Application of machine learning algorithms to identify cryptic reproductive habitats using diverse information...

Area of Science:

• Marine ecology
• Computational biology
• Conservation science

Background:

• Ecological data is often complex and uncertain, requiring advanced analytical methods.
• Identifying spawning aggregation sites is crucial for conserving marine fish species.

Purpose of the Study:

• To identify potential spawning aggregation sites for permit (Trachinotus falcatus) in the Florida Keys.
• To integrate diverse data sources using machine learning for ecological modeling.

Main Methods:

• Applied supervised (Random Forests; RF) and unsupervised (Fuzzy K-Means; FKM) machine learning algorithms.
• Utilized fish tracking data, environmental data, and visual surveys.
• Compared predictions from different algorithms and data sources.

Main Results:

• Both RF models showed similar predictive performance but differed in variable importance and site predictions.
• Fuzzy K-Means clustering identified site groupings consistent with RF predictions.
• Identified potential permit spawning sites characterized by deep-water reefs and high residency periods.

Conclusions:

• Multiple machine learning algorithms effectively integrated diverse data for ecological system modeling.
• Machine learning offers valuable tools for ecologists and conservation practitioners dealing with complex data.
• Conservation efforts for aggregate spawning species can benefit from identifying and protecting key sites identified through these methods.