AI Sky Sea

The Problem

To conserve the marine environment it is important to know how many marine animals there are, where they are, and how they are using their marine resources through time. We can determine this as well as the interaction with factors such as disburbance or over-fishing leading to starvation create behaviour changes by collecting rigorous distribution and abundance information. 1. The estimation of animal abundances and ecological parameters such as range, habitat use, and behaviour, of large marine organisms such as mammals and birds (e.g. penguins, petrels) is extremely challenging. 2. The application of AI to conservation is incredibly promising but needs input and application to real data to evolve. 3. We also have an enormous yet under-utilized resource in our volunteer/citizen scientists, who value contributing to meaningful data collection, research, and conservation.

Our Proposal

Data source: Coastal, harbour, estuarine, and boat operated drone footage can provide time and location linked records of marine animals, including mammals and birds. This data can be collected strategically, e.g. on a drone survey-day focussed upon a single harbour (to capture a snapshot of animal distribution and abundance), or incrementally with contributors uploading visual recordings as these are taken across season, time and location. Crucially, recordings with no target animals visible would still provide valuable information. Open science repository: Creating an open science repository for this time and location stamped footage means that it is available to other citizen scientists, biologists, managers. Central collation of data would allow facilitate AI analysis of information. Data forming part of a tertiary study, or similar, would be encouraged by allowing protected classifications for this research data, until completion/publication, at which time information would be available for open use. AI analysis and training: In addition to this proposal, this project will seek a Microsoft AI for earth grant. The training and development of AI marine animal identification (species and also individuals, where applicable) through citizen science and expert input and feedback will mean that all footage will be able to be assessed and provide meaningful information.

We Assume that...

We will gather enough information to yield density and ecological information: By linking location, time and date to all footage this means that every piece of data is providing useable information.

That an AI tool will be effective in surveying footage and yielding information: Input from volunteers will be essential for developing AI accuracy and improving data quality.

Constraints to Overcome

Proof of concept: Engaging citizen scientists with AI by building confidence and building networks, e.g. engaging schools and making the system easy and inviting to use. A question as to whether we will garner widespread interest: The first location is to be the Hauraki Gulf, New Zealand. There is an active community working to restore and conserve the Waitamata Harbour and Gulf which shows promising commitment to conservation initiatives. Long term repository host: Ensuring data is secure for long-term use (e.g. museum level curation of information).

Current Work

People: Developing a team, achieving buy-in from marine researchers and computer science. Communication: Circulating a proposal to interested parties, including conservation end-users. Applying to Microsoft for AI for Earth funding. Developing proof of concept.

Current Needs

Drone skills, computing resources, a test drone.