A novel method to estimate prey contributions to predator diets

Abstract

Stomach content data are frequently used to characterize predator feeding habits, often by describing the proportional contribution by mass or number of each prey type (diet fractions). These data pose several statistical challenges for analysis and estimation that have hindered our ability to create quantitative diet fraction estimates from stomach content data. To address these challenges, we developed a novel, likelihood-based mixture model to quantitatively estimate diet fractions. Simulation testing indicated that estimated diet fractions from the mixture model were more precise than those estimated either from a (stomach mass) weighted mean or the sample mean and were more accurate than a sample mean. Additionally, we applied the mixture model, a weighted mean, and sample mean to stomach content data for multiple types of predators. For three of four of these data sets, the mixture model demonstrated higher precision than and similar accuracy to a weighted mean and similar precision and better accuracy than a sample mean. The mixture model represents an important step in advancing statistical methods to address the challenges of stomach content data. Résumé :Des données sur le contenu stomacal sont fréquemment utilisées pour caractériser les habitudes d'alimentation de prédateurs, bien souvent en décrivant la proportion de la contribution, en poids ou en nombre, de chaque type de proies (fractions du régime alimentaire). Ces données posent plusieurs problèmes statistiques pour l'analyse et l'estimation qui ont limité la capacité d'établir des estimations quantitatives des fractions du régime alimentaire à partir de données sur le contenu stomacal. Dans le but de régler ces problèmes, nous avons mis au point un nouveau modèle de mélange reposant sur la probabilité pour estimer les fractions du régime alimentaire de manière quantitative. Des essais de simulation indiquent que les fractions du régime alimentaire estimées à l'aide du modèle de mélange sont plus précises que les fractions estimées à partir d'une moyenne pondérée (de la masse stomacale) ou de la moyenne de l'échantillon, et sont plus exactes qu'une moyenne d'échantillon. Nous avons en outre appliqué le modèle de mélange, une moyenne pondérée et une moyenne d'échantillon aux données sur le contenu stomacal pour différents types de prédateurs. Pour trois ensembles de données sur quatre, le modèle de mélange présente une plus grande précision et une exactitude semblable à une moyenne pondérée, et une précision semblable et une plus grande exactitude qu'une moyenne d'échantillon. Le modèle de mélange représente une importante étape dans l'établissement de méthodes statistiques permettant de surmonter les difficultés que posent les données sur le contenu stomacal. [Traduit par la Rédaction]

Unexpected food web responses to low dissolved oxygen in an estuarine fjord

“…The fraction of diet (by mass) consisting of euphausiids was calculated using the estimation model of Moriarty et al (2017): this model accounts for potential covariance between diet composition and feeding rate and is robust to outliers. We created a Bayesian version of this model, which provided more robust estimates of observation error and variance.…”

Feeding ecology of fishes associated with artificial reefs in the northwest Gulf of Mexico

“…The %W of dominant prey groups was used as the dependent variable for all statistical analyses of diet, as it is a useful proxy for estimating the nutritional contribution of prey groups [ 24 ]. Prey contributions to species’ diets were estimated from maximum likelihood estimates using a diet mixture model described by Moriarty et al [ 49 ]. Because this model assumes that the prey group for which contributions are being estimated makes up 100% (%W) of the contents for a minimum of one consumer’s gut (where the proportion (p) = 1), a reduced model (where this probability was assumed to be 0 instead of 1) was used when none of the samples met this assumption [ 49 ].…”

“…Prey contributions to species’ diets were estimated from maximum likelihood estimates using a diet mixture model described by Moriarty et al [ 49 ]. Because this model assumes that the prey group for which contributions are being estimated makes up 100% (%W) of the contents for a minimum of one consumer’s gut (where the proportion (p) = 1), a reduced model (where this probability was assumed to be 0 instead of 1) was used when none of the samples met this assumption [ 49 ]. Percent composition by weight was square-root transformed to reduce the importance of dominant prey groups and used to create a Bray-Curtis similarity matrix.…”

Spatio‐temporal analyses of marine predator diets from data‐rich and data‐limited systems

“…a workaround is needed) if any of the entries of the dataset at hand are associated with diet proportions that are exactly 0 or 1, because these values are outside the support of the Dirichlet distribution. Moriarty et al (2017) implemented a likelihood-based mixture model for estimating diet proportions (hereafter referred to as the "Moriarty method"). The Moriarty method uses a mixture of Bernoulli, Gamma and Beta models to deal with the specific cases where the proportion of a prey item in sampled stomachs is 0, ranges between 0 (exclusive) and 1 (exclusive), or is 1.…”

“…The contribution of different prey categories to predators’ diet by weight, generally referred to as “diet proportions,” is the most popular metric used in fisheries studies of trophic interactions (Baker et al, 2014; Elliott & Persson, 1978; Liao, Pierce, & Larscheid, 2001; Livingston et al, 2017). Diet proportions are typically determined from analyses of fish stomachs collected by fisheries‐independent surveys or obtained from fishers (Ainsworth, Kaplan, Levin, & Mangel, 2010; Glaser, 2010; Livingston et al, 2017; Moriarty, Essington, & Ward, 2017). Importantly, diet proportion estimates are needed to parameterize the diet matrices of most of the ecosystem modelling platforms that are instrumental in advancing EBFM (e.g.…”