Ms Harriet Mason

VISUAL REPRESENTATIONS OF SPATIOTEMPORAL UNCERTAINTY IN MONITORING AUSTRALIA’S ENERGY NEEDS

Ms Harriet Mason

1st year, PhD Candidate, Faculty of Business and Economics, Monash University (Zema Scholar)

Research interests: Data visualisation, exploratory data analysis, visual explanations, Machine Learning

Pictured left to right: Professor Dianne Cook, Harriet Mason, and Dr Sarah Goodwin
Pictured left to right: Professor Dianne Cook, Harriet Mason, and Dr Sarah Goodwin

Harriet holds a Bachelor of Economics (Honours) in Business analytics from Monash University, and has teaching experience as an assistant for the Introduction to Machine Learning at the Faculty of Business and Economics, Monash University.

Harriet Mason’s work is focused on visualising uncertainty in high dimensional data, specifically for forecasted energy demand that has both spatial and temporal components. Visualisations that capture the uncertainty in these complicated relationships will support more informed decision making.

Harriet is co-supervised by Professor Dianne Cook and Senior Lecturer Dr Emi Tanka from the Department of Econometrics & Business Statistics; Dr Sarah Goodwin, a Lecturer with Monash's Data Visualisation and Immersive Analytics research group ( Faculty of Information Technology); and Dr Ursula Laa, a Researcher at the University of Natural Resources and Life Sciences in Vienna.

“When I transferred to Monash I did a Bachelor of Economics with a specialisation in econometrics and business statistics, and a major in mathematical statistics. Last year I extended that bachelors with an honours year in business analytics. My thesis was writing an R package that calculated scagnostics (scatter plot diagnostics). They are a method of finding a numerical summary of the shape of a scatter plot. While this topic was in the field of visualisation, it was very applied. I enjoyed it a lot but I wanted to change my topic slightly for my PhD, so when I saw the Zema Energy Studies Scholarship my interest was piqued. The opportunity to look further into visualising uncertainty is something I jumped at.”

Research aims
An important part of AEMO's operations is the ability to strategically plan for potential and developing changes. Short-term and long-term forecasts intrinsically incorporate both spatial and temporal data uncertainty, which need to be understood for analysts and operators
to make informed and sound decisions on a daily basis. Communicating uncertainty clearly and accurately to enable decision making is a well-known challenge in the data visualisation community. This project would aim to incorporate user-centred design approaches, including observational studies, interviews and user studies to ensure prototype designs are tailored to AEMO's operations and specifically at times of critical stress.

Statistical graphics are established for exploring and communicating uncertainty. Plots such as histograms and dot plots display distributions of numbers. Side-by-side boxplots can be used to compare key elements of distributions across groups. Scatterplots are used to explore the association between pairs of numerical variables. Variation not explained by other factors is quantified as error and used to compute confidence intervals, or credible intervals, for estimation and prediction. There are numerous methods of uncertainty visualisation in common use for communicating between statisticians or quantitative professionals, but their effectiveness has not been compared or shown to
be more or less effective, or assessed for communicating to broader audiences and decision-makers.

Temporal forecasting has established methods for quantifying uncertainties (prediction error), and conventions for plotting prediction intervals. Energy usage typically also has a spatial context. Some solutions for visualising uncertainty in spatial data have included dynamic graphics, such as flickering. This reminds us that with today’s technology we can make clever use of interactivity and dynamic graphics to communicate uncertainty. In addition, energy use is typically multivariate, with supply coming from a variety of sources, and with various covariates, such as available infrastructure, or even local weather patterns. The comparison of multiple variables adds another level of complexity on the problem. Assessing the effectiveness of visualisation methods for communication depends on the human visual system as well as the purpose and audience. Developing effective visualisation relies on user-centered design and statistical inference for data graphics, that use computational resampling methods, and the line-up protocol as the measuring instrument.

Understanding, monitoring and managing energy use are critical factors for a functional society, and in tackling the challenges of climate change. Further to this, as renewable energy sources are increased into the energy market new sources and types of uncertainty are introduced that impact operators to manage the market. This research aims to tackle the problem of communicating forecast uncertainty of the systems operations related to energy. It will build on the growing body of work to summarise current practices, identify gaps, develop new techniques, test the techniques to determine which are the most effective for different purposes and audiences, and deploy the best methods. Ultimately, the work aspires to evaluate uncertainty communication methods for forecasted energy demand, that support more informed decision making.