Nuclear Power Is Not Safe

Should we subsidize uncompetitive old nuclear plants? Should we build (uncompetitive) new ones?

Most robotic vehicles used for cleanup at Chernobyl were quickly destroyed because radiation ruined their electronics, so human liquidators were sent back in. (Clay Gilliland, Wikimedia Commons)

One nuclear industry safety assessment figure claims risks are only one-tenth as great as the historic record shows they are. If the people in the industry know it, they are not admitting to a serious safety problem. If they don’t know it, they are simply ignorant of their industry’s record. Either way, I think it is clear that nuclear industry safety analyses should be regarded as untrustworthy.

To deal with the nuclear power industry, we have to understand that some numbers it produces are nearly always accurate, and some are nearly never so. The accurate numbers include the power rating of a plant, the amount of energy produced by the fission of an atom of a given isotope, the approximate percentages of specific isotopes among products of fission, and the amount of time it takes to refuel a plant, if everything goes normally.

There are other numbers that seem usually to be off by a factor of two. When the industry supplies the cost to build a new reactor and the amount of time it will take it, it is probably best to take those figures and double them. If a plant is projected to cost $6 billion and take six years to build, experience tells us that a safer estimate would be $12 billion and twelve years.

The really big problem, however, comes with statements about safety.

Whenever I think about this question, I recall an email I got from a nuclear power advocate, in which he appeared to gloat about how safe the industry is. Compared with the deaths associated with coal, he claimed, nuclear power produced almost none. He “proved” this by citing the explosion at Chernobyl, which he said killed 31 people. He compared this with the hundreds of thousands killed by the coal industry each year.

Most pro-nuclear advocates admit that in addition to those 31, who were mostly killed by concussion or flying debris, at least several dozen people died of radiation poisoning from the Chernobyl accident. One source at the United Nations put the number of additional cancer-related deaths in 2008 at 64 and rising. This is a far cry from the estimate in the Chernobyl Forum, also from the UN, which is that something like 4,000 early deaths could ultimately result. Greenpeace puts the number at many times that. Radical anti-nuclear activists and some European governments put the number at well over 100,000 and possibly as high as 200,000. While the numbers related to safety, they seem to be matters of opinion, rather than fact. The high numbers are well over a thousand times as great as the low numbers. (Wikipedia: Chernobyl Disaster: Deaths Due to Radiation)

Fortunately, safety can be assessed from other points of view, with specific, comparable safety-related numbers, and at least one of these is pretty much indisputable. Based on them, we can learn something about the accuracy of the nuclear industry’s other safety statements in general. I will focus on one number, which relates to the likelihood that a nuclear plant will suffer a “core damage event.” (Note that “core damage event” (CDE) is the term used for what people call “meltdown.” It includes events involving melted or deformed fuel, ranging from “partial meltdowns” to Chernobyl-scale events or worse, regardless of the amount of damage.)

Early on, when our oldest reactors were new, probabilistic risk assessments for nuclear plants indicated that a core damage frequency of one event in 10,000 reactor years could be expected. This means that in any given year, any given reactor has a one in 10,000 chance of melting down. More recently, that number was increased to one event in 20,000 reactor years, because newer reactors are thought to be safer. With some newer reactors, the numbers are said to be more like one event in 50,000 reactor years, and, in the case of the proposed small modular reactors, the claim is that some reactors cannot melt down at all.

The World Nuclear Association, in an article updated in February of 2019, says civil nuclear plants have run through approximately 17,000 reactor years, worldwide. According to the earliest risk analysis, which was valid for only the earliest reactors, that could have been expected to produce 1.7 CDEs. With most reactors having been built with designs that had much reduced risk of one in 20,000 years or more, we really should have expected one CDE, or possibly none.

There is another way to look at these figures. If the core damage frequency is one in 10,000, that means that for a reactor with a 40-year life span, the likelihood of the reactor melting down during its lifetime is forty in 10,000, one in 250, or 0.4%. If the reactor is designed to a core damage frequency of one in 20,000, then the likelihood of meltdown over its lifetime is one in 500, or 0.2%.

These numbers do not reflect what has happened in the real world. Having gone through 17,000 reactor years at civil reactors, we have experienced three meltdowns in Japan, all at Fukushima Daiichi; at least one meltdown in the Soviet Union, at Chernobyl (though given the Soviet inclination to cover things up, there might have been others); one in Scotland, at Chapelcross; two in France, both at Saint-Laurent, but on different occasions; one in Czechoslovakia, at Jaslovské Bohunice; and three meltdowns in the United States, one each at Three Mile Island (Pennsylvania), Fermi (Michigan), and SRE (California).

In other words, instead of the projected number of CDEs, which we might have expected to be one or fewer, there were at least eleven in the real world. Instead of having a frequency of one in 20,000 reactor years, or more, or even of one in 10,000 for the oldest plants in the world, the number was about one in 1,550. And we can calculate that the likelihood of a CDE in the lifetime of a given plant is certainly not 0.2% or even 0.4%. In the real world, it has proven to be about 2% for the time the reactors have served, which is, on average, about three-quarters of their service lives. To calculate for the full service life, divide that figure by three-quarters, and you get 2.66%. So based on experience, the likelihood that any randomly chosen nuclear plant will have melted down when its time is up is 2.66%, or one in about 37.6.

If you went before a group of citizens and told them, “The reactor we want to build here is really safe, there is only a one-in-forty chance that it will ever melt down,” how do you think you would be received?

This miscalculation can be attributed to a specific problem in the math used in risk analysis, and in my opinion it is serous enough to reduce its value of the risk analysis for nuclear plants to zero. In every case of a CDE, the event was beyond the design basis, meaning that there was no way to include it properly in the calculations. The causes of the events were not assessed properly in the calculations because they are unpredictable. The Three Mile Island meltdown happened because of human error. The Chernobyl disaster happened because of human error. How do you predict human error?

Actually, I believe all of the others depended on human error, as well. For example, I will claim that the Fukushima Disaster was due to human error. It was certainly beyond design basis, but in this case the industry seems to have exclaimed, “How could anyone have known?” Their excuse was that the event was unprecedented, that a tsunami of the magnitude that ruined the plant was unpredictable, arising from a record-breaking earthquake, also unpredictable.

But was it unpredictable? The Tohoku Earthquake of 2011 caused a 5.7 meter (18.7 feet) seawall at Fukushima to be overwhelmed by a tsunami whose waves were locally 14.5 meters (47.5 feet). While that is impressive, I would argue that it was not unpredictable. Along the coast in 2011, the highest wave was 40.5 meters (133 feet), enough to submerge a 10-story building easily. But on the same northeast coast of Japan, the Sanriku Earthquake of 1933 produced a tsunami with a maximum wave of 28.7 meters, and the Sanriku Earthquake of 1896 produced a tsunami of 38.2 meters.

When I think of these data, my mind goes to a meeting I can imagine between design engineers for the Fukushima Daiichi Nuclear Plant and their management. Someone asks how tall the seawall needs to be, and someone else answers with the question, “How tall can you make it within the budget?” Such is my imagination. I think the sea walls should have been at least six or eight meters higher, the Fukushima Disaster can be attributed to bad planning (possibly due to greed), and that can be correctly called “human error.”

The point is that the nuclear industry safety problems in question, serious safety problems that caused CDEs, developed more than an order of magnitude more often than they were expected. The simple reason is that certain problems are not correctly calculated in probabilistic risk assessments. These problems, of which human error is one, are not correctly calculated for a simple reason: they are impossible to predict, and so their probability cannot be used accurately in a calculation.

It seems clear that the greatest cause of the most destructive nuclear accidents is human error. The risk should not have been stated in the form of “one in 20,000.” It should have been stated as “one in 20,000, unless someone does something stupid.”

Question: What is the chance of someone doing something stupid?

Answer: Approximately 100%; given the current US administration, it is precisely 100%.

The relevant question here is not whether a stupid thing will happen, because it will, but whether it will cause a meltdown. Whatever the cause of failure is, the nuclear industry has to face the fact that until it addresses the failure of its risk analysis, its safety calculations are utterly without value.