Can machine learning be used to accurately identify wildlife in remote camera trap images in a rapidly changing world?
Machine learning (ML) techniques provide a powerful method to automate image processing. However, due to rapid environmental change, image algorithms may not perform well after major disturbance events. This project will work closely with WildCount, run by NSW National Parks and Wildlife Service to refine ML algorithms for identifying species in camera trap images and test the impacts of the 2019/20 mega-fires on the accuracy of species identification.
DigiFarm: incorporating biodiversity into farming decisions (with the Sydney Institute of Agriculture)
We aim to develop a digitally enabled network which will monitor native flora and fauna to inform sustainable agricultural practices. A unique combination of methods will be used: We will test new methods in camera trapping and acoustic recorders (birds and bats) in quantifying on-farm biodiversity and develop spatial models to identify biodiversity hotspots.
Simpson Desert Insights: designing Citizen Science programs for identifying wildlife in remote camera trap images (with Australian Museum)
This project will work closely with DigiVol at the Australian Museum, and the School of Life and Environmental Sciences, University of Sydney. It will determine the level of uncertainty in using Citizen Scientists to identify species in remote camera trap images.
Agricultural land suitability that includes ecological and cultural values (with Dr Floris Van Ogtrop, Primary, & Dr Ayesha Tulloch, USYD)
Land suitability maps for agricultural production are based on agronomic and climate data. The availability of spatial ecological/cultural data is improving. This project will incorporate these new data sources to create maps that are ecologically and culturally sensitive to assist decision makers.
Using digital technologies to track farmland ecological condition in remote arid Australia (with Dr Ayesha Tulloch, Primary, USYD)
Digital acoustic monitoring is increasingly used in wildlife studies around the globe. Acoustic monitoring devices are a powerful and cost-effective method to survey wildlife due to their ease in deployment and ability to continually monitor populations across time through the use of “soundscapes”. Digital monitoring is particularly important for tracking ecosystem condition in remote, poorly accessible locations, such as much of arid Australia where livestock grazing predominates. Numerous metrics can be derived from ecoacoustic monitoring datasets, including sound diversity (a potential surrogate for wildlife diversity) and sound abundance (a potential indicator for wildlife abundance or activity). However, the recency of these technologies means that there is little scientific evidence for a clear link between acoustic monitoring metrics and the condition of wildlife and landscapes.