"Snake Linda": The Field Science of Drone Photogrammetry

Kenzie McAdams

Purdue University

A new morning was upon us, and the familiar misty cloud that lives over Camp 17 greeted us and the new day. After a delicious breakfast of spacon (spam bacon) and oatmeal, we learned of our upcoming “Plan of the Day”. With safety training winding down and afternoons starting to become free for academics and field science, we were all given the choice of a variety of different opportunities to take part in for the day.

We started our day in the library, with a lecture by academic director Matt Beedle that introduced us to photogrammetry. Photogrammetry. The term was familiar, but I really had no firm grasp on what it meant. Matt described photogrammetry as “the science or art of making reliable measurements using photography”. During lecture, we explored the two types of photogrammetry: terrestrial (on land) and aerial (from the air). The focus of this lecture was on exploring and using Structure-from-Motion (SfM) Photogrammetry. SfM is a low-cost and effective tool that we utilized to obtain high-resolution datasets. The SfM method relies on multiple overlapping images that photogrammetric software uses to match points taken from the same feature in each image. With that in mind, Matt prompted us to think of some difficulties we may have with using this method on the icefield. One of the primary issues we would face in our attempts to image the icefield would be the lack of these identifiable features on the surface, due to it being a vast white expanse. Nearby Lake Linda, on the other hand, would pose as the perfect target for our experiment. The lake is a meltwater feature that drains subglacially down the Lemon Creek Glacier into the Lemon Creek valley. Due to its identifiable ridgelines and rock surfaces, it includes many reference points for the SfM software to do its analysis. Moreover, it is an ideal location to use aerial photogrammetry to model the elevation of the lake basin because it is not an easy task to obtain these measurements terrestrially.

 Team Drone in front of Lake Linda on the Lemon Creek Glacier. From left to right: Uwe Hoffman, Chris Miele, Matt Beedle, Deirdre Collins, Kenzie McAdams, Cézy Semnacher, Louise Borthwick, Alex Burkhardt, Lucas Beem. (Photo credit: Matt Beedle)

Team Drone in front of Lake Linda on the Lemon Creek Glacier. From left to right: Uwe Hoffman, Chris Miele, Matt Beedle, Deirdre Collins, Kenzie McAdams, Cézy Semnacher, Louise Borthwick, Alex Burkhardt, Lucas Beem. (Photo credit: Matt Beedle)

After a discussion on the pros and cons of drone flight path design, our initial task was to create two flight paths. During this discussion we chatted of techniques related to camera angle, area coverage, and possible difficulties with the surrounding topography. We students collaborated to design two paths: one that flew in a spiral pattern radially outward, with the camera facing at a 45-degree angle, and the other in a snake-like pattern with the camera facing straight downward. We had to think carefully about our camera angle choices, because if the angle of the camera were at the same angle as the slope of a feature, the image would not capture it. Student Alex Burkhart explains that by creating the spiral pattern, the 45-degree camera angle would continuously capture the rim of the lake as well as always capturing a portion of the concave basin. I helped to design the snake pattern, with the intention that by having the flight zig-zag over the surface with the angle facing straight downward, we would cover the most area and have the smallest chance at missing any features.  After we had our paths set, we journeyed out on our skis towards Lake Linda.

 A screenshot of the 3D image of Lake Linda with the “Snake Linda” flight path photo positions marked as blue squares. (Photo credit: Matt Beedle)

A screenshot of the 3D image of Lake Linda with the “Snake Linda” flight path photo positions marked as blue squares. (Photo credit: Matt Beedle)

Once in the field, our first priority was placing Ground Control Points (GCPs) around the rim of the lake. These GCPs served as exact locations for JIRP faculty surveyor, Uwe Hoffman, to collect GPS coordinates. With these GCPs geo-referenced, we can later locate our whole model in space. We wanted to space the GCPs consistently around the rim of the lake, so we decided to make the hike up and around the large moraine on the back side. The hike consisted of a side-hill ski up to a small ledge where we removed our skis from our feet and continued to bootpack up the ridge. In order to test the different methods of photogrammetry, students Auri Clark and Cézy Semnacher took photos terrestrially along the hike using their iphones.

 Kenzie McAdams preparing to position a GCP. (Photo credit: Cezy Semnacher)

Kenzie McAdams preparing to position a GCP. (Photo credit: Cezy Semnacher)

After placing all of our GCPs, we started our drone flights. During Spiral Linda, everything was going as planned until we noticed the drone getting pretty close to the rocks. As it flew closer, we could start to hear the concern in Matt's voice: “Is the drone going to hit the cliff? Is it worth continuing the flight?” After some speculation, Matt made the final decision to abort the mission. For our second flight, “Snake Linda”, we changed the flight elevation of each point to ~40m higher than previously done. We started our flight, everything going as planned, until we noticed an unexpected visitor: above the ridge we could see an eagle flying curiously around Lake Linda. Now we were not only on patrol to make sure the drone didn’t crash into the rocks, but also to keep an eye out for an airborne eagle attack. We suddenly noticed the drone once again flying precariously close to the rocks: How could this be? We had just updated all of the settings to make sure our flight ran smoothly. We re-checked the flight path, and then realized our mistake: we had neglected to adjust one point by 40 m. To prevent disaster, Matt took over manual control of the drone. All in all, after a few close calls with the rocks and one pesky eagle, we had collected our field data.

Once back at Camp-17, Matt uploaded our data into a program called Photoscan.  Photoscan generates a point cloud, which creates a 3-D surface that produces a model from all of the aerial photos taken in the field. We can use this 3-D model of Lake Linda to visualize and obtain surface elevation measurements. Unfortunately, the images from Spiral Linda weren’t recorded by the drone, so we only created a model using the Snake Linda flight. If we were to do this field experiment again, we would try to recreate the Spiral Linda flight so we could compare it to our current model. We can continue to use this technique in the future to observe the surface elevation changes over time of our favorite meltwater lake, Lake Linda.

 A screenshot of the final 3D image of Lake Linda in Photoscan. (Photo credit: Matt Beedle)

A screenshot of the final 3D image of Lake Linda in Photoscan. (Photo credit: Matt Beedle)