University of Tasmania, Australia

Species distribution modelling in the Southern Ocean

Level: PhD

Techniques for spatial modelling of species and community distributions continue to evolve, with relatively recent methods such as boosted regression trees and generalised dissimilarity modelling. The intent of such modelling is to take species observations made at discrete points or along transects, and to model the dependence of these observations on environmental predictor variables. The model can then be used to predict the species occurrence over a wider spatial region. This project would investigate various aspects of such modelling techniques that are particularly relevant to their application to Southern Ocean biota, such as:

a. Preprocessing of predictor variables. Species habitat modelling in the Southern Ocean often involves the use of synoptic remote-sensed or modelled predictors (e.g. SST, sea ice concentration, sea surface height & anomalies). The biological relevance of such predictors, and hence the subsequent model performance, can be improved by preprocessing and feature extraction. A simple example might be the calculation of time since melt from sea ice concentration data, which could potentially be more relevant than the sea ice concentration information itself. More complex examples effectively become modelling processes themselves - for example, combining melt information from sea ice concentration data with surface current data to estimate the spatial movement of melt water. However, there is often little guidance on the most appropriate preprocessing for a particular application.

b. Correlated predictors. Predictors in Southern Ocean models are often highly correlated (many with a strong latitudinal structure). This correlation has implications for model fitting (e.g. preventing overfitting) and model interpretation.

c. Spatial scale of interest. Southern Ocean modelling may be driven by a desire to investigate patterns of relatively large spatial scale (e.g. ocean-wide distribution patterns), and yet the data display high variability at small spatial scales. Work in this area could address methods for fitting models to such data, and for assessing model fit with respect to the spatial scales of interest.

d. Methods for handling uncertainty in predictors.

e. Integration of different biological data. Habitat modelling approaches typically use a single, reasonably consistent set of biological observations as the dependent variables. However, the ability to integrate disparate data sets into a unified model would be a valuable step forward in the data-poor Southern Ocean. An example might be the integration of animal tracking data with ship-board observations or samples.

f. Integration of habitat modelling techniques (i.e. prediction of species distributions based on environmental layers) with dynamic models that incorporate spatial movement (due to ocean currents or sea ice movement) and/or population dynamics e.g. Improving methods for prediction under climate change scenarios. Given a scenario of changes in environmental parameters (e.g. from IPCC models), is it reasonable to assume that environmental changes will be the dominant driver of species distribution changes, or should we also consider the potential effects of biological interactions (e.g. competition, predator-prey relationships).

The project would focus on specific modelling situations (e.g. particular taxa), while also considering the issues in a more general sense (broader frameworks for solutions, rather than specific solutions). Existing data from the Australian Antarctic Data Centre, SCAR-MarBIN, and other sources would be used.

Nominal supervisor: Ben Raymond