Point process models for spatio-temporal distance sampling data
Y. Yuan, F. E. Bachl, F. Lindgren, D. L. Brochers, J. B. Illian, S. T. Buckland, H. Rue, T. Gerrodette
Distance sampling is a widely used method for estimating wildlife population abundance. Design-based distance sampling methods allow abundance to be estimated within any unit that has adequate replication for reliable estimation, which constrains the spatial resolution at which animal density can be estimated, but there is often interest in the underlying spatial structure of populations at finer spatial resolution than this. More sophisticated model-based approaches are required to draw inferences in this case. We formulate the process generating distance sampling data as a thinned spatial point pro cess and propose model-based inference based on a spatial log-Gaussian Cox process. Our method uses a flexible stochastic partial differential equation (SPDE) approach to account for the spatial autocorrelation, and integrated nested Laplace approximation (INLA) for Bayesian inference. We illustrate the method using distance sampling data from a series of shipboard line transect surveys in the eastern tropical Pacific (ETP).
Striped dolphin grouping behaviour
F. E. Bachl, Y. Yuan, D. L. Brochers, J. B. Illian, S. T. Buckland, H. Rue, T. Gerrodette, F. Lindgren
Download ISEC 2016 presentation: PDF
Distance sampling is a well-established scheme for wildlife abundance estimation. With the advent of more sophisticated statistical methods recent studies have overcome restrictive design based approaches that do not model the spatial distribution of animals in the study region. In particular, point processes have been shown to provide a flexible framework for modeling species distributions and covariate dependence thereof. In this contribution we show how to overcome an important intricacy of inference in point process models. Covariate estimates like group size are intrinsically biased by their influence on detectability are often spatially dependent. By coupling a latent spatial model of the covariate with a family of detections functions we obtaina fully probabilistic estimate of the unbiased covariate distribution in space. Inference on detection functions, covariate and group distribution is drawn simultaneously using integrated nested Laplace approximation (INLA), where the latter two are modeled using the stochastic partial differential equation (SPDE) approach to account for the spatial autocorrelation. Moreover, we present an approach to directly translating a point process animal groups to single animals. The method is illustrated using distance sampling data from a series of shipboard line transect surveys in the eastern tropical Pacific (ETP).
Harbor porpoise distribution modeling using aerial survey data
L. D. Williamson, D. Sadykova, F. E. Bachl, J. Illian, B. E. Scott, P. M. Thompson, K. L. Brookes
A project in collaboration with the University of Aberdeen. Video surveys of harbour porpoise, which were performed along the east coast of Scotland in 2010 and 2014, have been analysed to map the distribution of the species. The study introduce the use of inlabru for species distribution modelling to an ecological audience. The data have therefore been analysed both using GAMs, which are often used by ecologists, and also inlabru to compare the results and illustrate the flexibility of the newly developed technique.
Results were presented at the Marine Alliance for Science and Technology for Scotland Annual Science Meeting 2016.
Differences in the distribution of porpoise and dolphin vs. their foraging locations
L. D. Williamson, F. E. Bachl, J. Illian, P. M. Thompson, K. L. Brookes, B. E. Scott,
A thorough understanding of species distributions is vital for effective spatial management of potential threats. Harbour porpoise are a species of conservation concern in UK and EU waters. Therefore, their distribution has been studied previously. However, most surveys have focussed on where the animals are detected either using visual, video or acoustic surveys. These detections do not necessarily reflect areas of importance for activities such as foraging. Using an array of 65 moored passive acoustic monitoring devices (C-PODs) deployed throughout the Moray Firth, Scotland in August 2010, we investigate if the overall distribution of harbour porpoise is different than their spatial patterns of foraging. The C-PODs recorded the times of echolocation detections of porpoise, from which presence/absence and foraging buzzes can be extracted. These data were analysed using flexible state of the art spatial statistical methods that account for the sampling strategy to model the spatial distribution of porpoise, as well as the distribution of their foraging. Model fitting was based on integrated nested Laplace approximation (INLA), which is computationally efficient and hence makes fitting realistically complex model fitting feasible. Overall porpoise distribution broadly matched what has been observed previously, with highest densities observed on the Smith Bank. However, porpoise foraging is more evenly spread throughout the study area, with some areas of highest foraging detected in areas with relatively low density. A strong diurnal pattern was also observed, with shifts in distribution and locations of foraging observed between day and night. A better understanding of the environmental or behavioural drivers of this shift in foraging is important for effective management of threats to this species. This research has shown that porpoise are indeed using the entire Moray Firth, perhaps for different purposes; therefore conservation strategies that focus on a single area may be insufficient to protect the species.
Results will be presented at the Scottish Ecology, Environment and Conservation Conference 2017
Harped and hooded seal whelping ground estimation
M. Jullum, F. E. Bachl, T. Thorarinsdottir
A project in collaboration with the Norwegian Center for Scientific Computing. The aim of the project is to estimate the abundance of harp and hooded seals in the North Atlantic. These species have been harvested in this area for centuries and regular abundance estimation is critical for the management of the species. Data from aerial photographic surveys over the Greenland Sea whelping areas conducted by the Norwegian Institute of Marine Research are combined with sea ice information from satellite data to estimate the abundance of seal pups. Subsequently, total species abundance estimates are derived.