Pattern recognition in long‐term Sooty Shearwater data: applying machine learning to create a harvest index

Area of Science:

• Ecology
• Conservation Biology
• Indigenous Knowledge Systems

Background:

• Rakiura Māori have a multi-generational tradition of harvesting Sooty Shearwater (Tītī; Puffinus griseus) chicks.
• Annual harvest diaries, some originating in the 1950s, provide a historical data source for Tītī populations.
• Traditional ecological knowledge from Rakiura Māori is integral to understanding the harvest.

Observation:

• Generalized boosted regression models, a machine-learning approach, were employed to analyze harvest data.
• Harvest indices were calculated, accounting for variables influencing catch numbers.
• Model performance was assessed by comparing predicted versus observed harvest values.

Findings:

• Machine learning models demonstrated strong predictive accuracy for both nanao (daytime burrow harvest) and rama (nighttime surface harvest) seasons.
• The 'day of season' was a significant predictor for rama harvests, with peak catches around 10-15 days into the season.
• Rama harvest data exhibited greater consistency, suggesting its suitability for long-term Sooty Shearwater population monitoring.

Implications:

• Machine learning offers a robust method for standardizing harvest data, correcting for extrinsic factors like effort and weather.
• This approach can be adapted for monitoring other harvested resources, including fisheries and terrestrial wildlife.
• Integrating indigenous knowledge with advanced analytical techniques enhances conservation and resource management strategies.