Open Access Research Article

Improving the Predictive Power of Marine Species Distribution Models by Incorporation of Modelled Prey Data

Lars O Mortensen1*, Jonas B Mortensen1, Ingrid Ellingsen2, Mathias Singsaas Frøseth1 and Henrik Skov1

1DHI A/S, Denmark

2SINTEF Ocean, Norway

Corresponding Author

Received Date: December 10, 2021;  Published Date: February 01, 2022


Species distribution modelling (SDM) often relies on hydrodynamic processes as proxy for prey items, when estimating animal distributions at sea. This is due to the fragmented nature of prey data available, which makes the data ill-suited for SDM. However, the incorporation of prey data remains an unresolved potential with the outcome of studies so far being ambiguous. To assess the potential effect of including modelled data on prey density in marine species distribution models developed solely on dynamic variables we used a dynamic modelling framework MARAMBS for prediction of the fine-scale densities and movements of seabirds in the Barents Sea. We tested the effect of incorporating synoptic modelled data on calanoid zooplankton from the SINTEF SINMOD model into this SDM framework for the planktivorous Little Auk (Alle alle). Copepods constitute the main prey of the little auk which breeds in the northern parts of the Barents Sea and makes seasonal migrations out of the region to/from wintering grounds further south off the Norwegian coast.

The approach consisted of a three-step process. Hydrodynamic variables were derived from a 3d oceanographic model developed, while prey information was derived from the SINMOD model. The derived information was coupled with observation data on the Little Auk in a generalized additive model (GAM) to predict the spatio-temporal distribution of the auk. Predictions were made in three-hour time steps, yielding a spatial timeseries on species density across the study area. GAMs were conducted with and without prey information to estimate model improvements. Furthermore, predicted species densities from GAMs were converted into spatio-temporal habitat suitability index (HIS), which served as a prey information for an Agent Based Model (ABM), simulating the migratory patterns of the Little Auk in the Barents Sea. Simulations were subsequently performed with initial HSI (no prey), prey HSI (HSI with prey information) and a simulation with baseline HSI and prey information separate but parallel. Simulations were validated against observations using Goodman and Kruskal’s gamma.

Results showed that both GAM’s and ABM performed significantly better with the inclusion of prey information. The most significant improvement was seen when incorporating prey HSI, where the prey information was included in the GAM and used to drive predicted movements of Little Auks by the GAM. Thus, the results show that the addition of a prey item as a predictor variable is highly likely to improve SDM predictions in comparison with existing models driven alone by oceanographic variables. At the same time, replacing oceanographic predictors with the prey information resulted in poorer model performance indicating inefficient detection of prey patches by the seabirds.

Signup for Newsletter
Scroll to Top