Share | 02/07/2018
As unmanned aerial vehicles (UAVs or drones) find their place in the forestry sector, we met with a pioneer in this field, researcher Stefano Puliti of the Norwegian Institute of Bioeconomy Research (NIBIO). In this wide-ranging interview, we discuss his commercial forestry learnings—including the always-hot topic of LiDAR versus aerial photogrammetry—and we delve into Stefano’s ongoing research, which could apply to several industries, combining partial-coverage eBee UAV observations with freely-available satellite imagery.
Hi Stefano, let’s help our readers get to know you first, shall we? What is your background and what was your route into remote sensing?
Stefano (LinkedIn): Sure. I started my forestry studies with my Bachelor’s degree in forestry management in Italy (I’m Italian). Then I decided I needed to position myself in a larger international context, so I did my Master’s degree in Finland and Sweden, where forestry is a more relevant topic and where the industry is quite large.
As soon as I arrived in Scandinavia I realised that remote sensing was one of the topics that interested me the most. I soon started to work with LiDAR data, working on remote sensing and collecting 3D point clouds, mainly from airborne platforms; I used these point clouds to model and to map forestry resources. Alongside my Master’s work, I also started working for a Finnish company called Arbonaut in order to apply this knowledge on a commercial basis.
After completing my Master’s, I did my PhD in Norway, again using remote sensing to map forest resources, this time at the Norwegian University of Life Sciences (NMBU). The team there is world-renowned for using laser scanning data to map forests—Erik Næsset, one of my supervisors, was one of the first researchers to use LiDAR data to map forests.
Stefano assembling his eBee UAV pre-flight.
Could you tell us more about Arbonaut and the work the company does within forestry?
Yes. It is a geomatics company, which mainly relies on external contractors to acquire LiDAR data. Then, the team would acquire ground reference data and develop models, linking it to LiDAR. We were delivering forest inventories to our customers in the form of accurate maps of an organisation’s forest resources, customers that included state forest enterprises and large, private forest owners.
I found drones to be very interesting tools that enable a wide audience to rapidly acquire high-quality three-dimensional data in-house
So, why the switch from remote sensing with LiDAR on manned airplanes to aerial photogrammetry with drones?
The beginning of my career was more focused on LiDAR data, but during my PhD, under the supervision of Terje Gobakken, I focused more on aerial photogrammetry as a cheaper and more widely available data source. Within this topic, I found drones to be very interesting tools that enable a wide audience to rapidly acquire high-quality three-dimensional data in-house.
I started by using photogrammetric data in a similar fashion to how I was using LiDAR data—obtaining information about tree heights and tree density—before using this information to map other forest variables such as timber volume or the biomass of carbon stocks.
What I found during my PhD was that photogrammetric data can indeed perform as well as LiDAR data
If you compare using LiDAR to aerial photogrammetry within forestry, what have you learned are the pros and cons of these respective approaches?
Well, LiDAR data, of course, has the great advantage of being able to penetrate vegetation and retrieve information about the terrain, even under forest cover. Once we are able to detect the terrain, we can create high-resolution digital terrain models, which are used to obtain relative tree heights. But, of course, laser scanning data is more expensive than photogrammetric data. What I found during my PhD was that photogrammetric data can indeed perform as well as LiDAR data.
Here in the Nordic countries, over the past 20 years, LiDAR has become the operational standard. It is, today, how forests are measured by the industry. However, organisations are starting to think more and more about whether the future will actually be based on aerial photogrammetry, since high-resolution terrain models from national LiDAR acquisitions already exist in most of these countries. For this reason, I think aerial photogrammetry, by drone, could become the new operational standard for small forest properties over the next 10 years.
Timber extraction, photographed during one of Stefano’s UAV missions.
To clarify, that prediction is based on already having the DTMs—generated from LiDAR data—and using these alongside new photogrammetry data as the base for tree height calculations?
Yes, exactly. Since there are national acquisitions of LiDAR data we can use this publicly available data to obtain the relative height of trees from photogrammetric measurements.
For your research, what type of mapping drones are you flying?
I’ve done most of my research using an eBee, although one of the main advantages of photogrammetry is that it can be done from many different platforms—unmanned aerial vehicles, manned aircraft and also via satellite. In that sense, it’s a very versatile approach.
When you started focusing on aerial photogrammetry, why did you start working with drones rather than sticking with, say, manned aircraft?
Actually, for my first study I did use manned aircraft because NMBU didn’t have the eBee at the time. Then, my interest in using drones came about because, as remote sensing acquisition platforms, UAVs are: one, very available, and two, the data we can obtain from them is very high in detail.
We can really rely on the drone’s acquisitions, the measurements that we get. In terms of UAVs being largely available, whenever required, I like to say that drones are allowing the development of do-it-yourself 3D remote sensing.
I like to say that drones are allowing the development of do-it-yourself 3D remote sensing
Acquiring airborne data with manned aircraft requires that you have a very large area to survey and you need to hire a contractor to do that. In contrast, drones are a very versatile tool. They allow any user, wherever they are in the world, to purchase a relatively cheap platform and to acquire their data and even process it on their own desktop. UAVs provide anyone with an instrument that is capable of obtaining high-resolution 3D data.
Why did you adopt the eBee drone specifically? What did your decision-making process look like?
Me and my supervising team realised that a fixed-wing drone could be a very suitable UAV platform because in forestry we typically need to cover large areas. So, that was mainly what drove that first decision. Then we had a demo with an AgEagle sales representative and we were impressed—it seemed like a very reliable platform in terms of both the hardware and its eMotion software.
I can just go out, fly, and that same day I can generate very accurate information about how much timber I have in my forest
From your forestry work and research to date, what do you see could be the most valuable or useful applications of drone-based aerial photogrammetry?
There are two key types of applications, which represent two different spatial scales. If we look very locally, it could be that a forest owner needs information on his forest stock or other variables of interest at a specific point in time. Let’s say I have to go to harvest a forest. This job is often sold to a contractor who comes in, harvests, and then sells the timber. So, what we need is an accurate estimate of how much timber there actually is. This calculation would be a primary small-scale application—going out in the field, flying, and that same day generating very accurate information about how much timber I have in my forest, and therefore how much I can sell it for.
The advantage of these methods is that they enable a precise estimation of large-scale forest resources while limiting the costs of UAV data acquisition
On a larger scale, there are ongoing efforts in many developing countries to establish systems to monitor, report, and verify forest carbon stock changes. To address these issues we developed an application using small mapping drones to provide estimates of forest stocks and I represented this work in the last paper I published in Remote Sensing of Environment [read the paper]. This is based on the acquisition of patches of UAV photogrammetric data over vast landscapes and using this to precisely estimate forest resources on a large scale. The advantage of these methods is that they enable a precise estimation of large-scale forest resources while limiting the costs of UAV data acquisition.
Monitoring the eBee’s flight from a hunter’s perspective.
A continuation of this work was carried out in another recently-published paper [read it for free, for a limited time, here]. In this second paper I augmented the UAV data with a layer of freely-available satellite imagery, such as the imagery that’s now available from the European Space Agency. I have been combining UAV data with Sentinel 2 imagery, which allows us to produce large-scale maps with precision comparable to full-coverage LiDAR.
I have been combining UAV data with Sentinel 2 imagery, which allows us to produce large-scale maps with precision comparable to full-coverage LiDAR
When you talk about employing both drone and satellite data to cover larger areas, could you explain a little more about how these two data sources are used?
We produced squares where we had high-resolution drone data, within which we had also made some field observations. Then, we had satellite imagery covering the entire study area.
Like this, we’re getting the benefits of the high-resolution drone data and, at the same time, getting the satellite’s overview of the entire area. What we do is to combine these two levels of information and thus we derive better information, continuously, on a large scale.
Stefano’s recent study employs ground truth data and partial-coverage eBee drone data alongside larger-coverage, lower resolution satellite imagery.
Right at the start of our conversation, you mentioned airborne LiDAR having been your starting point. We already talked about the difference between this and photogrammetry in relation to ground penetration, but aside from that factor, on a practical level, how would you compare the two approaches in terms of the cost and efficiency of collecting this data?
Well, I always say that UAVs and photogrammetry are a great tool and a cheap platform, but have larger costs per hectare than manned airborne data. At least, this is true in countries like Norway where there are commercial providers that acquire LiDAR data cheaply because it became a kind of operational standard.
In terms of the precision and accuracy of our forest resource estimates, we can pretty much get the same level of accuracy that we can get with wall-to-wall LiDAR
Conversely, in tropical countries where LiDAR technology is not readily available. There, you sometimes need to import the LiDAR aircraft and operations are more difficult because of the need of operating in environments that are often cloudy. The costs there might be very different, so in those cases UAVs could likely lower costs when compared to LiDAR. In terms of the precision and accuracy of our forest resource estimates, we can pretty much get the same level of accuracy that we can get with wall-to-wall LiDAR.
If national LiDAR data had not existed in Norway, do you think drones would have been less relevant to your work? In other words, would you have ended up still needing to commission airborne LiDAR services to create that first digital terrain model?
Well, one of the other studies I’ve been working on, in cooperation with the University of Florence and NMBU has actually been dealing with using UAV data without a LiDAR-derived DTM. This is still unpublished material, but that could answer your question and make a nice follow-up story in the future. The short version is that we are using only photogrammetric data, yet we are seeing that we can produce the same level of accuracy as we can get with LiDAR data, without having any information on the terrain. That’s going to be a revolutionary study, also because we applied this approach in two different areas: in Norway and in Italy, the second of which featured very steep terrain and more dense forest then here in Norway. So, to answer your question, I would definitely still use drones even if there was no LiDAR data available.
I would definitely still use drones even if there was no LiDAR data available. So, to answer your question, I would definitely still use drones even if there was no LiDAR data available.
Lastly, we wanted to ask about the positional accuracy of your drone data—how important is that?
That is very important. We set ground controls because we need as good correlation as possible between what we measure in the field and our remote sensing data. The smaller the geometric errors, the more correlation there will be. If we start losing the correlation between what we measure in the field and what we see from the sky, this will affect the precision of our maps and therefore our estimates.
Although I developed the methodologies we’ve been discussing—the idea of looking at UAV data not only as full coverage data, but potentially partial coverage, as a means of reducing the costs of using UAVs—for forestry, they are directly applicable to any other sector too
Thanks for speaking with us Stefano. Do you have anything final to add?
Actually, yes. I would just add that although I developed the methodologies we’ve been discussing—the idea of looking at UAV data not only as full coverage data but potentially partial coverage, as a means of reducing the costs of using UAVs—for the forestry sector, they are directly applicable to other sectors too. We can imagine applying this approach to any natural resource management-type question that users might have. It is, potentially, a very interdisciplinary approach.
Understood. Stefano, thanks so much for your time.
Read: Use of photogrammetric 3D data for forest inventory (Stefano’s thesis)
Sign up to receive updates directly to your inbox.