Improving our understanding of rare but extreme events that impact critical infrastructure

Improving our understanding of rare and extreme hazards is key to protecting critical infrastructure such as energy systems. Although such scenarios are very unlikely, their occurrence could have major consequences. As a result, energy companies must invest resources in safeguarding these systems and strengthening their resilience.

Data limitations on climatic hazards

A statistician at King’s College London and expert in modelling hazardous events, Dr Claudia Neves, has used her unique knowledge of data-driven methods to develop new statistical tools capable of predicting the unpredictable – all within the confines of limited observational data. Following up from a former PhD project with EDF R&D UK, a further collaborative R&D project was set up, for Claudia to explore and validate current statistical approaches to estimate the magnitude of rare events.

EDF R&D UK, said “When it comes to designing energy infrastructure, consideration of the nature and frequency of extreme hazards is an integral part of the design process from the very beginning. This helps ensure that our infrastructure is built and operated with an appropriate understanding of low-likelihood high-impact events."

“Climate change adds further urgency and uncertainty to the characterisation of such events. Alongside established practices and continuous development, it is also important to engage with emerging research approaches to broaden our understanding, particularly where observational data are limited.”

The challenge for statisticians like Claudia, is that there just haven’t been that many extreme incidents in the past to help inform the future. Most datasets stretch back less than a century, raising fundamental questions about how to develop and apply statistical tools under data-limited conditions. The answer lies in Extreme Value theory.

A new approach to statistical modelling for the unpredictable

Through this project with EDF, Claudia turned to a particular suite of methods to quantify different types of extremes in a probabilistic way. It was innovative, blue-skies research, that had never been done before, an important contribution to advancing significantly both theoretical and applied statistics for extreme events. The first step was trying to harness the bias in the data that was coming from incumbent estimation methods in a systematic way.

Claudia said, “natural hazards characterisation often uses well-established statistical models to estimate how rare an event is - like a "100-year flood". These prescribed extreme value models work well for typical risks, but when we need to extrapolate to a very rare event which does not exist in historical records (like that with an occurrence probability of 0.0001, also known as the once in a 10,000 event), these models often fall apart. Sometimes, they even lead to estimates that are so high that they exceed physically plausible levels, which hinders confidence amongst practitioners.”

Claudia turned to a far less-explored area: non-parametric extreme value statistics. This is an advanced field of Statistics that deals with evaluating rare, extreme events without relying on assumptions about the specific probability distribution generating the data. This makes sense because real-world phenomena are not really gaussian or generalised Pareto (the archetypical distributions for averages or extremes, respectively).

Claudia said, “By avoiding rigid distributional assumptions, non-parametric methods aim to provide more robust and trustworthy estimates for these extremely rare events, including by naturally respecting the data's boundaries and underlying physical constraints.”

Using her methods, she carried out a case study based on real data recorded by the Met Office across several gauging stations in England, generating new insight into the frequency and extremity of rainfall events around key infrastructure sites. The findings showed, as expected, that different statistical approaches lead to substantially different quantification of very rare events.

This work forms part of a broader and ongoing effort within EDF R&D UK to explore various statistical approaches for assessing natural hazards.

EDF R&D UK said, “Tools like those proposed by Claudia are an example of how collaborative research between industry and academia can further advance understanding of rare and extreme events and the consideration of climate change.

“Such exploratory studies help inform our understanding and enable careful evaluation and robust analysis, before industry practices are further developed or evolved.”

A statistical model for safeguarding critical infrastructure against climate change

Whilst Claudia’s work draws primarily from critical infrastructure contexts, the underlying methodological insights are relevant to a wide range of sectors where understanding low-likelihood, high-impact events is important. Such insights can inform a range of infrastructure systems such as renewable energy, transport or the built environment.

EDF R&D UK said, “Engaging with academic research allows us to explore different perspectives and emerging ideas, which is particularly valuable at the R&D stage. This in turn helps maintain a strong emphasis on methodological rigour within industry."