Colorado LiDAR Findings

[greenonion]

For you map lovers and geospatial folks… I just stumbled across this beautiful map producer, Muir Way: https://muir-way.com/collections/colorado

Taint cheap but they are gorgeous!

[bdloftin77]

In the Sangres, the saddle between UN 13060 and UN 13028 might warrant a looksee. I was on top of Rito Alto recently, looked to the west, and it made me wonder. It could've been an optical illusion with a ridge in the foreground. Nonetheless, if you're itching for another peak...

This doesn't have TNM coverage. If I download Saguache from the Colorado Hazard Mapping website (OIT) and it does have coverage, I'll let you know what the result is. (or if you download Saguache, and send me the tiles they're there, I'd definitely give a quick response).

One more that would be really interesting.

What does the LiDAR say about Kings Peak and South Kings Peak?

Since 1966 the northern summit has been considered the highest.

What if the pre-1966 survey was right all along? That would mess up a lot of state highpointers. I have to admit that South Kings certainly does look higher from some directions, but of course that isn't definitive.

That would be really interesting! Looks like Utah doesn't have TNM coverage there. If you happen to find a LAZ or LAS tile for Kings and South Kings and send it over, I'd definitely take a look.

https://prd-tnm.s3.amazonaws.com/LidarE ... dex.html#/
^ This map shows all the areas where you can easily download single LAZ tiles and analyze them. For Colorado, the only other website that I know of is the Hazard Mapping one (OIT) (go to Lidar Request Form, then FTP download page link). To look at peaks from their data though, you have to download entire counties (usually a couple hundred GB) of data. And sometimes after downloading, you realize that it only has parts of the county available, others are missing/not covered by their tiles.

[bdloftin77]

What happened to the 13ers right around the 4000m line? Such as Cupid and Isolation peak in the Front Range (both currently just 5-6ft shy of the 4K mark).

For "Cupid," I got 13,126'. For Lizard Head, I got 13,112'. I'll include these in my next submission to John, and hopefully almost all the available 14ers as well.

Isolation isn't yet available via TNM or OIT unfortunately. 13,122 is supposedly available via OIT, but when I tried to find the tile for its coordinates via an FTP download, it didn't exist. Might have to either do a special request for that one, or wait til they finish uploading all their data to their new website. (13,122 might not even be ranked)

It looks like the others around the 4000m boundary are available. I'll add those to my analyzation list. I'm trying to get the 14ers set up and sent out, then the remaining Centennials, then it's more up in the air. For the Centennials at the very end of the list, I'll probably spend some time looking at pictures and videos confirming the summits are as I was expecting. For the rest, it might take a decent amount of time if I end up looking at summit videos for each one. It's hard to find good videos/pictures of the summits themselves, and not of the surrounding scenery or selfies. I'll see what I can find.

[glenmiz]

What happened to the 13ers right around the 4000m line? Such as Cupid and Isolation peak in the Front Range (both currently just 5-6ft shy of the 4K mark).

For "Cupid," I got 13,126'. For Lizard Head, I got 13,112'. I'll include these in my next submission to John, and hopefully almost all the available 14ers as well.

Isolation isn't yet available via TNM or OIT unfortunately. 13,122 is supposedly available via OIT, but when I tried to find the tile for its coordinates via an FTP download, it didn't exist. Might have to either do a special request for that one, or wait til they finish uploading all their data to their new website. (13,122 might not even be ranked)

It looks like the others around the 4000m boundary are available. I'll add those to my analyzation list. I'm trying to get the 14ers set up and sent out, then the remaining Centennials, then it's more up in the air. For the Centennials at the very end of the list, I'll probably spend some time looking at pictures and videos confirming the summits are as I was expecting. For the rest, it might take a decent amount of time if I end up looking at summit videos for each one. It's hard to find good videos/pictures of the summits themselves, and not of the surrounding scenery or selfies. I'll see what I can find.

This sounds like a labor-intensive effort and very much appreciated. It's intriguing to see play out. Do you think it will it ever get to the point to be automated and, therefore, systematically done across large areas? For instance, I finished the Jefferson County ranked class 1-4 this year, and I'm curious if any "new" entries might pop up and push me back to real bushwhacking again.

Thanks again for all of the work you and John are doing. Glen

[bdloftin77]

It looks like the others around the 4000m boundary are available. I'll add those to my analyzation list. I'm trying to get the 14ers set up and sent out, then the remaining Centennials, then it's more up in the air. For the Centennials at the very end of the list, I'll probably spend some time looking at pictures and videos confirming the summits are as I was expecting. For the rest, it might take a decent amount of time if I end up looking at summit videos for each one. It's hard to find good videos/pictures of the summits themselves, and not of the surrounding scenery or selfies. I'll see what I can find.

This sounds like a labor-intensive effort and very much appreciated. It's intriguing to see play out. Do you think it will it ever get to the point to be automated and, therefore, systematically done across large areas?

No problem! As far as automation, John is doubtful. Summits could be relatively easy - look for the highest point within a certain radius of the former high point coordinates. For summits above treeline, both classes could be used, and for summits below treeline, class 2 only could be used. Rocky summits below treeline might be missed, and tall cairns/radio towers above treeline would be incorrectly included (or if class 2 only is used, basically any rocky feature above treeline would also be missed). These could be error-checked afterward, but those are potential issues. Downloading the data might be an issue - we don't have all the lidar tiles in Colorado (thousands) pre-downloaded; we have to find the summit and saddle locations on The National Map, then download those specific tiles. Maybe a program could download certain tiles from that website based on coordinates, I'm not sure. Saddles are much trickier - there are sometimes multiple saddle candidates (multiple areas with the same contour elevations), and each of these has to be looked at. On LoJ, only one of these has coordinates for each peak. For John's program, he has to input either horizontal or vertical rectangle bounds specifically tailored for each saddle to avoid including false high points.

Analyzing DEMs would possibly be easier as each DEM square touches several others. But these are less accurate than lidar point-clouds. For lidar point clouds, the points are generally close to each other, but there are sometimes spurious high or low points. There's also areas without data, so this might cause issues.

For peaks where high accuracy is wanted or needed (eg fringe 12ers and 13ers, soft-ranked peaks, error-range peaks), lidar point-cloud data (LAZ or LAS tiles) are desired and probably mandatory. For other peaks though, perhaps John might eventually settle with an automated process. LAZ or LAS tiles would be preferred, but perhaps he might be okay with DEMS. For the saddles, regardless of if a process is able to be automated or not, all the saddles would probably have to be at least glanced over manually to make sure saddle candidates aren't missed.

If anyone is interested in creating an automated process for analyzing Colorado's peaks and come up with a good workflow/program, let us know! 4000+ peaks will definitely take a very long time to analyze by hand, especially since both John and I have full-time jobs.

For instance, I finished the Jefferson County ranked class 1-4 this year, and I'm curious if any "new" entries might pop up and push me back to real bushwhacking again.

Thanks again for all of the work you and John are doing. Glen

For this, we'd probably only have to look at soft-ranked peaks. I could bump this up on my priority list, since there might be other folks who would be interested in new Jefferson county summits. Looks like there might be ~30ish soft-ranked peaks in Jeffco.

Boulder and Larimer counties might have a lot of interest, but these don't yet have full downloadable lidar coverage. It still might be worthwhile to finish what we can in those areas sooner as opposed to later.

[Boggy B]

I've been fiddling with code to consume a point cloud and return all points with a given amount of prominence.
This is a slightly different challenge from locating exact summit/saddle points within a predetermined area. It only took an evening to get a working algorithm for DEMs, but as you mentioned, DEMs aren't terribly accurate as they only use ground class and lose horizontal granularity due to mangling the geometry. And therein lies the main difficulty with LiDAR data: returns have nondeterministic adjacency, so writing an efficient algorithm to discover whether a path exists to a higher point which doesn't drop below the given prominence threshold is tricky (trickier than a nearest-neighbor calculation, since what you actually need is the tightest perimeter of neighbors to enable search of a perimeter that recursively expands from a given point).
For above-treeline peaks, totally bogus points should be easy enough to filter out programmatically, but results would still need to be manually checked to rule out summit cairns (or people, though 12ers and 13ers tend to not have people on them). Below-treeline peaks introduce more complexity.
I don't see point classification ever distinguishing between a cairn and a natural pile of rocks, so some level of manual checking will always be required unless we're ok counting a cairn as the highpoint (personally, I don't care).

[bdloftin77]

I've been fiddling with code to consume a point cloud and return all points with a given amount of prominence.
This is a slightly different challenge from locating exact summit/saddle points within a predetermined area. It only took an evening to get a working algorithm for DEMs, but as you mentioned, DEMs aren't terribly accurate as they only use ground class and lose horizontal granularity due to mangling the geometry. And therein lies the main difficulty with LiDAR data: returns have nondeterministic adjacency, so writing an efficient algorithm to discover whether a path exists to a higher point which doesn't drop below the given prominence threshold is tricky (trickier than a nearest-neighbor calculation, since what you actually need is the tightest perimeter of neighbors to enable search of a perimeter that recursively expands from a given point).
For above-treeline peaks, totally bogus points should be easy enough to filter out programmatically, but results would still need to be manually checked to rule out summit cairns (or people, though 12ers and 13ers tend to not have people on them). Below-treeline peaks introduce more complexity.
I don't see point classification ever distinguishing between a cairn and a natural pile of rocks, so some level of manual checking will always be required unless we're ok counting a cairn as the highpoint (personally, I don't care).

Nice. I'm curious.. could you explain the filtering points by prominence part? Do you find the saddle, and then it filters out points below 300' of prominence above the saddle? Or is there some other "zero point" from which you measure the prominence (maybe the summit, then filtering downward)?

I think John honestly isn't too concerned about rock cairns or small man-made features. Most of the peaks he's looking at probably have smaller or non-existent cairns, and it'd be hard for his algorithm to determine if the high point found is a cairn or not. I'm probably getting into the weeds too much with trying to filter out cairns for the 14ers and Centennial 13ers.. but those seem to be the most important peaks to many people here, so I'd like to try to get as accurate of elevations as possible. Unless there's a known large structure on a certain peak (radio tower, huge cairn on Phoenix peak, etc), or if the peak is a borderline peak (edge of 13ers), or if it's competing with other peaks for the edge of a popular list (lowest Centennials), it probably doesn't matter a whole lot if a cairn is caught or not.

[Boggy B]

Nice. I'm curious.. could you explain the filtering points by prominence part? Do you find the saddle, and then it filters out points below 300' of prominence above the saddle? Or is there some other "zero point" from which you measure the prominence (maybe the summit, then filtering downward)?

Every point that is higher than all adjacent points is a peak with >0 prominence. The algorithm searches an expanding perimeter from this point until either all points on the perimeter are lower than the given prominence threshold, or one of the perimeter points opens a path to a point that is higher than the original. If the former, the point has at least the given prominence; if the latter, the point that opened the path is the saddle.

[seano]

Nice. I'm curious.. could you explain the filtering points by prominence part? Do you find the saddle, and then it filters out points below 300' of prominence above the saddle? Or is there some other "zero point" from which you measure the prominence (maybe the summit, then filtering downward)?

Every point that is higher than all adjacent points is a peak with >0 prominence. The algorithm searches an expanding perimeter from this point until either all points on the perimeter are lower than the given prominence threshold, or one of the perimeter points opens a path to a point that is higher than the original. If the former, the point has at least the given prominence; if the latter, the point that opened the path is the saddle.

Would Andrew Kirmse's algorithm that follows the steepest gradient up from saddles be faster than searching in all directions down from peaks?

[Boggy B]

Would Andrew Kirmse's algorithm that follows the steepest gradient up from saddles be faster than searching in all directions down from peaks?

The algorithm described there frontloads processing, first building a graph of all peaks and saddles, which is computation-intensive.

My approach is likely much faster for one iteration on a tile for a given prominence threshold, but it would be vastly slower in the aggregate, since it doesn't generate any structure that can be iterated with different inputs. I'm sure it's not the fastest way to process a single tile, either.

It would be nice to see that project revived for processing LiDAR data. I'd guess the geometry implies a non-trivial refactor on peak/saddle discovery.

[seano]

The algorithm described there frontloads processing, first building a graph of all peaks and saddles, which is computation-intensive.

My approach is likely much faster for one iteration on a tile for a given prominence threshold, but it would be vastly slower in the aggregate, since it doesn't generate any structure that can be iterated with different inputs. I'm sure it's not the fastest way to process a single tile, either.

It would be nice to see that project revived for processing LiDAR data. I'd guess the geometry implies a non-trivial refactor on peak/saddle discovery.

Thanks -- now I think I understand your goal. If your code works how I'm imagining it does (expand perimeters highest-point-first until they meet or drop below the prominence threshhold), then nothing better comes to mind. But I'm ignorant about the details of LiDAR data, so feel free to ignore me.

[Boggy B]

The algorithm described there frontloads processing, first building a graph of all peaks and saddles, which is computation-intensive.

My approach is likely much faster for one iteration on a tile for a given prominence threshold, but it would be vastly slower in the aggregate, since it doesn't generate any structure that can be iterated with different inputs. I'm sure it's not the fastest way to process a single tile, either.

It would be nice to see that project revived for processing LiDAR data. I'd guess the geometry implies a non-trivial refactor on peak/saddle discovery.

Thanks -- now I think I understand your goal. If your code works how I'm imagining it does (expand perimeters highest-point-first until they meet or drop below the prominence threshhold), then nothing better comes to mind. But I'm ignorant about the details of LiDAR data, so feel free to ignore me.

Ha. I wasn't aware someone had actually implemented a prominence algorithm(s), though I'm not surprised. They're way more credentialed than me (I've never worked for Google and slept through my algorithms courses in college), and they took a more scientific and scalable approach to solving the problem. I'm just trying to do something quick and dirty, though it's not proving to be quick.