“What on earth was I thinking? How am I going to fly a drone out here?”
So opened the first line of my field notebook. It was meant with humour but, I have to admit, there was an undeniable feeling that the entire project that had brought me to the beautiful forest of Danum Valley in Malaysia may be about to fall down around me.
The idea of collecting aerial data of tropical forests using a drone, aka an Unmanned Aerial Vehicle (UAV), seemed like a great one. By using a UAV over ground-based censuses or traditional airborne and satellite data capture, all the benefits of increased spatial and temporal resolutions are yours and you can side-step problems such as obscuration of the imagery by clouds. Plus it’s cheaper than purchasing high-resolution satellite imagery or chartering a plane. So it was with some – or more honestly, a lot of – naivety that I had sat planning this fieldwork from my cosy office in England. Then, I arrived at my field site and was faced with the reality of rugged terrain topped with very dense, astonishingly tall forest: the tallest tropical trees in the world have just been found here. Armed with only a small, plastic UAV and standing in the forest with the canopy 70 m above me, how I ever thought my scheme to launch the UAV and fly it up through the undergrowth and out of a canopy gap only a few meters wide could be pulled off was beyond me.
The reason behind this madness is simple and important: to collect data on liana presence in the forest canopy. Lianas (woody climbers) are a key component of tropical forests. In recent years, however, many studies have found that liana abundance in tropical forests is increasing relative to trees and, as Schnitzer and Bongers (2011) note, any alteration to tropical forests has important ramifications for species diversity and productivity, as well as the global carbon cycle. Lianas can decrease the amount of carbon fixed by an area of tropical forest by ~76%/year (van der Heijden et al., 2015) via liana-tree competition: reducing tree growth by up to 84% (Schnitzer et al., 2014), as well as fecundity (Nabe-Nielsen et al., 2009), and survival (Phillips et al., 2005). Lianas also fail to replace this lost biomass, further reducing carbon uptake, as they rely on trees for structural support and allocate a higher proportion of biomass to foliage production over carbon-dense stem production (van der Heijden et al., 2013).
Despite this, lianas remain understudied, from ground censuses and remote sensing data, and much is left unknown about the scale, cause and impact of increasing liana populations (Marvin et al., 2016). The ability to repeatedly and accurately map lianas is crucial for analyzing and quantifying their effects on forest function, while helping uncover mechanisms behind their proliferation, as a continued increase may further reduce tropical forest carbon storage and sequestration, thus endangering the future of the tropical carbon sink. So, I was aiming to test whether UAV technology could be used to accurately identify liana canopy cover in tropical forests in Malaysia. I figured that UAV-based remote sensing provided a potentially inexpensive and accessible new tool to rapidly map and monitor lianas at larger spatial scales and faster temporal scales than possible with plot-based censuses while overcoming the limitations of traditional airborne and satellite techniques. But only if you can launch the UAV and get it above the canopy without crashing it.
The quest to do just this led me into some precarious situations in pursuit of appropriate launch gaps. Large canopy gaps are preferable, as they increase the ease with which the UAV is manually piloted up, and down, through the gap (minimising battery usage), as well as allowing a stronger GPS signal to reach the bottom of the canopy gap. When the GPS signal is weak, the UAV loses stability and can ‘drift’ when flying (making attempting to safely launch and land without crashing into a tree even more hair-raising). Of course, large canopy gaps weren’t always easy to find. I found myself climbing up almost sheer cliffs of mud and unstable rocks that would fall away underfoot, scrambling down landslips to small, relatively flat outcrops (and then trying hard not to fall off the side of them when looking up at the UAV), and spending more time than I would have liked grabbing lianas and tree trunks for support only to have the unpleasant realisation that they were not well fixed to the ground.
Once an appropriate canopy gap was found near each site, the UAV surveys could start. The UAV I used was an inexpensive, though not expendable, DJI Phantom 3 Advanced, which proved perfect for this work; its small size and light weight meant that I was able to carry it through the forest with relative ease. It is equipped with a high quality 12 megapixel RGB camera mounted on a gyro-stabilised gimbal which uses information from the on-board accelerometer to keep the camera level, even as the aircraft accelerates, decelerates, banks and turns, ensuring little-to-no blurring of the images. Image quality was further enhanced through the high-quality Sony EXMOR 1/2.3” sensor which has a relatively narrow field of view (FOV) reducing ‘fish-eye’ distortion in the images, and a fast shutter speed reducing image blur.
Launching and landing the UAV wasn’t as easy as putting it down on the forest floor and taking off from there. I am incredibly grateful to all of the people who helped me with the surveys, by holding the UAV over their heads, above the ground vegetation, and catching it again on its descent. Without their help, the UAV would never have made it above the canopy. Once there, automated software was used to conduct the surveys of each plot. The UAV possesses GPS and GLONASS positioning to enable this autonomous flight and each image taken is geo-tagged with the GPS location and elevation of the UAV at the point of capture. Where possible the surveys were carried out when there was even cloud cover to ensure diffuse radiation and less shadowing on the ground – resulting in slightly less time-intensive image post-processing. In these favourable conditions, the flight time of the UAV on a single battery is ~23 minutes, in which it can cover ~8 ha. Six batteries were taken into the field, meaning a total of ~48 ha could be surveyed in one day. This allowed the plots and a ‘buffer’ of surrounding vegetation to be captured in the images, minimising ‘edge effects’ in post-processing, when the individual images are stitched together to create orthomosaics of the plots in which liana presence in the tree crowns can be identified.
Despite my initial scepticism, the fieldwork was a success. There may have been one or two near misses, and a couple of trees may have a few fewer leaves than before the UAV propellers got near them, but, in the end, the forest is intact and the UAV and I made it home. The data are looking positive too – the images have a spatial resolution of ~2 cm, the orthomosaics of the plots have been created, and lianas in tree crowns can be identified from them. It looks like UAVs may indeed be an affordable and accessible new tool to rapidly collect data on liana canopy cover: you just need to work up your nerve to start flying them in tropical forests.