Developing a population model for Rum Manx Shearwaters for assessing offshore wind farm impacts and conservation measures

5 Discussion

The drivers of population size and the long-term trends of the Rum population of Manx shearwaters are complex and include both intrinsic and extrinsic variables operating within the colony and in the birds' marine range, which includes wintering and pre-breeding areas in the western Atlantic. Currently, uncertainties exist with respect to the relative importance of the various proximate population drivers (breeding success, adult survival, recruitment, immigration/emigration etc.) in determining population trends and how underlying environmental variables (including rat depredation, summer weather, winter weather, marine food supply, anthropogenic impacts etc.) interact to drive these. These uncertainties mean that it is difficult to predict population trends, a prerequisite for environmental assessments, as well as the predicted efficacy of any management interventions at the colony.

To address this uncertainty, Marine Scotland, and NatureScot, commissioned this study via the ScotMER programme to construct a robust model of the population dynamics of the Rum Manx shearwater population by applying Bayesian hierarchical modelling approaches to relevant available data. In doing so, this project aimed to provide vital information that will help support the sustainable development of offshore wind whilst also protecting Scotland’s unique marine environment.

The markdown code and documents for modelling viability of the Rum Manx Shearwater population including the fully-fitted, user-friendly and computationally efficient R-functions for generating these projections are available with the deliverables of this project via the following the link which will open the GitHub page: https://github.com/JasonMat/ManxRum.git. Future work could design more systematic factorial designs for investigating impacts on this population. Alternatively, case-specific questions can be asked if the exact profile of impacts for a particular development is known. In addition to producing a robust prediction tool for forecasts and counterfactuals, this project has reached several valuable conclusions that may be useful for future data collection or the management of impacts on the population. Specifically:

1. Available population and demographic data are sufficient to fit an informative population model for the internationally important Manx shearwater population on the Isle of Rum. The model can reconstruct past trends and provide informative predictions and scenario counterfactuals over periods of decades.

2. It is less clear whether the available data can inform us about the current strength of density dependence and potential consequences for recruitment. A detailed model of successively relaxing density dependence is described in the Technical Report, but the available data do not allow it to be embedded in the state-space model.

3. It is equally not clear whether the model can detect trends in response to covariates. The original ambition was that once the state-space model had decomposed the observed dynamics into individual demographic trajectories, it would become easier for the model to detect correlations with covariates. In the case of the one covariate (rainfall) for which data and prior knowledge existed to do this, the results were not conclusive. Other covariates (such as rat density) were not examinable due to data-sparsity.

4. Importantly, the simulation trials with this model, replicating the exact availability (types, sample sizes and time instances) of the real data, indicated that retrieval of information on density dependence and covariates may currently prove challenging.

5. Investigation of the sensitivity on the precision and bias of population estimates indicates that precision is less influential than bias. So, any additional data collection effort may better be spent in accurately mapping the shearwater greens, rather than increasing effort within well studied areas (i.e., spatial extent is probably preferable to local replication).

6. Low and intermediate population estimates for 2021 give similar results, indicating a robust family of models.

7. While the survey methods remain unstandardised, the precautionary modelling approach is to proceed with a model that permits both over- and under-estimation of the current population. This is based on the facts that the Inger et al. (2022) report did not validate the habitat modelling approach (potential upward bias) and did not propagate analysis uncertainty to final results (potential over-representation of precision).

8. Predictions from the fitted model anticipate a mostly stable population under the status-quo.

9. Press perturbations seem to be capable of turning the population to a declining trend. For example, a 5% annual reduction in adult survival leads to a 69% chance of decline over the next 25 and 100 years, whereas a more drastic (50%) annual reduction in fecundity would be required for a similar outcome.

10. Subsequently, from these indicative and highly selective investigations, we conclude that the population would be most vulnerable to long-term but modest reductions in the survival of breeders.

The strategy taken here to model environmentally induced variation was to start by associating an annually varying random effect to each demographic process. In this way, even though the baseline models did not contain extrinsic covariates, we were able to capture the levels of variability in each demographic process. This offers an advantage for forecasting, because covariate time series are, by default, historical data and are not known in the future. Our random effects are built-in to the forecasts and subsequent sensitivity conclusions drawn from this study. Hence, sensitivity here is concluded not just as the result of deterministic strength of one demographic process over another, but also upon their relative levels of stochastic variation.

Nevertheless, the stated objective of the project funders was to have a framework that can investigate proximate covariates. Technically, this task can be thought of as an apportioning of the empirically inferred dispersion (the random effects), into the contributions of covariates. We began this process here by looking at rainfall.

The finding that rainfall did not appear to have a negative impact on breeding success is contradictory to previous work by (Thompson 1987; Thompson and Furness, 1991) as well as the intuition of several experts who have worked on this population on Rum. The inferential approach to looking at this relationship is different here compared to previous work. The model essentially reconstructs the time series of breeding success across the full duration of the study and simultaneously draws on information from Rum and Tiree, to get a more complete picture of rainfall across the years. Fig. 3a indicates that rainfall is a weak covariate of breeding success. It may be that the relationship derived by previous workers, based as it was on a shorter segment of time, was in fact spurious. Alternatively, it may be that the reconstruction of breeding success in the present model is accounting for the effects of missing covariates, other than rainfall, that have a confounding effect. Or the effect of rainfall may require analysis of daily rainfall data rather than monthly totals as used in this study, in order to identify rare events of exceptional levels of rainfall. Finally, it may be that the linear incorporation of monthly totals of rainfall into our model was too simplistic to capture the physical processes of burrow flooding. For example, if optimal burrows characterised by a low risk of flooding are occupied first by breeders, then it is likely that the effects of flooding will be felt only in high-density populations where individuals are forced to occupy high-risk burrows.

From a modelling perspective, it is now straightforward to include additional covariates, on the condition that the covariate time series are sufficiently complete and that there is enough population survey data to detect correlations. Neither of these requirements currently holds, but it is important to catalogue here possible extensions with additional influential covariates.

The sparsity of data on rat abundance did not permit us to examine rat abundance as a covariate of breeding success, but this should be a monitored variable for future investigations with this model. On Canna (which can be considered a small offshoot of the Rum colony) the Manx shearwater population declined because of predation of eggs and chicks by rats. However, the rats on Rum do not seem to have a similar effect on the shearwaters. On Rum, the colony is unusual in that shearwaters nest about 1000 feet above sea level at the tops of the Rum hills but the colony may previously have also extended to lower altitudes. In addition, having burrows in the well-drained gravelly ground of the hilltops made burrows less prone to flooding during heavy rain (Thompson 1987, Lambert et al. 2021). Data availability for rats is very limited on Rum, covering the period 2010-13, which critically, does not overlap with the availability of data on breeding success. Nevertheless, it seems that there is something specific to Rum which until now has prevented the growth of a rat population to a sufficiently large size that could drive shearwater population decline. That could potentially be winter cold in the colony area limiting rat numbers.

Additional information on weather-related variables, such as days of frost and climate-related variables, such as sea-surface temperature and the North Atlantic oscillation have previously been used (Thompson and Furness 1991, Duff 2011) and could be examined again as determinants of adult survival in the present more integrated approach. The availability of those varies but can go as far back as 1958. Wood et al. (2021) found a significant correlation between adult survival at Skomer and wind strength. However, the annual variation in survival estimates at Rum is very small for the available series of years (varying only between 0.975 and 0.986 over the period 1994-2011) so the possibility to detect such covariate effects will be limited. Wood et al. (2021) were unable to find any environmental covariates that correlated significantly with breeding success of Manx shearwaters at Skomer. Note however that fledgling weights, as well as chick survival to fledging, are also important with respect to future survival and are affected by timing of breeding. Recent tracking work from Skomer suggests that pre-laying females forage in very different locations from adults later in the season provisioning young (Tim Guilford, pers. comm.). Taken together, this may indicate that there could be particular spring conditions within which females find it more challenging to accumulate the resources needed for egg production, resulting in delay to laying dates, which in turn will impact chick fledging dates and weights with associated impact on post-fledging survival.

If rat depredation is a factor impacting breeding success, there could potentially be a weather-related effect whereby harsher winters might be anticipated to reduce the rat population within the colony. However rainfall conditions in the preceding spring might also be a factor, such that in years with high levels of egg mortality associated with burrow flooding there may be more food supplies within the colony to sustain scavenging rats over the winter period. Incidentally, this is also the reason for added concern around Highly Pathogenic Avian Influenza (HPAI) impacts (Lane et al. 2023) and possibility of large-scale associated breeding failure, through either direct mortality of chicks or indirect mortality through adult mortality resulting in chick starvation, leading to a super-abundant winter food supply that could sustain a rat population in the colony area over winter and drive high depredation of eggs/chicks in the subsequent breeding season.

Fishing productivity, has previously been used to explore effects on adult breeding success (Duff 2011), but Manx shearwaters do not normally scavenge at fishing boats and their prey is not generally subject to fishing mortality in the west of Scotland waters. Disease, most recently, the occurrence of HPAI on Rum, is expected to impact the population with mortality surveillance and HPAI testing measures in place to assess the impact of an outbreak should this occur (NatureScot Scientific Advisory Committee Sub-Group on Avian Influenza 2023). Finally, the impact of marine renewable developments can only currently be investigated by scenario exploration (similar to the forecasts produced here). However, such scenarios could be made more realistic by distinguishing between the impact of construction (a pulse perturbation) and operation (a press perturbation) as well as the postulated effects on multiple demographic processes simultaneously. The models developed in this work, using the entirety of available data on this population offer a valuable baseline against which the effects of anthropogenic developments can be assessed.