Automatic forecasting for non-Gaussian time series

Automatic forecasting algorithms are crucial tools for many modern-day business operations. Yet nearly all of the available algorithms, and their associated software packages, assume that the series can be modelled with Gaussian errors. This makes it hard for end-users to employ these models when they routinely deal with non-Gaussian series, such as counts of sales, compositional data that must sum to 1, or time series that present as proportions. This project will develop extremely efficient and adaptable forecasting algorithms that can work with a huge variety of real-world time series, filling a crucial gap in the available software landscape for industry-level forecasting purposes.

The student will be based at The University of Queensland within the School of Veterinary Science, primarily supervised by Dr. Nicholas Clark. The candidate will work with a diverse group of international researchers. Additional support will be given by partners at the Ecological Forecasting Initiative and the Spatial Epidemiology Laboratory, including assistance in computer-based data analysis, model building and scientific communication. The selected student will have the opportunity to work with all partners on this project but will be based at UQ.

This project will help develop the candidate’s skills in critical thinking, project management, data management and analysis, writing and communication. Expected applications of the project are incredibly diverse, meaning the student will be well prepared for a future career in research or with government and non-government land management, biosecurity or conservation agencies.

Forecasting species’ responses to change

Using models to anticipate the future is vital to mitigate the impacts of environmental change on ecosystems. Yet most ecological models are one-off attempts to predict what ecosystems might be like in many years or decades. This makes it hard for decision-makers to use these models. It also favours models that are not easily scrutinised and improved. This project aims to employ an iterative cycle to forecast how species occurrences and abundances will change over short timescales, and will analyse forecasts to improve models so that we can continue to learn. This represents a new way of thinking in ecology that, like weather forecasting, has the power to advance our understanding of ecological processes.

The student will be based at The University of Queensland within the School of Veterinary Science, primarily supervised by Dr. Nicholas Clark. The candidate will work with a diverse group of international researchers. Additional support will be given by partners at the Ecological Forecasting Initiative and the Spatial Epidemiology Laboratory, including assistance in computer-based data analysis, model building and scientific communication. The selected student will have the opportunity to work with all partners on this project but will be based at UQ.

This project will help develop the candidate’s skills in critical thinking, project management, data management and analysis, writing and communication. Expected applications of the project are incredibly diverse, meaning the student will be well prepared for a future career in research or with government and non-government land management, biosecurity or conservation agencies.