Measurement of Fukushima-related radioactive contamination in aquatic species.

The Great East Japan (Tohoku) Earthquake and consequent tsunami of March 11, 2011, caused immense damage and loss of life in Japan. It is estimated that over 15,000 people died and in excess of 2,500 persons are still unaccounted for (1, 2). The tsunami that struck the Fukushima Dai-ichi Nuclear Power Plant (NPP) led to breakdown of the reactor cooling systems, resulting in overheating of fuel in the three reactors that were operating at the time of the earthquake (3). Substantial radionuclide releases ensued, both to the atmosphere (most of which was deposited on the Pacific Ocean) and directly to sea (4, 5). Concentrations of Cs in seawater, which are an indicator of contamination of marine biota, were very high near the site of the accident for a few weeks afterward, but decreased to relatively low levels within a year or so (6) (Fig. 1, refs. 7, 8). The paper by Okamura et al. in PNAS (9) builds on these estimates (6), and focuses on levels of contamination with Cs and Cs in a number of aquatic species, fitting a Weibull model to estimate the distribution of contamination in the sample. The paper uses extended aquatic monitoring data taken over the period from April 2011 to March 2015. More importantly, it takes account, as it has to, of the fact that many measurements are now below the limit of detection. To get around this problem, the authors substituted belowdetection limit estimates with the value of the detection threshold. An additional problem was that parameter estimates produced by likelihood maximization did not converge; Okamura et al. (9) dealt with this issue by use of a two-stage fitting process. They first fitted a randomeffects model, aggregating across groups of species and prefectures. Having estimated the necessary model parameters from the random-effects model, and fixing at the associated mean the ecological dispersion parameter, ρ, that controls the time decay in the scale parameter for the Weibull distribution, they then reestimated the remaining model parameters. Both of these modeling methods are less than optimal. Fixing measured estimates at the threshold limit will lead to bias, both in the central estimates (mean) and in their uncertainty. The mean is likely to be upwardly biased by this method, and estimates of uncertainties are likely to be downwardly biased. The problems introduced by the two-stage fitting process used by Okamura et al. (9) are less serious: It does not take full account of the variability in the data, but it would not be expected to bias the resulting parameter estimates. Bayesian methods of dealing with missing data are a better way of addressing both issues, although their computational demands can be substantial (10). Okamura et al. (9) suggest that contamination levels of both Cs and Cs are low overall, with the probability of contamination at levels>100 Bq/kg very low in bothmarine and freshwater species. Contamination levels tend to decrease with distance from Fukushima, suggesting that “contamination of marine fish has rapidly dispersed even at the bottom of the sea” since the accident (9). Because the authors’ estimates of mean are Fig. 1. Measured concentrations of Cs in seawater near the Fukushima Dai-Ichi NPP. JST, Japan Standard Time. Reproduced from refs. 7 and 8.