Statistical Methodology

Modelling the foraging distribution of seabirds

Movement is a vitally important aspect of animal behaviour. Whilst data on movement have traditionally been difficult to collect, recent technological developments – such as the increasing use of smartphones and the decreasing cost and weight of electronic tags – mean that automated data on movement are now becoming available in large quantities. Here we describe elements of three projects concerned with the analysis of automated data on seabird movement, all undertaken with the ultimate aim of protecting species for which Scotland holds important breeding populations.

foraging and speed graphs Observed speeds obtained from bird-based GPS records for an individual kittiwake (top), and the predicted behavioural states that are associated with fitting a two-state hidden Markov model to these data (bottom).

The first project, undertaken in collaboration with RSPB, involved analysing the foraging behaviour of five seabird species – guillemot, razorbill, shag, kittiwake and fulmar – at colonies located throughout the British Isles utilising data obtained from GPS tags that had been attached to individual birds. Since the data do not distinguish unambiguously between foraging and non-foraging behaviour, a key element of our analysis involved classifying behaviour on the basis of time series of observed flight speeds, which we did using hidden Markov models. Subsequent analysis was primarily based on using logistic regression to compare the environmental characteristics of observed locations against the characteristics of ‘control’ points drawn from the study regions.

A second project, being carried out in collaboration with CEH, involves analysing automated GPS tag data to provide inputs to a mechanistic model of foraging behaviour and energy balance for birds. This model is then used to study the impacts of offshore wind farms on seabird behaviour in the Forth-Tay area (e.g. the increased flight distances that birds may need to travel in order to avoid wind farms, and the reduced energy intake that they may incur as a result of being excluded from areas with high prey densities), and thereby upon survival and productivity. The analysis of the GPS data is solely concerned with describing, rather than explaining, spatial variations in bird densities, and therefore makes use of a semi-parametric modelling approach (GAMs – generalised additive models).

The final project involved the analysis of flight paths of individual terns for JNCC, collected without the stress of capture by following birds in a boat as they leave their colonies to feed. This has the additional benefit of enabling the birds’ activity to be recorded alongside the location information from the on-board GPS. Previous work used a form of random effects modelling, but we found better results were obtained using a weighted regression analysis, as the observed flight paths were of very different lengths. The data sets were large, and in order to properly account for spatial correlation our regression analysis was conducted using the INLA (Integrated Nested Laplace Approximation) software. We found associations between spatial locations of foraging and variables such as levels of chlorophyll and seabed depth; importantly, the best predictive model depended on both colony location and tern species.

map of the sea round Strangford Lough Map of the sea around Strangford Lough, coloured from blue (low) to red (high) according to the predicted relative usage by arctic terns.

Further details from: Adam Butler and Mark Brewer

Article date 2013


Statistical Genomics and Bioinformatics

Process and Systems Modelling

Statistical Methodology

PhD Opportunities

Meetings & Seminars