In addition, nuclear accidents are not independent—the current nuclear fleet is close
to the 1980s fleet for more than three quarters of reactor is over 25 years old.
Moreover, the evolution of safety performances and standards makes heroic to
assume that safety is time-invariant.
A second reason is that assessing the risk of nuclear accidents exclusively on
data from past observations implicitly assumes that no other knowledge is available
on nuclear safety. It ignores all the works carried out over the past 50 years by
thousands of nuclear scientists and engineers on safety. This knowledge has partly
crystallized in PSAs. The first large-scale probabilistic assessment was carried out
in the US in the 1970s. It was led by Norman Rasmussen, then head of the nuclear
engineering department at MIT. PSAs have now been carried out on all nuclear
power plants (hereafter, NPP) in the US and many others worldwide. Similarly
reactor vendors carry out such studies for each reactor model while it is still in the
design stage. For instance, the calculated core meltdown frequency for the UK EPR
is 10
−6 per year and the core damage with early containment failure is estimated to
3.9 Â 10
−8 per year.
As for observed frequencies, assessing the risk of nuclear accident based
exclusively on PSAs would be unsound. The use of PSAs has strong limitations,
too. Firstly, they are not mainly designed to provide a final single number. They are
designed to detect exactly what may go wrong, to identify the weakest links in the
process and to understand the failures which most contribute to the risk of an
accident. Secondly, PSAs have a limited scope. They study known initiating events
such as seism or loss of coolant but not all the possible states of the world because
the list of all causes and failures is unknown. Thirdly, PSAs assumes perfect
compliance with safety standards and regulatory requirements. An implicit
assumption is that safety standards are enforced thanks to an independent, competent and powerful safety regulatory authority. All these limitations can explain in
part why PSAs figures are much lower than observed frequencies.
5
If we want to make progress in estimating probabilities of nuclear accident, we
have to use all the current available quantitative knowledge and therefore to
combine information from PSAs and observed accidents. Escobar-Rangel and
Lévêque have made such an attempt [4]. The issue addressed in their paper is to
compute the post Fukushima-Daiichi global probability of a core-meltdown.
Different models are used including a Poisson Exponentially Weighted Average
model to capture the idea that recent accidents are more informative than past ones
and to introduce some inertia in the safety performances of the fleet. This model
shows that the Fukushima Daiichi accident results in a huge increase in the probability of an accident. The arrival rate in 2011 is similar to the arrival rate computed
in 1980s. To put it another way, this catastrophe has increased the probability of an
5
For a detailled discussion on the discrepancy between observed frequencies and calculated
frequencies in PSA models see [3], chapter “Consequences of Severe Nuclear Accidents on Social
Regulations in Socio-Technical Organizations”.
The Economic Assessment of the Cost of Nuclear Accidents
83
to the 1980s fleet for more than three quarters of reactor is over 25 years old.
Moreover, the evolution of safety performances and standards makes heroic to
assume that safety is time-invariant.
A second reason is that assessing the risk of nuclear accidents exclusively on
data from past observations implicitly assumes that no other knowledge is available
on nuclear safety. It ignores all the works carried out over the past 50 years by
thousands of nuclear scientists and engineers on safety. This knowledge has partly
crystallized in PSAs. The first large-scale probabilistic assessment was carried out
in the US in the 1970s. It was led by Norman Rasmussen, then head of the nuclear
engineering department at MIT. PSAs have now been carried out on all nuclear
power plants (hereafter, NPP) in the US and many others worldwide. Similarly
reactor vendors carry out such studies for each reactor model while it is still in the
design stage. For instance, the calculated core meltdown frequency for the UK EPR
is 10
−6 per year and the core damage with early containment failure is estimated to
3.9 Â 10
−8 per year.
As for observed frequencies, assessing the risk of nuclear accident based
exclusively on PSAs would be unsound. The use of PSAs has strong limitations,
too. Firstly, they are not mainly designed to provide a final single number. They are
designed to detect exactly what may go wrong, to identify the weakest links in the
process and to understand the failures which most contribute to the risk of an
accident. Secondly, PSAs have a limited scope. They study known initiating events
such as seism or loss of coolant but not all the possible states of the world because
the list of all causes and failures is unknown. Thirdly, PSAs assumes perfect
compliance with safety standards and regulatory requirements. An implicit
assumption is that safety standards are enforced thanks to an independent, competent and powerful safety regulatory authority. All these limitations can explain in
part why PSAs figures are much lower than observed frequencies.
5
If we want to make progress in estimating probabilities of nuclear accident, we
have to use all the current available quantitative knowledge and therefore to
combine information from PSAs and observed accidents. Escobar-Rangel and
Lévêque have made such an attempt [4]. The issue addressed in their paper is to
compute the post Fukushima-Daiichi global probability of a core-meltdown.
Different models are used including a Poisson Exponentially Weighted Average
model to capture the idea that recent accidents are more informative than past ones
and to introduce some inertia in the safety performances of the fleet. This model
shows that the Fukushima Daiichi accident results in a huge increase in the probability of an accident. The arrival rate in 2011 is similar to the arrival rate computed
in 1980s. To put it another way, this catastrophe has increased the probability of an
5
For a detailled discussion on the discrepancy between observed frequencies and calculated
frequencies in PSA models see [3], chapter “Consequences of Severe Nuclear Accidents on Social
Regulations in Socio-Technical Organizations”.
The Economic Assessment of the Cost of Nuclear Accidents
83