Bayesian stable isotope mixing models effectively characterize the diet of an Arctic raptor

Abstract

1. Bayesian stable isotope mixing models (BSIMMs) for δ 13 C and δ 15 N can be a useful tool to reconstruct diets, characterize trophic relationships, and assess spatiotemporal variation in food webs. However, use of this approach typically requires a priori knowledge on the level of enrichment occurring between the diet and tissue of the consumer being sampled (i.e. a trophic discrimination factor or TDF). 2. Trophic discrimination factors derived from captive feeding studies are highly variable, and it is challenging to select the appropriate TDF for diet estimation in wild populations. We introduce a novel method for estimating TDFs in a wild population-a proportionally balanced equation that uses high-precision diet estimates from nest cameras installed on a subset of nests in lieu of a controlled feeding study (TDF CAM). 3. We tested the ability of BSIMMs to characterize diet in a free-living population of gyrfalcon Falco rusticolus nestlings by comparing model output to high-precision nest camera diet estimates. We analysed the performance of models formulated with a TDF CAM against other relevant TDFs and assessed model sensitivity to an informative prior. We applied the most parsimonious model inputs to a larger sample to analyse broad-scale temporal dietary trends. 4. Bayesian stable isotope mixing models fitted with a TDF CAM and uninformative prior had the best agreement with nest camera data, outperforming TDFs derived from captive feeding studies. BSIMMs produced with a TDF CAM produced reliable diet estimates at the nest level and accurately identified significant temporal shifts in gyrfalcon diet within and between years. 5. Our method of TDF estimation produced more accurate estimates of TDFs in a wild population than traditional approaches, consequently improving BSIMM diet estimates. We demonstrate how BSIMMs can complement a high-precision diet study by expanding its spatiotemporal scope of inference and recommend this integrative methodology as a powerful tool for future trophic studies.

Reevaluating trophic discrimination factors (Δδ13C and Δδ15N) for diet reconstruction

“…SIDER, which uses existing TDF data and phylogenetic relatedness, is a relatively new method that can handle multiple tissue types and estimate TDF variance. Because diet class tends to be phylogenetically constrained (Cattin et al, 2004;Rom an-Palacios et al, 2019) and is the primary driver of variation in TDF-δ 15 N (Figure 5b), it may be appropriate to use SIDER to estimate TDF-δ 15 N. Additionally, the TDFδ 13 C of carnivores is not strongly influenced by diet source composition (Figure 5a), and SIDER performed well for B. brevicauda, as it has for other carnivores (e.g., Morgenthaler et al, 2020;Swan et al, 2020; but see Johnson et al, 2020). However, our field study highlights the problem of using phylogenetic relatedness to estimate TDF-δ 13 C for herbivores and omnivores, which are strongly influenced by diet source composition (Figure 5a).…”

“…Trophic discrimination factors are primarily derived from lab‐based feeding experiments, where individuals are fed fixed diets until tissues reach an isotopic steady state (del Rio et al, 2009), although occasionally studies derive TDFs from wild animals with well‐known diets (e.g., Fox‐Dobbs et al, 2007; Johnson et al, 2020; Newsome et al, 2010) or from stomach contents (e.g., Codron et al, 2012; Stephens & Rowe, 2020; Wang et al, 2020). Because of the difficulty of maintaining captive animals, TDFs are not available for most species, and ecologists use values from a closely related species, default to commonly observed trophic step increases based on pooled data from meta‐analyses (often ca.…”

Use of marine vs. freshwater proteins for egg‐laying and incubation by sea ducks breeding in Arctic tundra

“…The program MixSIAR (Stock & Semmens, 2016) has been used successfully for many diet analyses, including diets of birds (Johnson et al, 2020) and relative contributions of marine vs. terrestrial foods to feather production (Robillard et al, 2020). However, in various instances, similar Bayesian approaches have been less successful (Kolts et al, 2013; Robinson et al, 2018).…”

“…However, in various instances, similar Bayesian approaches have been less successful (Kolts et al, 2013; Robinson et al, 2018). Bayesian models can be quite susceptible to biases resulting from priors specified to incorporate preexisting information (Brown et al, 2018; Johnson et al, 2020). We used MixSIAR because it has high statistical power and allows user‐specified variances in isotopic inputs.…”

Gaining insight into the assimilated diet of small bear populations by stable isotope analysis

“…However, when dealing with stable isotope data, some degree of uncertainty is inherent in SIA, including natural variability in the sources' isotopic signatures, variable TEF within a given species, different stable isotope ratios in different tissues, variations in the isotopic source and mixture process errors 80 . It has been recently recognized that the use of informative priors may introduce bias in the posterior www.nature.com/scientificreports/ estimations 81,82 , leading to erroneous representations of diet composition [83][84][85] . Nevertheless, we believe that incorporating informative priors based on non-isotopic data, such as prior knowledge on diet composition obtained with traditional dietary methods, in some cases can improve the precision of dietary reconstruction by SIMMs.…”