A new dataset sees verified images from wildlife camera traps available in the Atlas of Living Australia (ALA).
But it wasn’t just a case of point and shoot to incorporate this new data into the ALA.
The Wildlife Observatory of Australia (WildObs) is a national initiative transforming how Australia collects, processes, and shares data from wildlife camera traps. WildObs is a partnership led by the Queensland Cyber Infrastructure Foundation (QCIF) and supported by our NCRIS colleagues the Terrestrial Ecosystem Monitoring Network (TERN) and the Australian Research Data Commons (ARDC).
For our part, the ALA developed a new process for ingesting data in Camera Trap Data Package (CamtrapDP), a new standard that allows richer data capture specifying deployments, observations and media. It’s been developed by the Biodiversity Information Standards organisation (TDWG) and adopted by the Global Biodiversity Information Facility (GBIF). The ALA transforms Camtrap data into Darwin Core events and occurrences.
WildObs works across many research projects to curate verified camera trap datasets, and we’ve collaborated with them to mobilise data using CamtrapDP. This then allows the ALA to provide the infrastructure behind the WildObs Tagged Image Repository and makes the camera trap data findable, accessible, interoperable and reusable (FAIR).
The ALA’s normal data processing pipeline prepares the data for indexing and mapping. We match records to our taxonomic backbone, and standardise dates and times and coordinates. We track globally unique identifiers for the records and datasets, and we apply threatened species statuses and responsibly manage obfuscation and access to sensitive records.
Peggy Newman is the ALA’s Data Manager.
“The ALA’s ingestion pipeline is ideal for handling most of the standardisation and data cleaning needed to produce AI-ready datasets, which is often the most time consuming and mind numbing part of building AI/ML models,” Peggy says.
“In the coming years, we’re expecting to see a huge influx of biodiversity datasets like this where the observations and identifications come about from automation and modelling.
“For the models to get better, they’ll need to be built with and informed by openly shared and verified datasets. When the models are better, they’ll be able to help produce better quality datasets in shorter timeframes.
“We see the ALA’s data pipelines and FAIR practices as supporting researchers to improve this kind of feedback loop in biodiversity monitoring.”
The first dataset recently published in the ALA features wildlife in the Wet Tropics captured and identified by quantitative wildlife ecologist Zachary Amir from the University of Queensland who is part of the WildObs team, hosted by TERN. Zach was looking at interactions between native and invasive vertebrates in the rainforests of North Queensland.
Acknowledgment:
WildObs is made possible through a national collaboration of National Collaborative Research Infrastructure Strategy enabled infrastructure. TERN provides standardised field protocols and hosts the systematic survey data across Australian ecosystems. The ALA hosts the tagged image dataset and occurrence records for public access and reuse. The Australian Research Data Commons (ARDC) through the Queensland Cyber Infrastructure Foundation (QCIF) is building the user-facing platform to support image upload, tagging, model training, and data sharing using advanced AI tools. It’s part of the ARDC’s Planet Research Data Commons, which provides national-scale data infrastructure for earth and environmental science research and decision making.
