The observation site, Fukushima University, is located in Fukushima city, in the northernmost basin (Fukushima Basin) in the Nakadori Valley, surrounded by the Ou Mountains to the west and the Abukuma Highlands to the east (Fig. 1). The distance of the observation site from the FDNPP is approximately 60 km. The Nakadori Valley was formed by the Abukuma River, which starts in the mountains in Fukushima Prefecture near the border of Tochigi Prefecture and flows northeast through the central parts of Fukushima city to the Pacific Ocean in Miyagi Prefecture. The major radioactive plumes arrived twice in Fukushima city, on 15 and 20 March (plume no. 3 and no. 8, as identified by Nakajima et al., 2017, respectively). These plumes were transported over the Abukuma Highlands (where the summits are mostly lower than 1000 m) but were blocked by the higher Ou Mountains (summits are 1000–2000 m) and, thus, transported along the Nakadori Valley (Nakajima et al., 2017). The land surface of Fukushima city was contaminated mainly on the afternoon of 15 March by plume no. 3. The air dose rate in Fukushima city started to increase at 17:00 LT (local time), associated with the weak rain that started at 13:00 LT, and peaked at 19:30 LT at a value of 24.0 µSvh-1.

The total deposition (dry plus wet deposition or fallout) samples were collected with a precipitation sampler (RS-20, Miyamoto Riken Ind. Co., Ltd.) with a funnel diameter of 20 cm. As a heating device was not installed on the sampler, any snow in the funnel was manually melted in a water bath in winter. The accumulated snow in the funnel never reached the top of the funnel during the whole observation period. A filtration device was installed in the sampler using membrane filters (4-880-03, Advantec) with a pore size of 0.45 µm. The radioactivities of 134Cs and 137Cs in the filtered water stored in the polyethylene bottle and those on the filters were both measured using high-purity germanium detectors at Fukushima University and were defined as the dissolved and particulate fractions of the deposition, respectively. It should be noted here that this separation does not perfectly differentiate between water-soluble and insoluble radio-Cs. The clogging of the pores of the membrane filter can occur during filtration. The measured total (dissolved plus particulate) deposition amounts were compared with those measured using the official method of Japan (MEXT, 1976) at the Fukushima Prefecture Nuclear Power Center (Fig. 1), which is located 6.5 km north-northwest of Fukushima University. Our method was found to be consistent with their official method: the correlation coefficient R was 0.81, with a slope of 1.16 (the values from Fukushima University were 16 % larger). Differences in locations and sampling intervals (daily at the Fukushima Prefecture Nuclear Power Center and monthly at Fukushima University) could also have contributed to the differences in the measured values at the two sites.

X-ray fluorescence (XRF) analysis was carried out using a RIX1000 (Rigaku Corp.) at Fukushima University. The measurement setup recommended by the manufacturer was used for the XRF. The major and trace element contents were analyzed by the fundamental parameter method and calibration curve method, respectively (Takase and Nagahashi, 2007). Measurements were conducted for PMf (see Sect. 2.2.2), the particulate fractions of precipitation (see Sect. 2.3), and the river sediments. River sediments that characterize the surface soils of the Nakadori Valley were also measured to assess the composition correlations with the airborne and deposition samples. River sediments were collected at 15 sites upstream and downstream of Fukushima city in the Abukuma River and its tributaries in 2010. Samples were taken from the gravel layer of the lower terrace at five sites, from alluvial fan deposits at one site, and from current riverbed sediments at nine sites. The dried sediment samples were sieved and divided into two grain size groups: particles smaller than 180 µm (defined as fine sediment particles) and particles 180 µm–2 mm (coarse sediment particles).

Kajino et al. (2016) used a Lagrangian model (LM) to simulate the atmospheric dispersion and deposition of 137Cs resuspended from bare soil and forest ecosystems from January to December 2013. As the resuspension fluxes and size distributions were unknown, they adjusted the flux from bare soil (forest ecosystems) so that the simulated airborne surface concentrations matched those measured in Namie (Tsushima; Namie High School, Tsushima Campus; 37.56∘ N, 140.77∘ E; 30 km northwest of the FDNPP) (Fig. 1) in the winter (summer) of 2013, and they adjusted the dry and wet deposition parameters (reflecting the size distributions and hygroscopicity) so that the simulated total (dry plus wet) deposition over land in March 2011 matched those measured by the aircraft measurements (NRA, 2012). Thus, note that the size distribution of the simulation was assumed to have submicron size ranges that were consistent with those of the primary emissions (the direct emissions associated with the FDNPP accident in March 2011) but that may not be applicable for resuspension events; the carrier aerosols are presumed to be soil dust or bioaerosols, which are usually in the supermicron size range. Kajino et al. (2016) concluded that their simulations are likely reliable because the simulated differences between the concentrations in the contaminated area (or emission source area) (i.e., Tsushima) and those in the downwind area (Meteorological Research Institute, MRI; Tsukuba city; 36.06∘ N, 140.13∘ E; 170 km southwest of the FDNPP) (Fig. 1) were consistent with the observed differences at the two locations.

However, Kajino et al. (2016) used only concentration measurements to validate the simulations. The current study also used concentration and deposition measurements from Fukushima University for model validation. The previous study compared only the two locations in the contaminated forest areas and in the downwind urban/rural regions; the current study includes an additional location in the urban/rural region near the contaminated forest of the Abukuma Highlands (60 km northwest of the FDNPP).

Figure 2 shows the time variations in the airborne surface concentrations of 137Cs from May 2011 to March 2019. Just after the accident, the 137Cs concentrations were higher than 0.01 Bq m-3, and the maximum concentration of 0.0169 Bq m-3 was detected on 23 May 2011. The concentration quickly decreased to a level of 10-4 Bq m-3, and the minimum concentration of 4.05×10-6 Bq m-3 was obtained on 5 December 2018.

It is remarkable that the decreasing tendencies in the earlier stage and the later stage were different. The regression lines of the raw data time intervals for the whole period (red; May 2011–March 2019), the earlier stage (blue; May 2011–December 2014), and the later stage (lime; January 2015–March 2019) are shown in Fig. 2, with the half-life (Th) in days and the decrease rate (Rd) in percent per year. The decreasing tendency (Th=0.753 years, Rd=92.0 % yr-1) of the earlier stage is approximately 3 times faster than that of the later stage (Th=2.07 years, Rd=33.5 % yr-1). This is shown later in Fig. 3 in Sect. 3.2 and discussed in Sect. 4.2, but it could be related to the relative abundance of the particulate and dissolved fractions of radio-Cs in the environment. The dissolved fractions of radio-Cs may discharge faster than the particulate fractions from contaminated environments, such as soils and plants. The relative abundance of the dissolved fractions was larger in the earlier stage than in the later stage such that the decreasing tendency in the concentration was faster than that in the later stage. In addition to the natural variability, decontamination work, which was completed by March 2018 in Fukushima city and the surrounding municipalities, may also have contributed to the difference in the decrease rates; this possibility is also discussed in Sect. 4.2.

The regression analysis is also performed over different time periods, but the results are not substantially different. The Th and Rd before December 2013, 2014, and 2015 are 0.670, 0.753, and 0.900 years, and 103 % yr-1, 92.0 % yr-1, and 77.0 % yr-1, respectively. The Th and Rd after January 2014, 2015, and 2016 are 2.05, 20.7, and 2.56 years, and 33.8 % yr-1, 33.5 % yr-1, and 27.1 % yr-1, respectively.

It is also interesting that our data show different seasonal variations from those measured in Tsushima by Ochiai et al. (2016) and Kinase et al. (2018). The levels in their studies were high in summer and low in winter; however, as depicted by the gray line in Fig. 2, the concentration in this work rose starting in October, with maxima in the spring season around March and minima in the summer. The maxima in the spring are approximately 1 order of magnitude larger than the minima in the summer. The measurements of their studies were conducted in high-dose areas in the mountain forest (approximately 400 m a.s.l., above sea level), and the high-volume samplers were set near the ground surface. In contrast, the current air sampling was conducted in a relatively low-dose area (10 times lower than that in Tsushima) located in an urban/rural region on a hill (approximately 200 m a.s.l.) at the southern end of the Fukushima Basin. The heights of the samples were 25 m from the ground surface. Such geographical and altitude differences could have caused these differences.

Figure 3 shows the monthly cumulative deposition of 137Cs from March 2011 to March 2019. The monthly deposition amount peaked in March 2011 at 202.2×103 Bq m-2, decreased to 1 % of the initial amount after 1 year, and decreased to an order of 1 Bq m-2 after 8 years. It also showed seasonal variation and was high from winter to spring. Nevertheless, the current level is 2–3 orders of magnitude larger than that before the Fukushima nuclear accident.

It should be noted here that the initial-stage data (i.e., March and April 2011) are excluded to obtain the decreasing rates to extract only the resuspension period and to exclude the influences of primary emissions. The decrease rates of deposition (Th=1.30–4.69 years) were generally slower than those of the concentrations (Th=0.753–2.07 years). It is tricky to identify the reason for this phenomenon. A perfect simulation could answer this question, but high uncertainties in atmospheric deposition modeling and land surface modeling inhibit a perfect understanding of these long-term circulations of radio-Cs in the environment. It is safe to presume here that the decreasing tendencies in deposition and concentrations are different because the contributions of major emission sources to deposition and concentrations are different. If the dominant source of the concentration is near (far from) the observation site and that for of deposition is far from (near) the site, the faster decrease rate in concentration is due to the faster (slower) reduction rate in the nearby sources of emissions than in the distant sources.

There is also a distinct difference in the decreasing tendencies before and after 2015. In addition to the effect of decontamination work, as previously discussed in Sect. 3.1, the relative abundances of the dissolved and particulate fractions of 137Cs could be a part of the reason. The particulate fraction made up 72.6 % of the deposition in March 2011 (Fig. 3), which is presumed to have been largely influenced by primary emissions. Here, it is interesting to note that most primary radio-Cs emissions are thought to be composed of water-soluble submicron aerosol particles (e.g., Kaneyasu et al., 2012, and almost all numerical simulations afterwards, such as Sato et al., 2020), while water-insoluble Cs-bearing microparticles (CsMP; Adachi et al., 2013; Igarashi et al., 2019b) may contribute somewhat to primary emissions (Ikehara et al., 2020; Kajino et al., 2021). If the primary radio-Cs species in aerosols were 100 % in water-soluble forms, the particulate fraction should have made up 0 % of the precipitation in March 2011 (although some of the water-soluble Cs could have been converted to a water-insoluble form through adsorption to soil particles accumulated on the membrane filter during filtration). After April 2011, as the contributions of resuspension were thought to be dominant, the dissolved fractions became larger. The ratio varied, but the dissolved fractions were generally higher before 2016, and the particulate fractions became dominant after 2016. There seemed to be a regime change in the physicochemical properties of radio-Cs circulating in the environment in the area around 2015, which could have changed the decreasing tendencies of both the concentrations and deposition before and after 2015. This result is consistent with the finding of Manaka et al. (2019), who reported that the exchangeable proportions of radio-Cs rapidly decreased in forest soils from 2 to 4 years after the accident (i.e., from 2013 to 2015).

The regression analysis is also applied over different time periods, and we found a remarkable change in 2015. The Th before December 2013, 2014, and 2015 is similar with values of 1.09, 1.30, and 1.56 years, respectively, but Th after January 2014, 2015, and 2016 is 3.98, 4.69, and 12.67 years, respectively. The ratios of half-lives (after to before) of the three periods are 3.64, 3.61, and 8.12, respectively, indicating that there could be a remarkable change in the tendency between January 2015 and January 2016. Time series of changes in the ratio before and after a particular date from 2014 to 2016 are illustrated in Fig. S1. Due to the lack of data, the half-lives after 2015 varied substantially depending on the start month (exceeding 100 years in some cases). However, it is obvious that the ratio is stable before January 2015 at a value of around 4.0, and it starts to increase from the spring to summer of 2015. We may be able to conclude that the regime change in the physicochemical properties of radio-Cs occurred during the year 2015.

The seasonal variations in particulate and dissolved 137Cs were slightly different from each other and different from those of the concentration. The concentration peaked in March in almost all years, and the total deposition peaked in January. The peaks of the total deposition in January coincided with those of the dissolved 137Cs before 2016, but the peaks of the dissolved 137Cs became unclear afterwards. The peaks of particulate 137Cs occurred in March before 2016, which coincided with those of the concentrations. After 2016, there were no clear seasonal variations in particulate 137Cs. There are clear and different seasonal variations in the concentration and deposition. However, at the current stage, we have no knowledge of or numerical tools to reveal the hidden mechanisms underlying these variations.

Figure 4 shows the time series of the seasonal mean concentrations of 137Cs obtained from the cascade impactor measurements. The sampling interval for the cascade impactor measurements was 3 weeks. The seasonal means included a sampling period if any part of the sampling period was included in the season; for example, the raw data from the sampling period from February to March contributed to the averages of both DJF (December–January–February, i.e., winter) and MAM (March–April–May, i.e., spring). The seasonal mean total (all sizes) concentrations of cascade impactor measurements during the sampling period agreed well with those of the “all size” observations as presented in Fig. 2, with R2=0.93. The same seasonal variation discussed for the all size observations (Sect. 3.1) was also observed; the atmospheric 137Cs concentration was relatively high in DJF and MAM compared with that in JJA (June–July–August, i.e., summer) and SON (September–October–November, i.e., autumn).

The most dominant size range in activity was the backup filter (<0.39 µm, or rebounded particles such as soil dust and bioaerosols; Kinase et al., 2018), and its seasonal variation agreed well with that of the total particle concentration (high in DJF and MAM). On the other hand, the second largest contribution was made by the size range of 4.2–10.2 µm, which showed the opposite seasonal variation and was relatively high in JJA and SON. The seasonal variations in the largest particle fraction, larger than 10.2 µm, are interesting: high in DJF and May (same as the backup filter) but high in JJA in 2016 and 2017. The reason for the variations and differences is not yet clear, but the current measurement indicates that the dominant particles and their sizes may be distinct depending on the season. The decrease rates of each size were different before and after approximately 2015, as discussed in Sect. 3.1 and 3.2, but the size distribution of the concentration did not change substantially before and after approximately 2015.

Cascade impactor sampling is associated with the bouncing effect, whereas filters for the impactor–cyclone system were immersed in silicone oil to prevent the bouncing effect. Thus, the cascade impactor and the impactor–cyclone measurement data are compared in Fig. 5. Figure 5a shows the data with the same measurement time intervals (3 weeks for the cascade impactor data and 1 month for the impactor–cyclone data). The concentrations of 137Cs in the backup filters were well correlated with those of PMf. No remarkable seasonality was observed in PMc, but some enhancements were observed in JJA in 2015 and SON in 2016.

Figure 5b shows the correlation coefficients among the seasonal mean size-resolved data from the cascade impactor and impactor–cyclone measurements. If we assume that the bouncing effect on the impactor–cyclone measurements was negligible, the cascade impactor data and the impactor–cyclone data were consistent. There was a positive correlation between PMf and the backup filter data. There were also positive correlations between PMc and the 1.3–2.1 and 4.2–10.2 µm data. There was a negative correlation between PMc and PMf, which strongly indicates that fine-mode particles are the dominant carriers of 137Cs in winter and spring and that coarse-mode particles are the dominant carriers of 137Cs in summer and autumn. However, there was also a contradiction in the data. There were low or negative correlation coefficients between the backup filter data and the cascade impactor data at smaller size ranges, such as 0.39–0.69, 0.69–1.3, and 1.3–2.1 µm, but the backup filter data were positively correlated with the impactor data for >10.2 µm. It appears that bouncing effect might have occurred: particles larger than 10.2 µm bounced in the latter stages and were captured in the backup filter. However, as previously discussed, the behavior of the >10.2 µm particle data was not consistent in time, i.e., they were generally high in DJF and MAM and were high in JJA in 2016 and 2017 (Fig. 4). Kinase et al. (2018) and Igarashi et al. (2019a) considered that the dominant carriers of resuspended 137Cs were coarse-mode particles such as soil dust and bioaerosols. Ochiai et al. (2016) conducted two-stage impactor sampling and measured the airborne surface concentrations of 134Cs and 137Cs above and below 1.1 µm from 2012 to 2014. They showed that the contributions of coarse-mode particles (>1.1 µm) were dominant, with maxima in summer. The contributions of the fine-mode particles (<1.1 µm) were much smaller, and no significant seasonal variations were found. All of their measurement sites were surrounded by contaminated forests in the Abukuma Highlands (Tsushima and the nearby sites), so the sampling sites were different from those in our study. Such larger particles may have contributed to the backup filter data in the current measurements; however, based on the fact that the backup filter data were positively correlated with PMf and not with PMc, fine-mode particles (<2.5 µm) should also play a key role in determining the concentrations in Fukushima city.

On the other hand, if we assume that the bouncing effect is also significant in the impactor–cyclone system due to the long sampling duration, as suggested by Okuda et al. (2015), the positive correlation between the backup filter particles and PMf was simply due to the bouncing effects of the larger particles in both systems.

Even though the emission sources of the dominant particles collected by the size-resolved measurements could not be identified in this study, the possible aerosol sources that would explain the differences in size and seasonality of the two locations are discussed later, in Sect. 4.1.

Figure 6 shows the relative abundance of the XRF-measured atomic number concentrations of elements in the PMf monthly sample from September 2014 to January 2018. Among the 15 detected species, PMf was mainly composed of SiO2, Al2O3, and SO3. The fractions of SiO2 show clear seasonal variations and were higher around May. The seasonal variations in Al2O3 and SO3 are the opposite of that in SiO2. A positive temporal correlation was obtained between the 137Cs in PMf and SiO2 (R=0.30). Negative correlations were obtained for Al2O3 and SO3, with correlation coefficients of -0.36 and -0.35, respectively. Note that these results do not prove that the SiO2-bearing aerosols are the carriers of resuspended 137Cs, but we can safely conclude that the origins of SiO2 and 137Cs may be close to each other (i.e., that both come from the same source or the same area/direction).

Figure 7 shows comparisons of the relative abundance of the periodic mean XRF-measured atomic number concentrations in different samples, fine sediment particles, coarse sediment particles, PMf, and particulate fractions of precipitation. The PMf and precipitation data over the same period, from October 2014 and December 2012, were averaged. The sediment samples were collected in 2010. The 10 species that were common to all samples are shown in Fig. 7. The composition differences are not very remarkable, and correlation coefficients for the compositions among samples are above 0.9 s. The features of the PMf composition were distinct from the others. PMf included SO3 (17.8 %) and Cl (2.65 %), whereas the others did not.

Weathered biotite is abundant in the soil in Fukushima, and it absorbs radio-Cs efficiently (Kogure et al., 2019). The compositional correlation coefficients between the weathered biotite (Takase, 2020) and the four samples were high: 0.73–0.87. However, when the two major components SiO2 and Al2O3 were excluded, the compositional correlation coefficients changed significantly. The eight (the common 10 species shown in Fig. 7 minus SiO2 and Al2O3) compositional correlation coefficients between the fine and coarse sediment particles were 0.98, but those between the sediments and PMf were 0.01 and 0.19 for the fine and coarse sediment particles, respectively. The eight compositional correlation coefficients for the particulate fractions of precipitation were moderate: 0.36, 0.44, and 0.45 for fine sediment particles, coarse sediment particles, and PMf, respectively. The eight compositional correlation coefficients for weathered biotite were 0.76, 0.71, 0.50, and -0.14 for fine sediment particles, coarse sediment particles, particulate fractions of precipitation, and PMf, respectively.

The findings from the current section are summarized as follows. The mean compositions of both fine and coarse sediment particles are similar to those of biotite, which absorbs radio-Cs efficiently. A similar composition feature was observed for the particulate fractions of precipitation. The composition of PMf was slightly different from those of the other samples, but the 137Cs concentrations in PMf became larger when the relative fractions of SiO2, the major component of biotite, increased. Thus, biotite may have played a key role in the environmental behavior of radio-Cs in Fukushima city since September 2014. However, the major carriers of radio-Cs before September 2014 and those in the dissolved fractions in precipitation are still unknown.

In Fig. 8, the concentrations of 137Cs in 2013 simulated by Kajino et al. (2016) are compared with the time-resolved observation data (Fig. 2). Kajino et al. (2016) included 137Cs resuspended from bare soil, 137Cs resuspended from forest ecosystems, and additional 137Cs emissions from the FDNPP. The additional 137Cs emissions were negligibly small in East Japan, with concentrations that were 2–3 orders of magnitude smaller than those from the two abovementioned sources; therefore, the additional 137Cs emissions are not depicted in the figure. The simulation was successful in explaining the magnitude and seasonal variations in concentrations at Tsushima and Tsukuba, but the simulation in Fukushima city disagreed with the observations. The simulation showed an enhancement of 137Cs from forests in the summer, but this was not detected in the observations. The observed magnitude and seasonal trends are rather similar to those simulated for 137Cs from soil dust.

Kajino et al. (2016) used only the observed concentrations to estimate the regional budget of resuspended 137Cs in the air, but we used the observed deposition to evaluate the model, as shown in Fig. 9. Suppose there is a simple nonlinear relationship between the deposition (D) and airborne surface concentration (C): 1D=aCb, where a represents a removal rate, and b represents nonlinearity, such as spatial and temporal variabilities. If one can take a long-term average of D and C, Eq. (1) may hold. Equation (1) is reformulated as 2log⁡(D)=blog⁡(C)+log⁡(a). The log–log scatterplot between the monthly mean concentrations and monthly cumulative deposition of observed (purple) and simulated (orange) 137Cs is depicted in Fig. 9a. The coefficient of determination of the observation was 0.68, with a risk factor <0.1 %. Equation (1) holds for the monthly mean resuspended 137Cs at Fukushima University. As seen in Eq. (2), the intercept of the y axis indicates the removal rate a. log(a) is dimensionless, but if b is close to 1, the unit of a can be meters per second (m s-1). From Fig. 9, the value of b for observation is close to 1. Therefore, the ratios of the monthly deposition amounts to the monthly mean concentrations are referred to as the climatological deposition velocity (m s-1). Time series of the climatological deposition velocity are presented in Fig. 9b. Note that the concept of climatological deposition velocity differs from that of dry deposition velocity. The dry deposition velocity is defined as the ratio of the mass flux divided by the concentrations, but this climatological deposition velocity is only the ratio of total (dry plus wet) deposition amounts divided by concentrations without the concept of mass flux. To account for the wet deposition flux, both in-cloud and below-cloud concentrations are needed, but such vertical measurement data are not available.

Figure 9a clearly shows that the removal rate (a) used in Kajino et al. (2016) can be underestimated by 1–2 orders of magnitude. The deposition velocities used in Kajino et al. (2016) were estimated from the observation of 137Cs in March 2011, which was supposed to be mainly composed of submicron water-soluble particles. However, the current study and the series of previous studies regarding resuspended 137Cs indicated that the host particles of 137Cs could be substantially larger (e.g., soil and bioaerosols). This may be the reason for the overestimation of the simulated 137Cs concentration from forests during summer in Fukushima city. If the deposition velocities of the model increased by 1–2 orders of magnitude, the transport of 137Cs from the contaminated forest to Fukushima city in summer may decrease such that the simulated concentration in Fukushima city agrees with the observation. Certainly, their simulated regional budget needs to be reassessed using the realistic deposition velocities indicated in the current study.

The observed climatological deposition velocity varied by more than 1 order of magnitude over time. There are two main deposition mechanisms: dry deposition and wet deposition. Wet deposition is associated with precipitation. The variations in the climatological deposition rate seem to agree with the observed precipitation, but almost no correlation was observed (R∼0.10). The mean climatological deposition velocity was 5.3×10-1 m s-1, and the peak values occurred in January. The maximum value was 4.9 m s-1 in January 2013, when the monthly precipitation was not very high (81.2 mm). Possible reasons for these peaks in January are discussed later, in Sect. 4.3. The typical order of the dry deposition velocity of supermicron (1–10 µm in diameter) particles is approximately 10-3–10-2 m s-1 (e.g., Petroff and Zhang, 2010), which is substantially lower than the values in our climatological deposition velocity analysis. As mentioned above, the magnitudes of the instant deposition velocity and our climatological deposition velocity are not directly comparable, but it demonstrates the efficacy of wet deposition as compared with dry deposition. Wet deposition plays an important role in the removal of resuspended 137Cs-bearing particles from the air.

Figure 10 compares the deposition amounts of 137Cs at Fukushima University (60 km northwest of the FDNPP) and the MRI (170 km southwest of the FDNPP) from March 2011 to March 2019. The deposition data at the MRI are available from Environmental Radioactivity and Radiation in Japan (https://www.kankyo-hoshano.go.jp/data/database/, last access: 14 June 2021). There was a significant positive correlation between the deposition amounts of 137Cs at the two sites, but the ratios varied substantially over time. Figure 10b indicates that the deposition ratios at the two sites were approximately 10, which is almost the same level as the initial amounts (202×103 Bq m-2 at Fukushima University and 17.6×103 Bq m-2 at the MRI), with a variation of more than 1 order of magnitude and peaks in winter (especially January) that decreased slightly over time. Figure 9b shows that the January peak is typical in Fukushima city and was not observed at the MRI. The possible reasons for the January peak in Fukushima city are discussed later, in Sect. 4.3. The slight decreasing tendency was probably due to decontamination, which was ongoing in Fukushima during the period until 2018, as shown later in Table 1. Certainly, natural variations could also have contributed to the decreasing tendency.

The current study clearly shows that the concentrations of 137Cs are high from winter to spring, with peaks in March, and low from summer to autumn in the urban/rural area of Fukushima city (60 km northwest of the FDNPP). It also shows that the deposition amounts of 137Cs are high in the winter, especially in January, and low from summer to autumn. This seasonal trend is the opposite of that observed in a forested area in the Abukuma Highlands (Tsushima, 30 km northwest of the FDNPP), which was high in the summer (Ochiai et al., 2016; Kinase et al., 2018). From winter to spring, northwesterly winds prevail over the region associated with migrating disturbances, while southeasterly winds prevail over the region associated with the Pacific high pressure system. The three simulated monthly mean surface wind fields for January to March and June to August are shown in Kajino et al. (2016).

In summer, Fukushima city is downwind of Tsushima. The concentrations of 137Cs at Tsushima are approximately 10 times greater than those in Fukushima city, but there is no enhancement of 137Cs in summer. The traveling distances of carrier aerosols depend on their aerodynamic diameters. The distance between the two sites is approximately 30 km. The traveling distances of aerosols below <10 µm are not very different and are larger than 100 km because their gravitational deposition velocities are negligibly small. On the other hand, the traveling distances rapidly decrease proportionally to a square of the diameter above 10 µm, and the traveling distance of an aerosol with a diameter of several tens of micrometers is on the order of 10 km (Kajino et al., 2021). Igarashi et al. (2019a) reported that the major proportions of bioaerosols in forests in summer are smaller than 5 µm in diameter and can travel a fairly long distance. Pollen is much larger than 10 µm, but pollen emission is limited in summer (Igarashi et al., 2019a). Consequently, there was a significant enhancement in concentrations in the forests in summer but no enhancement in the downwind urban/rural areas, probably because the carrier aerosols were efficiently deposited onto the ground surface by wet deposition in addition to dry deposition before significant amounts of atmospheric 137Cs reached the downwind areas. Consistent with our findings presented in Fig. 5, 137Cs in PMc was more abundant than that in PMf and the backup filter particles in summer. To obtain a quantitative understanding of the regional cycle of atmospheric 137Cs in the northern part of Fukushima Prefecture, accurate simulations are required in the future.

In winter and spring, the concentrations of 137Cs are probably enhanced due to local emissions from nearby sources because the location of the sampling site is upwind of the Abukuma Highlands and the ground surface in areas upwind of the sampling site in the season (northwest directions) is less contaminated than the site. In winter and spring, 137Cs in the backup filter particles and PMf is pronounced in Fukushima city. These characteristics are somewhat different from those reported in previous studies. Miyamoto et al. (2014) measured the size distributions of radio-Cs with a cascade impactor for two periods, from 17 March to 1 April and from 9 to 13 May 2011, at a site 120 km southwest of the FDNPP. They showed that the peak size ranges were 1.2–2.1 and 0.65–1.1 µm in the former and latter periods, respectively. Doi et al. (2013) reported that the peak diameters of the 137Cs concentration from 4 to 11 April were 1.0 and 1.5 µm at Tsukuba, 170 km southwest of the FDNPP. Kaneyasu et al. (2017) measured the size distributions of 137Cs and other chemical components six times at Tsukuba from April to September 2011. The peak diameter ranges were 0.49–0.7 µm in the earlier stages (before 9 June), but the contributions of coarse-mode particles (>1 µm) increased after 9 June, and the second modes appeared in the ranges of 3.5–5.2 and 7.8–11 µm in July and September, respectively. Judging from their measured mass size distributions of Ca, which is assumed to originate from mineral soil, Kaneyasu et al. (2017) concluded that soil particles could be the major carrier of resuspended radio-Cs in Tsukuba. Our XRF analysis indicated that radio-Cs is carried mainly by soil particles in Fukushima city, but the size distributions are greater in PMf (<2.5 µm) and in the backup filter particles. If radio-Cs is carried by soil particles, it is natural to presume that the fractions of radio-Cs in PMc would be large (e.g., Fig. 3 of Kaneyasu et al., 2012, or Fig. 4 of Kaneyasu et al., 2017). One could argue that the bounced coarse-mode soil particles are observed in the backup filters, but, in fact, the seasonal mean 137Cs concentrations in the backup filter are positively correlated with PMf and negatively correlated with PMc (Fig. 5). One could further argue that bounced large particles are also collected in PMf despite the special procedures employed to prevent rebound in the impactor–cyclone system.

There are four possible explanations for these results:

If the bouncing effect did not occur in either system, the major sources of radio-Cs in Fukushima city are probably related to combustion (a mass peak below 0.39 µm means that the number peak is approximately 100 nm).

As described in Sect. 3.1 and 3.2, distinct decrease rates were observed before and after approximately 2015 in both the concentrations and the deposition. There may be two main reasons for this: natural variation and decontamination. Natural variation (the dissolved fractions of precipitation or the exchangeable proportions of forest soils discharging faster than other forms from the local ecosystems; Manaka et al., 2019) has been previously described, and the effect of decontamination is presented in some detail here.

Table 1 summarizes the achievement ratios of the scheduled decontamination of different land use types in Fukushima city and the surrounding municipalities (available at https://www.pref.fukushima.lg.jp/site/portal/progress.html, last access: 14 June 2021). The municipalities are in the northern part of Fukushima Prefecture, which comprises 55 % forest area, 15 % farmland area, 6 % residential area, and 23 % other areas (https://www.pref.fukushima.lg.jp/uploaded/attachment/42042.pdf, last access: 14 June 2021). More than 94 % decontamination was achieved for the farmland area by March 2014. For the residential and public facility areas, some parts were decontaminated by March 2014, but others were not fully decontaminated until March 2018. For the road and forest areas, decontamination was not completed in most areas by March 2014, but extensive decontamination was conducted from 2014 to 2018. Note that only a part of the forest (20 m from the forest edges) was decontaminated, which accounts for approximately 1 % of the whole forest area of the northern part of Fukushima Prefecture. Additionally, only the litter layer of the forest was removed, and the soil layer remained.

If one assumed that contamination occurred independently of the land use type, approximately 30 % (farmland and half of the residential and other land types) of northern Fukushima was decontaminated by 2014, and an additional 15 % (half of the residential and other land types) was continuously decontaminated by 2018. The difference between the decrease rate from May 2011 to December 2014 (93.1 % yr-1) and that afterwards (30.7 % yr-1) was higher than the decontamination rate (30 %–45 % every 3–7 years). If the concentration at Fukushima University was affected mainly by the emissions from nearby sources (i.e., within the northern part of Fukushima Prefecture), decontamination would not be the sole reason for the change in the decrease rates before approximately 2015 and after this period. Natural variation (i.e., regime changes in the chemical forms of radio-Cs) would likely have occurred during that period. As previously discussed, biotite may have played a key role in the environmental behaviors of radio-Cs in Fukushima city after approximately 2015, but the current study could not identify the key aerosol particles that carried dissolved (or exchangeable) radio-Cs and were abundant in Fukushima city before approximately 2015.

The climatological deposition velocities (or the ratios of the deposition rate to the mean airborne surface concentration) in Fukushima city were remarkably high in January 2013, 2014, 2015, and 2017 (Fig. 9). They were approximately 1 order of magnitude larger than those in the other months. The ratio of the deposition in Fukushima to that at the MRI was approximately 10 on average, but the ratios in January of those years exceeded 100 (Fig. 10). On the other hand, no peaks were observed in January of 2012, 2016, or 2018.

There are two possible explanations for these results: vertical distribution and the existence of superlarge particles. In terms of the former, the substantial proportions of 137Cs in the upper air may have caused lower concentrations but higher deposition due to the wet removal of 137Cs aloft. However, due to the northwesterly winter monsoon, the upper air over Fukushima city is also upwind of the Abukuma Highlands; thus, this possibility is less likely. In terms of the latter, superlarge particles (∼100 µm or larger in diameter) have settling velocities that are too high (as high as those of drizzle droplets) to enter the high-volume air sampler but that allow them to settle efficiently in a deposition sampler. Unfortunately, however, analyses of the surface meteorological observational data for Fukushima city from the JMA, such as temperature, precipitation, snow cover, and wind speed data, did not reveal the differentiating features between the years with (2013, 2014, 2015, and 2017) and without (2012, 2016, and 2018) high deposition peaks.