The recent announcement that some UAS companies have been granted a beyond line of sight waiver for commercial operations by the FAA is a clear indication that drones are gaining more of a foothold in the geospatial industry. The idea of fleets of UAVs autonomously flying all over the world is incredibly exciting for environmental monitoring even if it is terrifying for social monitoring. The technology is making it possible to image crops down to the leaf-scale, generate 3D models of construction sites in mere hours, and perform virtual inspections of infrastructure from a desktop computer thousands of miles away. If you happen to be an avid consumer of tech press releases, you might be forgiven for thinking that UAVs can do everything from collecting remote sensing imagery to delivering burritos, and that the era of manned aircraft is effectively over. The possibilities are exciting for sure, but our recent experience has given us some perspective on when UAVs are a good fit – and when they are not – for the kinds of problems our clients want to solve.
Quantum Spatial has been working with partners in the UAS industry to push the development of automated environmental mapping and facility inspection, among other efforts. These projects range from electrical infrastructure inspection to wide-area habitat mapping, with a goal of increasing resolution and reducing costs compared to manned flights. From a purely technical perspective, our efforts have been wildly successful. With some of the best drone pilots in the business, we are able to snap photos of anything we want at any imaginable scale. If someone wants centimeter resolution imagery of a building, no problem. If you want to find hot spots on buildings with thermal imagery, also no problem.
If, however, you want high resolution imagery of hundreds of square miles of eastern Oregon, you have a problem. The sticker price on that one might be a shock.
It turns out that flying drones for most large-scale remote sensing projects doesn’t pencil out, at least right now – although true beyond line of sight permission will help a lot.
That’s the first little secret about UAVs doing traditional remote sensing. Drones typically are more expensive than manned flights on a per-mile basis. In one day, a UAV can cover maybe as much as a couple thousand acres with a visible or multispectral camera. In the same time, a manned aircraft can cover hundreds of thousands of acres with hyperspectral imagery, LiDAR and and orthophotos all at the same time. Those differences relate to cost. What we are trying to do is find the sweet spot in this airborne vs. UAV game.
What drone flight can do, that manned flights cannot, is cover a very small area for a very reasonable price, provide ultra-high resolution data, and fly at very specific times of day or with very short notice.
These are the projects that we are focused on optimizing. UAVs are much more efficient at doing remote sensing in ways that were never possible with airborne or spaceborne systems in the first place. They can collect oblique cm-scale resolution imagery, carry fine-scale LiDAR systems and operate in close quarters.
The second often unaddressed issue with UAV mapping is the need for ground control. Due to the fact that UAVs have much higher resolution, you need much more precise ground control in order to create quantitative data products. The plethora of UAV software companies that advertise “push-button” solutions rarely say anything about the fact that you need to collect a well-distributed array of survey points for any project that you want to measure better than to within a few meters. Again, the main question is: What do you want to be able to do with the data?
Even more interesting are the agricultural UAV companies that are selling push-button analytics that will automatically deliver all your agronomy needs. While that sounds great, it often isn’t as easy as that. Farmers may be able to adjust fertilizer application based on NDVI imagery, but that is only one small piece of the crop health puzzle.
In 2014, we spent the better part of a year working with one of the world’s largest agricultural firms to build out a massive R&D precision agricultural program using airborne hyperspectral imagery. It was amazing what we could tell about crops. We could identify the differences in hundreds of corn strains, find pigweed in soy fields and correlate high-precision analytics collected from the tractor with spectral response. We did multiple acquisitions throughout the growing season and multiple states throughout the country. In the end, any quantitative results that we were able to derive required expert ground truth and statistically distributed sample points in every plot. Then there was the fact that three Ph.D. scientists were needed in order to interpret the data. This doesn’t mean that some elements of UAV-based precision agronomy might not be useful, but our conclusions were that the program was incredibly expensive not from the imagery collection, but from the ground truth and domain expertise required to turn that imagery into actionable information.
As technology gets cheaper, UAVs become more widespread, and companies and researchers develop more automated methods, there is little doubt that the limitations I've outlined will become less significant. We at QSI are frequently called upon to be the architects of solutions for our customers who want to solve specific problems and think that drones are the answer. In some cases, they are not. When they are, we have to design the entire program to account for
- The application.
- The actionable analytics that needs to be extracted from the data.
- How it integrates in with the business processes that use the date to make decisions.
- The sensors and acquisition protocol that will collect data at the precision required by the end application.
As the technology and ecosystem around it matures, organizations would do well to be skeptical of claims about the availability of shrink-wrapped solutions. Instead, I would advise them to look at their needs holistically and carefully select the different elements of their program.