By Thiago Silva, University of Stirling, and Sam Hughes, Forest Research. Around 13 thousand hectares of newly created woodland were reported in the UK for just 2022/23 [1], an area roughly the size of Bristol. Monitoring the development of these woodlands through field work alone would be a massive endeavour, as highlighted last year by Laura Braunholz on this very blog. So as part of our research in TreE_PlaNat, we are also testing a different approach for monitoring woodlands – by shooting lasers at them. Seeing the world from above using dronesTraditionally, satellite imagery has been used to monitor large extents of woodland and other habitats, but it has some limitations. The best freely available imagery, from the Sentinel-2 platform, gives us images with 10 metres per pixel - to go beyond that we would need to purchase expensive, commercial satellite imagery. Moreover, satellite images can be obscured by clouds, which may be a bit of a problem for monitoring land surface in the UK… Drones have exploded in popularity as consumer devices for video and photography, but have also revolutionised the way we use images for studying woodlands and restoration [2]. Fairly good aerial mapping can be done using off-the-shelf drones costing £500 - £1000, and a very capable, mapping-specific drone paired with a high-precision GPS system (for increased mapping accuracy and repeatability) can be had for less than £10,000. That will give you the ability to map up to 200 hectares (almost 500 acres) in a single flight of about 40 minutes, with a pixel size of about one inch per pixel. Especially when we consider the size of most new woodlands, the difference between drone and satellite imaging is stark. Here is a comparison between a drone airphoto and a Sentinel-2 image of one of TreE_PlaNat’s woodland sites. You’d be hard pressed to even tell there was a woodland there based on the satellite image alone. But even with this amazing level of detail, all we get from aerial photography is the top of the canopy. This is useful for mapping woodland extent, identifying individual tree crowns and gaps, and maybe even picking up species. But when it comes to monitoring woodlands, the vertical component of the canopy is very important. How tall are the trees? How dense is the canopy? How much understory vegetation is there? This vertical complexity is a key component for understanding habitat quality, carbon stocks, and the overall quality and health of a woodland. Beam me up... I mean, down!To overcome the limitations of aerial photography, enters LiDAR. An acronym for Light Detection and Ranging, LiDAR works in a similar way to radar (Radio detection and ranging) and sonar (sound navigation ranging). We know radio waves travel at the speed of light, and sound waves at the speed of sound, so by sending out a radio or sound pulse and precisely measuring the timing of its echo after bouncing on a surface, we can tell how far the surface is. The same principle applies to LiDAR, but what we are sending is a laser beam instead – which gives us pinpoint accuracy. LiDAR sensors will send thousands of beams per second in multiple directions, and each beam will pinpoint the position of what/where it hits, creating the main output of a LiDAR scan – the point cloud. Lidar point clouds can range from thousands to billions of points, depending on sensor characteristics, area covered and scanning distance. LiDAR is ideal for creating 3D models of woodlands, as the narrow beams can go through tiny gaps in the canopy and hit different parts of the trees – some even reaching the ground. The technology itself is not new – the first developments came in the early 80s, and by the late 2000s LiDAR was already in regular use for both research and commercial forestry [3]. Still, until very recently LiDAR sensors would cost from a few hundred to about a million pounds. And being large and heavy sensors, they needed a crewed airplane to fly them – which are also very costly to obtain and operate/maintain. The rise in drone technology and the push for cheaper/smaller lidar sensors for self-driving cars and navigation in general created the ‘perfect storm’ for a new generation of laser scanners, called solid-state LiDARs. These are cheaper and simpler to manufacture and are much smaller and lighter as they require fewer moving parts. They do trade off some range and accuracy for this, but that means a fully capable drone system equipped with a LiDAR sensor and GPS station can be had for less than £30,000, at least 10x cheaper than the previous generations of LiDAR drones. Add in another £1,000 for pilot training and certification, and a few thousand more for a good PC and required software. But once you have the system, the cost of flight is only the cost of getting to the site, plus the energy cost for charging the batteries. A drone-mounted LiDAR system also has the advantage of flying much closer to the canopy and at a lower speeds, meaning it can cover the surface with a lot more beams. Airplane-based LiDARs will be flown at 500-3000 ft, yielding about 20-50 points per square metre, while our drone flies at 200-400ft and yields between 100-200 points per square metre. Airplane LiDARs still have the advantage of covering very large areas at once, while drone coverage is limited by battery life and flying regulations. Looking inside the woodlandOne useful way to look at LiDAR data is by taking a cross-section (a ‘slice’) of the point cloud, so we can look at it from the side. On the figure below, the red line shows the direction of the slice on a top and a perspective view of the point cloud, and the actual slice is shown at the bottom. It is easy to see there how, from left to right, we transition from a more mature, taller woodland to an open gap with a single standing dead tree, and then to a younger stand with shorter and denser trees. We can also see where the beams hit the ground, revealing the curved shape of the underlying terrain. Although we can understand a lot about the structure of each woodland by just looking at the full 3D point cloud and the different slices, to be more consistent and repeatable we also try to be quantitative about it. Over the years, scientists have devised several lidar metrics, calculations based on the position and distribution of points, which enable us to measure and compare different woodlands. One of the most basic metrics is the actual canopy height. To obtain it, we can’t just measure the tops of the trees in the point cloud. A shorter tree at a higher elevation would seem taller otherwise. So one of the first steps when processing LiDAR data is to isolate ground points and interpolate them to create a Digital Terrain Model (DTM), a representation of the ground surface as if it was bare. We can also interpolate the top points of the point cloud to generate a Digital Surface Model (DSM), as if we draped a cloth on the land surface and followed its contours. We can then subtract the DTM from the DSM to get a Canopy Height Model (CHM), a representation of the woodland as if it was standing on perfectly flat ground. From the CHM, we can then calculate the height of every tree, and see how tall the woodland is on average, as well as how variable tree heights are. Tree height is strongly associated with woodland age and with wood volume and carbon stocks, so it is a very important metric. Other metrics will look at the vertical complexity of the woodland, for example by looking at how skewed the distribution of point heights is towards the top or bottom of the woodland. A point cloud where most points are clustered at the top means a dense canopy with no underbrush, while a more even distribution means there is a well-developed subcanopy. A grassland with sparse trees would be skewed towards the bottom, as most of the plant mass is close to the ground. And we have even more creative metrics, such as the ‘rumple index’ (we always chuckle saying it). It is calculated is by estimating the total surface area (again, imagine a cloth being draped over the woodland, and then measuring the area of that cloth), and then dividing it by the corresponding flat ground area. A high number will mean a very, well…rumpled surface with lots of consecutive ups and downs, meaning a woodland with trees of uneven height and lots of gaps. A smaller number means a more even, ‘tidy’ surface, for example for a planted tree monoculture.
As we also took measurements of habitat structure on the ground, measuring tree stem density and diameter at breast height (DBH) in plots across our sites, we can compare drone metrics with field-based assessments. At TreE_PlaNat, we are using these metrics to compare how naturally colonized, planted, and mixed woodlands differ in their structure, and consequently, their habitat quality and complexity. To see some of our early results and ask any questions you may have, watch our webinar (held 17th May 2024) here.
0 Comments
|
AuthorsLaura Braunholtz, Ecology post-doc, University of Stirling Categories
All
|