Abstract
Displacement mortality rates are challenging to quantify empirically and currently lacking empirical literature to provide evidence (WP1). Expert elicitation suggests that there are high levels of uncertainty regarding the values of these rates, but that they could potentially encompass values representing large changes to demographic rates (see WP2).
An alternative method for quantifying displacement mortality rates, and sources of variation in these rates, is via a mechanistic model. Simulations from the mechanistic model can be used to estimate the mortality rates associated with different scenarios. Within Work Package 3 we used SeabORD (Searle et al. 2014, 2017), an individual-based model of seabird behaviour, energetics, demography and windfarm interactions during chick-rearing, to estimate the levels of displacement mortality for breeding adults and their dependents, and for mass change in breeding adults, associated with different colonies under different wind farm scenarios, for three species (black-legged kittiwake, common guillemot and razorbill). However, SeabORD is a computationally intensive model to run, and so to provide a framework for approximating the outputs that SeabORD would have provided under other scenarios, we developed a statistical emulator – a statistical model that is designed to approximate a mechanistic model. Emulation uses a “training set” of mechanistic model inputs and outputs to build a general model for the relationship between the mechanistic model inputs and outputs, and, as such, provides an approximation to the mechanistic model that can be used to predict the likely outputs that the mechanistic model would have produced under alternative scenarios (e.g., in this context, for alternative SPAs and/or wind farm scenarios).
Within WP3 a “training set” was generated by running SeabORD for each of three species (guillemot, kittiwake and razorbill) at three SPAs, and a range of windfarm scenarios. These runs were then used to build emulators, which aimed to capture the key characteristics of the SeabORD runs, and thereby to predict the results that would be obtained by running SeabORD at SPAs/windfarms that were not included in the training set. Emulators were constructed in relation to three key outputs from SeabORD: chick mortality (per nest), adult mortality (as a proportion of breeding adults) and percent mass loss over the chick-rearing season. In each case, the response variable used for the emulator was the difference between these values under a windfarm scenario and the values obtained under the baseline. The emulator links these response variables to a range of explanatory variables that summarise key characteristics of the SPA, windfarm and SPA-windfarm interaction.
The results of the emulation work in WP3 need to be interpreted cautiously because they are based on a relatively small training set of SeabORD runs, but the key findings from the work were that the impacts of windfarm scenarios on adult mass loss and adult and chick mortality over the course of chick rearing all show a very strong positive relationship to the proportion of birds displaced at each time point (“ptdisp”). This proportion is simply the multiple of the displacement rate with the proportion of the bird distribution that lies within any footprint (“totalpinords”). There was no clear evidence for non-linear effects, or for effects of other explanatory variables, but there was evidence in almost all cases that the magnitude of the relationship varied between SPAs.
The SeabORD outputs focused on are simulations of overall changes in adult mass and adult and chick survival rates within the entire breeding population being simulated within SeabORD. In contrast, displacement mortality rates focus on the excess mortality rate experienced by a subset of the whole population – i.e., those birds that have been displaced. The exact definition varies by context: the expert elicitation exercise, in particular (WP2 report), defined the adult mortality rate in relation to any adult birds that are affected by displacement, including those affected indirectly rather than directly, and defined an impact rate for chicks as well as adults.
Displacement mortality rates are of interest for two key reasons:
1. Displacement mortality rates are a key input to the Displacement Matrix, were a focus of the literature review (WP1) and expert elicitation exercise (WP2) and are the main focus of this project.
2. Displacement impacts on birds subjected to displacement effects are likely to be more generalisable across scenarios and colonies than population level impacts on overall mortality, since the latter will be heavily influenced by the level of baseline spatial interaction with the wind farm footprints.
In this WP we consider ways in which the outputs from SeabORD and the associated emulator from WP3 can be used to provide information on displacement mortality rates. SeabORD does not directly use a “displacement mortality rate”, and the main outputs from it relate to population-level outcomes, but a range of other metrics can also be extracted from SeabORD, and some of these are related to the “displacement mortality rate”. In particular, the emulator within WP3 was explicitly constructed in a way that allows the parameters of it to be interpreted in relation to displacement mortality rates.
In this work package we define “model-based displacement mortality rates” in two possible ways and calculate these using the SeabORD outputs and estimated emulator parameters from Work Package 3. We use the phrase “model-based displacement mortality rates” to make it explicit that the rates we define using SeabORD outputs are not directly comparable to the “displacement mortality rate” used in the Displacement Matrix – we outline the connection between the different definitions of “displacement mortality rate” considered in the Displacement Matrix, Expert Elicitation and SeabORD/emulator in WP5, and the challenges in translating between these. The key advantage of using the emulator, as well as the SeabORD runs themselves, is that the emulator provides a natural framework for investigating variability and uncertainty in rates.
We quantify mean rates and variability in the resulting “model-based displacement mortality rates” for each of the three species.