Comparing distribution of harbour porpoise using Generalized Additive Models and hierarchical Bayesian models with Integrated Nested Laplace Approximation

Laura Dawn Williamson* (Corresponding Author), Beth Scott, Megan R. Laxton, Janine B. Illian, Victoria L.G. Todd, Peter I. Miller, Brookes Kate L.

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

2 Citations (Scopus)
3 Downloads (Pure)

Abstract

Species Distribution Models (SDMs) are used regularly to develop management strategies, but many modelling methods ignore the spatial nature of data. To address this, we compared fine-scale spatial distribution predictions of harbour porpoise (Phocoena phocoena) using empirical aerial-video-survey data collected along the east coast of Scotland in August and September 2010 and 2014. Incorporating environmental covariates that cover habitat preferences and prey proxies, we used a traditional (and commonly implemented) Generalized Additive Model (GAM), and two Hierarchical Bayesian Modelling (HBM) approaches using Integrated Nested Laplace Approximation (INLA) model-fitting methodology. One HBM-INLA modelled gridded space (similar to the GAM), and the other dealt more explicitly in continuous space using a Log-Gaussian Cox Process (LGCP).

Overall, predicted distributions in the three models were similar; however, HBMs had twice the level of certainty, showed much finer-scale patterns in porpoise distribution, and identified some areas of high relative density that were not apparent in the GAM. Spatial differences were due to how the two methods accounted for autocorrelation, spatial clustering of animals, and differences between modelling in discrete vs. continuous space; consequently, methods for spatial analyses likely depend on scale at which results, and certainty, are needed.

For large-scale analysis (>5–10 km resolution, e.g. initial impact assessment), there was little difference between results; however, insights into fine-scale (<1 km) distribution of porpoise from the HBM model using LGCP, while more computationally costly, offered potential benefits for refining conservation management or mitigation measures within offshore developments or protected areas.
Original languageEnglish
Article number110011
Number of pages9
JournalEcological Modelling
Volume470
Early online date5 May 2022
DOIs
Publication statusPublished - 1 Aug 2022

Bibliographical note

Acknowledgments
We thank colleagues at the University of Aberdeen, Moray First Marine, NERI, Hi-Def Aerial Surveying Ltd and Ravenair for essential support in the field, particularly Tim Barton, Bill Ruck, Rasmus Nielson and Dave Rutter. L.D.W. was supported by the Marine Alliance for Science and Technology for Scotland (MASTS), the University of Aberdeen and Marine Scotland Science. Collaboration between the University of Aberdeen and Marine Scotland was supported by the Marine Collaboration Research Forum (MarCRF). Digital aerial surveys in 2010 were funded by Moray Offshore Renewables Ltd and 2014 by Marine Scotland. Additional funding for analysis of the combined datasets was provided by Marine Scotland. Collaboration between the University of Aberdeen and Marine Scotland was supported by MarCRF.

Data Availability Statement

Supplementary materials
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.ecolmodel.2022.110011.

Keywords

  • Bayesian model
  • Generalized additive model (GAM)
  • Integrated nested laplace approximation (INLA)
  • Harbour porpoise
  • Species distribution model

Fingerprint

Dive into the research topics of 'Comparing distribution of harbour porpoise using Generalized Additive Models and hierarchical Bayesian models with Integrated Nested Laplace Approximation'. Together they form a unique fingerprint.

Cite this