GIS Gurus! Skiing-related project help please

[always wright]

I'm currently producing my final project for a GIS class. I'm trying to plot an avalanche safe least-cost route through the Whistler backcountry. There are two operations I'm having trouble with, and I know there are some GIS guys on this board. I would really appreciate some help.

First problem:
I'd like to tell ArcMap to stick to ridgelines whenever possible, but I'm not sure how to go about it at all. Any suggestions?

Here's how I propose to find all the ridges:
- I'm starting with a 30m DEM.

- If I smooth it out to 50m, then it will give me a more generalized picture of the terrain, smoothing out all the ridgelines and bowls a little bit.

- If I then subtract the 30m DEM from the 50m DEM, I will have a map full of positive and negative values. The positive values should be convexities like bowls or gullies, and the negative values should be ridges. (I'm pretty sure that's right, but I might have this backwards)

I can then make this one of the factors in my least-cost analysis (i.e. assign all negative DEM values a low friction).

Does that make sense, or is there an easier way?

Second problem:
I've produced least-cost paths before, but they were always isotropic surfaces. Mountains clearly aren't (i.e. higher friction values going uphill than skiing downhill). How do I account for this? I'm sure ArcMap is sophisticated enough to know when my path is going uphill or downhill, but I don't know how to get the program to cooperate.


Thanks! (and to pre-empt everyone else, NEEEERRRDDS. Sorry to steal your thunder.)

[geomorph]

I'm no guru, just a hack that uses GIS. For your first point, I don't think you need to re-sample the DEM and then subtract values. I would create an accumulation array (in spatial analyst tools, you have to create a flow direction array first). This accumulation array will have high values in the valley bottoms and low values on the ridges, you could define some threshold value that would define the ridges and use that as an input to your project.

Sorry, no help on the second problem.

[elesquiador]

Google 'Terrain Shape Index'

[Telenater]

Another very worthwhile terrain index is the topographic position index.

I'm sure that there's a way to weight based on the direction of movement, but the standard least cost path won't be able to do it in a stock form. Most least cost path analysis tools consider only the cost of moving/residing in a cell not the directionality of movement.

As mentioned flow direction can be used to get the most downhill cell from each other cell, but for what you're doing, you might do better using slope and aspect.

You might also look into a gated least cost path as a possible improvement over a pure least cost path, because it demonstrates a range of costs (ie, select all cells that can be on a path within 2% of the cost of the very lowest cost).

[always wright]

Great stuff, thanks a lot. That definitely gave me a good starting point for google searches. I'm still stuck on the anisotropic surface though. I may have to forget about that part... if I only include the criteria that make the path 'safe' then it becomes isotropic, and then it's a simple least-cost analysis. I'm basically getting graded on how complicated and accurate my analysis is though, so I'd love to be able to include it.

edit: I think I actually may have found something that works. Path distance calculates cumulative cost while accounting for horizontal and vertical factors. This may be what I'm looking for.

[Alpenglo]

No guru either but here's my advice for what its worth. I wouldn't try to reinvent the wheel and smooth/subtract the two DEM's to find the ridges. As geomorph mentioned the accumulation function will be better anyways. Also the slope curvature (convexity vs. concavity) will be usefull as well. Keep in mind that it some non-ridge features will have a similar value though.

Is the 30m data the best available? I would be extremely cautious about using such a course dataset for any type of avalanche/ complex mountain terrain related study. Depending on scale of course. which is what by the way?

My advice would be to not to worry too much about building a one-size fits formula/function to determine the route. Unless you have a very high resolution DEM and mad programming skills that is. Use several functions/formulas - slope/aspect/accumulation/curvature, etc. - as input into an ultimately human-powered decision making process to plan the route. Avalanche/mountain terrain and conditions are ridiculously complex and a experienced mountain traveller will do a better job interpreting the various function outputs than any grand master GIS formula will. Remember that you are smarter than arcview (at least when it comes to avalanches anyway.) Maybe use photos as a reality check to validate the model?

[ak_powder_monkey]

maybe use the slope function in spatial alalyst

[Scrubadub]

Are those 30 m DEMs the highest res you can get? Not sure in Canada but you can get free 10 m in the US, and with a little cash (and some know how) you can make high res DEMs using radar interferometry of ALOS data.....get the files from Alaska Satellite Facility

[mcsquared]

I once wrote a FORTRAN program that attempted to do the same thing you are doing. So fuck you for making me feel old.

Subtracting the two DEMS doesn't sounds right to me though. Because it seems to me that you would want to stick to flat areas as well not just ridges. Unless I am misunderstanding the concept. I would probably stat by calculating the slope and aspect and then do some other stuff and then give up and then google a little bit and then find this: http://gislounge.com/how-to-select-r...atial-analyst/

[always wright]

No guru either but here's my advice for what its worth. I wouldn't try to reinvent the wheel and smooth/subtract the two DEM's to find the ridges. As geomorph mentioned the accumulation function will be better anyways. Also the slope curvature (convexity vs. concavity) will be usefull as well. Keep in mind that it some non-ridge features will have a similar value though.

Is the 30m data the best available? I would be extremely cautious about using such a course dataset for any type of avalanche/ complex mountain terrain related study. Depending on scale of course. which is what by the way?

My advice would be to not to worry too much about building a one-size fits formula/function to determine the route. Unless you have a very high resolution DEM and mad programming skills that is. Use several functions/formulas - slope/aspect/accumulation/curvature, etc. - as input into an ultimately human-powered decision making process to plan the route. Avalanche/mountain terrain and conditions are ridiculously complex and a experienced mountain traveller will do a better job interpreting the various function outputs than any grand master GIS formula will. Remember that you are smarter than arcview (at least when it comes to avalanches anyway.) Maybe use photos as a reality check to validate the model?

30m is the highest res. available to me. The scale is 1:50,000. This project is for school, and it's more just proof-of-concept, so it won't actually be implemented. I'm well aware that any software model is insufficient for either avalanche prediction or route planning in hazardous terrain. Besides proof-of-concept, the point of my project is to demonstrate the limitations of the program, so that's fine.

[Adolf Allerbush]

I think the approach you've listed is not going to yeild the results you want. Like others have said I would use slope/aspect tools of either spatial or 3D analyst to do this. And then some human decision making.

Also, maybe it would be better to choose an area where you have better DEM data? Most of the U.S. has better than 30 m accuracy. If your project is to show the limitations of ArcGIS/ArcINFO/ArcVIew then using this relatively course surface will only indicate that the data sucks...instead of showing that the program has limits. I suggest revising your study area to somewhere that has better data, then using some slope/aspect tools etc.

As for your second problem it sounds like you'll have to write some code for that one IMO.

[always wright]

I once wrote a FORTRAN program that attempted to do the same thing you are doing. So fuck you for making me feel old.

Subtracting the two DEMS doesn't sounds right to me though. Because it seems to me that you would want to stick to flat areas as well not just ridges. Unless I am misunderstanding the concept. I would probably stat by calculating the slope and aspect and then do some other stuff and then give up and then google a little bit and then find this: http://gislounge.com/how-to-select-r...atial-analyst/

This method seems to be inaccurate; perhaps the problem is that I'm using a 30m res. It identified low points very well; I overlaid a hydrology layer on it, and it reflects the position of streams and rivers very well, but it identifies everything else as a ridge. Even Whistler Village is identified as a ridge, and it's in the valley bottom.

[khakis]

I think it's identifying your valleys better than your ridges because valleys are generally wider than ridges. Small features don't show up on coarse DEMs. As for using a least cost path to follow ridges, have you thought about using raster calculator to turn all your data negative? Then all your ridges turn into valleys. If 3d analyst doesn't like negative values, add 5000 (meters, right?) to your negative raster. Then you complete your analysis, convert the z-values in your output back by adding the 5000 back and multiply by -1 again. Sorry I can't give any more specific advice, haven't used 3D analyst in a while. Sounds like a sweet project...wish I'd have thought of something similar when I had this kind of assignment

[wanker]

general data info - the 10m srtm dem's are generated from the 30m data. so they're really not that good. 30m has +/- 16m vertical accuracy, 10m has +/- 20m accuracy. for a general proof of concept like this the 10m would probably be ok, but for scientific measurements and analysis stay away.


always_wright - problem #2. what layers are you using to build your final cost surface? are you not using the dem as a weighted input?

[gordons]

could you create a feature class and manually trace the ridge lines with the edit function and then convert vector to raster and use that as a cost layer with the highest elevation the least cost?

p.s. i am a student at Georgia Southern and i have to say that your project is very interesting

[Telenater]

As an interesting, side note, when I was learning GIS, many years ago (ArcInfo 7 on Unix) I did a somewhat similar project looking at trail routing for erosion mitigation, but mostly succeeded in demonstrating that 30m resolution DEMs and the available soils data couldn't do what I wanted.