2020-02-11 Planning & Environment Committee Agenda

PLANNING & ENVIRONMENT COMMITTEE MEETING AGENDA                                          11 FEBRUARY 2020


10.    Noosa & District Landcare Group
Project title -   Drones and Artificial Intelligence for broad scale monitoring of koala abundance and
                  evaluation of revegetation - Yurol and Ringtail Forest

Brief project overview

The objective of this project is to establish a baseline estimate of koala abundance at representative sites in
the Yurol/Ringtail area, to digitize remnant, regrowth and replanted areas, and to evaluate how revegetated
and remnant landscapes are utilised by koalas.

Full project description

There is an urgent need to find smarter ways to manage koala populations across Australia, and re-
establishment of native vegetation with the intention of providing high quality koala habitat provides one
means of doing this. The capacity to plan and monitor this effectively is currently limited since it depends on
knowledge of what constitutes good koala habitat, the ability to evaluate habitat cost effectively through time
after replanting, and the capacity for cost effective broad scale monitoring of koala utilisation of the area.
Questions around the biophysical attributes of good koala habitat, the mix of food trees vs shade trees, the
density of these trees and the age that which the revegetated patch becomes useable are currently unknown.

To answer these questions we first need a baseline estimate of koalas that occur in a region. We then need
to be able to measure characteristics of re-vegetated areas through time, and at the same time evaluate the
utilisation of the area by koalas. It is essential to develop a means to do this efficiently, over large areas, and
through time. The opportunity now exists to leverage recent advances in artificial intelligence and vegetation
modelling, combined with state-of-the-art aerial sensing capabilities, to create a new approach to monitoring
koala presence in combination with their habitat structure– and the capacity to evaluate these in combination
is vital. This will lead to more affordable data collection, vastly improved effectiveness in the use of the data
collected, and deliver new approaches to better plan for re-establishment of native vegetation and manage
koalas and koala habitat in a cost-effective manner.

QUT has recently developed an approach for detecting koalas that is not only more efficient but more
accurate than ground based expert observers (Corcoran et al, 2019). When put together with our capacity
to produce digital vegetation models using drones and modelling, these become powerful tools to evaluate
the success of site re-vegetation through time. The research team includes members of the Australian Centre
for Excellence in Robotic Vision, and remote sensing vegetation modelling experts.

We propose to use these advanced drone based detection methods together with new remote sensing
methods for modelling vegetation. In effect, we plan to digitize the landscape, whether existing or
revegetated, to determine how the revegetated habitat develops over time, and how koalas use the
landscape as it grows and becomes more complex after revegetation. Finding a threshold for complexity at
which koalas can make effective use of vegetation will provide important guidelines for the effectiveness of
future revegetation and habitat restoration.

While the newly developed methodology for koala detection with drones and AI is more cost effective and
accurate than human based experts, there is still an overhead in time – the time taken to retrieve the drone,
and bring the data back for processing. While processing time is not great, this lag could be problematic
when live estimates are required – for example, during land clearing. Recent changes in computational
hardware means that, with adaptations, our koala detection AI algorithm could be processed live, so that
accurate estimates of koala location could be determined within minutes of a drone being airborne. This kind
of technology is known as Edge computing, but as yet has not been applied in wildlife monitoring. In a world
first, we intend to apply this using an ARC Linkage to koalas in the Yurol Ringtail area.

The support received from a Noosa Council Grant will allow us to take the additional step of drone surveys,
to achieve live results through an ARC research grant application. Should an ARC research grant not be
approved, the non-live version will be the results for this Noosa Council grant. Still a great outcome and
baseline dataset.




                                                                                                Page 96 of 119