Abstract

Topic: Moose management and monitoring

Spectral properties of moose dung pellet piles – first results of a pilot study aiming to detect dung piles in multispectral drone images

Frida Linder¹, Annika Felton¹, Sarah Gore², Luiz Henrique Elias Cosimo³, Lukas Graf¹

Southern Swedish Forest Research Centre, Swedish University of Agricultural Sciences, Lomma, Sweden
Department of Wildlife, Fish and Environmental Studies, Swedish University of Agricultural Sciences, Umeå, Sweden
Department of Forest Resource Management, Swedish University of Agricultural Sciences, Umeå, Sweden

In Sweden, moose populations are relatively dense compared to other boreal countries. The browsing damage they cause on Scots Pine, an important species for timber production, costs society large sums of money and some stakeholders call for adjusted management of moose populations. As such, efficient cervid management requires population estimates through monitoring . The estimation of moose populations in Sweden requires a large amount of work and partially are estimated by counting dung pellets in field plots by ground teams. The implementation of automated technologies in monitoring systems bears chances to streamline data collection efficiently, potentially increase pellet detection rates and therefore efficiently guide management decisions. As such, we aim to develop a more time efficient method to detect moose dung pellets piles in young pine forests using drones equipped with multispectral sensors. We are aiming to identify the most important parameters to use multispectral sensors on drones to maximize detection accuracy of moose dung pellets in young pine forests. We collected data in four young pine stands (mean height of 2m) in southern Sweden . We followed established protocols to identify moose pellets and georeferenced, alongside dung pellet piles of other cervids, using EMLID Research RS3+ . We obtained multispectral imagery from a DJI Mavic 3m in 5 different bands (Red, Green, Blue, Near-Infrared and Red Edge) and created orthomosaics with 0.4cm spatial resolution using DJI Terra Pro. In total, we georeferenced 253 moose dung piles, as well as 108 roe deer dung piles . We extracted values of several multispectral indices (e.g., NDVI, green NDVI, OSAVI, etc.) and compared how they relate to, for example, age of the dung pile (old/new) or estimated vertical cover (in %). The results of this study will help set a baseline and aid feature and variable selection to train models that will be able to detect dung piles from multispectral imagery in the future.