Project and Master Theses for Bob O'Hara
Most of my work is with ecological data, both developing method to fit ecological models to the data, and analysing the data. We generally used Bayesian approaches, with extensions of GLMs and mixed models.
A particular focus has been on methods to integrate different data sets into a single model of a species' distribution. We have a set of analyses that we have run, but there are several clear ways to improve the models. This means we have projects to develop better models for species (e.g. for all Norwegian birds), or to develop the methods to improve the modelling process.
- Observation Biases To model where species are based on where they are seen, we have to have a model for where people go to see species. Some models have been developed but we currently do not understand how well these models work for real data. This project could look at different approaches to modelling observation effort (covariates, random fields) and how well each works: is one better, or are there trade-offs?
- Spatio-temporal Models Our current models ignore time, but this is important in practice. We are starting to develop spatio-temporal models, and this project could either focus on (1) applying the models to Norwegian species, (2) developing the computational tools to fit the models quickly, or (3) explore model development (e.g. how much data do we need?).
- Data Quality A long-running question with integrated models is how much of each sort of data do we need? Surveys are better quality but expensive, so can we replace them with citizen science observations? Can we optimise where we should survey, or highlight regions that need more observations? We can use ideas from Value of Information theory to ask where we should try to collect data (or, if we have data, what would be the cost of ignoring it).