Highlights of 2016 – Seascapemodels

Papers I read that changed the way I thought in 2016

That isn’t causation, THIS is causation…

It’s not every day that someone discovers something as fundamental as a new type of causality. In 2012, Sugihara and colleagues did, well almost. They describe a new method for detecting non-linear causality in time-series. The up-shot is, causality might have been hiding in our time-series for years and we have never noticed it.

Basically causality in highly connected non-linear systems - read ecosystems - doesn’t look like a linear correlation

The impact of this work has probably been modulated somewhat by limits of available data - you need long and high resolution data-sets for their method.

Well they have followed up this year with a solution - you can leverage multivariate time-series to detect causality. Multivariate data drastically reduces the number of samples you need.

I recommend reading both these papers. They are pretty heavy, but it is worth the time, so give yourself a good few hours to really absorb the implications of their work.

** Include image: this isn’t causation (correlation graph), THIS is causation (graph of simplex map)

Community analysis with abundance models

Multivariate statistics have leapt forward and it has never been been a more exciting time to do stats on community data….

I was never that interested in community ecology, or at least the stats involved. You know all what I mean, all the stuff about 4th root transforms and dissimilarity indices. I found I never really knew how to interpret results once you abstracted a community to some sort of ‘distance’ measure.

Enter model-based methods.

Now we can infer species interactions, test for environmental associations and make predictions all without ever leaving the comfort of natural ecological units - counts of individuals.

There is a rapidly growing literature on this stuff, but I recommend any of the papers from David Warton’s group, which got me into model-based community analysis. Try his TREE review on joint modelling, Hui’s comparison of traditional vs model-based ordination and if you like trait analysis Brown’s (not me!) approach to trait*environment interactions.

Machines with intuition

Google research has built an algorithm that can beat the world’s top Go players.

If you don’t known Go it is a game a bit like checkers, but more complex. Go is widely considered a major challenge for machine learning, far harder than chess, because there are far to many possible moves for today’s computer simply to predict every possible outcome. A machine that can play Go well needs to have intuition. Somewhat scarily Google’s Alpha Go algorithm can be trained to develop intuition about how to win.

It makes me wonder how long before they automate the job of scientist… Several recent works (read this and this) have already automated (parts of) the job of statistician.

Measuring fish habitats

Ferrari habitat complexity paper

Do another post on talks I heard that changed things:

Megan’s SLR image of habitats near Sea level.

Resilience talk at ICRS - inspiring b/c no slides

Here is what I wrote about Emma Lee’s talk:

“A great talk from Emma Lee changed the way I think about science. I come from a very quantitative background, where the point of science is knowledge and improving the future for nature and humanity. But Lee convinced me that telling a story is a suitable goal in itself. People love to tell stories, so much so it is almost a basic human need. Science is the way to come closer to the truth in our thinking and in doing so provide an outlet for people to communicate authentic stories.”

Maria Gardener