Seminars

Current Seminars

On Wednesday 31st March 2021, 13:30 GMT

Speaker: Oscar Rodriguez de Rivera Ortega, School of Mathematics, Statistics and Actuarial Science, University of Kent

Title: Joint Species Distribution Modelling with HMSC: Community matters.

Abstract:
An ecological community is characterised as a group of species that are usually found together. Ecological communities can be animal or plant assemblages with similar habitat requirements and contain species which may influence each other or rely on similar processes in their environment. A community’s occurrence or presence is usually a result of underlying environmental conditions.

Typical community data include field observations on the occurrence or abundance of species in a set of temporal and/or spatial replicates. Community data on species can be measured in many kinds of units, such as presence/absence or abundance. Community data are usually accompanied by environmental data consisting of a set of covariates that ecologist hypothesises to be important in explaining community composition.

Since the introduction of species distribution modelling package BIOCLIM, SDMs have become highly popular, not only in community ecology but also in geography, conservation and wildlife management (also there are some “braves” applying these methods to define pandemic distribution… but that is another story and shall be told another time). SDM frameworks are classified as single-species distribution models (SDMs), that model each species separately, and joint species distribution models (JSDMs), that model all species at the same time, focused directly for communities comprising many species.

There are two strategies to analyse community data using JSDM. These two ways are known as the “predict first, assemble later” (SSDM) and the “assemble and predict together” (JSDM). Hierarchical modelling of species communities (HMSC) belongs to the second strategy.

HMSC is a multivariate hierarchical generalised linear mixed model fitted with Bayesian inference. This is a very general and widely applied statistical framework and thus the novelty of HMSC is not in the statistical framework itself but in how the framework is applied to combine information from many types of data to infer community assembly processes.

In this talk I will show with examples how Hierarchical modelling of species communities (HMSC) is a useful and interesting tool as a Joint Species Distribution Model.

For joining details for this Zoom meeting please contact either:
Nicole Augustin, Nicole.Augustin@ed.ac.uk  or Oscar Rodriguez de Rivera Ortega, o.ortega@kent.ac.uk

Past Seminars

On Wednesday 3rd March 2021, 13:30 GMT

Speaker: Ben Swallow, School of Mathematics and Statistics, University of Glasgow

Title: Parallel tempering as a mechanism for facilitating inference in hierarchical hidden Markov models

Abstract: The study of animal behavioural states inferred through hidden Markov models and similar state switching models has seen a significant increase in popularity in recent years. The ability to account for varying levels of behavioural scale has become possible through hierarchical hidden Markov models, but additional levels lead to higher complexity and increased correlation between model components. Bayesian approaches being less favoured due to computational demands. Given these demands, it is vital that efficient estimation algorithms are developed when Bayesian methods are preferred. We study the use of various approaches to improve convergence times and mixing in Markov chain Monte Carlo methods applied to hierarchical hidden Markov models, including parallel tempering as an inference facilitation mechanism. The method shows promise for analysing complex stochastic models, especially with multimodal distributions or with high levels of correlation between components, but our initial results show that it requires careful tuning in order to maximise that potential.

Wednesday 20th January 2021, 13:30 GMT

Speaker: Vianey Leos Barajas (University of Toronto).

Title: Learning through failing with hidden Markov models
Abstract:
The hidden Markov model (HMM) is a ubiquitous modeling framework in statistical ecology. In this talk, I will go back to the basics of applying HMMs to ecological data by taking a journey through the computational and inferential challenges of fitting mis-specified HMMs in practice. I will also detail the practical consequences that arises from these challenges in terms of model building and model interpretation. By taking a closer look at an established modeling framework in statistical ecology, we will find that there’s always more to learn and discover!

Wednesday 16th December 2020, 13.00

Stylianos Taxidis (Data scientist)

Title: The data scientists process and how can data science help in the field of climate change.
Abstract:
Data science and AI have lately become commonly used words to describe the progress of the field in analysing and creating new value of our digital world. How can we ensure a successful and scientifically rigorous process while remaining goal and time focused to resolved the critical problems of our time. What can data science offer to help with the resolution of environmental issues and more specifically climate change. What are the techniques we can employ from NLP, classical machine learning and deep learning to measure, monitor, quantify the impact of climate change but also suggest solutions becoming an action aggregator and accelerator. This will be a presentation of a holistic review on techniques that have been utilised in other industries and can provide value in the data to information to environmental action lifecycle. The audience will learn about the stages of a data science process and how they can be applied in the environmental domain. The presentation also aims to close highlighting both risks and opportunities of this application

On Wednesday 25th November 2020, 13:00

Speaker: Megan Laxton (University of Glasgow).

Title: Taking the Observation Process into Account: Modelling Species Distributions and Abundance in Space using R-inlabru

Abstract:
Often in ecology, it is impossible to fully observe the study domain, and the uniformity of sampling methodologies is limited by real-world constraints. The spatial distribution of resulting data is therefore produced by a combination of the underlying ecological process and the method of observation. In species distribution modelling, failing to take the observation process into account could lead to spurious significance in the interpretation of covariate effects, and a generally poorer understanding of population spatial structure. The R-package inlabru was created for realistically complex ecological data, so is equipped to handle spatially varying detection probabilities; helping to prevent misled inference and inaccurate predictions. In this talk, I will discuss some ways in which complex observation processes can be accounted for within inlabru, to more accurately represent input data in ecological modelling. This will include examples such as orangutan nest distribution in Borneo, Eurasian crane population spread in the UK, and estimating abundance of wildebeest in the Serengeti.
On Wednesday 14. October at 13:00

Speaker: Emiko Dupont (University of Bath)
Title: Spatial+: a novel approach to spatial confounding
Abstract:
In spatial regression models, collinearity between covariates and spatial effects can lead to significant bias in effect estimates. We encounter this problem, known as spatial confounding, modelling forestry data to assess the effect of temperature on tree health. Reliable inference is difficult as results depend on whether or not we include spatial effects in the model. The mechanism behind spatial confounding is poorly understood and methods for dealing with it are limited. We propose a novel method, spatial+, a modified version of the spatial model. Using a thin plate spline model formulation, we are able to analyse the asymptotic behaviour of effect estimates to provide a theoretical explanation for why spatial confounding happens and why spatial+ works. These results are also demonstrated in a simulation study. Spatial+ is straight-forward to implement using existing software and standard model selection criteria can be used for comparing models. Another advantage of the method is that it extends to models with non-Gaussian response distributions. Finally, while our results are derived in a thin plate spline setting, the spatial+ methodology transfers naturally to other spatial model formulations.

For joining details please contact either:

Nicole Augustin, Nicole.Augustin@ed.ac.uk  or Oscar Rodriguez de Rivera Ortega, o.ortega@kent.ac.uk