Modelling biodiversity in the oceans for conservation and management

How are marine bioregions, regions of distinct marine communities, distributed in our oceans, and how should we allocate Marine Protected Areas (MPAs) to protect the full range of biodiversity into the future? Inferring biodiversity over large areas from point samples of marine species requires an understanding of how environmental conditions drive biology. Regression techniques are frequently used in marine ecology to find the relationships between environmental predictors and species responses and then infer responses from environmental conditions alone. However, modelling very large marine bioregions poses a challenge for regression techniques that require a single species response matrix as input. Regression techniques can handle a small amount of unknown data using approaches such as imputation or dropping rows, but each marine survey can only cover a small subset of species in the ocean, leaving all other species as unknowns. I am applying the Gradient Forest technique, which mines split parameters from random forest models to construct the relationship, to the problem of modelling marine biodiversity over large scales. I will discuss the challenges of grouping spatially-correlated, smoothly-varying data into discrete clusters, when the true number of clusters is unknown and may not even exist in a strict sense. I will also show some applications of these techniques to planning and evaluating an MPA network around Australia. I will also discuss further plans for the project including a global bioregionalisation and designing MPA networks that continue to represent all bioregions under climate change.