Suggested Software and Workflows

[KevinKepp_Cowlitz]

I work at a utility that is trying to utilize a Matrice 300 rtk drone and a Zenmuse P1 camera to map change detection on a power canal. We would primarily be looking for erosion along the banks. There was a failure some years ago and we would like to proactively mitigate any future failures. I was curious if anyone had any experience doing similar work or has suggestions on the efficacy of photogrammetry vs Lidar in this use case, as well as what software was necessary post processing. Thanks!

[JeremiahJohnson_Esri]

I think that photogrammetry would work great for this use case providing the banks are visible from the air. If shrouded by trees and shrubs then a more expensive option like lidar would be required. The ArcGIS Flight app is compatible with that drone and camera. It would be awesome to try to fly a "corridor scan" with terrain follow turned on. It would track up and down the canal taking photos at an interval appropriate for Site Scan or Drone2Map to turn into a point cloud/mesh/DTM.

[Troy_Griffin]

Hi Kevin. Let me preface this with the fact that I have only been doing this for a short time, so please don't consider this a definitive answer. But I am doing some similar tasks. I work for a public works dept and we are monitoring some erosion problems as well along a river bank. I have the M300 and P1 camera too and have found it to be an awesome combo. I experimented with Site Scan on a 2 week trial and liked it enough to get it. I have only had it a couple weeks now, so am still learning. I am using the "compare" feature. If you're not familiar, you can basically lay 2 models on top of each other and use a slider bar to compare images. You can also show a cut away or profile of the river bank just by drawing a line across the section you're interested in, and you can move that line on the fly as you look at a specific location. Right now, we aren't taking any intricate measurements, just comparing images visually, but you could.

I don't know if Site Scan is "the" answer for you, and after attending the User Conference last week I will be looking at a few other solutions as well, but so far Site Scan has been great for this use case. I have the L1 LiDAR camera on order, mostly because we have some vegetation issues in other locations as well as along our river banks. I think that decision depends on the degree of accuracy you desire vs. budget constraints. The P1 is doing the trick for us at this point. I hope that helps.

[DerrickWestoby]

Hey Kevin, 

I do some work like this (also with the M300 & P1) and the approach I take can work for quick/low-effort identification of areas, or more involved quantification.  The actual analysis is done by our water resources department, though.  

As long as the erosion is visible in the Z/elevation axis, you can create a derivative DEM by calculating the elevation difference between two temporal DEMs.    You can do this with the digital surface models (DSMs) directly from your photogrammetry software, and could have something to look over about 15 minutes after processing of your 2nd temporal dataset is done.  You're going to see a lot of noise and erroneous areas if you're using DSMs, especially next to water features and if there's vegetation present, but it's such a quick process and useful visual that I do the same thing for a lot of our construction projects.     If  you need to quantify the erosion though, you'll need to do some cleanup (and lidar likely would help this, depending on the site). 

Again, that's assuming you can detect the erosion in the Z axis.  I don't know of any way to auto-magically identify it in the XY axis, but I'm looking forward to reading other replies and suggestions to your question.  

If you have data from two different flights over the same area, I'd be happy to give you a quick step by step over a screenshare.  derrick.westoby@pbsusa.com 

edit: 
I forgot I posted a quick script for this on the global mapper forums, if you want to try.  In the post I talk about how I use it to check derivative surfaces for erroneous points or missing breaklines before finalizing, but it's the same idea for quick identification of change.  

I don't have any hydro projects on the computer right now, but here's an example on a new subdivision.  On-site features of interest that have changed are easy to identify visually, and the noise is simple to interpret and ignore.  

This isn't a deliverable, this is just what I see when I quickly interpret the "Elevation Change DEM". I ignore areas like "grass mowed" and "car", and focus on areas like "sidewalk", "stockpile" and "home" when updating survey drawings.

On a hydro project, I'd use this to quickly identify areas of interest (erosion region candidates), then I'd work on cleaning up the data and building a true DTM in those areas to quantify the change, before sending on to WR for analysis. 

[KevinKepp_Cowlitz]

I appreciate all of the replies, lots to learn. Thanks!

[CodyBenkelman]

There's no mention above regarding ground control.  If you're seeking to compare datasets from different days, it will be critical that you have good 3D control points distributed across the project area.  Since your area of interest is a linear feature, make sure you have control near the ends and multiple points along the canal.  "How many?" depends on the length captured in a single flight but I'd suggest a minimum of 10, using 6-7 for control and withholding 3-4 as check points.  Of course, more is better.  You'll want the accuracy of your control to be ~3 times greater than what you're seeking to measure, so if you want to detect changes of ~10 cm, aim for control accurate to 4-5 cm.   Last, I would recommend you survey permanent features (concrete corners) so you can use the same control points for multiple dates (and you can go back and measure control *after* the flight has been completed - e.g. if you decide you need more control points).  Temporary control markers are certainly usable but it may be visually confusing if you have temporary control markers in different locations on different dates.

[Jonah_Hall]

Hello Kevin! I haven't seen anyone comment on the lidar aspect much so I may try and add some information on that process. My colleague Jeremiah pointed out two considerations when it comes to leveraging lidar. The first being if there are any vegetation that could obscure the ground from the camera (grasses / shrubs) and the other being the cost. 

If you have used or looked at photogrammetry software (Site Scan for ArcGIS, Drone2Map, Pix4D, etc) then you may have noticed that they offer the ability to export not only digital surface models (DSM) but digital terrain models (DTM) as well. Now while each software or software suite may have a different way of generating those surfaces one common way is for the photogrammetric derived point cloud to be classified and then have those surface derivatives generated. Now, when your dataset has dense vegetation that obscures the ground ideally the software will not classify any of that area as ground. This will not effect the quality of the DSM as that dataset is only reporting the top elevation at the location, but areas with no classified ground points will have their elevations interpolated for the DTM. As you can imagine the larger the area obscured, the more interpolation and potentially more errors introduced into the dataset. 

One of the big advantages for lidar as a remote sensing dataset is the ability to penetrate vegetation and provide better coverage over these obscured areas. There are a host of other benefits that come with lidar as a data source but in terms of your application I want to simply highlight the penetration and z value precision. I do want to paint an accurate picture for this data source and say that with "great scanning power comes great processing". Generating raw lidar datasets will require additional steps compared to the streamlined workflows you may be use to with creating photogrammetry datasets post-flight. These processes are handled either by the lidar scanner manufacturer or from the sensor integrator you purchased the scanner. 

One approach which was highlighted above by Derrick Westoby that I would like to emphasize is this idea of identify then or quantify. While it is dependent on the scale of your operation, a tip 'n cue approach which leverages both datasets might be another viable solution. Leveraging photogrammetry or UAS imagery inspections along canals at a higher frequency you may be able to locate areas of concern and then capture lidar over said areas. The inverse would be infrequent lidar captures and then intermediary imagery collects at set intervals or post-event to provide a more up to date information. With either case and really any approach I will also highly emphasize what my colleague Coty said about the necessity of ground control for any form of comparison analysis. 

Best of luck and I am always happy to discuss more!

Jonah