Cabot Institute blog

Find out more about us at www.bristol.ac.uk/cabot

Thursday, 2 February 2012

Learning lessons from Fukushima

When disasters happen scientists pretty much have a duty to try to understand what happened and why, and to try to learn the lessons. This week the catastophist Gordon Woo of Risk Management Solutions gave a seminar here at the Cabot Institute and suggested that the question that we should really ask is not "why did this happen?" but "why did this not happen before?". This is also one of the ideas that emerged from a recent exercise that we undertook to try to understand the recent events at the Fukushima nuclear power plant in Japan. The range of skills available within Cabot allowed us to take a fundamentally holistic approach to the analysis that wouldn't have been possible for any single individual. The results of the analysis are here, but two main points emerge.

First, there is the need to tackle is "chained" or "cascaded" hazards, which, as very low probability events, have traditionally been treated as independent random events and hence have too low a likelihood of coinciding together. There may be hidden dependencies, which are not always either obvious or intuitive, requiring careful analysis to tease out or recognise. This is particularly the case for complex infrastructure like nuclear power stations.

Second, it is no longer adequate to rely on deterministic assessments of hazards and risks from natural hazards as these cannot account properly for uncertainty. Dealing with uncertainty requires a probabilistic analysis that looks at the full range of possible situations that may arise, not just a single one that a company or regulator has (perhaps somewhat arbitrarily) decided is the 'worst case'. Probabilistic approaches should now be regarded as mandatory, and application of rigorous, structured approaches to assessing risk are needed. Such assessments must include evaluation of all credible alternative models for natural processes, rather than just adopting particular models that happen to support inherited views.

No comments:

Post a Comment