Droning for Seeds

By Stephen Elliott

First Published in Samara  Issue: 32 July - December 2017 (The newsletter of the Millennium Seed Bank Partnership (MSBP)

The UN is calling for the restoration of 350 million hectares of forest to degraded land around the world by 2030, but where will the seeds come from? Targets like WWF’s “Trillion Trees Partnership” will require seed collection on an almost industrial scale. With most countries still lacking native seed supply chains, seed collecting in remnant forest remains essential, but current methods can be restrictive. Collectors must push their way through dense forests, with binoculars pointed aloft, searching for ripe fruits amongst the minute fraction of the forest canopy that is visible from the ground. Even when a fruiting tree is found, the seeds may not be ripe, necessitating a tedious return trip. So, conventional seed collection can be inefficient, unpredictable and consequently expensive.

So why not look for seed trees from above? This has been attempted by scanning high resolution photos, taken from planes 500-1,000 m up, for distinctive crown features to devise simple dichotomous keys (Gonzalez-Orozco, et al., 2010). Other researchers have achieved some success using hyperspectral imagery and lidar (Baldeck, et al., 2015), but such technologies are expensive and the details discernible from planes are limited. With the advent of affordable unmanned aerial vehicles (UAV’s or drones), high-resolution digital cameras can be flown much closer to the forest canopy, so the development of much finer, cost-effective tree identification systems becomes possible.

Kris sets up the drone to fly over restored forest where the species of all trees are known.At Chiang Mai University, Thailand, a graduate student in the international Environmental Science Program, Krishna Rai (from Bhutan), is trying to spot the crowns of target seed tree species within a seasonally dry tropical forest, using an off-the-shelf drone (DJI Phantom 4 Pro), an automated flight planner and opensource, image-processing software. Flying a drone low over an undulating forest canopy is, of course, risky, but the Phantom 4’s collision-avoidance sensors have so far prevented crashes. Rai is experimenting with three types of data: i) crown morphology, ii) leaf characteristics and iii) image filtering. Leaf type, shape and arrangement can be discerned easily and are highly distinctive for some species. Image filtering can also dramatically distinguish the crowns of some species, but its effectiveness varies among species and seasonally. A species that is distinctive, when flushing new leaves may become indistinguishable just a few weeks later. Consequently, we envisage drones being used to locate target species, when they are at their most distinctive and to monitor them subsequently for fruit-set and ripeness using auto-flight plans.

Kris used leaf characteristics and whole-crown features from drone images, and image filtering, along with traditional dichotomous keys (a concept developed by Linnaeus in the 1700s), to visually identify seed trees.Who knows – one day soon, drones might even be able to collect fruits autonomously, with sensor-rich robotic arms: not so farfetched since robots that are capable of picking fruits in orchards already exist. As the use of drones becomes routine, the way we find and identify tree species is about to change fundamentally. Although conventional taxonomic species descriptions will remain a staple of tree guides, they may shortly become complemented with drone photos of tree crowns and the image-filter settings that distinguish tree crown species at various seasons.

 REFERENCES

• Baldeck, C. A., G. Asner, R. Martin, C. Anderson, D. Knapp, J. Kellner, S. Wright, (2015). Operational Tree Species Mapping in a Diverse Tropical Forest with Airborne Imaging Spectroscopy. PLoS ONE 10(7): e0118403. doi:10.1371/journal.pone.0118403 Gonzalez
• Orozco, C., M. Mulligan, V. Trichon & A. Jarvis, (2010). Taxonomic identification of Amazonian tree crowns from aerial photography. Applied Vegetation Science. 13: 510–519, DOI: 10.1111/j.1654-109X.2010.01090.