We are delighted to announce David Wilkinson as the winner of the Robert May Prize 2021, awarded to the best paper by an early career researcher in the 2021 volume of Methods in Ecology and Evolution.

In this interview, David shares insights on his winning article ‘Defining and evaluating predictions of joint species distribution models’. Congratulations to all the shortlisted authors, whose articles you can read in our Robert May Prize 2021 Virtual Issue.

Tell us your career stage, what you work on, your hobbies and interests

I’m a Research Fellow at the University of Melbourne and I’m nearly two years post-PhD. I sit across two research labs at the moment: Quantitative & Applied Ecology Group (QAECO) and MetaMelb. I conduct my ecology research in QAECO where I work on applying species distribution models in a variety of contexts, most recently in response to the Australian bushfires of 2019/20 and for arctic marine forests. In MetaMelb I lead the Data Management and Analysis team for the repliCATS project where we crowd source evaluations of the credibility of published science. Outside of work I enjoy reading a lot of fantasy and science fiction, playing board games, and building things. I’ve been building and painting model kits for nearly twenty years, and over the past few years I’ve been getting into woodworking to build my own furniture.

How would you pitch your article to someone if you had just 30 seconds in an elevator with them?

In the wild species exist in communities and there are a lot of benefits to modelling them as such. When we do so it opens up several options for how we can predict species distributions and community assemblages.

Where did the idea to develop this method come from?

My PhD became a deep dive into joint species distribution models (JSDM) and what we could do with them. We started by comparing different JSDM implementations in a standardised framework so that we could identify where they were the same and how they differed. Single species models are used for prediction all the time, so it was a natural progression to explore how this could be done in the multi-species context.

What were the major challenges in developing this method? How did you overcome this?

The two biggest challenges here were my lack of a statistical background and just how computationally intensive these methods can be on larger datasets. I hadn’t calculated an integral since high school, so this meant that I had to do a lot of reading to get my head calculating integrals in multivariate space. Drawing physical representations of the probability distributions was a huge help in being able to understand what the different prediction types were doing. To be able to apply these methods to larger datasets I had to learn how to use a supercomputer, and that has arguably been one of the most useful skills that I developed throughout my PhD.

How do you plan to apply the method you published/what have you been working on since its publication?

We’ve been working on a comparison of the predictive performance of different JSDMs using the different prediction types we defined in this paper across several real and simulated datasets. We’re getting close to finalising the manuscript for submission, so you’ll hopefully be hearing from us very soon…

Who will benefit from your method (researchers, species, habitats, community groups)?

Anyone working with species or communities can benefit from this method. You might need multi-species data to fit a JSDM, but you can generate predictions at the species or community level so they can be used for a variety of conservation applications. These predictions will be useful to researchers, land managers, policy makers, and more.

If you could travel back in time, would you add to or change anything about your method?

I’d definitely like to improve the computational performance of these methods. For larger datasets it can take days or weeks even on a supercomputer to generate the predictions and it will definitely be a barrier to the method’s wider uptake for high dimensional datasets.

You can read David’s full paper here.

Read the full Robert May Prize 2021 Virtual Issue here.