I’ve finally finished (I hope!) adding all the rest of the houses in Teversham parish. Doing the Foxgloves estate was quite a long job, so I hope no-one builds any more estates round here for a while!

Many surrounding areas have been getting fields filled in, so I thought I’d better do Teversham before it started looking like an embarrassingly empty gap in the map. This turned out to be a more complicated exercise than I’d expected!

There are several kinds of field boundary round here: tracks, roads, hedges & fences, drainage ditches, sometimes a change in ground level, or sometimes no boundary at all. Sometimes it’s ambiguous and I’ve had to make judgement calls: often a fence will have become sufficiently overgrown to constitute an effective hedge for some of its length. Ditches, too, often have bushy stuff growing, more or less impenetrably, along at least one side.

Sometimes, too, I can’t get close enough to a field boundary to be sure what it consists of, and nor is it clear from the aerial photgraphy. Mostly I’ve settled on a likely guess, but occasionally I’ve just had to leave the barrier out (but leave a gap between the fields).

Then there’s the question of whether the field areas should share nodes with the barriers which surround them. After a bit of experimentation, I’ve settled on generally sharing nodes with hedges and fences, but to offset from ditches, roads & tracks. This reflects the reality that farmers generally seem to plough pretty much right up to hedges and fences, but not to ditches, roads or tracks. It also renders reasonably attractively.

So there’s been a Bing imagery update since I last did any detailed work round here. Good news: the most-detailed-but-one level seems to be much better registered with reality than it used to be. Bad news: the most detailed layer is aligned worse, and the tiles vary widely in quality and alignment.

Anyway, I’ve added building outlines for all the houses in Teversham Village. What I’ve been doing is leaving the old nodes (which mark the front door locations) holding the address information, and putting new building=residential paths down which go through those nodes, but are not themselves tagged with address information, nor divided further than physical buildings. That seems more honest than guessing how the buildings are divided internally when a building contains more than one dwelling.

I've been working on Teversham Parish, including the Foxgloves estate adjacent to Cherry Hinton.

I started by improving the accuracy of the existing mapping, correcting a few errors and adding missing detail as I go. My basic strategy is to gather enough GPS data to be able to calibrate the alignment of the aerial imagery, but then to take most of the detail from the imagery.

Firstly I'm gathering a few well averaged waypoints: I'm following Garmin's advice here, and averaging 16 separate samples, taken at least 90 minutes apart, with each sample averaged for 30 seconds. I'm picking points which are clearly identifiable on the aerial imagery, and which have a good clear view of the sky (and specifically with a southern horizon good enough to pick up EGNOS corrections).

Secondly, I'm gathering many track logs (typically ten or more) along a few routes within the area of interest. These act as a cross check on the position of the averaged waypoints, and give me an idea of how accurate (or at least reproducible!) the GPS data is.

Then I'm firing up Potlach2, using space-drag to bring the imagery into alignment with my GPS data, and then adjusting the OSM data to match the (now calibrated) imagery. That's pretty much done now for Teversham Parish, though I may make a few slight adjustments out towards Fulbourn Tesco once I've finished getting another good reference point out that way.

I'm also checking against ground truth by visiting the mapped areas.

I just wish there was a way to leave a note for other mappers to warn them that the raw Bing data is a few metres out, so that they don't come along and unknowingly undo all the good work! I can't even work out what the best thing is to put in the source tag: I've been using "source=survey;bing" which matches some recommendations I found in the help pages (though I noticed that Potlach2 does warn me that I might want to think twice about giving the tag multiple values).

… See full entry