Point distance transect sampling using iterated integrated nested Laplace approximations

This repository contains the data and code to reproduce the results and figures in the Chapter 2 of my thesis "Expanding the use of spatial models in statistical ecology" (link to thesis will go here once it is available online to download).

I have attempted to make the code reproducible by using the renv package. renv creates a local cache of all packages used for the analysis. Details of these package versions and the R version used for the analysis can be read in the renv/renv.lock file. renv will install the required packages in a local repository and all R sessions within the repository will automatically use these packages. This means you can clone this repository and run this analysis without changing the packages you use for other projects, which may require different versions than those used here. Note that renv does not guarantee reproducibility as different machines will have different local operating system and compilers.

If you would prefer not to use renv and to set up the dependencies yourself then run

renv::deactivate()

in the project root directory. This will deactivate the renv auto-loader and you can inspect renv.lock to see the dependencies required. If you already have a version of INLA, sf, and rgdal on your machine this might save you some time by avoiding installing these hefty packages again in a local cache (if they are a different version that is, if they are the same version on your machine then they are copied to the local cache which is quick but does use more disk space).

If you have Rscript installed you can run the analysis by running the following commands from the project root directory:

Rscript Analysis/fit_model.R
Rscript Analysis/eval_model.R
Rscript Analysis/eval_spde.R
Rscript Analysis/posterior_N.R
Rscript Analysis/excursions.R

fit_model.R should be run first, the rest of the scripts can be run in any order.

This code will produce results very similar to the figures in the thesis. However, there is a technical difference in the approximate inference technique used in this analysis. This does not seem to affect the results in this example. inlabru has been updated in the time since this the thesis was written and the criteria for convergence of the iterated INLA approach in this repo uses the more recent version.

See the iterated INLA vignette available here, which details the new approach and the footnote which contains the older method used for the thesis.