Abstract
This project has developed an individual- based model (IBM) of seabirds for the non- breeding season which allows time- energy budgets, and consequent impact on body mass and survival, to be simulated under baseline (current) and future (with offshore wind, OW) scenarios.
The development of a non- breeding season model of displacement is crucial in allowing the quantification of displacement impacts within the assessment process to be as transparent as possible, and to be based on the best available scientific evidence, and thereby plays an important role in filling a key evidence gap in relation to the assessment process.
The development of a mechanistic model- based approach to quantifying displacement impacts, and the uncertainty associated with this, will also be of direct use in reducing consent risk by providing an improved understanding of the mechanisms underpinning displacement risk and the ability to incrementally incorporate new evidence into assessments as it becomes available.
The IBM has been applied (within WP2) to two case studies of UK seabird populations - common guillemot on the Isle of May, and red- throated diver in the Outer Thames Estuary SPA – using a hypothetical but plausible scenario of North Sea OW development.
This work package has focused on the quantification of uncertainty and variability using the IBM, and begins by outlining the approach taken to treatment of uncertainty within the model.
Uncertainty and sensitivity analyses (UA and SA) are then used to identify key sources of uncertainty within the model, which underpin recommendations around future research and data collection.
Since the model is, as with other IBMs, relatively computationally intensive, a high- performance computing cluster is used to undertake the model runs that underpin these analyses
The uncertainty analysis (UA) focuses on evaluating overall levels of variation in mass at the end of the non- breeding season between agents and between parameter combinations
The sensitivity analysis (SA) focuses on sensitivity of model outputs to variation in several key model inputs, specifically 7 key population- level parameters that relate to initial mass, energy costs of activities, proportion of individuals susceptible to OW displacement effects, and the body mass threshold for adult mortality.
These analyses focus on three key model output metrics of direct relevance to offshore wind assessments: difference in population- level average (mean or median) mass at the end of season between impacted and baseline scenarios, which provides a proxy for OW impacts on productivity, and difference in population- level adult survival between impacted and baseline scenarios.
Results of SA show the mean or median OW impact on mass at the end of the non- breeding season are most sensitive to three input parameters: the energy cost of “active” behaviour, followed by the energy costs of “inactive” and “dive” behaviours.
SA also showed that OW impact on adult survival is most sensitive to two input parameters: mean initial body mass and the mortality threshold.
Investigations of the relationships between outputs (R- squared values for SA models, bootstrap standard errors on key output metrics) and the number of agents per combination suggest that a large number of agents (potentially much larger than 8000) may be needed to obtain stable summary statistics of OW impacts, because of high levels of inter- individual variability (e.g., in initial mass and energy costs of activities) relative to levels of variability resulting from changes in parameter values.
However, the qualitative results of the SA were the same regardless of the SA method used (random forest or regression tree) and regardless of whether 8000 or 4000 agents were considered, suggesting that key qualitative results of the analyses may be robust to the number of agents used.
SA results showed relatively low sensitivity to the proportion of individuals from the colony of interest that are susceptible to displacement, but this is likely to be because the effects of competition on the population of interest are dominated by displacement effects on individuals from other colonies, and the fundamental rebalancing of time spent feeding within the IBM in response to energy intake.
Uncertainty would be reduced through an improved understanding of the energetic costs incurred by non- breeding seabirds, which might be obtained through deployment of appropriate biologging devices and/or biophysical modelling.
Future work would be valuable to (a) better understand and incorporate uncertainty in the populationlevel values of parameters (e.g., via expert elicitation) and (b) evaluate sensitivity to bird distribution maps and sea- surface temperature maps.