Initial spread of 137 Cs from the Fukushima Dai-ichi Nuclear Power Plant over the Japan continental shelf : a study using a high-resolution , global-coastal nested ocean model

The 11 March 2011 tsunami triggered by the M9 and M7.9 earthquakes off the T ōhoku coast destroyed facilities at the Fukushima Dai-ichi Nuclear Power Plant (FNPP) leading to a significant long-term flow of the radionuclide 137Cs into coastal waters. A high-resolution, global-coastal nested ocean model was first constructed to simulate the 11 March tsunami and coastal inundation. Based on the model’s success in reproducing the observed tsunami and coastal inundation, model experiments were then conducted with differing grid resolution to assess the initial spread of 137Cs over the eastern shelf of Japan. The 137Cs was tracked as a conservative tracer (without radioactive decay) in the three-dimensional model flow field over the period of 26 March–31 August 2011. The results clearly show that for the same137Cs discharge, the model-predicted spreading of 137Cs was sensitive not only to model resolution but also the FNPP seawall structure. A coarse-resolution ( ∼ 2 km) model simulation led to an overestimation of lateral diffusion and thus faster dispersion of 137Cs from the coast to the deep ocean, while advective processes played a more significant role when the model resolution at and around the FNPP was refined to∼ 5 m. By resolving the pathways from the leaking source to the southern and northern discharge canals, the high-resolution model better predicted the 137Cs spreading in the inner shelf where in situ measurements were made at 30 km off the coast. The overestimation of 137Cs concentration near the coast is thought to be due to the omission of sedimentation and biogeochemical processes as well as uncertainties in the amount of 137Cs leaking from the source in the model. As a result, a biogeochemical module should be included in the model for more realistic simulations of the fate and spreading of 137Cs in the ocean.


Introduction
The 11 March 2011 Tōhoku magnitude 9.0 and 7.9 earthquakes caused a massive tsunami with ∼ 16 m wave height nearshore and tsunami-induced inundation that devastated the eastern coast of Japan (Fig. 1).Unlike previous dumping from the storage room contained low-level radioactive water, while the leaking from reactors contained highlevel radioactive water, with a concentration of 9.4 × 10 14 Bq for 137 Cs and 134 Cs as well as 2.8 × 10 15 Bq for 131 I from Unit-2 over the period of 1-6 April and of 9.8 × 10 12 Bq for 137 Cs, 9.3 × 10 12 Bq for 134 Cs, and 9.5 × 10 12 Bq for 131 I from Unit-3 over the period of 10-11 May.In contrast to the Chernobyl disaster in 1986, this was not the most serious radionuclide-leakage in the past.The difference is that the Chernobyl Nuclear Power Plant was located inland and its impact on the Black and Baltic Seas was through atmospheric deposition with a value of 10 5 Bq, much smaller than what happened at the FNPP.As a result, following the 11 March 2011 tsunami event and in addition to the wet and dry deposition from the atmosphere, the coastal water was contaminated by discharges of a large portion of high-level radioactive water out of the FNPP from leaking sources (Honda et al., 2012) and from inland-polluted rivers (Oura and Ebihara, 2012).
Among these radioactive isotopes, 137 Cs was of particular interest because of its long 30.2 yr half-life.The accumulation of 137 Cs in marine food chains could exert a profound impact on marine biota and human health and thus the local to regional ecosystem (Buesseler et al. 2011;Grossman, 2011).Accurately determining the initial dispersion of 137 Cs off Japan's coast was a prerequisite for assessing its long-term impacts in the interior of the Pacific Ocean.After the leaking occurred, many efforts have been made to monitor the spread of 137 Cs off Japan's coast.The Min- istry of Education, Culture, Sport, Sciences and Technology (MEXT) (http://radioactivity.nsr.go.jp/ja/list/238/list-1.html) and the Tokyo Electric Power Company (TEPCO) (http:// radioactivity.nsr.go.jp/ja/list/239/list-1.html) started measuring the 137 Cs concentration around the FNPP and in the offshore coastal waters (Figs. 2 and 3).In addition to these two government-established monitoring programs, several field surveys were carried out in an offshore region to assess the spreading of 137 Cs by oceanic currents, lateral diffusion and vertical mixing (e.g.Honda et al., 2012;Behrens et al., 2012;Dietze and Kriest, 2012).The research team led by K. Buesseler (Woods Hole Oceanographic Institution) made a comprehensive survey in the shelf and deeper waters off FNPP in June 2011 (Buesseler et al., 2012).Their survey measured 137 Cs concentration over the inner-shelf area 30 km away from the coast and then along several transects across the Kuroshio pathway in the deep ocean (Fig. 3).The data collected from these monitoring and field surveys have provided a direct assessment of temporal change and spatial distribution of 137 Cs concentration in the coastal waters.Due to the complex nature of advection and mixing in this coastal region, however, these data cannot alone be used to predict the spreading processes of 137 Cs from FNPP in the shelf waters after the leaking started.This is one of the key reasons why an ocean model was proposed for this purpose.
It is not a trivial task for a model to simulate and predict accurately the spatial distribution and temporal change of the 137 Cs concentration off the Japan coast.Since advection and mixing are two key physical processes that control the spread of 137 Cs in ocean waters, we need an ocean model that is capable of resolving an integrated coastal and regional circulation system over scales from a few meters (small scale, e.g.around FNPP) to a few kilometers (mesoscale) over the shelf.The flow around FNPP and near the coast is mainly controlled by tidal exchange, winds and local coastal geometry and bathymetry.The circulation in this shelf region includes the Kuroshio Current on the south, the Oyashio Current on the north, the Tsugaru Current from Tsugaru Strait, and multiple eddies formed in the intersection area of these currents (Fig. 1).To simulate the outflow from FNPP, we need a model with accurate fitting of complex coastal geometry within and around FNPP.The water over the shelf was always stratified so that water temperature and salinity must be included in the model simulation.Several regional-scale ocean model exercises have been made to simulate the 137 Cs spread from FNPP, e.g.Kawamura et al. (2011) and Tsumune et al. (2012) with a spatial resolution of 2 km or larger, and Estournel et al. (2012) with a resolution of 0.6 km.However, the water exchange between FNPP and the surrounding ocean is through a ∼ 200 m-narrow entrance between the two breakwaters.The distance between the two discharge canals (namely, the north and south discharge canals) is 1300 m.Without sufficient model resolution to accurately capture the complex pathways of 137 Cs from FNPP, assessments made by these regional-scale models could be biased with large uncertainty.It is not clear, however, to what degree this bias could be.Could the bias caused by model resolution and geometric fitting issues led to a significant different conclusion about the dispersion of 137 Cs off the Japan coast or reproduce the same distribution with just a small difference in accuracy?To our knowledge, this issue has not been well addressed yet in previous modeling experiments.
Geometric fitting of complex coastlines around FNPP and in coastal regions is a critical factor to resolve multi-scale geometrically controlled nearshore advection while sufficient model resolution is prerequisite of capturing a realistic lateral dispersion.Chen et al. (2008) conducted a model-dye comparison experiment over Georges Bank, with an aim of examining the impact of model resolution on lateral dispersion in the coastal ocean.They found that in order to simulate accurately the observed lateral dispersion within a tidal mixing front with a spatial variation scale of a few kilometers, model resolution down to ∼ 500 m or less was required.The overestimation of lateral dispersion due to model resolution varied in space and time, which could be 4-10 times larger as the model grid size is greater than 2-4 km.As a result, the model could point to an unrealistic conclusion that differed significantly from the dye observations.The dynamical processes off the FNPP coast are more complicated than on Georges Bank, so that failure to adequately resolve the important spatial scales in this region might lead to a large lateral dis- persion rate and thus overestimate the offshore spreading of 137 Cs over the eastern Japan shelf.
The biggest challenge for a model to provide an accurate simulation of the spatial distribution and temporal change of 137 Cs over the Japan shelf is the large uncertainty in the estimation of the total amount of 137 Cs leaking into the water.The leaking lasted for months, so that the source was both spatially variable and time-dependent (Estournel et al., 2012).One approach to solve this problem is to treat 137 Cs as a conservative tracer (without radioactive decay) and inversely determine its source amount by tracking it in the flow field for a relatively short period during which the modelpredicted tracer field had the best match to observations.This method was used to evaluate the total amount of 137 Cs from FNPP in the previous modeling experiments made by Kawamura et al. (2011), Tsumune et al. (2012), and Estournel et al. (2012).This method is generally sound, but an adjustment in this type of inverse tracking could vary from model to model, particularly for the case with different model resolutions and setups.Due to this uncertainty, the more interesting model problem, in our opinion, is on gaining knowledge of the sensitivity of the model assessment results to model skill and configuration rather than on evaluating how well a model simulates the observed 137 Cs concentration.
We, an international research team with members from the University of Massachusetts-Dartmouth, Woods Hole Oceanographic Institution, and Yokohama National University (a member now at the University of Tokyo), have developed a high-resolution global-regional-coastal integrated seismic-ocean-tracer FVCOM model system to simulate the March 11 earthquake-induced tsunami, coastal inundation and initial spread of 137 Cs.Taking advantage of the geometric flexibility of the unstructured triangular grid, the model has a local resolution of up to 5 m around FNPP and near the coast.Nesting with the global-FVCOM hindcast field with data assimilation of satellite-derived sea surface temperature and sea surface height, the high-resolution regional-coastal FVCOM model not only resolved a realistic regional circulation but also provided a better representation of the water exchange between FNPP and the surrounding ocean.Built on our success in simulating the observed tsunami and coastal inundation (Chen et al., 2013), we applied this model system to track 137 Cs over the period of 26 March-31 August 2011.
Our studies aimed at assessing the impact of multi-scale physical processes on the initial spread of 137 Cs in the coastal region of Japan, which form the foundation for more realistic simulations with the inclusion of biogeochemical processes.

The model and design of numerical experiments
The 137 Cs was tracked as a conservative tracer in the threedimensional (3-D) flow field predicted by the high-resolution nested global-coastal FVCOM model over the period of 26 March-31 August 2011 (Fig. 4).Hereafter we refer to the global FVCOM model as Global-FVCOM and the Japan coastal FVCOM model as JC-FVCOM.FVCOM is the prognostic, unstructured-grid Finite-Volume Community Ocean Model originally developed by Chen et al. (2003) and upgraded by the FVCOM team (Chen et al., 2006a(Chen et al., , b, 2012)).FVCOM solves the flux form of the governing equations in control volumes constructed with multi-triangular meshes using a second-order accurate discrete flux scheme, which provides accurate fitting of irregular coastal geometries and flexibility in adjusting the grid resolution to capture the key physical processes (Chen et al., 2007).The finite-volume approach ensures local mass, heat, salt, and tracer conservation in the sense of numerical computation, which is suitable to trace 137 Cs for this study.The tracer module of FVCOM was validated through a model-dye comparison experiment made on Georges Bank in the northern North Atlantic Ocean by Chen et al. (2008).
Global-FVCOM is a fully ocean-ice coupled model covering the entire global ocean with a grid resolution of ∼ 2 km along the eastern Japanese coast (Fig. 4).The vertical grid discretization was implemented using a hybrid terrainfollowing coordinate with a total of 45 layers (Chen et al., 2013).The s coordinate was used in regions with depth greater than 225 m, in which 10 and 5 uniform layers with a thickness of 5 m were specified near the surface and bottom, respectively.The uniform thickness σ coordinate was employed in regions of depth less than 225 m.The coordinate transition occurred at the depth of 225 m where all layers have a uniform thickness of 5 m.Global FVCOM was driven by astronomical tidal forcing with eight constituents (M 2 , S 2 , N 2 , K 2 , K 1 , O 1 , P 1 , and Q 1 ), the NCEP reanalysis meteorological forcing fields (surface wind stress, net heat flux/shortwave irradiation, air pressure gradients, precipitation minus evaporation (P − E)), and freshwater discharge from all major rivers along the coast.Initialized with the assimilated model fields at the end of 31 December 2010, we ran Global-FVCOM for the period of 1 January 2011-31 August 2011 for this 137 Cs tracking experiment.To ensure that Global-FVCOM was capable of capturing the regional circulation along the eastern Japanese coast, satellite-derived sea surface temperature (SST) (http://www.nodc.noaa.gov/SatelliteData/ghrsst)and AVISO sea surface height (SSH) (http://www.aviso.oceanobs.com/en/data/products/sea-surface-height-products.html) were assimilated into the model.Global-FVCOM has been validated through a 50 yr spin-up simulation and a 33 yr  hindcast assimilation (Gao, 2011;Hu et al., 2011).
JC-FVCOM was configured with horizontal resolution varying from 2 km near the boundary nesting with Global-FVCOM to 5-10 m in the nearshore coastal region (including the FNPP) (Fig. 4).JC-FVCOM had the same hybrid vertical coordinate system with 45 layers and was forced with the same meteorological forcing as Global-FVCOM.
To assess the importance of resolving the detailed geometry around the leaking facility on the temporal and spatial distribution of 137 Cs in the coastal region off Japan, we tracked 137 Cs in two types of flow fields: One from Global-FVCOM with a horizontal resolution of 2 km along the coast around the FNPP and the other from the nested JC-FVCOM-Global-FVCOM system with a resolution of 5 m in and around the FNPP.The first set of experiments used the same approach as previous studies (Kawamura et al., 2011;Tsumune et al., 2012;Estournel et al., 2012), in which the 137 Cs discharge was treated as a point source at the coast in a regional model, while the second set of experiments had sufficient resolution to simulate the 137 Cs discharge from the FNPP through the different pathways into the ocean.
We adopted a similar approach as used by Tsumune et al. ( 2012) and Estournel et al. (2012) to determine inversely the amount of 137 Cs at the source based on the best fit with observations made at the northern discharge canal (1F-N) and the southern discharge canal (1F-S) of FNPP (Fig. 2).The model-predicted 137 Cs field was then validated with comparisons to the monitored concentrations at 2F and Iwasawa south of 1F-S and other TEPCO, MEXT and WHOI measurements data made in the nearshore and offshore regions.Our estimated amount of the total direct ocean release was 14.5 PBq, which is closer to the amount of 16.2 ± 1.6 PBq estimated by Rypina et al. (2013)  In this study, direct atmospheric loading was not considered.As reported by Kawamura et al. (2011) and Tsumune et al. (2012), almost all of the 137 Cs atmospheric deposition into the ocean occurred in March, with very little deposition afterwards.While atmospheric loading can be easily included in a tracer model, an accurate estimate of the time-and spatialdependent 137 Cs loading that occurred during March was not available, so previous modeling assessments were focused solely on the direct water discharge at the coast.We followed the same strategy in our tracer experiments.To avoid underestimation due to the lack of atmospheric loading, we started tracking 137 Cs on 26 March 2011, with the understanding that the model-data mismatch in late March would likely be caused by atmospheric loading.Since the 137 Cs loading from leaking sources were adjusted by best fitting with measurements at sites 1F-N and 1F-S at the two canal exits, the model simulation should indirectly account for some of the actual atmospheric deposition occurring in late March.Our focus here is to examine the importance of resolving the complex structure of the FNPP on predicting the initial spread of 137 Cs in the Japanese coastal region.This approach could help us separate the impacts of the water source on the oceanic environment from the atmospheric source.

Comparisons with observations
The results of the high-resolution model case predicted by the nested JC-FVCOM/Global-FVCOM will be presented first, followed by a comparison with the Global-FVCOM coarse-resolution model case.
The 137 Cs released at the nuclear reactor sites within the FNPP facility flowed out of the FNPP mainly through the channel bounded by the northern and southern breakwaters (Fig. 2; Ohnishi, 2012).In our high-resolution model case, which direction and how much 137 Cs flowed to the 1F-N and 1F-S sites outside the breakwaters depended on whether or not the model was capable of resolving the local water flushing processes around FNPP.The observed 137 Cs concentration at these two sites varied with the same trend but slightly different amplitudes, reached two peaks around the end of March and early April, and then rapidly decayed with time after 12 April 2011 (Fig. 5).By tuning the amount of 137 Cs at the source, the high-resolution model reproduced these variations reasonably well.This suggested that the model could provide a realistic flow exchange process between FNPP and the adjacent ocean.
Sites 2F and Iwasawa are located near the coast about 9 and 14 km south of 1F-S, respectively.The model-data comparison at these two sites again captured the observed rapid increase in 137 Cs concentration in late March and the gradual decay trend in April (Fig. 6).However, at these sites the peak concentration of 137 Cs was about 1 ∼ 2 orders of magnitude smaller than that at 1F-S.
A further comparison was made at eight MEXT sites over the shelf (Fig. 7).The measurements suggest that, excluding atmospheric loading, a significant amount of 137 Cs in the radioactive water from the FNPP arrived in this shelf area about two weeks after the leaking started.Although the MEXT sites were only about 30 km away from the coast, the maximum 137 Cs concentrations were about 10 2 -10 3 lower compared with the peak values observed at the monitoring site 1F-S.The 137 Cs concentrations at MEXT sites reached a peak value of ∼ 100 Bq L −1 at the end of April and then rapidly decreased to ∼ 10 Bq L −1 or less during May.The model-computed 137 Cs concentration and its decay trend with time were in reasonable agreement with observations.For example, at site MEXT-3, the observed 137 Cs concentration was slightly lower than 100 Bq L −1 in early April and lower than 10 Bq L −1 in early June.These two values and varying trends were well captured by the model.For the same given time period, the model suggested that the 137 Cs concentrations were higher at the northern sites than at the southern sites.It is thought that the 137 Cs detected at the MEXT sites before 8 April was due to atmospheric deposition (Kawamura et al., 2011;Estournel et al., 2012;Tsumune et al., 2012).Since neither atmospheric deposition nor an initial field of 137 Cs concentration was set up in the current study, no direct comparison with observational value recorded during that period should be made.
We then compared the model-computed 137 Cs concentrations with observations at the same location and time where/when measurements were made for all available data sources, including TEPCO, MEXT, and WHOI.At the surface (Fig. 8), starting in early April, the model-computed 137 Cs concentration showed a good match with MEXT and TEPCO measurements, even though a large uncertainty exists about the initial 137 Cs distribution, loading from atmospheric deposition and point source as well as the complex biogeochemical processes in seawater.A mismatch appeared in late March that was likely a result of atmospheric deposition, which was omitted in our study.TEPCO measurement sites were located in the nearshore regions close to FNPP.Reasonably good agreement between model-computed and observed 137 Cs concentrations for TEPCO data in April through August suggests that the high-resolution model succeeded in resolving the advection and dispersion processes near the coast.A good match was also found for MEXT data from April to May, indicating that the model was also robust to capture these physical processes in the inner shelf region over a time scale of a month after leaking started.The model, however, tended to overestimate the 137 Cs concentrations recorded during the WHOI June survey and at MEXT sites in August.The WHOI survey started with one transect roughly around the 200 m isobath 30 km from the coast followed by other transects across the slope and eastward  Kuroshio main stream.Placing both model-computed and observed 137 Cs concentrations at measurement sites and creating images based on these data (Fig. 9), we can see that the model was robust in predicting the spatial distribution of 137 Cs concentration that were observed during the WHOI June survey, but it tended to overestimate the size of the 137 Cs concentration plume and its values in the mid-shelf and slope regions.
Near the ocean bottom (Fig. 10), however, the modelcomputed 137 Cs concentration was generally lower than the observed values at both TEPCO and MEXT monitoring sites.The overestimation at the surface and underestimation near the bottom imply that in addition to vertical diffusion and mixing, there were other physical processes that were responsible for a downward flux of 137 Cs in the water column.By adding a sinking term in the 137 Cs tracer model, one should be able to improve the simulation results.The critical issue is that such a sinking term is related to sedimentation over the shelf with the sinking velocity varying significantly with different types of sediments.We will discuss this issue in the next section.

Comparisons between high-and coarse-resolution models
Most of the 137 Cs was transported into the Japan's coastal shelf through a pumping-like process from the narrow exit between the northern and southern breakwaters of FNPP.As a result of tidal flushing and dumping of cooling water into FNPP, the maximum outflow at the exit was about 2 m s −1 .This strong outflow was jet-like and formed a cyclonic vortex initially due to shear instability (Fig. 11).With a continuous supply of water from the exit, this vortex became large and then separated into several large cyclonic and anticyclonic vortexes in late March before entering the continental shelf where the regional-scale circulation became dominant.
Once the 137 Cs tracer was over the shelf where the water depth was 50-100 m or deeper, its spread was strongly influenced by the local wind and regional circulation (Fig. 12).In April, the tracer appeared like a coastal plume, which moved back and forth in the south-north direction along the coast.
In May, the 137 Cs plume was still constrained within the coastal region but had a significant southward transport.It arrived in the southern region about 180 km south of FNPP in mid-May, where a portion of the 137 Cs was carried offshore by the eastward-flowing Kuroshio Current.At the same time, the northward wind caused coastal upwelling and windinduced Ekman flow advected and dispersed the 137 Cs plume offshore.In July and August, the plume was predominantly To our knowledge, all previous model assessments of the 137 Cs spreading were made with a regional-scale model without resolving the geometry of FNPP.Key questions here include is the initial pumping process from FNPP critical for a model to produce a realistic spread of 137 Cs from the FNPP, or could the nearshore be ignored if one is only interested in predicting the 137 Cs spread over a regional scale?To address these questions, we followed the methods used in previous model assessments and tracked 137 Cs in the flow field predicted by Global-FVCOM.In this regional model case, because the 2 km resolution grid was unable to resolve the FNPP facility and breakwater complex, the leaking 137 Cs was treated as a point source with the same rate of release used in the high-resolution model.The resulting spread of 137 Cs predicted by Global-FVCOM differed significantly from the high-resolution model case.At 04:00 GMT 26 March, for example, the high-resolution nested model showed that the 137 Cs tracer was still around the FNPP exit, but the Global-FVCOM-computed 137 Cs con- centration was distributed symmetrically surrounding the point source and covered a much larger area with a width of ∼ 0.2 • in latitude along the coast (Fig. 12: upper panels).At 00:00 GMT 1 June, the 137 Cs concentration predicted by Global-FVCOM had spread in the entire shelf and slope region up to 43 • N, while the 137 Cs concentration computed by the high-resolution nested model remained high near the coast and no tracer was found north of 41 • N (Fig. 12: lower panels).It is clear that Global-FVCOM significantly overestimated the size of the plume.As a result, the modelcomputed 137 Cs concentrations were significantly lower than observations at both nearshore and offshore measurement sites (Fig. 13).
The high-and coarse-resolution model results can help us understand why previous modeling efforts failed to reproduce the temporal variation of the 137 Cs concentration over the shelf region.Applying a 2 km-resolution Regional Ocean Model System (ROMs) to the Japanese coast, Tsumune et al. (2012) conducted a tracer experiment to predict the 137 Cs spread over the shelf.The model did capture the 137 Cs concentration peak at MEXT-8 in mid-April, but significantly underestimated 137 Cs concentrations at the other MEXT-1 to MEXT-7 sites.Coincidentally, Estournel et al. (2012) reported that their model also underestimated 137 Cs concentrations at all MEXT sites, even though they increased the grid resolution to 600 m.They attributed this underestimation to the lack of information on the river discharge in the model which could cause a thin, low-salinity surface layer and enhance the offshore transport under the influence of wind.Our results, however, suggest that in order to reproduce the observed spread of 137 Cs over the shelf, a model needs to resolve the realistic coastal geometry of the FNPP and adjacent region.In order to reproduce the dispersion process from the leaking source to 1F-N and 1F-S, a model must be capable of resolving the complex small-scale vortex current field that controlled the water exchange or pumping around FNPP.Failure to capture this initial pumping process could lead to the unrealistic 137 Cs spreading over the shelf.
As we pointed out in the introduction, the spread of 137 Cs is controlled mainly by advection and dispersion processes.Chen et al. (2008) analytically derived the governing equations controlling the movement of the center of a small-scale dye patch in the coastal ocean.The equations indicate that after the dye is released, the movement of the dye patch is driven by the ensemble velocity integrated through the dye patch and the concentration flux related to the vertical shear of the horizontal velocity of the dye patch.Considering a dye patch that moves conservatively in the ocean, the total amount of the dye remains unchanged, but its concentration can change significantly as a result of deformation of the dye patch due to vertical and lateral dispersion that are related to velocity shears and turbulent diffusion.In order to capture the dye spreading, it is critical to resolve the realistic vertical and lateral diffusion processes.For many coastal ocean models, the horizontal diffusion is parameterized using a Smagorinsky eddy parameterization method (Smagorinsky, 1963), which depends on the model resolution and velocity shears.
Our results indicate that an underestimation of 137 Cs concentration over the shelf predicted by Global-FVCOM was mainly due to the overestimation of 137 Cs spreading in the coastal region.This overestimation was caused by insuffi- cient grid resolution to capture realistic lateral diffusion.This explanation can be applied to previous regional model simulations and emphasize the critical importance of model resolution in the parameterization of lateral diffusion.

Discussion
Like previous modeling efforts, we treated 137 Cs as a dissolved conservative tracer without including biogeochemical processes through interactions with suspended matter and releases from re-suspended sediments as well as sedimentation.The overestimation at the surface and underestimation near the bottom in the model-predicted 137 Cs concentration suggest that sedimentation processes should be included if one tries to make a more realistic prediction of the spread of 137 Cs over the shelf.This finding was also anticipated by Estournel et al. (2012) who suggested that an overestimation of the predicted 137 Cs concentration can probably be caused by ignoring a sinking term in the tracer equation.The suggestions from these modeling experiments are seemingly supported by the 137 Cs concentration levels found in the bottom sediment layer at monitoring sites along the Japanese coast.At many monitoring sites in the shelf region between the 50 m and 200 m isobaths, the observed 137 Cs concentration in sediments increased significantly with time (Fig. 14).In monitoring sites around FNPP, the sediment 137 Cs concentrations showed high values before July and then decreased rapidly with time afterward (Fig. 15).A simple estimation indicates that during April-June, the model-data discrepancy values at the surface and near the bottom were about 48 % and −39 % in the coastal area.Therefore, without adding a sinking term in the 137 Cs tracer equation, about 9 % more of the total amount of 137 Cs remained in the seawater than what was measured.Assuming these extra amounts were all deposited in the sediment through sedimentation, this means that about 9 % of the discharged 137 Cs could be removed by sedimentation.This value is much higher than the 0.1-2 % estimated by Honda et al. (2012) in traps and Kusakabe et al. (2013) in the upper 3 cm of surface sediment at monitoring sites, implying that the 9 % unbalanced amount of 137 Cs was not caused only by missing sedimentation.Since the uncertainty of 137 Cs loading at 1F-S and 1F-N was at a level of 15 % relative to the total release amount, the missed sedimentation processes in the tracer model could be at the same level as that observed by Honda et al. (2012) and Kusakabe et al. (2013).

Summary
A high-resolution, global-coastal nested ocean model was developed to simulate the initial spreading of 137 Cs over the Japan shelf after the 11 March 2011 Fukushima Dai-ichi Nuclear Power Plant failure.With sufficient resolution to resolve the complex water exchange process between the FNPP and adjacent coastal ocean, this nested model succeeded in reproducing the temporal variation and spatial distribution of 137 Cs over the shelf during the April-August 2011 period.The comparison between high-resolution nested and regional-scale models clearly showed that given the same discharge of 137 Cs, the model-predicted spreading of 137 Cs was sensitive not only to model resolution but also to geometric fitting.Failure to capture this initial dispersion process from the leaking source to the 1F-N and 1F-S monitoring sites could lead to an unrealistic prediction of 137 Cs spreading over the shelf.A coarse-resolution (∼ 2 km) regional scale model overestimated lateral diffusion and thus caused faster dispersion of 137 Cs from the coast to the deep ocean.
The 137 Cs spreading process predicted by the highresolution nested model was in good agreement with measurements over the inner shelf, but showed an overestimation at the surface and underestimation near the bottom in the offshore region.These model-data discrepancies were mainly due to an uncertainty of 137 Cs loading from sources and the assumption that treated 137 Cs as a dissolved conservative tracer without inclusion of biogeochemical and sedimentation processes.

Fig. 6 .
Fig. 6.Comparisons of model-computed and observed 137 Cs concentrations at 2F and Iwasawa over the period 26 March-8 June 2011.