Diet analysis using generalized linear models derived from foraging processes using R package mvtweedie.

Area of Science:

• Ecology
• Quantitative Biology
• Statistical Modeling

Background:

• Diet analysis commonly treats prey data as compositional percentages, posing analytical challenges.
• Existing methods often require complex models or data transformations for compositional data.
• Predator foraging behavior can be modeled as a point process, providing a theoretical basis for diet analysis.

Purpose of the Study:

• To present a novel approximation for diet compositional data analysis using a Tweedie generalized linear model (GLM).
• To provide a practical tutorial and software (mvtweedie package) for implementing this Tweedie GLM approach.
• To demonstrate the utility of the method with ecological case studies.

Main Methods:

• Developed a Tweedie GLM approximation derived from a thinned and marked point process model of predator foraging.
• Created the R package 'mvtweedie' to extend 'mgcv' and 'ggplot2' for prey composition calculations.
• Applied the method to diet data from Tufted Puffins and wolves in Alaska.

Main Results:

• The Tweedie GLM offers a practical solution for analyzing complex diet compositional data.
• Demonstrated decadal prey switching in Tufted Puffins (e.g., sand lance to herring).
• Showcased dietary variations in wolves based on habitat (upland vs. coastal).

Conclusions:

• The Tweedie GLM provides a robust and accessible framework for ecological diet analysis.
• The mvtweedie package facilitates the application of this novel statistical approach.
• This method can advance our understanding of predator-prey dynamics and ecological responses.