network tool on very dense trail networks

[mtbikernate]

I'm accumulating tracks for a trail network in town and I'm running into some troubles.  I'm trying to use the network function to merge everything together.

This trail system has probably 5-8 miles (I haven't ridden the advanced stuff with the stunts yet) of trail on 64 acres, for reference.  This trail doubles on itself a lot.  The network function is having a lot of troubles determining intersections and deciding what to reduce.  In some areas, it's telling me there's an intersection where none exists...just that two trails pass so close together, there might be a little overlap from one ride to another.  Another place I'm having trouble is that there's a couple loops on the trail that TF is reducing entirely.  On one, the loop is reducing to a dead end, and on another, the loop (which serves as something of a central ending of all the trails at the bottom of the hill) is getting reduced to nothing more than a single point.

I've dropped the tolerances as low as possible on the algorithm.  If I make them any lower, it won't even reduce two tracks on the same trail from two different rides with different reception.

Any suggestions about how to deal with such a dense network of trails?

I used to do this sort of thing by hand, and if the network algorithm can't handle such a high density, them I'm going to have to continue doing it by hand.

[ScottMorris]

Hmm.  This is an interesting problem.  I haven't dealt with too many dense trail networks, so my experience is limited.

If the density of the trails is quite often tighter than the expected GPS error, the method is pretty likely to fail to represent the actual trail network accurately.  The problem is, as you noted, when the tracks of two close trails actually intersect.  The method considers them to be the same trail.  There's no provision for considering the elimination of an intersection that might have been erroneous.

I could think about how this could be done.  It's certainly a weakness of the method and one worth thinking about.

As for the loops being eliminated or turned into dead-ends... it would be best to turn the contraction strength all the way down.  You can still leave the reduction strength higher than zero, so that you do get reduction of multiple traces into a single track.

You might be able to get some good results by just hand editing to eliminate intersections of trails that don't actually intersect.  But there may be too many of those, as well, so it might not be feasible.

I'll let you know if I come up with anything.  Maybe request some test data if I make some progress on it.

Thanks for the post.

[mtbikernate]

It is difficult because the trails are indeed sometimes so close together, they're within the margin of error for a GPS position fix, and when you have two tracks on different days where the two actually do intersect, that seems to be where the problem arises.

I've tried reducing the contraction strength, but it ends up giving me results with hundreds of intersections where each individual track for the same trail crosses...to get the contraction to an appropriate level so it doesn't excessively overdo the intersections, I am left with an unsatisfactory result otherwise.

It does give me the loop, but it puts shortcuts and intersections and other things in there that don't really exist just because there's some degree of error inherent in each individual track.

Right now, I'm working with making a network out of maybe 9 different files that more or less all follow the same trails (in different order).  Like I said, the network is even more dense than that, but I want to get what I have merged before I go and add the other trails.

[kevroc]

I know this is an old post, but I'm just researching the network feature more and thought I'd put this on record for future searchers...

It sounds like the best thing to do is to use the recorded tracks as a guide and draw the track lines by hand, once they are manually drawn (where you could eliminate the gps errors and track overlaps) and get relatively good tracks, you could then create the network.