Performance evaluation of cetacean species distribution models developed using generalized additive models and...

Area of Science:

• Ecology
• Marine Biology
• Conservation Science

Background:

• Species distribution models (SDMs) are crucial for managing mobile marine species, offering spatially and temporally explicit distribution data.
• Generalized Additive Models (GAMs) and Boosted Regression Trees (BRTs) are common SDM frameworks, but comparative studies are scarce, especially those using presence/absence or density data.
• Few studies explore how species distribution characteristics influence SDM performance.

Purpose of the Study:

• To compare the performance of GAMs and BRTs for marine species distribution modeling.
• To evaluate GAMs predicting species density against those predicting presence/absence and against BRTs.
• To assess both explanatory and predictive power of different modeling approaches.

Main Methods:

• Compared GAMs (presence/absence and density) with BRTs using systematic survey data (1991-2014) for cetaceans in the California Current Ecosystem.
• Evaluated model goodness of fit (explanatory power) and performance on novel datasets (predictive power).
• Utilized a robust dataset for a taxonomically diverse suite of marine mammals.

Main Results:

• Both BRTs and GAMs effectively described overall distribution patterns for most species.
• Density GAMs demonstrated substantially greater predictive power on novel data compared to presence/absence GAMs and BRTs.
• Differences in predictive power are likely due to response variables and fitting algorithms.

Conclusions:

• Density GAMs offer superior predictive capabilities for marine species distribution modeling compared to presence/absence GAMs and BRTs.
• Understanding the strengths and limitations of different modeling techniques is vital for effective marine spatial management.
• Modelers and resource managers can use these findings to select optimal SDM techniques for specific research and management questions.