Leave Town or Hunker Down?

NRC versus EPA on Evacuation

An earlier version of this post had a bad error in fraction of Fukushima inventory released numbers in the Release Amount section. Mea maxima culpa.

The NRC and EPA have both formulated policy on how to handle a release of radioactive material from a nuclear power plant. The NRC urges rapid evacuation. The EPA argues for shelter in place combined with deliberate relocation. Both are wrong, but the EPA's position is far more sensible.

As usual, I assume some background including the meaning of the acronyms LNT and SNT.

The NRC Approach to Evacuation

In 2012, the NRC published the three volume results of its State of the Art Reactor Consequence Analysis (SOARCA).\cite{nrc-2012} SOARCA concluded that the consequences of a melt down are tolerable, although it would never use that adjective. Suppose there are 500,000 people living within 10 miles of the plant. SOARCA claims the worst case number of premature LNT public deaths will range from 30 to 100 depending on the scenario.

SOARCA came up with this conclusion despite using LNT and an unrealistic plume model, which resulted in highly concentrated plumes. The SOARCA result depends on:

1) We have 8 or more hours between loss of cooling and the first release, as happened at both TMI and Fukushima.

2) Almost all the inventory of radioactive material stays within the reactor containment, as happened at TMI, but not at Fukushima.

3) Quick and effective evacuation results in practically everybody being 10 miles or more from the plant at the first release. Nobody is allowed to return until dose rates are well below background on much of the planet.

Release Timing

SOARCA uses a detailed model of the casualty progression within the plant called MELCOR. According to MELCOR, STBSO core damage starts in 1 to 3 hours, reactor vessel failure in 8 hrs, and containment failure 8 hours BWR and 25 for PWR. At Fukushima, core damage probably started in about 4 hours.\cite{unscear-2013} Unit 1 Reactor Pressure Vessel failed in about 11 hours, when we started to see a rapid increase in dose rates at the Main Gate. The Unit 1 hydrogen explosion occurred almost exactly 24 hours after the blackout. Except for the H2 explosion's, the MELCOR timing is reasonably consistent with what happened at Fukushima.

At Three Mile Island, where there was a rapid, major dump of reactor coolant --- an event that SOARCA did not analyze --- the uncovered core was probably badly damaged within an hour of the first alarm. However, there was only a tiny bit of seepage to the atmosphere until hour 27 when there was a puff from the auxiliary building. Even then the dose rates at the plant boundary were background level numbers.

The takeaway is we will usually have about 8 hours of warning time between something that causes a loss of cooling and the first release.

Release Amount

Due to the hydrogen explosions, the Fukushima releases were much larger than the SOARCA MELCOR prediction. In almost all the SOARCA scenarios, the iodine and cesium releases was less than 2% of the inventory, in most cases less than 0.5%. Most of the iodine and cesium condensed out on plant surfaces or was captured in the one of the water streams. The TMI numbers were much smaller. At TMI, only 0.0002% of the iodine and no detectable cesium was released.\cite{rockwell-2004}[p 271] But at Fukushima, 2 to 8% of the I-131 was released and 1 to 3% of the Cs-137 in Units 1 to 3 were released, well above the MELCOR numbers.\cite{unscear-2021}[p 14] My takeaway: most future releases will emit a tiny to small fraction of the core inventory, but not all.

Modelling the Plume

SOARCA used the MACCS2 software to model the release plume. MACCS2 assumes a straight line Gaussian plume. In any run of this program, the wind direction and speed is constant both in time and space for the entire release, as is the atmospheric stability, mixing height, and dry deposition velocity. This is nonsense. In the real world, the wind is constantly shifting, as are the other weather variables. The deposition velocity depends on the local weather conditions at each point.

A far better approach is a puff model. These programs model a release as a series of puffs. The location of the center of each puff is tracked through time. At each time step, Gaussian diffusion around this point determines the concentration of each pollutant throughout the area. All the weather variables can be time and space varying.

Figure 1 shows how different the results can be. These figures track a hypothetical release from a Hanford processing plant, which started at 1994-01-02 at 0300, when the wind was westerly. The puff model used in this exercise was CALPUFF which is the EPA standard atmospheric pollution model. In the CALPUFF run, the wind changes on a hourly basis spreading the plume around. The MACCS2 run is forced to stick to the initial wind speed and direction resulting in an entirely different, far more concentrated plume.

Preposterously, for LNT this difference need not be important. Dilution gets you nothing. If the population distribution is uniform, the LNT harm will be the same in both the top and bottom of Figure 1. To first order, the LNT harm depends only on the size of the release, not how it is spread around. For a non-linear model such as SNT, how spread out the release is is all important. For SNT, spreading out the release by a factor of five will reduce the overall harm by better than a factor of 30.

In the SOARCA analyses, the authors tried to compensate for MACCS2's limitations by repeating the plume generation step about 1000 times, sampling from the locale's (one year) wind roses, atmospheric stability indices, and precipitation. Replacing one very unrealistic plume with 1000 such plumes is not much of an improvement, and for LNT it does not make much difference.

It is a mystery to me why NRC chose MACCS2 over CALPUFF, which does a far better job of approximating reality. EPA and NRC are clearly not on the same page here. But once again, as long as LNT is your harm model, you are already so far from reality, it does not make much difference.¹

Radiation Harm

Following bureaucratic dogma, SOARCA uses LNT as its radiation harm model. This combined with rapid, effective evacuation meant that the great bulk of the cancer risk was caused by the tiny dose rates (less than 5 mSv/year) experienced by evacuees returning to their home well after the release, and then being exposed to mostly groundshine radiation over an assumed 50 years. There are many places on this planet where the back ground dose rates are much higher than 5 mSv per year. In none of these areas can we detect any increase in cancer.

SOARCA does depart slightly from the LNT party line. In a sensitivity analysis, it tosses out dose rates below 6.25 mSv/yr, the US average including medical exposures. Since so much of the dose was acquired after return, this reduced the computed fatalities by a factor of 15 to 150 depending on scenario. This result did not make it into the Executive Summary.

SOARCA's final risk metric is very strange. It's the average probability of fatal cancer for someone who lived within 10 miles of the release. There is no distinction between someone living 1 mile from the plant and someone living 10 miles from the plant. I guess this makes sense to a true LNTer, who only cares about total collective dose. But it is not consistent with NRC's focus on the Maximally Exposed Person elsewhere. And for a non-LNTer, it makes no sense at all. Finally, there is no way you can come up with a reasonable compensation system with such a metric. Can the NRC do anything right?

Evacuation

The NRC is a firm believer in evacuation. After Fukushima, Chairman Jaczko said he would have evacuated out 50 miles, increasing the trauma by an order of magnitude. SOARCA spends a great deal of effort on modelling the evacuation in each of its scenarios in great detail. The basic assumption is that 99.5% of the population within ten miles evacuates to 30 miles away for the plant. In addition,there is a ``shadow" (voluntary) evacuation of 20% of the people in the 10 to 20 mile ring moving to 30 miles out.

SOARCA divides the 0 to 10 mile population into six groups: school kids, special facilities (hospitals, prisons, etc), the 0.5% who refuse to evacuate, an evacuating tail (last 10% to leave), and everybody else. The school kids are evacuated in 2.5 hours, the special facilities and tail in 6.25 hours, everybody else in 5.25 hours. In almost all the SOARCA scenarios, the evacuation is complete before the release starts. At that time, everybody except the refuseniks are 10 miles from the plant.

For the BWR plant SOARCA analysis, the evacuees were not allowed home until the dose rate was less than 5 mSv/y.² Tellingly, we are not told the exile times. Reading between the lines, not only was the evacuation quick and effective, but also the exile times were very long.

SOARCA assumes a 10 mile evacuation is recommended by NRC immediately upon loss of cooling and ordered immediately after NRC recommendation. Neither happened at Three Mile Island. SOARCA assumes a 10 mile evacuation will go smoothly. At Fukushima a 6, then 12 mile evacuation was a disaster.

But the aspect of SOARCA that blows my mind is that there is no mention of the costs of evacuation, nor the far higher costs of the multi-year exile times. SOARCA implicitly assumes they are negligible. In fact, essentially all the harm to the public at Three Mile Island and at Fukushima was due to evacuation and exile. The appendix on Fukushima says not a word on evacuation and exile costs, despite the fact they must have been aware of at least the 50 nearly immediate deaths of frail, elderly people from the uprooting. SOARCA also ignores the 106 people that died in the 2005 Gulf Coast hurricane evacuations.

In NRC's world view, making people poorer is unimportant. Global warming is not their business. Harm from alternate sources of electricity is not a consideration. Evacuation and exile costs can be disregarded. The only thing that counts is radiation harm, based on a harm model that defies biology. It is a preposterously blinkered outlook; and it is costing humanity dearly.

EPA and Evacuation

EPA's position on evacuation is laid out in its PAG Manual, Protective Action Guides and Planning Guidance for Radiological Incidents.\cite{epa-2017} Except for its dependence on LNT, the PAG manual is a pretty reasonable document. EPA divides the response into Early Phase, Intermediate Phase, and Late Phase.

Early Phase is while the release is on going and the plume has yet to move passed a point. During the Early Phase most of the dose is from plume inhalation. During the Early Phase. there is a great deal of uncertainty about the amount and temporal and spatial distribution of the dose rate. Intermediate Phase starts after the release has stopped and ends when clean up starts. During this phase, the amount and spatial distribution of the radionuclides are pretty well known. Most of the dose is from material deposited on the ground, known as groundshine. The Late Phase is when clean up starts. The split between Intermediate and Late strikes me as arbitrary and unnecessary.

The EPA Guidelines correctly put a great deal of emphasis on shelter-in-place. The manual has a nice drawing, Figure 2, showing the order of magnitude reduction in photon dose from staying indoors during the plume. It says ``shelter-in-place should be preferred to evacuation whenever it produces equal or greater protection".

Figure 2. EPA estimate of dose reduction from staying in doors during Early Phase. 10 means the indoor dose is one-tenth the outdoor dose.

For the early phase, the PAG manual recommends shelter-in-place or evacuation if the projected dose rate is greater than 10 mSv in 4 days. This is based on accepting an LNT probability of 0.0002 of mortal cancer from the dose. That's an LNT Lost Life Expectancy(LLE) of about 1 day. The manual also says when the projected dose is less than 10 mSv for the first four days, evacuation is NOT recommended, although shelter-in-place should be considered. The manual references the 106 evacuation deaths from the 2005 Gulf Coast hurricanes and the several thousand at Fukushima, which apparently changed EPA's view on the risks of evacuation.

For the Intermediate Phase, the guideline is relocation (aka exile), if the projected whole body dose is greater than 20 mSv in the first year or greater than 5 mSv in any subsequent year. The document distinguishes between evacuation and relocation and emphasizes that relocation should be done deliberately, not hurried. This shift in EPA policy makes the concept of an Emergency Planning Zone (EPZ) obsolete. The main purpose of the EPZ is to facilitate rapid evacuation, which the EPA no longer supports.

The problem with the EPA PAG is the relocation/exile guidelines. According to LNT, 20 mSv cumulative increases your probability of developing fatal cancer by 1 chance in 1000. The corresponding LLE is 4.2 days. Per Table 1, the EPA would have relocated most of Okuma, Futaba, and Namie, and parts of Tomioko and Iitate to avoid the first year dose. That's bad enough.

But the EPA's Guideline to relocate if the dose rate in the follow on years is more than 5 mSv per year, is far, far tighter, ridiculously so. By this logic, a wide swath of the US mountain west including Denver should be depopulated. According to EPA's own numbers, eight states average more than 5 mSv/y.\cite{mauro-2005} And expect some very long exiles. The Okuma hardest hit group cannot go home for 38 years.

This nonsense is a product of LNT's obsession with cumulative dose. Unless we adopt a dose-response model that recognizes our ability to repair radiation damage, a large release really is catastrophic in terms of dislocation cost. If we accept a dose-rate based model, the dislocation costs disappear, even in a Fukushima size release.

Nonetheless, EPA's position on evacuation is radically different from and far more reasonable than NRC's. Congress must decide which policy should be followed. At Three Mile Island and after Fukushima, it was the NRC policy that spoke for the feds. Fortunately, neither Governor Thornburgh nor the Japanese followed the NRC recommendations. We won't be as lucky next time.

Postscript on Masking Up

During the Early Phase, most of the dose results from plume inhalation. Both SOARCA and the EPA PAG talk about distributing potassium-iodide pills to reduce radioiodine uptake to the thyroid. SOARCA assumes this will cut the thyroid dose by 35%. But strangely neither consider masking up during the plume passage. The radioactive aerosols in a release range in diameter from about 0.5 to 20 microns or larger, with a peak around 2 um.\cite{nrc-2013}[pages 7-23, 7-28] This is a lot bigger than the Covid virus (about 0.1 micron), and in a range where N95 masks offer protection factors of 100 to 1000, Figure 3, virtually eliminating the inhalation harm. This would be valuable insurance against a much larger than expected release.

Figure 3. N95 mask protection as a function of aerosol size, reference \cite{cho-2010}}

So here's the Gordian Knot Group's response to a release:

1) Give a 20 pack of N95's to every household within 10 km of the plant. N95's cost less than a dollar apiece. Also issue a dosimeter to each such household. In bulk, a dosimeter should cost less than $50. For 50,000 households, this package would cost 3.5 million.

2) When a release threatens, tell everybody including the people in the buffer zone to go home and, if instructed by phone/TV, mask up. People can also check their dosimeter and decide for themselves. (Communications may be down.)

3) Periodically issue reports showing the dose rate pattern and advising which locations should mask up, which need not, and which need no longer shelter in place.

4) Monitor the dose rates through time throughout the area. Subsequently, estimate each individual's LLE via SNT and pay the corresponding compensation, assuming no evacuation and no masking.

People would be free to evacuate if they want, but their compensation would not change.

1. CALPUFF was originally developed for the California Air Resources Board in the 1980's. Over time it has acquired all sorts of marginally useful bells and whistles, such as attempts to model flow around buildings. It has a pretty complete pollutant chemistry module, tracking the development of smog, ozone, etc. CALPUFF has grown to over 100,000 lines of mostly archaic Fortran. But it has no radioactive decay capability? This could easily have been added.

2. For the runs in which they truncated the harm at 6.25 mSv/y, this results in the nonsensical situation in which people are not allowed to return home despite the fact that the harm in doing so is zero.