Govan, Emma (2024) Flexible Models and New Algorithms for Fitting Stable Isotope Mixing Models. PhD thesis, National University of Ireland Maynooth.
Preview
Emma_Govan_phd_thesis.pdf
Available under License Creative Commons Attribution Non-commercial Share Alike.
Download (2MB) | Preview
Abstract
Stable Isotope Mixing Models (SIMMs) are important for ecologists. They allow
for the study of animal diets via measurement of biologically relevant stable isotopes.
These measurements can be used to estimate the contribution of different
food sources to an animals diet. Knowledge of an animals diet is important when
we wish to conserve species, as we need to know what food they rely on. Knowledge
of an animals diet can be used to quantify an animals niche and to assess
competition between species. SIMMs are also widely used in studies on pollution
and air quality, where they may be referred to as ‘source apportionment’, ‘end
member analysis’, or ‘mass balance analysis’ models.
However, SIMMs are currently mainly run using Markov chain Monte Carlo (MCMC),
which, while guaranteed to converge, can be prohibitively slow, requiring millions
of iterations in order to reach convergence if the model is complex. In this thesis
we have developed new tools for running SIMMs via Variational Bayes. This allows
for a speed improvement ranging between two and one hundred times when
compared to MCMC while still obtaining comparable results.
The work in this thesis is divided into 3 chapters, each focusing on a different R
package. Separate R packages were implemented for ease-of-use for non-expert
users as well as to allow past work using these packages to be replicable in future.
The packages are all designed for ecologists and individuals without a robust
statistical background, with detailed vignettes and examples included in the packages.
Chapter 3 in this thesis focuses on simmr, an R package for running SIMMs. simmr
allows users to choose between running models through MCMC or through Fixed Form Variational Bayes. simmr is designed for ease of use for non-expert users and
has built-in plotting and summary functions.
Chapter 4 in this thesis focuses on cosimmr, an R package developed for running
SIMMs with fixed covariates included. This package is developed using Variational
Bayes and offers up to a one order of magnitude speed improvement over other
packages. cosimmr has built-in predict and plotting functions and allows for users
to easily visualise their results.
Chapter 5 focuses on cosimmrSTAN, an R package developed which utilises STANs
Variational Bayes functionality in order to run complex SIMMs with fixed and/or
random effects, as well as allowing the hierarchical fitting of food sources or the use
of raw source data. cosimmrSTAN offers between 70-100 times speed improvement
over other packages.
These speed improvements mean that ecologists can use SIMMs more easily, with
accessible packages and quicker turnaround for results. This also means model
comparison becomes more accessible, with users able to run multiple models
quickly and compare results between them in order to make better informed decisions
about covariate inclusion. Ultimately, use of these packages will allow for
more comprehensive analyses of animal diets, and will allow users to gain insights
into species’ role in the ecosystem.
Item Type: | Thesis (PhD) |
---|---|
Keywords: | Flexible Models; New Algorithms; Stable Isotope Mixing Models; |
Academic Unit: | Faculty of Science and Engineering > Research Institutes > Hamilton Institute |
Item ID: | 20666 |
Depositing User: | IR eTheses |
Date Deposited: | 09 Oct 2025 14:54 |
Funders: | Insight Science Foundation Ireland (SF1/12/RC/2289_P2_Parnell) |
URI: | https://mural.maynoothuniversity.ie/id/eprint/20666 |
Use Licence: | This item is available under a Creative Commons Attribution Non Commercial Share Alike Licence (CC BY-NC-SA). Details of this licence are available here |
Repository Staff Only (login required)
Downloads
Downloads per month over past year