Precipitation Redistribution Method for Regional Simulations of Radioactive Material Transport During the Fukushima Daiichi Nuclear Power Plant Accident
Abstract
To reproduce more accurate deposition maps of radioactive materials released in the Fukushima Daiichi Nuclear Power Plant accident in March 2011, our study focuses on the uncertainty of atmospheric transport simulations caused by precipitation. In our new method, simulated wet deposition distribution of 137Cs is modified by high-resolution radar rain gauge data of observed precipitation. Sensitivity experiments are conducted to examine the impact of using both different reanalyzed meteorological data sets as boundary condition and the observation data of precipitation as the redistribution of simulated precipitation. Among the results, the experiment modified by high-resolution radar rain gauge data realized the most accurate cumulative 137Cs deposition from 18 to 27 March. While the meteorological field is reasonably simulated in the atmospheric transport model in the experiments, the results showed that applying observed precipitation also contributes to improve the accuracy of simulated wet deposition amount.
Key Points
- Wet deposition of 137Cs simulated by the regional isotope tracer model is redistributed in correspondence with observed precipitation
- To reproduce wet deposition, both high-resolution boundary condition in the model and highly accurate simulated precipitation are necessary
- In our results, validation is evaluated differently depending on the time boundary of the measurement of 137Cs fallouts
1 Introduction
During the Fukushima Daiichi Nuclear Power Plant (FDNPP) accident caused by the great East Japan earthquake and tsunami in March 2011, large amounts of radioactive materials were released and spread over the eastern side of Japan (Nuclear Emergency Response Headquarters Government of Japan, 2011). During March and April 2011, the estimated emissions of 137Cs were approximately 9–37 PBq (Science Council of Japan, 2014). The released radioactive materials were transported by atmospheric advection and diffusion and were deposited over the ocean and on the land directly and by precipitation (Yoshida & Kanda, 2012). From 12 to 23 March, air concentrations of 137Cs exceeding 10 Bq/m3 from FDNPP were often observed at the monitoring sites in the Fukushima prefecture and Kanto region, including those in the metropolitan and southern areas (Tsuruta et al., 2014). On 21 March, the observed concentration of 137Cs rose to 309 Bq/m3 at the monitoring site at Tsukuba in the Kanto region (Tsuruta et al., 2014). Some of the deposited radioactive materials on land in the Fukushima prefecture and other nearby prefectures were transported into environmental waters, and radioactive iodine and cesium were monitored in tap water, although their concentrations decreased below detection levels beginning in April 2011 (Ikemoto & Magara, 2011). It is important to investigate the precise distributions of deposited radioactive materials to know how the accident affected citizen's health and the environment.
Atmospheric transport simulations have been conducted to investigate the spatiotemporal deposition distributions of radioactive materials in the early phase of emissions partly because of insufficient monitoring data for radionuclides (Tsuruta et al., 2014; Yasunari et al., 2011). According to the previous studies, the formation process of the highly concentrated distribution of 137Cs in the Kanto region from 20 March through 23 March, hereafter called the hot spot, was such that the radioactive plume was transported from the FDNPP to the south by a northerly wind, passed through the metropolitan area via a northeasterly wind, and was deposited by precipitation (Kinoshita et al., 2011; Morino et al., 2013; Srinivas et al., 2012; Terada et al., 2012). In addition, the more precise process was investigated by Yoshikane et al. (2016). They found that the northeasterly wind was associated with a mesoscale low-pressure system, and they concluded that the transportation of the radioactive materials was related to the diurnal regional climate cycle in this area (Yoshikane et al., 2016). However, the numerical calculations using atmospheric transport models had uncertainties that depended on each model's scheme and data set.
A major source of uncertainty is found in issues of source amount, location, and timing. In previous studies, the emissions were estimated using the inverse method with combined atmospheric dispersion model results and observation data of radionuclides (Chino et al., 2011). These estimated values were updated, and the emission rate in the early phase of the accident was estimated using additional monitoring data (Katata et al., 2012; Terada et al., 2012). The estimate of Terada et al. (2012) was validated using sea surface concentration data of 134Cs and was improved (Kobayashi et al., 2013). In addition, Stohl et al. (2012) used fuel inventories as the basic emission scenario and estimated the amounts released using an atmospheric dispersion model and measurements of air concentration and deposition (Stohl et al., 2012). Another study used gamma dose rate monitoring for estimation (Saunier et al., 2013). Although several groups have studied source term estimation as noted above, the best method has yet to be decided.
Representations of the modeled physical processes, particularly those related to dry and wet deposition, are also regarded as major sources of uncertainty. Concentrations of radioactive materials contained in precipitation are highly dependent on those processes. In our previous study, a set of sensitivity tests for the parameters in dry and wet deposition processes, that is, dry depositional speed and washout coefficient, was conducted. It could be concluded from the results that the washout coefficient in the model formulation of the wet deposition process clearly impacted the wet deposition concentrations and their spatial and temporal distributions, whereas changes in that parameter in the dry deposition process did not influence the dry deposition as much as for wet deposition (Saya et al., 2015).
Another major source of uncertainty derives from meteorological conditions. Because the movement of a radioactive plume is dominated by advection-diffusion processes, it is essential that meteorological conditions are accurately simulated. In an earlier study, various atmospheric dispersion models with different meteorological data sets were compared (Draxler et al., 2015). As far as the meteorological conditions are concerned, precipitation must be simulated as realistically as possible because the spatiotemporal distribution of wet deposition is highly dependent on precipitation distribution. However, Draxler et al. (2015) found that the result of the deposition distribution applying high-resolution observed precipitation data did not perform better but the result using the precipitation field that is obtained by the meteorological model was improved. As these are different results, some transformation processes of the precipitation data set as applied to the dispersion also affected the model performance (Draxler et al., 2015). In another previous study, the simulated results of the spatial distributions of radioactive materials showed similar sensitivity to the meteorological data obtained by meteorological models as sensitivity to the wet deposition parameter in the wet deposition process (Leadbetter et al., 2015). In the study by Leadbetter et al. (2015), the mesoscale meteorological model along with radar precipitation data was applied to the transport simulation. Similar to Draxler et al. (2015), they also concluded that more analysis of the dispersion model simulations using high-resolution radar data was necessary (Leadbetter et al., 2015). Therefore, to further investigate the case of applying high-resolution precipitation data, the high-resolution radar rain gauge observation data values were replaced with other meteorological analysis data sets for wet deposition model calculations (Arnold et al., 2015; Saito et al., 2015a). However, results obtained using replaced precipitation values did not always present the most accurate deposition distribution; the 137Cs concentration time series was deteriorated (Saito et al., 2015a).
In this study, we focused on the third major issue: The uncertainty caused by precipitation needed to improve the accuracy of simulating wet deposition. Our approach was to use high-resolution radar rain gauge data to modify the wet deposition amounts on each model grid that were calculated via numerical model simulations using a reanalyzed meteorological data set. The aim of our study was not merely to find the best model and meteorological data set pair. To more accurately reproduce the simulated deposition distribution, we analyzed how such a hybrid method that combines both observed and simulated precipitation would contribute. In addition, for the FDNPP accident, the hot spot mainly formed when precipitation was observed. Therefore, improving the reproduction of the wet deposition distribution itself is important.
Referencing sensitivity tests employing different atmospheric boundary conditions, we examine the impacts on simulated precipitation and simulated deposition. We then suggest a method to improve the hourly scale variations in the wet deposition distribution using observed precipitation data. The potentially biased perception of the model results is also discussed based on validation exercises using hourly and daily observation data.
In the next section, the atmospheric model used in our study and its settings are described. Details regarding the meteorology data set and correction method for the wet deposition values with observed precipitation are also presented in the next section. The results of sensitivity tests and a comparison between the simulated concentration of 137Cs and observed radiation dose rate in hourly scale are presented in the third section. In the last section, the results are discussed and the study is summarized.
2 Materials and Methods
2.1 IsoRSM
We conducted simulations using the isotope-incorporated regional spectral model (IsoRSM: Yoshimura et al., 2010). The stable water isotopes HDO and H218O are incorporated into the regional spectral model (RSM; Juang et al., 1997). RSM was developed by the National Centers for Environmental Prediction (NCEP) and the Scripps Institution of Oceanography for use as a numerical weather and climate prediction model. A characteristic of RSM is its use of a spectral nudging technique (Kanamitsu et al., 2010; Yoshimura & Kanamitsu, 2009) that improves accuracy in the predicted meteorological field (Yoshimura et al., 2010). The overall performance of RSM as a regional climate model has been well demonstrated in many studies (e.g., Ham et al., 2016; Miller et al., 2009). The advantage of using IsoRSM is that both calculation of meteorological field and transport process of tracers can be conducted in the same model, whereas a typical dispersion model handles the meteorological field as input and not-so-often updated (hourly to several hourly interval). Because the meteorological field varies every time step of the model (i.e., 20–40 s), cloud microphysical variables vary nonlinearly. Furthermore, the mass/energy/momentum balances are always conserved in the IsoRSM, whereas a typical dispersion model cannot have them all conserved. Therefore, we used IsoRSM instead of dispersion models. By using IsoRSM, examining the best combination of dispersion model and input meteorological model is not necessary.
For the results reported herein, we used a version of IsoRSM in which aerosol forms of 131I and 137Cs and their deposition processes were implemented by Saya et al. (2015). Yoshikane et al. (2016) showed that the radioactive material version of IsoRSM was reasonably skilled at simulating radioactive materials produced in the Fukushima accident. In addition to the advection of radioactive aerosols, wet and dry deposition processes are incorporated in that version of IsoRSM.
(1)In equation 1, P and q are water condensation and water vapor, respectively. Both values can be obtained at any height during the precipitation process. C is the concentration of the radioactive material. Its temporal change is proportional to P/q α is the washout coefficient and represents how many particles can be incorporated into condensed water. In this study, α is set as 0.5. Regarding the dry deposition process, the formula is based on Maryon et al. (1991). The noniteration dimensional-split semi-Lagrangian advection scheme (Juang, 2007, 2008) was implemented to reduce the severe noise caused by the spectral representations of the horizontal distributions for all tracers (including water vapor; Chang & Yoshimura, 2015).
2.2 Data
2.2.1 Observed Precipitation Data
To validate the simulated precipitation and redistribution of wet deposition using observed precipitation, Radar Automated Meteorological Data Acquisition System (AMeDAS) precipitation data (hereafter, RAP) developed by the Japan Meteorological Agency (JMA) were used. RAP consists of composite precipitation radar system data calibrated using an in situ rain gauge data set (AMeDAS) that covers all Japanese islands (Oki & Sumi, 1994). Currently, RAP is broadcast every 30 min at a 1-km resolution (Takido et al., 2016). The data used in this study are distributed in GRIB format on DVD (Japan Meteorological Agency, 2011), and the hourly precipitation amounts in the following sections are the averaged values for the previous hour. The details of this data set and the format are also supported in Saito et al. (2015b).
2.2.2 Observed Radioactive Material Data
Daily observed data for the fallout of 137Cs in March 2011 after the FDNPP accident, as observed in each prefecture, are available to the public from the Ministry of Education, Culture, Sports, Science and Technology (MEXT; Ministry of Education, Culture, Sports, Science and Technology (MEXT), 2011). Using precipitation sampling devices, the fallout was sampled, and the radionuclides were analyzed using a germanium semiconductor detector in each prefecture (Ministry of Education, Culture, Sports, Science and Technology (MEXT), 2011). The daily cumulative fallout data values were recorded at 9:00 JST (0:00 UTC) every day, and the daily total amounts are cumulative values from 9:00 JST to 9:00 JST on the following days. It is worthwhile noting that the hourly precipitation rates in RAP are the previous 1-hr averaged values, whereas the observed daily 137Cs fallouts are the total amounts from 9:00 JST to 9:00 JST on the following days.
We used the radiation dose rate data that were maintained by each prefecture. The number of monitoring spots, equipment heights, and starting days of the observations varied depending on prefecture. In this study, the radiation dose rate data at the Tokyo Metropolitan Institute of Public Health located in Shinjuku-ku in Tokyo (Tokyo Metropolitan Institute of Public Health, 2011a) were used to investigate the time when the radioactive plume reached Tokyo. In the data, Grays are used as the radiation dose rate units. However, it is assumed that 1 Gray can be converted to 1 Sievert (Tokyo Metropolitan Institute of Public Health, 2011b). The dose rate data are the hourly averaged values.
2.2.3 Meteorological Data
For the atmospheric boundary conditions, we used NCEP and GPV-MSM. NCEP is reanalysis data that assimilates forecasted values from numerical models with various sources of observation data (Kalnay et al., 1996). Currently, the updated version of NCEP, NCEP/DOE reanalysis, is provided by the National Oceanic and Atmospheric Administration, with more observational data added and some errors of the previous version corrected.
GPV-MSM presents the results of the mesoscale climate model, whose domain covers the islands of Japan and its near seas, and it is provided by JMA. In this study, applying GPV-MSM data is based on the study by Yoshikane et al. (2016). The GPV-MSM data used in this study are provided by the GPV Data Archiving in Kitsuregawa laboratories, IIS, The University of Tokyo. In our model calculations, spectral nudging was applied to constrain the model by correcting the dynamical fields inside the domain (e.g., wind, air pressure, temperature, and humidity) toward corresponding atmospheric boundary data. The different nudging scale is set for each sensitivity experiment. For the experiment using NCEP reanalysis, the scale is 1,000 km. For the experiment using GPV-MSM, the scale is 10 km. The difference of the nudging scale is due to the difference of horizontal resolution of the two data sets.
By choosing these two different resolution data sets, we can analyze how the deposition distribution is sensitive to the boundary condition itself and how the coarse or fine horizontal resolution in the boundary conditions improves the results.
2.3 Simulation Settings
We first conducted two numerical IsoRSM experiments using different meteorological boundary conditions provided by the NCEP reanalysis data and GPV-MSM data. The horizontal resolution of the NCEP reanalysis data was approximately 180 km, and the time resolution was 6 hr, whereas those for GPV-MSM were 10 km and 3 hr, respectively. In these experiments, a common emission amount estimated by Terada et al. (2012) was used. The simulation period was from 0:00 UTC on 11 March to 21:00 UTC on 31 March in 2011. The calculation domain and the resolution are 136°29′31.2″E–145°37′19.2″E, 33°36′21.6″N–42°11′16.8″N, and 5 km. The domain and locations of the observation stations of fallouts mentioned in section 2.1 are shown in Figure 1.
The main physical processes for the atmospheric simulations were as follows. The relaxed Arakawa-Schubert scheme was used for convective parameterization (Moorthi & Suarez, 1992), the Noah land surface model was employed (Ek et al., 2003), the Chou scheme was used for radiation (Chou & Suarez, 1994), and the nonlocal scheme (Hong & Pan, 1996), which is based on parameterization using the first-order closure model considered with boundary layer heights and velocity scale, was employed as the planetary boundary layer scheme. A new method was used in both experiments to amend the precipitation fields (described in the next section), from which we obtained four different simulated radioactive material results, N1, N2, G1, and G2, where N stands for NCEP, G stands for GPV-MSM, and 1 and 2 stand for without and with the new method described in the next section, respectively (see Table 1).
| Experiment | Atmospheric boundary condition | Redistribution by RAP |
|---|---|---|
| N1 | NCEP | Not applied |
| N2 | NCEP | Applied |
| G1 | GPV-MSM | Not applied |
| G2 | GPV-MSM | Applied |
2.4 A Method for Improving Simulated Wet Deposition Using Precipitation Data
Precipitation is among the most difficult quantities to simulate in regional atmospheric models. Compared to precipitation, the skills of other atmospheric variables such as wind speed, air temperature, and air pressure are inherently well reproduced by the model. Therefore, it is reasonable to conclude that a significant part of failures in simulations of radioactive material deposition derives from the precipitation process. With regard to this issue, we suggest a method for improving the spatiotemporal distributions of radioactive material deposition by only changing precipitation.
(2)
(3)The advantage of this method is that the transformations inside the model for applying the method are not necessary because our redistribution method is conducted to simulated wet deposition not during the model calculation but after all the calculation. While the transformation or combination between dispersion model and meteorological model had to be considered to use the radar precipitation data in the previous studies (Arnold et al., 2015; Draxler et al., 2015; Saito et al., 2015a), this method can be applied simply to the model outputs. The disadvantage is that the area-total amount of the wet deposition is always conserved before and after this method. Thus, if the area-total deposition amount is wrong at some moment, that cannot be fixed. However, only changing the area-total amount of deposition depending on the precipitation should not be conducted because it cannot keep the balance with the air concentration of 137Cs inside the domain. To solve the uncertainty of this area-total wet deposition with keeping the balance, a more advanced data assimilation technique, which incorporates the input of high-resolution precipitation data during the model calculation step, would be a possible candidate for further study.
3 Results
3.1 Validation of Precipitation
In this section, the precipitation fields in the N1 and G1 experiments are validated. Figure 2 shows the observed precipitation, RAP (Japan Meteorological Agency, 2011), and the precipitation distribution simulated in G1 and N1 in the Kanto region every 3 hr from 6:00 to 15:00 on 21 March in 2011 (JST) when precipitation is simulated and also observed clearly at the observation stations in the Kanto region inside the domain. As shown in Figure 2, G1 and N1 produced clearly different characteristics, and both were more or less different from RAP. In RAP, over 4 mm/hr of precipitation was distributed on the western edge of the domain and moved to the Pacific Ocean after passing the Kanto region. At 9:00 and 12:00, a small area but large amount of precipitation, that is, over 10 mm/hr, was distributed over the domain. In addition, from 9:00 to 15:00, less than 1 mm/hr of precipitation was broadly observed, primarily in the Kanto region and in the southern area of the Tohoku region. However, no precipitation was observed at some stations and times: Saitama, Shinjuku, Ichihara, and Chigasaki at 6:00 and Utsunomiya, Saitama, and Shinjuku at 12:00.
For G1, similar to that for RAP, over 4 mm/hr of precipitation was simulated, including over 10 mm/hr that was simulated at 9:00 and 12:00. Weak precipitation of 0–1 mm/hr also appeared broadly on the Japan Sea side of the Tohoku region. In addition, unlike RAP, precipitation was continuously simulated in the Kanto region for G1. At each time and at most of the stations, with the exception of Maebashi at 6:00, 12:00, and 15:00 and Saitama at 6:00, the distribution of precipitation can be found in G1 results.
N1 showed some similarities with RAP and G1, but there were also some unique and changing patterns. Similar to RAP, no simulated precipitation occurred at some stations at 6:00, and the precipitation at 9:00 was not distributed over the entire Kanto region. In addition, a rain field with intensities of 2–4 mm/hr passed the Kanto region from west to east from 9:00 to 12:00, but that passage was not clearly seen for N1, although similar fields were seen for RAP and G1. As shown in Figure 2, weak precipitation of 0–1 mm/hr predominantly occurred through 9 hr. This weak precipitation field appeared widely not only in the Kanto region and over the Pacific Ocean but also over the northern area of the Tohoku region and Japan Sea.
3.2 Simulated Deposition of 137Cs
The sensitivity tests of the atmospheric boundary conditions and observed precipitation data undertaken by comparing the simulated 137Cs depositions resulting from G1, N1, G2, and N2 are addressed in this section. Figure 3 shows the total 137Cs deposition from 18 to 27 March in 2011 at seven stations in the Kanto region as simulated in the four experiments. These were compared with the observed 137Cs fallout at the same location (Ministry of Education, Culture, Sports, Science and Technology (MEXT), 2011). In Figure 3, each scatterplot has five diagonal lines. The solid black diagonal line in each figure indicates that the simulated deposition was the same as the observed fallout. The gray dashed lines next to the solid black lines indicate the matching ranges: a factor of 2 (0.5–2; F2) and a factor of 10 (0.1–10; F10). Regarding the time period in Figure 3, at the beginning of the accident, for example, on 11 or 12 March, the deposition in the Kanto region was under the detectable value in some areas in the observation by MEXT. Therefore, before this period, it is difficult to validate the model. In addition, our main target of wet deposition of 137Cs in the Kanto region mainly occurred after 20 March.
Table 2 shows how many stations are included in F2 and F10 in each experiment. In F10, all seven stations are included in N2, G1, and G2. Even in N1, six stations are included. This indicates that our model performed well in the F10 range regardless of resolution of the boundary conditions and existence of redistribution by observed precipitation. On the other hand, in the range of F2, G2 clearly became the best result among the experiments. Six of seven stations are included with Shinjuku included in F2 only in G2. Even the result at Hitachinaka in F10, its position is located relatively closer to the line of F2 in G2. Although high-resolution boundary condition, GPV-MSM, is applied to both G1 and G2, the score of G1 is almost same as N1 and N2. From this result, it is inferred that both high-resolution boundary conditions and the modification by the precipitation observation data are necessary to obtain the highly accurate results fulfilling F2 matching criteria. In addition, it is also found that the RAP can contribute to improve the accuracy of simulated 137Cs deposition distribution by using our method that does not require the transformation of the observation data to the atmospheric models. However, the result that is relatively inaccurate, like Hitachinaka, is not largely improved even when the redistribution method is applied. That means that there is the limitation of the method for the case which is still clearly inaccurate by applying high-resolution boundary conditions.
| Experiment | The number of stations in F2 | The number of stations in F10 |
|---|---|---|
| N1 | 4 | 6 |
| N2 | 3 | 7 |
| G1 | 4 | 7 |
| G2 | 6 | 7 |
To compare the results on a daily basis, Figure 4 shows the daily total amounts of 137Cs for the four experiments from 18 to 27 March. To see the effect caused by precipitation, Figure 4 also shows the RAP time series. At all the stations, the observed fallout increased on 20 and 21 March and rapidly decreased on 22 March. After 22 March, except on 25 March at Hitachinaka, the fallout amounts were less than 0.1 kBq/m2. The trends for all of the experiments were approximately similar to those of the observations. However, there were some discrepancies in the amounts and delays of the increase. Table S1 in the supporting information shows the numerical values of the graph in Figure 4.
Discrepancies clearly appeared on 20 or 21 March. On 20 March, the observed 137Cs amounts exceeded 0.1 kBq/m2 at all stations, especially for Hitachinaka and Chigasaki, where the amounts were 13 and 1.6 kBq/m2, respectively. However, except for Utsunomiya and Hitachinaka, all of the simulated results were below 0.05 kBq/m2. At Hitachinaka, the observed amounts were 10 times greater than those from the four experiments. On 21 March, the results were smaller than observed at Hitachinaka, Shinjuku, and Ichihara, but the simulated deposition amounts rapidly increased in most of the cases.
From the results of four experiments, both overestimation and underestimation can be found in various ways between 20 and 23 March at all of the stations. In that period, precipitation was observed and is illustrated with the bar graph in Figure 4. Regarding those precipitation days, G1 and G2 showed slightly better results. G2 achieved the best performance, especially on 21 March at Hitachinaka and Shinjuku, when the amount of observed depositions was significantly larger than in other days and exceeded 5 kBq/m2. From the results, we can conclude that the experiments using GPV-MSM generally achieved better results in comparison with NCEP, and the experiment in which the redistribution method was applied was more accurate than those that did not when the observed depositions are higher like on 21 March at Hitachinaka and Shinjuku.
In Figure S1 in the supporting information, the wet deposition distribution maps of 137Cs from each experiment are shown. The date and time of the maps are the same as in Figure 2. By comparing both the figures, it is possible to visually understand how the redistribution method worked to correct the simulated deposition distribution toward the observed precipitation distribution.
3.3 Effect on the Evaluations Caused by the Definition of Daily Observation
Figure 4 shows the temporal variations in the total deposition of 137Cs validated using the daily observation data. As shown, there were almost no simulated radioactive deposits at Maebashi, Saitama, Shinjuku, and Chigasaki on 20 March, whereas the observed fallouts suddenly began to increase. Even for G2, whose accuracy was better than the others, as the investigations documented in the previous section showed, the amounts of 137Cs were almost zero. In Arnold et al. (2015), the results using RAP also did not always improve the accuracy of the model performance. They concluded that highly developed precipitation observation data are inconsistent with the meteorological scheme in the model without the modification. In addition to this problem about the implementation of the high-resolution radar precipitation data to the model, we speculate that this discrepancy was caused by the delay of the simulated atmospheric advection process because it has been shown that the simulated depositions rapidly increased on 21 March.
To further investigate the time series of changes on 20 and 21 March, we used the hourly observed dose rate data. Figure 5 shows the observed radiation dose rates (μGy/hr) and simulated concentrations in air at a height of 2 m at Shinjuku on 21 March for G1/G2. The radiation dose rate was taken from observational data collected by the Tokyo Metropolitan Institute of Public Health (2011a). It should be noted that these two sets of values are inherently different; the concentrations of 137Cs in the air at the surface in the simulation only include the amounts of 137Cs, and the observed radiation dose rates do not distinguish 137Cs from other radionuclides. In addition, the simulated air concentrations were detected only when there was a plume at the observation spot, whereas the observed radiation dose rates included both cloud shine and ground shine. However, in this section, to focus on when the radioactive plume approached the Kanto region, the timings when the values increased are compared.
As shown in Figure 5, the observed radiation dose rate rapidly increased from approximately 8:00 to 10:00 in the morning. The rapid increase in the simulated 137Cs surface concentration did not occur until 11:00. At the other observation stations in the Kanto region, the differences were approximately 2–3 hr (figure omitted). According to these results, we conclude that the arrival of the radioactive plume was delayed by a few hours in the simulation.
Generally, this model error of a few hours delay is not critical behavior in the regional atmospheric modeling development. Thus, it is not incorrect to conclude that the model performed quite well in terms of atmospheric advection. However, these small differences produced the significant consequence found in Figure 4, in which it is seen that the daily model results were evaluated to be poor. Observed radiation dose rate began to increase during the morning of 21 March and increased rapidly more than 0.01 μGy from 8:00 to 9:00. However, for G1/G2, the 137Cs surface concentrations continued to slightly increase from 9:00 to 10:00. The daily boundary of the Japanese meteorological measurements of deposition exists at 9:00 each day. Therefore, the increase of the deposition formed by increased concentration from 9:00 to 10:00 is included in the daily deposition on 21 March. However, if the concentration increased before 9:00, the increase of the deposition would be included in the daily data on 20 March, the same as the observation in Figure 4. In other words, the large depositional amounts observed on 20 March were not well reproduced by the model because of 2–3 hr delay of the simulated concentration. From this fact, we want to emphasize that not only the model accuracy but also the recording method used for the observation can affect the model evaluation.
4 Summary and Conclusions
To accurately obtain data on the distribution of 137Cs released during the FDNPP accident, this study focused on the deposition of 137Cs that is simulated by a regional atmospheric model and on improving that model. Among the uncertainties arising in the numerical simulation, atmospheric boundary conditions were our main target because they affect plume transport in the atmosphere and precipitation, which influence the spatiotemporal distribution of wet deposition, including the hot spot observed in the metropolitan area after the accident.
The RSM incorporating isotope tracers, IsoRSM, was used. For comparison and to investigate the differences arising from their use, two different atmospheric boundary conditions, NCEP and GPV-MSM, were used. Furthermore, AMeDAS precipitation (RAP) was used to modify the spatial distributions of wet deposition, and the impact of that modification was evaluated.
Among the four sets of results, the experiment driven by GPV-MSM and modified by RAP was the most accurate in terms of the cumulative 137Cs deposition from 18 to 27 March. This indicates that although the atmospheric advection process was reasonably well reproduced, improvements to the precipitation process are needed, especially when conducting atmospheric transport simulations of radioactive materials.
During the validation on a daily basis, it was found that the four sets of results showed the increase in 137Cs deposition on 21 March, but the observations showed that increase a day ahead on 20 March. After validation using the hour-scale radiation dose rates in air, it was identified that this large inconsistency was derived from a delay in atmospheric advection of only 2–3 hr. The daily boundary of the observed deposition in Ministry of Education, Culture, Sports, Science and Technology (MEXT) (2011) is at 9:00 on each day. In the case of 20 and 21 March, the plume reached around a few hours before 9:00. The simulated concentration is increased around that time, but it is a few hours after 9:00. Therefore, observed daily deposition was increased on 20 March, not on 21 March in the data, while simulated increased deposition is included in the daily deposition on 21 March. As such, the manner in which observations are recorded can affect model evaluation.
Simulated deposition maps of released radioactive materials have not been used widely as disaster prediction tools because the manner in which those maps should be used has not been well discussed. Due to uncertainties, simulation results do not perfectly realize past and future distributions. In discussions with the public, this problem of uncertainties must be explained properly to reduce misunderstandings. In conclusion, simulation models for nuclear power plant disasters must be developed by solving problems of uncertainty, including those arising from the use of data assimilation meteorological field accuracies, and by discussions among scientists, local government officers and local residents on how to use risk maps.
Acknowledgments
This research was supported by the “A tracer simulator of fallout radionuclides for safe and sustainable water use” project under the Core Research for Evolutional Science and Technology (CREST) program, the Japan Science and Technology Agency (JST); the Environment Research and Technology Development Fund (S-12) of the Japanese Ministry of the Environment; and the Data Integration and Analysis System (DIAS) funded by the Ministry of Education, Culture, Sports, Science and Technology (MEXT) and through JSPS KAKENHI Grant Numbers JP15J10465, JP16H06291, JP18H03794 and JP15KK0199. IsoRSM can be installed at https://g-rsm.wikispaces.com/. The GPV-MSM data are provided by the GPV Data Archiving in Kitsuregawa Laboratory, IIS, the University of Tokyo at http://apps.diasjp.net/gpv/. The NCEP Reanalysis data are obtained from NOAA/OAR/ESRL PSD, Boulder, Colorado, USA at http://www.esrl.noaa.gov/psd. AMeDAS precipitation data (RAP) are provided by the JMA. The data used for drawing figures in this paper are available. The location and terms and condition for these are shown at http://hydro.iis.u-tokyo.ac.jp/~s.akane/data/Readme_jgra2018_redistribution.
