Quantifying the dilution of the radiocesium contamination in Fukushima coastal river sediment (2011–2015)

Fallout from the Fukushima Dai-ichi nuclear power plant accident resulted in a 3000-km2 radioactive contamination plume. Here, we model the progressive dilution of the radiocesium contamination in 327 sediment samples from two neighboring catchments with different timing of soil decontamination. Overall, we demonstrate that there has been a ~90% decrease of the contribution of upstream contaminated soils to sediment transiting the coastal plains between 2012 (median – M – contribution of 73%, mean absolute deviation – MAD – of 27%) and 2015 (M 9%, MAD 6%). The occurrence of typhoons and the progress of decontamination in different tributaries of the Niida River resulted in temporary increases in local contamination. However, the much lower contribution of upstream contaminated soils to coastal plain sediment in November 2015 demonstrates that the source of the easily erodible, contaminated material has potentially been removed by decontamination, diluted by subsoils, or eroded and transported to the Pacific Ocean.

dilution of radiocesium contamination in sediment transiting rivers draining the main plume of fallout from the FDNPP accident.
Accordingly, a modified sediment tracing technique is used to quantify the relative contributions of contaminated and depleted radiocesium sediment sources in two paired catchments in the Fukushima region. The paired catchments include the Niida catchment (275 km 2 ), where decontamination started in 2013, and the neighboring Mano catchment (175 km 2 ), where decontamination works started in 2014-2015. The only significant difference between these catchments is a major dam located between the upstream -more contaminated -section of the Mano catchment and the coastal plain -where soils contain very low radiocesium concentrations (Fig. 1). In contrast, the only major dam in the Niida catchment is situated on a tributary and it will therefore have a limited impact on sediment transport in the main river stem.
Due to the proximity of the Niida and Mano catchments, we hypothesize that they are exposed to similar climatic conditions and therefore potential differences in sediment radiocesium contamination in their upper reaches will reflect the different timing of the implementation of the remediation works. The validity of this hypothesis is supported by the fact that initial radiocesium soil contamination levels were the highest in upper parts of both catchments and lowest in their coastal plains. Furthermore, both catchments were shown to receive similar rainfall erosivity with similar spatial distributions of rainfall during major typhoon events 17 . Distribution models were used to quantify the relative contributions from the two end-members (i.e. contaminated and depleted sources) to sediment in these two catchments at 41 sites that were sampled every 6 months from November 2011 to November 2015.

Results
Spatial patterns of decontamination works in the Niida and the Mano catchments. Remediation started late in 2013 in the headwaters of the Niida catchment, along the main stream of the main Niida tributary (Fig. 2). Thereafter, the decontamination works moved progressively downstream along this tributary. Decontamination expanded along the Iitoi River, the central tributary of the Niida, in 2014. In the southern tributary, the Hiso River, the majority of the decontamination started in 2015.
In the Mano catchment, remediation works started in late 2014. As remediation works started only in late 2014 in this catchment, variations in initial contamination proportions observed in upper parts of the Mano catchment between autumn 2011 and spring 2014 should reflect variations in erosion intensity and sediment supply due to heavy rainfall, rather than soil erosion that may have been accelerated during initial decontamination operations.

Evolution of contamination levels in the Niida River. Contamination of sediment transiting the Niida
River decreased from 2011 (median -M -65% of contribution from the upstream contaminated soils) to 2015 (M 10%; Fig. 3). The progressive transfer of contaminated material from the upper catchment to the coastal plain -where it may be stored temporarily in the channel -is reflected by the higher contribution of upstream soils recorded in the downstream river sections in autumn 2013 (M 41%, median absolute deviation -MAD 22%) and in spring 2015 (M 37%, MAD 28%). After a first flush of contamination during the year that followed the accident (from a maximum median contribution of upstream soils of 40% to a minimum of 16%, in spring 2012 (see Fig. 3 Evolution of contamination levels in the Mano River. The sediment disconnectivity induced by the Mano Dam (with the exception of a major dam release that occurred in 2011) provides an opportunistic method to estimate the decrease in upstream contaminated soil median contributions (from 68% to 6%) that  would have been observed in the coastal plains of the Niida catchment in the absence of continuous supply of contaminated material originating from upper catchment parts (Fig. 4). After spring 2012, the median contribution of these upstream contaminated soils remained systematically below 20% in the coastal plains, although two low-magnitude peaks of contamination were observed in autumn 2013 (M 16%) and in spring 2015 (M 10%; Fig. 4). When comparing the Niida and Mano contamination level evolution, it becomes very apparent that a large quantity of contaminated sediment is likely stored in the Mano Dam reservoir.

Evolution of contamination levels in sediment collected in upstream tributaries of the Niida River.
When the model results are grouped for each upstream tributary of the main Niida River, the evolution of the contaminated soil contribution is different than what was observed for the entire upstream Niida catchment (Fig. 5A). The median contribution of contaminated soils to sediment increases from the north to the south (from a median maximum of 69% in the upstream Niida mainstream to 98% in the Hiso River), reflecting the spatial pattern of the main radioactive pollution plume. Peaks of contaminated soil contributions are observed during spring 2014 (M 99% in the Iitoi River) and spring 2015 (M 98% in the Hiso River; M 63% in the Niida River). These peaks occurred after the majority of remediation works in paddy fields connected to these rivers was completed. For both the Niida and Iitoi Rivers, this peak in contamination is followed by a sharp decrease of the contribution of contaminated soils to sediment and their associated uncertainties.

Discussion
These results illustrate the strong spatial and temporal variability of radioactive contamination levels of sediment transiting the Niida and the Mano Rivers between 2011 and 2015. This is likely explained by the occurrence of various erosion processes in the catchment and the collection of sediment deposits in different hydro-sedimentary contexts. Despite these spatial differences and the uncertainties associated with the model results, there is a major decrease in sediment contamination levels during this period. This trend is relevant with other observations reported in the literature 6,18,19 .
In particular, during the first two years that followed the nuclear accident, it was estimated that 1-5% of the initial catchment radiocesium inventory was exported by Fukushima coastal rivers to the Pacific Ocean 19,20 . This trend corresponds to the sharp decrease in the proportion of contaminated soil contribution modelled in sediment transiting the Niida and the Mano Rivers in this study. This observation is consistent with the typical radionuclide wash-off behaviour expected after a nuclear accident, with a short term intense 'leaching' phase during the weeks and months following the radionuclide emissions 21,22 .
Thereafter, the export of radionuclides bound to sediment is much slower, and should not exceed 1% of the initial radiocesium inventories in soils according to the estimations available from the literature 6 . Our modelling indicates that the decrease in the contribution of contaminated soil material found in sediment slowed down from 2012 onwards. However, in contradiction with the theoretical models describing the radionuclide wash-off with a negative exponential form, our results show the periodic increase in contamination levels following heavy flooding events, such as typhoons or spring floods (Table 1).
There were sediment-bound radiocesium peaks evident that were directly related to decontamination works. Although this increase is associated with large uncertainties, our modelling results suggest that remediation efforts may lead to a temporary increase in these levels during or shortly after the start of these operations. This may be explained by the fact that vegetation is removed to prepare these works 11 , which leaves the soils unprotected during heavy rainfall and enhances the production of overland flow. This observation had already been made in upper parts of the Niida River 16 . The large uncertainties may be explained by the integration of recently eroded soils originating from the remediated fields with material transiting in the river. However, when the investigation is prolonged during several seasons following the start of the remediation works, contamination levels in nearby river sediment decrease sharply, illustrating the removal of the easily erodible radiocesium source. According to the model results, decontamination is therefore likely efficient over the medium term, once areas have been completely remediated. Indeed, the uncertainties associated with the modelled contributions of contaminated soils to sediment decrease with time, which likely reflects the export of the most contaminated material to the Pacific Ocean and the progressive homogenization of the contamination level of the material transiting these coastal rivers.
In general, Fukushima river systems appear to be very reactive to both rainfall events (Table 1) and the radiocesium remediation (Fig. 5). This behaviour is likely explained by the high sediment connectivity of paddy fields, which were shown to be the main sources of contamination to the rivers shortly after the accident 23 . Transfers of radiocesium-bound sediment from paddy fields are exacerbated during different periods of the year, depending on the rice cultivation cycle. For instance, the puddling period, when the paddy soil is mixed with irrigation water down to a depth of 15 cm, is a very erosive period 24 . Together with the occurrence of a flood in early March in 2015 (Fig. 6), this probably explains the high contributions of contaminated soils modelled in sediment collected in spring 2015. In addition, runoff during typhoons that may occur at the end of the cultivation period may also trigger export of contaminated soil from paddies, as modelled in autumn 2013.
The very low proportions of contaminated soil median contributions modelled in sediment collected in November 2015 (6% in the Mano coastal plains vs. 10% in the Niida) indicate that the most erodible contaminated soils have already been flushed away to the Pacific Ocean. Although contaminated material is likely stored   17 .
for a long time in the major dam in the Mano River catchment, the supply of contaminated sediment is more and more unlikely in catchments where remediation works have been completed. In future, the potential contribution of forests as a perennial low-magnitude source of radioactive contamination to the river systems should be investigated, as forests cover two-thirds of the surface area in this region. Although both subsurface sources were also modeled as a depleted radiocesium source owing to the paired catchment design, the relative contributions of subsurface sources and decontaminated sources should be examined through the application of a different suite of appropriate fingerprints.
Catchment land use mainly consists of forest (75% SD 4%) and cropland (21% SD 4%) 10 . The main catchment features include an upstream coastal mountain range (< 900 m) and a broad, more densely inhabited, coastal plain (i.e. < 100 m) that occupies on average 25% (SD 6%) of both catchments. Soils in the upstream areas of these catchments were heavily contaminated, with soil radiocesium ( 137 Cs) inventories ranging from 20 kBq kg −1 to 75 kBq kg −1 . In contrast, soil radiocesium inventories in the lowland coastal plains were less than 20 kBq kg −1 7 . Maps showing the progress of remediation works in the study area were developed from field observations and a compilation of data available from the local municipality (Iitate Village) (Fig. 2).
Cumulative rainfall reached 5057 mm at a gauging station in the Niida catchment monitored between March 2011 and December 2015. This monitoring period included six typhoons and one tropical storm that resulted in > 100 mm of rainfall. Mean annual rainfall in the Fukushima region is ~1400 mm 17 . Sampling. Two approaches were used to develop a 137 Cs source dataset for modelling. First, soil samples (n = 160) were collected between 2011-2015 in locations reported to be highly connected to the stream network 7 . At each of these locations, ten subsamples (~5 g per subsample) were scraped from the soil surface randomly in a 10 m 2 quadrant using a non-metallic trowel and composited into one sample. Second, 137 Cs activities were included for 100 soil samples collected in these two catchments in June and July 2011 by the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) 10  Nine sediment sampling campaigns were conducted between November 2011 and November 2015. Sampling occurred bi-annually at 41 locations: in fall after the typhoon season, and in spring, after the snowmelt runoff. The goal was to sample deposited sediment that was transferred during these main erosive periods. In total, 327 sediment samples were obtained. Fine sediment samples were taken from material deposited after the last major event at the same sites, during each of the nine campaigns. These lag deposit samples were comprised of fine particulate material that settled on channel banks, inset benches and floodplains during the falling limb of the last significant hydro-sedimentary event. Ten subsamples (~5 g per subsample) of recently deposited material were taken with a plastic spatula over a 5 m reach and composited into one sample.
Laboratory analyses. All samples were dried at 40 °C for ~48 h, sieved to 2 mm and ground to a fine powder in an agate mortar, and pressed into 15 mL polyethylene containers for measurement. 137 Cs activities were determined with gamma spectrometry using coaxial N-and P-type HPGe detectors (Canberra/Ortec). 137 Cs activities were measured at the 662 keV emission peak. More information on these methods can be found elsewhere 7 . All activities were decay-corrected to March 14, 2011, the date of the main radionuclide fallout deposition 25 . Distribution modelling. A distribution modelling approach was used to incorporate distributions throughout the entire modelling framework, including the source contribution terms 26 . Distribution models were used to quantify the relative contributions from two end-members, i.e. soils contaminated and depleted in radiocesium ( 137 Cs). A summary of radionuclide activities characterizing both source types in the Niida and Mano catchments is provided in Table 2. A summary of the model results is given in Table 3.
For the two-source model, it is assumed that sediment samples constitute a discrete mixture of their sources, with the source contribution of A being x, and the contribution of B being 1−x where: C is the in-stream sediment distribution, A and B are the two source distributions, and x is modelled as a truncated normal distribution (0 ≤ x ≤ 1) with a mixture mean (μ m ) and standard deviation (σ m ) 27 . The model is solved by minimizing the median difference between the distributions of both sides of Eq. 1 (i.e. C and Ax + B (1 − x)) with the Optquest algorithm in Oracle's Crystal Ball software with a randomly generated mixture mean (μ m ) and standard deviation (σ m ). The two source distributions were fit with the Crystal Ball software which selected the optimal distribution based on the results of three statistical tests: Anderson-Darling, Kolmogorov-Smirnov and Chi-Square. For the depleted source distributions in both catchments, log-normal distributions were modeled and for the contaminated sources, Maximum Extreme Distribution (also known as Gumbel distributions) were modeled.
Source contributions (x) were determined by simulating Eq. 1 with 2500 Latin Hypercube (500 bins) samples drawn from source and sediment distributions and varying the mixture mean (μ m ) and standard deviation (σ m ). This model simulation and solving process was then repeated 2500 times with the median proportional source contribution from these 2500 additional simulations, reported as the contribution of each source. Model uncertainty was calculated through summing three quantifiable model uncertainties: (1) the median absolute deviation of the individual source median contribution for the additional 2500 simulations; (2) the modelled standard deviation; and (3) the median absolute deviation of this modelled standard deviation for the 2500 model additional simulations 28 .