I'm having trouble with generating service areas (SA) in network analyst and spent some time playing with the tool, the data and reading posts on this forum
Firstly, my SA polygon layers are very holey when I don't want them to be - and opting to use the generalized polygon building method doesn't make much of a difference other than moving the holes about. Holes range from small to medium sized though all are much smaller than the break point for the polygon building. In this case I'm trying for exclusive (non-overlapping) SAs across and extent and anticipate that the whole extent should be covered. Sometimes the holes include network edges (which I've checked are both connected, relevant and well within the SA breakpoint), but more often describe gaps between network elements and mark the points at which neighbouring SAs should touch.
Ideas? - this is for a critical process and simply using some automated non-contextual method to fill/simplify the holes in the coverage of SAs is not enough.
Next while poking this problem about I've been trying different numbers of facilities across the same network and alarmingly achieve different SA output depending on the number of facilities used (obviously ensuring they are contiguous across successively smaller extents). Its nothing obvious like edge effects (intriguingly the boundary SAs are often identical regardless of facilities used) but there is enormous variation in how polygons around the same facilities in the interior are constructed and the holes that appear between them)
The fewer facilities I use the fewer holes there are
This suggests some computational issue (i.e. memory / math?) but I'm well short of the limits of what this machine should theoretically be capable of.
So the final analysis was hoped to include 4000+ facilities across a single road network
I've been playing with a test subset of 490 facilities (from one area) and been successively reducing this number until I get to 75 but even then I've got persistent holes within the interior area of the facilities.
1. to remove holes 2. to create stable SA building behavior so that I can faith in the output 3. how I can run or chunk the 4000 facility problem
As far as the holes in the coverage of service areas goes, it seems to have something to do with the very variable nature of the network I'm using with incredible density in some parts and very sparse in others. In practice I've given up trying to figure why the holes are there and solved the issue by using a uniformly more rich network. Tonnes of extra processing and work but I'll end up with no functionally important holes and a more robust map. the SA polygon generator still produces weird small missing areas in polygons but these are small enough for me to 'fill' them non-contextually by merging them into their smallest neighbour.
As far as the inconsistency in SA areas when using different numbers of facilities, its still foxing me but I'm finding that network analysis is a bit fragile on my machine - and once it's kicked up an error it seems to wobble in subsequent operation afecting everything else. Thus at the moment I'm rebuilding a test network for the third time today - each time its built differently dropping edges at random, for no reason. Restarting Arc with a fresh instance and picking up a clean version of the data seems to work wonders, so it seems that I'm back to where I was back in day with arcview 3.0 - constantly saving individual steps because the application would wobble at the slightest provocation, not work or corrupt data.