fit.
I think I’m right in saying that because fences are shorter than the buildings they will naturally need a different offset to line up compared to the houses, because of angles™. So I don’t think that it is unusual for them to need different offsets, aka. there is no one “global offset” that will work for all structure heights.
Thankfully fences usually follow the cadestral boundaries so I tend to just use one offset for the buildings, then follow cadestral boundaries for the fencing unless satellite imagery suggests a clear deviation.
I would probably focus on ensuring that that houses offset prevents houses from crossing cadestral boundaries, and that any semi-detached or terraced splits follow the cadestral overlay too.
You should align the imagery so the ground-level aligns with the Land Registry parcels. Parallax (from the imagery being taken at an angle rather than vertically) means that the tops of buildings (and to a lesser extent fences) won’t then be aligned. Normally you can see enough bases of fences or buildings to be able to make this alignment.
When mapping, if you want to trace a building, if you think the cross-section at the top is the same as the footprint, then trace round the top, and then move the resulting way to correct for the offset of the top relative to the bottom. For multiple houses of the same height you can save time by tracing a lot of them, selecting all the outlines and moving them all in one go since the offsets should all be the same.
If the ground is flat, then the imagery offset once corrected should be reasonably good over a few hundred meters, unless it’s been stitched together with a different image source, and then there might be a visible jump in the offset along a line. If the ground is not flat, then parallax may mean the offset can’t be relied upon when the ground level changes by more than a couple of meters.
For residential housing estates I like to align with pavement edges around junctions. They’re effectively at ground level so parallax isn’t a concern, and intersecting streets improve the accuracy in both dimensions. Of course this only works with the right vintage of housing estates.
When tracing building outlines from aerial imagery, I follow exactly the same process as described by @Robert_Whittaker.
I have noticed that stitching lines can run down the middle of a road. Then you have different offsets on the two sides of the road.
I have found that offsets are typically good for a couple of hundred metres.
There is another technique which can assist with aligning imagery if you are using JOSM. This uses the 1:1,250 and 1:2,500 Ordnance Survey plans from the National Library of Scotland. You set up three layers in JOSM, the aerial imagery at the bottom, the relevant OS plan in the middle, and the cadastral overlay on top. First align the OS plan to the overlay. They usually match very well so the alignment is easy to do. Then hide the overlay and make the OS plan partially transparent so you can see the aerial imagery underneath. Finally align the aerial imagery to the OS plan. The OS plan potentially gives you a lot more details against which you can align the aerial imagery.
I like to get the imagery aligned to within one pixel in each direction (on a non-HiDPI display) at zoom 19. One pixel equates to 20cm on the ground. (JOSM and iD report the offset in projected coordinates which comes to 30cm per pixel. To convert from on-the-ground to projected, you need to divide by the cosine of the latitude.) In JOSM, if you click on the background, you can then use the arrow keys to move the imagery in one-pixel steps.
Have I understood correctly that you move the imagery until the pavement edges are at equal distances from the parcel boundaries on the two sides of a street? Or until the edge is at a constant distance from the boundary as you go round a corner? Usually a parcel boundary is at a wall, fence or hedge or at the edge of a front garden (rather than at the kerb).
I came across one case where I could not convincingly establish the offset. This was in spite of getting assistance from the OS plan. The offset varied by up to three pixels at zoom 19 depending on which object I used for the alignment. This was in the area from Brown Hart Gardens to Oxford Street in central London. In the end I went round with an RTK GPS to nail down the offset.
I don’t want to discourage your meticulous pixel-level work, which is an admirable level of commitment.
Rather just for general consumption, I was persuaded by a post in the dim and distant past by @SK53 that accuracy within 5 metres was just fine, and the most that could be hoped for, or needed. I think I recall him talking about multiplying errors/tolerances, but I may be mixing it up with something else. I more clearly remember mention of all mapping points being relative anyway - the relationship between nearby elements is much more important than the relationship between each element and Newlyn (or choose your own datum).
Still, I have to restrain myself sometimes from mass aligning an area to +/-5 metres, when it’s already good to +/-10m. Doing so would undoubtedly improve the map - but in most cases I feel it would improve it less than, say, an outing to sort out a couple of bus stops (which would probably take less time).
As you say, the relative distances between features is the most important part of the map geometry. But I would suggest that ±5m is not good enough accuracy for most things. A terraced house could be 5m wide, so operating at that error margin would mean that (for example) UPRN overlays could be out by 1 or 2 houses along a terrace.
Personally I try and get the alignment to within 0.5m, ideally closer to 0.1m if the Cadastral Parcels line up well with other features in the area I’m editing.
That makes sense. It may well be the discussion was pre-cadastral parcels data, and was the best that was then possible. idk, but your UPRN is a good example of why it matters - or maybe better, can matter.
EDIT - I think what I was trying to drive at was a general sense of “don’t let rough and ready alignment stop you mapping, because you can still be useful”. But now I’m wondering if that’s not quite right. Is it always going to store up problems for later if you don’t, so you should always spend (what seems to me at the moment) disproportionate time getting it right first time?
Yeah, I think that it’s more important to take the time to get geometry right, because it’s more of a pain to fix it afterwards (especially as more of the map gets built up around it, co-aligned). Tagging is relatively easy to improve afterwards.
I don’t think I’ve done it in the UK, but I’ve seen areas where it’s easier to create a blank layer in JOSM, trace all the buildings from scratch and then use the conflate tool to merge all the new geometry across.
