Meeting report: Point Processes for Ecology June 2016

Much of the data we collect as ecologists can be plotted as dots on a map. These dots might represent the locations of individual trees or plants, or places where more mobile animals have been spotted. Spatial point pattern data like these are widely used to estimate abundance, map species distributions, and to understand movement behaviour and habitat preferences. Ecology has developed a suite of methods for analysing these point pattern data, often by aggregating the points together into counts, or considering the point locations as presences in presence-absence analyses (with workarounds to conjure up appropriate absence records).

Over roughly the same period, the statistics research community has developed a cohesive framework for considering point data: point process models (PPMs). Spatial PPMs consider the entire pattern of dots as arising from some spatial process. This process might be a continuous ‘intensity surface’ describing how many dots we would expected to see in a given area, or some mechanism by which the points interact with one another to generate the pattern.

IMG_20160702_101409.jpg

From the middle distance, Janine Illian (University of St. Andrews) recaps some of her recent PPM research

 

In the last few years, papers in the ecology literature have started to draw links to these PPM approaches, and apply them to ecological problems. The popular species distribution modelling (SDM) approach MaxEnt is equivalent to a PPM, and many other SDM approaches are similar to PPMs. Spatial capture recapture and density surface models have been developed which use point processes to estimate abundance from observations of individual locations. Point process models have also been applied to infer habitat preferences from telemetry data.

In June of this year, the Quantitative Ecology SIG sponsored a working group meeting to discuss recent advances in ecological PPMs and identify hurdles to the models being used more widely in ecology. The meeting; Point Processes for Ecology: State of the art and next steps (held in Seattle to coincide with the ISEC 2016 conference) brought together 20 experts in PPMs, analysing ecological point data, and developing statistical software. We wanted to unite people working on PPMs in different corners of ecology, so we deliberately invited researchers who hadn’t previously worked together. Judging by the co-authorship network below, I think we did a decent job of that.

Co-authorship network of workshop attendees – links between people indicated they had previously published papers together. Most of the workshop attendees hadn’t worked together before and were meeting one another for the first time.

coauthorship_network.png

Co-authorship network of workshop attendees. Links between people indicate they had previously published papers together. Most of the workshop attendees hadn’t worked together before and were meeting for the first time.

After kicking off the meeting with the usual introductions and presentations of our own PPM interests, we had an open discussion of what we thought is stopping ecologists from using these advances in their work. This brainstorming session led to a whole lot of messy scribbling on blackboards and, after the chalk dust had settled, we arrived at two broad areas that most needed work to advance ecological PPMs: increasing understanding, and developing tools.

Increasing Understanding

One of the major barriers to uptake of PPMs in ecology is a lack of familiarity with the PPM literature. Despite the high level of statistical literacy in ecology, and ecologists’ wide uptake of similar statistical approaches, most of the ecologists we had spoken with are not at all familiar with PPMs. Getting the message out there with ‘explainer’ papers in the ecological literature is one way of doing that. These sorts of papers have been hugely effective for communicating new methods in the world of species distribution modelling, and a recent paper (by some of the workshop attendees) has done the same to communicate PPM methods to SDM researchers.

We reckon there’s plenty of space for more PPM explainer papers in other areas of ecology, and we settled on a couple of key topics. These include: the benefits of interpreting and explicitly modelling species presence/absence data as arising abundances, rather than some loosely defined probability of presence (a fundamental principle of the PPM approach); ways in which different data types (such as counts, presence/absence surveys and camera trap data) can be combined into a single analysis using PPMs; and the methods that have been developed in the statistical literature for evaluating point pattern models and data.

Developing Tools

Understanding how PPMs work and why they are useful is a critical first step for helping ecologists use these methods. However new methods are useful if ecologists have the tools to apply them. In the case of PPMs, there’s some work to be done making software both available and usable for ecologists. Whilst there is already some great software out there for modelling of point patterns (such as the spatstat and INLA R packages), most of this is targeted at statisticians. Developing software that fits ecologists’ needs, and can be used to answer ecological questions would help us to overcome the technical barriers to PPM uptake.

Specifically, there’s a need to be able to fit PPMs to large, broad-scale ecology datasets efficiently, so a comparison of current and novel methods in that area would be useful. Providing a user-friendly interface to existing and new PPM methods in R (the lingua franca of quantitative ecology) would also be helpful. Linking up a package like that with the ‘explainer’ papers on data integration and model evaluation would ease the route into PPM modelling for ecologists.

IMG_20160703_142114.jpg

Just some of the scribbling on one of many blackboards.

After defining these broad areas for future work, we started fleshing and setting up some of these projects, and even broke some ground writing manuscripts. Hopefully it won’t be long before the co-authorship network becomes a lot more connected. As well as working on these projects, we’re also very keen to hold more events on ecological PPMs in the future; both more working groups and symposia that so that more people can join us. So keep an eye out for more PPM events and papers in the near future!

IMG_20160702_191007.jpg

The nearest we got to a group photo; relaxing in the sun at Gasworks Park, Seattle

Workshop 12th March: writing a field guide to predictive methods in ecology

Badger culls, fishing quotas, global warming and freakish weather. All things which have been in the news recently; and all things which require computer models to make predictions to inform decision making.

If we take the badger cull, there has been huge criticism over whether the cull will work. Targets indicated that Gloucestershire badger populations needed to be cut by 70%, otherwise TB could be spread more than if there was no cull. Apparently only 708 badgers were killed, an estimated 30% of the population (does this mean there are only 2360 badgers in Gloucestershire?).

Is it practically possible to provide more certainty about population numbers, effective cull percentages and even whether badger culls will work?

Perhaps using modern computational methods will help.

For example, fishing quotas (or at least the scientific targets, before the politicians become involved) largely rely on estimations of population size derived from accounting methods, developed many decades ago. If a fish species typically lives five years, and you estimate the number of each age group caught each year, then you can work backwards to estimate the population size five years ago. These methods, however, don’t consider biological effects such as competition and predation from other fish, because these mathematical parameters are hard to define (some methods do, and are based on repeated sampling of fish stomach contents to find out what they eat).

However, given most fish eat things which generally fit into their mouths, a computational approach, perhaps involving an agent-based model where fish move around randomly interacting with other fish, and consuming smaller ones, may give a good indication of predation rates?

Of course, scaling up an individual-based model to predict species interactions in the North Sea fish stocks is a massive task involving huge amounts of computing time, and even so, without an accurate knowledge of how fish behave, may prove much less accurate than the simple accounting methods currently used.

The key is, different predictive models do different things. They may be able to provide parameter estimates, but not predict population change; or they may provide good short-term predictions but not long-term; or excellent temporal predictions but with no spatial information.

So how do you know what to choose?

Workshop

That question is the focus of a workshop to take place at Microsoft Research (Cambridge) on Wednesday 12th March. We want scientists involved in predictive ecology to attend, and to give an honest critique of their methods. What are the advantages? What are the disadvantages and the limitations of these approaches? We’d also like the end users to attend, whether you are policy makers, conservationists or industry based. What do you need from predictive ecology? The aim of the workshop will be to develop an honest ‘field-guide’ to predictive methods in ecology. This will make it easier for people to identify suitable methods for particular problems, know the benefits of particular methods and their pitfalls, and make it much easier to know how such predictions stack up against end user requirements for these predictions.

For more information, or to register interest in attending please contact computational@britishecologicalsociety.org

link round up no. 1

Here’s a round up of some interesting things and job opportunities we found on the web recently:

Things:

Petr Keil wrote a great blog post showing how to fit the same ecological model using R’s glm function, numerical optimisation and MCMC

accidental aRt showcases the unexpectedly beautiful works of art that can arise when plotting in R goes awry. They’re on twitter at @accidental__aRt

There’s a huge online index of free programming books (including guides for R, Python and Matlab). The list is maintained in a github repository, so you can contribute if you know of any which aren’t on there!

This crossvalidated thread contains some great examples for understanding expectation-maximisation algorithms

Openrefine is an open-source tool for visualising, cleaning and transforming data

The Santa Fe Institute are running a Summer school and an online course on Complex Systems in 2014 

Jobs:

post-doc modelling sleeping sickness control at the Liverpool school of Tropical Medicine

software developer in ecological networks at the University of Liverpool

spatial ecologist at CEH Wallingford

post-doc in spatial/space-time analysis at Southampton

programmer at CBER at UCL

BES Computational Ecology SIG: the “something for all” SIG

Quantitative methods are evolving fast in ecology, way faster than any of us can keep up with. Many of us do not have the foundational training in mathematical, statistical or computational skills to pick these up easily. Otherwise we’d work for banks, obviously. One consequence is that many of us spend a lot of our time feeling frustrated by quantitative methods.

The Computational Ecology SIG exists to help members with the quantitative techniques that involve some form of computation (the vast majority of them). This includes data entry, storage and delivery, statistical analyses and modelling. Our priority is to enable the widest possible community of ecologists understand and use the best quantitative computational methods. We’ve not been clear about that over the last couple of years because we ourselves were not entirely sure how best to serve you. Therefore we resorted to doing what any self-respecting ecologist would have done and ran a set of experiments; a number of different events pitched at various audiences with different levels of quantitative expertise. The clear winner was to provide people with opportunities to learn how to implement quantitative methods well. Hence, our most popular events in the past year have been training courses on integrated population modelling, species distribution modelling, spatial analysis and good coding practices. At those, people clearly made the most of the opportunity and got on with the learning, discussions and asking challenging questions.

So moving forward we intend to do more training events, and build upon them: broadening out to an even wider community of people aiming to get started with the methods, wanting to understand how to do them well, or simply understand what they are all about. This year we have big plans. We’re going to expand on our online presence by setting up a website to serve you with useful updates, tutorials, guides, advice and blog posts on quantitative methods. We’re also going to provide you with an online Field Guide to Ecological Models; to tell you all those things about the different methods you never get told in any undergraduate ecology degree as well as those things you might have done. As well as expanding this presence we aim to continue to host training events and we’ll provide more information about those as they emerge (likely a software carpentry bootcamp, an ‘ecological models in conservation applications’ workshop and an event again with the International Biometric Society and Royal Statistical Society).

Our SIG does not have the largest membership but we could potentially serve the largest proportion of BES members: those aiming to make sense and use out of the quantitative methods. You don’t need to be a computer nerd to join (and NO it doesn’t help… that’s the point!)