Closed Hornbydd closed 5 years ago
dsear
commented
6 years ago Nice work Duncan. Thinking forward to extraction of data for each segment I think the 500m plus bridges segments works best, otherwise you're going to get alot of really weird artefacts of the segmentation carried forward into any analysis / calculation of metrics.@joewheaton I think this was the issue about segmentation scale influencing outcomes that you and I were talking about yesterday.
Hornbydd
commented
6 years ago @dsear and @joewheaton - Good news I have worked up the logic to segment the network based upon bridge and regularly spaced points. It's a meaty model calling several sub-models to ensure the data is carrying the correct attribution. The tool is working very well. An example of output is shown below:
Red squares are bridges, black points are regularly spaced points, each point is given a unique ID, point type and a measure distance from the end of the branch. The river has been colour coded by segment ID to show change in ID.
The model is this:
The model actually generates tributary points too but does not use them in the splitting process.
I also attribute the segments with a Y/N field indicating if the polyline is a result of a split process. Below is a zoomed in section of the network colour coded by this attribute. Red indicates a line that was generated by a split. I've included this as this (if YOU think it OK) is a way of simplifying the network, all the blue lines are less than 500m and were never split so this could be a criteria to delete them out as we will never (?) be attributing against them.
Some simple statistics on the segmentation process are (from my sample data): 2,154 segments. 1,293 (60%) are segments created by a split point, which means 40% of the network was never split. A simple sum of distance by wasSplit is 136,645m is N and 533,445m is Y. Thus by total distance non-split segments account for 20% of the network.
So another way of saying it is if you dropped all the blue lines in the image above which is 40% of the segments, it still leaves 80% of the network by length.
This model works because the OS waterLayer has been pre-processed.
The image below shows a sub-catchment of the Derwent labelled by BranchID, a single branch has been selected.
These branches are calibrated Polyline M features which can be used by linear referencing tools to measure and map against. They CANNOT be used for classic trace like functionality as they are nothing more than a bunch of lines that happen to look like a river network to us but are really lines that start and end somewhere but importantly do not connect at nodes.
Hornbydd
commented
5 years ago Tool is complete and no further development is currently planned. It is likely that it will require further development if we try to roll it out for national level processing.
@joewheaton and @dsear I have been working up the logic to segment a river network. I am unlikely to use the existing segmentation tools because of their reliance to pre-populated USA attribution and that David has specified that the segmenting of the network should include points that are:
So all 3 sets of points will be used to cut up the network. I've had a quick chat with David to make him aware of the implications of this and would like your opinion. We think dropping tributary junctions as split points is the way forward for the following reasons...
Below is a screen shot of all 3 split point types (tributaries, regular distance (500m) and bridges)
If you use ALL 3 point types then in the sample data I am using you go from 1,189 branches to 3,239 segments, their distribution of lengths are:
Segmenting with all types of points creates segments ranging in length from 7cm up to 500m with 12% of segments attaining 500m.
Compare this with splitting only with bridge points and 500m distances. You go from 1,189 branches to 2,154 segments.
One still gets small segments (to be expected) but you have at least 43% of the data achieving a 500m length.
If I zoom into some random part of the Derwent, this is all split points:
This is the same area with tributary points turned off:
So the question is do we segment with just bridges and regular 500m spaced points or do we segment using bridges, regularly spaced points AND tributary junctions?