Abstract
The Isle of Rum (Scotland) holds 25-30% of the world’s Manx shearwater population. It is currently unclear whether this internationally important and relatively isolated sub population is increasing or not, and whether it would be vulnerable to proposed marine renewable developments in its vicinity. In this work, we undertook a detailed review and collation of available historical and current data for Manx shearwaters on Rum and synthesised them into an integrated population model that allowed us to reconstruct population trends and quantify future sensitivities.
We found extensive and precise information for several demographic processes (especially breeding success and adult survival) and baseline rates. The availability of survey data, directly estimating the population size is limited, intermittent and of variable (often low) precision. Availability of weather covariates is unexpectedly patchy (i.e., not collected from the same sources), but extensive. Data for key covariates of concern, such as proxies of rat abundance are very recent (hence of limited value, since their consequences may not yet have become manifested in the population data) and do not coincide with data on breeding success.
For our modelling, we adopted a Bayesian state-space model because it allowed us to 1) integrate diverse data types, collected by different field methodologies and 2) represent different forms of uncertainty and stochasticity in our results and projections. This approach therefore achieves the fine balance of maximising precision (by using all data available) without overestimating confidence (by correctly propagating uncertainty to the final results).
The population part of our state-space model incorporates essential facts about Manx shearwater life history (e.g., generational lags) as well as environmental and density-dependent constraints to growth. The simplest version of the model contains no environmental covariates, opting instead to represent extrinsic drivers by random effects in breeding success and adult survival. More elaborate extensions of the process component of the model illustrate how a partially observed covariate (rainfall) can be incorporated into demography.
The observation part of our state-space model considers the features (limitations and advantages) of all the field data. We consider bias and precision in population estimates, particularly regarding the most recent and statistically influential studies. We perform a sensitivity analysis by relaxing the reported level of precision in different parts of the observation data and proceed with a precautionary version of the model that includes previously omitted sources of uncertainty.
The modelling approach was validated by fitting the model to synthetic data of a similar nature and extent to the currently available real data. This validation exercise suggests that the available data are sufficient to accurately and precisely retrieve hidden parameters and to reconstruct latent population trends in partially observed or wholly unobserved demographic time-series.
Such reconstructions, generated from the real data indicate that the population has been increasing since the 1980s but may now be starting to experience the effects of density dependence. We extend the temporal horizon of the model to a 100-year forecasting range and run several counterfactual scenarios relating to anthropogenic impacts on adult mortality and fecundity. These experiments indicate that the population would be robust to very strong pulse perturbations (e.g., 50% adult mortality and complete failure in fecundity), but would be vulnerable to sustained, press perturbations in the survival of adults.
We conclude by discussing possible approaches to survey data imprecisions, the future inclusion of other covariates and the development of a detailed model for density dependence more appropriate for the biology of burrowing seabirds.