Atmospheric removal times of the aerosol-bound radionuclides 137 Cs and 131 I measured after the Fukushima Dai-ichi nuclear accident – a constraint for air quality and climate models

Caesium-137 ( 137Cs) and iodine-131 ( 131I) are radionuclides of particular concern during nuclear accidents, because they are emitted in large amounts and are of significant health impact.137Cs and131I attach to the ambient accumulation-mode (AM) aerosols and share their fate as the aerosols are removed from the atmosphere by scavenging within clouds, precipitation and dry deposition. Here, we estimate their removal times from the atmosphere using a unique high-precision global measurement data set collected over several months after the accident at the Fukushima Dai-ichi nuclear power plant in March 2011. The noble gas xenon-133 ( 133Xe), also released during the accident, served as a passive tracer of air mass transport for determining the removal times of137Cs and131I via the decrease in the measured ratios137Cs/133Xe and131I/133Xe over time. After correction for radioactive decay, the 137Cs/133Xe ratios reflect the removal of aerosols by wet and dry deposition, whereas the 131I/133Xe ratios are also influenced by aerosol production from gaseous131I. We find removal times for137Cs of 10.0–13.9 days and for 131I of 17.1–24.2 days during April and May 2011. The removal time of 131I is longer due to the aerosol production from gaseous 131I, thus the removal time for 137Cs serves as a better estimate for aerosol lifetime. The removal time of 131I is of interest for semi-volatile species. We discuss possible caveats (e.g. late emissions, resuspension) that can affect the results, and compare the 137Cs removal times with observation-based and modeled aerosol lifetimes. Our137Cs removal time of 10.0-13.9 days should be representative of a “background” AM aerosol well mixed in the extratropical Northern Hemisphere troposphere. It is expected that the lifetime of this vertically mixed background aerosol is longer than the lifetime of fresh AM aerosols directly emitted from surface sources. However, the substantial difference to the mean lifetimes of AM aerosols obtained from aerosol models, typically in the range of 3–7 days, warrants further research on the cause of this discrepancy. Too short modeled AM aerosol lifetimes would have serious implications for air quality and climate model predictions.


Introduction
During nuclear accidents, radionuclides can be released into the atmosphere and transported over long distances.Caesium-137 ( 137 Cs) and iodine-131 ( 131 I) are the radionuclides of greatest concern, because they are highly volatile and therefore quickly released into the environment, can be easily measured and constitute a significant risk to human health (e.g.Balter, 1996;Bleuer et al., 1997;EPA, 2012).While 131 I, with a half-life period of 8.02 days, is important for short-term exposure of the population, 137 Cs determines the long-term effect of a nuclear accident due to its longer half-life period (30 yr).Therefore, transport and removal of these species from the atmosphere are of major importance, as seen after the Chernobyl accident (NEA, 2002).These radionuclides attach mainly to the ambient accumulation-mode (AM) (∼ 0.1-1 µm diameter) aerosols and share their fate during transport and removal from the atmosphere (Chamberlain, 1991).Thus, studies of these radionuclides are not only of interest per se, but can also be used to evaluate the Published by Copernicus Publications on behalf of the European Geosciences Union.

N. I. Kristiansen et al.: Atmospheric removal times of aerosols
behavior of AM aerosols which are detrimental for air quality and influence the global climate (Friedlander, 1977;Seinfeld and Pandis, 1998).
The radionuclide 137 Cs attaches mainly to the inorganic AM aerosol fraction, e.g.ammonium, sulfate and nitrate (Jost et al., 1986).For the emissions of radionuclides from the Fukushima Dai-ichi nuclear power plant (FD-NPP) accident in March 2011, there is direct evidence that the 137 Cs was attached to aerosols in the size range 0.1-2 µm diameter, identical to that of simultaneously measured sulfate aerosols (Kaneyasu et al., 2012).This was found by comparing the measured activity size distributions of 137 Cs to the mass size distribution of several aerosol components.An overlap of the two size distributions was found for non-sea-salt sulfate aerosols suggesting that 137 Cs primarily traced the fate of sulfate aerosols.These aerosols grow by coagulation with other particles during transport (Jost et al., 1986) and are removed by wet and dry deposition.Thus, the removal rates of 137 Cs should be representative for the AM aerosols in general.If 137 Cs removal times can be determined from measurement data, this provides also a valuable constraint on the AM aerosol lifetime, for which otherwise few observational constraints exist. 131I attaches to AM aerosols as well but, in contrast to 137 Cs, is released both as gas and in particulate form.The gaseous release fraction is typically as high as the particulate fraction.For the FD-NPP emissions about 70 % of the released 131 I was gaseous (Masson et al., 2011).During transport, there is an exchange between the gas and particle phases.The gas-to-particle conversion for 131 I typically occurs on time scales of about 2-3 weeks (Masson et al., 2011;Uematsu et al., 1988).Therefore, 131 I is less suitable for tracing the fate of non-volatile AM aerosols but can still impose upper limits on the AM aerosol lifetime.In addition, 131 I can be of interest also for considerations of the fate of semi-volatile species in the atmosphere.
Reported aerosol lifetimes derived from 137 Cs and other radionuclides produced by cosmic rays, radon decay or nuclear bomb tests vary from 4 days to more than a month (Giorgi and Chameides, 1986), reflecting the different origin (e.g., surface or stratospheric) of radionuclide tracers.Aerosol residence times of ≤ 4 days in the lower troposphere and ≤ 12 days in the middle to upper troposphere may be seen as typical (Moore et al., 1973) but higher values of eight days for the lower troposphere have been reported as well (Papastefanou, 2006).Following the Chernobyl nuclear accident, the exponential decline of the 137 Cs concentrations indicated a residence time of 7 days (Cambray et al., 1987).Other observation-based methods not using radionuclide data suggest aerosol lifetimes from a few days to about one month in the troposphere (Williams et al., 2002;Paris et al., 2009;Schmale et al., 2011).Models give global average residence times of AM aerosol in the atmosphere on the order of 3-7 days for species emitted mainly from the surface (Chin et al., 1996;Feichter et al., 1996;Stier et al., 2005;Berglen et al., 2004;Liu et al., 2005;Bourgeois and Bey, 2011;Chung and Seinfeld, 2002;Koch andHansen, 2005, Textor et al., 2006), which is rather shorter than the lifetimes obtained in most observation studies.
In this study we derive removal times for 137 Cs and 131 I in the atmosphere by using concentration measurements in a global network of background radionuclide monitoring stations operated by the Preparatory Commission for the Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO; see Fig. 1) in Vienna.These measurements are unique, since the stations are globally nearly uniformly distributed, the data are globally inter-calibrated, and their high accuracy allows quantifying the radionuclide activity concentrations over several orders of magnitude (Schulze et al., 2000;Werzi, 2009).We used CTBTO measurements taken during three months following the accident at the Fukushima Dai-ichi nuclear power plant (FD-NPP) in March 2011 (Stohl et al., 2012a), which released a pulse of radionuclides to the atmosphere.The accident had been triggered by an earthquake on 11 March at 05:46 UTC and a related tsunami one hour later.While the earthquake led to an automatic emergency shutdown (scram) of the three running reactor blocks and the complete loss of off-site power, the tsunami caused the failure of the emergency cooling systems.Consequently, there was a rapid melt-down of the reactor cores and a massive injection of radionuclides into the atmosphere.
During the accident, the whole inventory of the noble gas 133 Xe was set free from the reactors.Xenon measurements are available from a subset (about 50 %) of the CTBTO stations measuring 137 Cs and 131 I.While both 137 Cs and 131 I attach to the ambient AM aerosol, the noble gas 133 Xe, with a half-life of 5.25 days, can be considered as a passive tracer of air mass transport.Thus, the aerosol removal times can be directly determined, to the authors' knowledge for the first time, by the decrease in the concentration ratios between two aerosols ( 137 Cs, 131 I) and a passive tracer ( 133 Xe) as a function of time.We compare the obtained removal times with observation-based and modeled aerosol lifetimes and discuss the implication of using the removal times as an estimate of AM aerosol lifetime.

Data and methods
We have used measurements of atmospheric activity concentrations of the noble gas 133 Xe and the aerosol-bound radionuclides 137 Cs and 131 I available from several stations operated by the CTBTO covering the whole Northern Hemisphere (Fig. 1).The stations are part of the International Monitoring System built up during the last 15 yr to measure signals (seismic, hydroacoustic, infrasound and radionuclides) from underground or atmospheric nuclear explosions.Measurements in the time period August 2010 to December 2011 were used for this study.The CTBTO stations are equipped with high-volume aerosol samplers.About 20 000 m 3 of air is blown through a filter, collecting particulate radionuclides with a collection period of 24 h. 137Cs, 131 I and other aerosol-bound radionuclides are measured with high-resolution germanium detectors (Schulze et al., 2000;Medici, 2001).The minimum detectable activity concentration (MDC) of 137 Cs varies for the different stations, but is on average about 1 µBqm −3 .A similar MDC is obtained for 131 I. CTBTO stations, however, only measure the aerosol-bound 131 I fraction, because the gaseous fraction is not trapped by the currently used filters.About 50 % of the CTBTO stations are also equipped with xenon detectors which measure four radioxenon isotopes.The isotope 133 Xe is the most prevalent and important one and therefore used here.The measurement accuracy for 133 Xe also varies for the different stations, but is on average about 0.1 mBqm −3 , and the collection period is typically 12 h.Atmospheric background levels of 137 Cs, 131 I and 133 Xe were defined as the mean activity concentrations before the FD-NPP accident (August 2010 to 11 March 2011), and subtracted from all measurement values after the FD-NPP accident.The impact of this correction was negligible at most stations because of very low background values which rarely exceed the level of detection.The measured 133 Xe (half-life 5.25 days), 137 Cs (half-life of 30 yr) and 131 I (half-life 8.02 days) enhancements over the background were corrected for radioactive decay to the time of the earthquake.Figure 2 shows an example of the uncorrected and corrected time series data for the station Oahu.It is seen that the emission pulse of 137 Cs and 131 I is observable at this station until late May.The background of all three radionuclides is very low so the effect of the background subtraction is negligible for 137 Cs and 131 I while a small effect can be seen for 133 Xe in late May when the enhancements over background are slightly lower than the uncorrected values.
The low background in combination with the high measurement sensitivity facilitates quantification of the radionuclides over a period of almost three months.This is long compared to the period of major emissions of about eight and four days for 137 Cs and 133 Xe, respectively (Stohl et al., 2012a).Emissions of 131 I had a similar temporal behavior as 137 Cs (Chino et al., 2011;Katata et al., 2012a), thus allowing us to consider all the radionuclide emissions as one single pulse.Although the 133 Xe emissions ceased before the 137 Cs and 131 I emissions, the highest emissions occurred during the same three days (Stohl et al., 2012a;Katata et al., 2012a), ensuring a high level of correlation between the radionuclides.Thus, the removal of 137 Cs and 131 I can be gauged against an inert noble gas ( 133 Xe) tracer.
We use two different approaches to estimate the removal times of the aerosol-borne radionuclides.For both approaches the emissions of 137 Cs, 131 I and 133 Xe are treated as a single pulse, referenced to the time of the earthquake, allowing to determine the age [days] of the air mass containing the radionuclides when it reached each CTBTO station.Since all measurement samples use the same reference date, the removal time calculation does not depend on the assumed The first approach uses a multi-box model to estimate the total atmospheric burden of 137 Cs, 131 I and 133 Xe (Stohl et al., 2012b).This method uses all available measurements of 137 Cs, 131 I and 133 Xe from all stations (Fig. 1).This is contrary to our second approach which relies on co-located measurements of the three components.Such co-located measurements are only available for 11 of the stations.If we assume that the measured 137 Cs concentrations at the ground are representative for the depth of the tropospheric column and for the latitude band a certain station is located in, the total atmospheric burden of 137 Cs follows from where N is the number of stations (latitude bands) used, A i the area of latitude band i, H i its tropospheric scale height, and 137 Cs i the decay-corrected enhancement over the background at station i, averaged over a suitable time interval (here 4 days).Likewise, the calculation is done for the total atmospheric burdens of 133 Xe and 131 I. Using meteorological analysis data from the Global Forecast System (GFS) model of the National Centers for Environmental Prediction (NCEP), monthly mean tropospheric scale heights were obtained by dividing the air column density up to the last pressure level below the tropopause height with the surface density.The box model extended from 20 • S to 90 • N; latitudes of a few stations were shifted by a maximum of 3 • latitude, to reduce clustering of stations at particular latitudes.The box model assumption that radionuclides are relatively well mixed within latitude bands is not fulfilled during the first weeks after the accident, so it was only used from 1 April.The results were not sensitive to variations of this date.
The second approach takes direct advantage of co-located measurements of 137 Cs, 131 I and 133 Xe without using the multi-box model.Only stations measuring all three radionuclides were used in this approach (Fig. 1).The 12-hourly 133 Xe data (after background subtraction and decay correction) were linearly interpolated to the sampling times (24 h) of the 137 Cs or 131 I data in order to calculate the ratio of the radionuclides.In addition to using all measurement data points individually, we also calculated the ratios of the mean activity concentrations for each day over all stations following where J is the number of stations performing simultaneous measurements at sample time t.This considerably reduces the scatter found for individually measured ratios and is more similar to the multi-box model approach.
The ratios 137 Cs/ 133 Xe and 131 I/ 133 Xe (and similarly, values of ω and the ratios for the atmospheric burdens) decrease with time as 137 Cs and 131 I are removed from the atmosphere, whereas 133 Xe is conserved.The time scale of the decrease is referred to as removal time throughout this paper and is based on the e-folding time scale.The removal time is calculated based on a fitted trend line through the data, assuming that the data follow a model of exponential decay where t is the sample time, τ is the removal time, and ε is the effective emission ratio at the time assumed for the emission pulse.Equally, the calculations were done using 131 I measurements.The fraction of variance in the ratios explained by the exponential model is given by the squared correlation coefficient R 2 .Since all measurements are referenced to the time of the earthquake, the actual temporal variability of the emissions is not considered, however it was accounted for in a sensitivity study included in Sect. 4.

Results
In the first approach, we estimate the total atmospheric burdens of 137 Cs, 131 I and 133 Xe, [ 137 Cs], [ 131 I] and [ 133 Xe], using the simple multi-box model.The suitability of the underlying assumption that the FD-NPP emissions of 133 Xe are relatively well mixed in latitude bands is confirmed by the small variation in [ 133 Xe] with time (Fig. 3a, blue line).The variability is particularly small after day 30 (10 April) when enough time had passed for a nearly complete mixing in the extratropics and before day 60 (10 May), after which measurement uncertainty and/or background subtraction are becoming more substantial due to the fact that radioactive decay has removed most of the FD-NPP emission pulse.This also demonstrates that ocean uptake of the slightly watersoluble 133 Xe is negligible on the time scale considered.The small overall decrease with time is more likely due to leakage of 133 Xe into the stratosphere and into the Southern Hemisphere, outside of the box model domain.
In contrast to [ 133 Xe], [ 137 Cs] decreases with time due to wet and dry deposition (Fig. 3a, green line).By fitting an exponential model to the change of [ 137 Cs] with time (similar to Eq. 3) we find the removal time τ b (b for "box model") of 137 Cs removal from the atmosphere to be 9.3 days (95 % confidence interval (C.I.) 8.7-10.0days).A more correct estimate of τ b can be obtained from the ratio [ 137 Cs]/[ 133 Xe], as leakages to the stratosphere and Southern Hemisphere, which are not considered in the multi-box model, would affect estimates of [ 137 Cs] and [ 133 Xe] similarly.This gives a slightly longer τ b of 10.0 days (C.I. 9.3-10.9)(Fig. 3a, black line).Likewise for 131 I, τ b is found to be 15.3 days (C.I. 12.8- 18.9) from the decrease of the [ 131 I]/[ 133 Xe] ratio (Fig. 3b).The τ b is longer for 131 I than for 137 Cs due to gas-to-particle conversion of 131 I, as discussed further in Sect. 4.
The other approach for estimating the removal time of 137 Cs and 131 I takes advantage of directly co-located measurements of the radionuclides.For each pair of simultaneous measurements, we calculated the 137 Cs/ 133 Xe and 131 I/ 133 Xe ratios of the decay-corrected values over time (as illustrated in Fig. 2).Furthermore, time was counted relative to the reference time (the time of the earthquake).The decrease of the 137 Cs/ 133 Xe and 131 I/ 133 Xe ratios over time vary for the different measurement stations and removal times range from 8.8 to 18.1 days for 137 Cs and 11.1 to 26.1 days for 131 I (Table 1).For 137 Cs, the shortest removal times (< 10 days) are found for the tropical sites (Wake Island and Oahu) which are affected by strong wet scavenging due to tropical precipitation (see Figs. 16 and 19 of Stohl et al., 2012a).
For Ulan-Bator, located at high altitude (∼ 1300 m a.s.l.), the short removal time can probably be explained by the transport across high mountain chains and strong scavenging due to orographic precipitation.For the North American stations, the removal times for 137 Cs are longer and range from 13.1 to 18.1 days.The European stations give homogenous results with a removal time around 15 to 16 days.For Ussuriysk, a station mostly upwind but closest to FD-NPP, the removal time estimate has a large uncertainty range because variations in the 137 Cs/ 133 Xe emission ratio were occasionally transferred to the station directly without the damping effect of air mass mixing during long-range transport, which tends to eliminate short-time variations of the emission ratio.For 131 I the picture is not as clear because gas-to-particle exchange during transport is playing a role (see Sect. 4).
Combining the global set of measurements from all stations, the 137 Cs/ 133 Xe ratios decrease with a removal time Table 1.Removal times (τ ) for 137 Cs and 131 I estimated from the exponential decay of measured 137 Cs/ 133 Xe and 131 I/ 133 Xe ratios.τ a when combining measurements from all 11 stations as in Fig. 3c, d (grey trend lines), τ ω as calculated from the ratios ω (the mean activity concentrations for each day averaged over all stations using Eq. 2), as in Fig. 3c, d (black trend lines), and τ b from the ratios obtained with the box model (Fig. 3a, b).The 95 % confidence intervals for the removal times are given (see Appendix A). scale τ a (a for "all") of 13.9 days (C.I. 12.8-15.2days) (Table 1, Fig. 3c, grey line).This removal time estimate has larger uncertainty than the multi-box model estimate due to more scatter in the individual data points, which probably also explains the somewhat longer removal time obtained.

Discussion of uncertainties
To assess the impact of some of the assumptions made in the analysis, we carried out two sensitivity tests for the method using the direct measurements of 137 Cs and 133 Xe.First, we used only measurement data after 1 April.This represents a relatively well-mixed case, as an emission pulse from East Asia typically is quite homogeneously mixed across the troposphere in the extratropical Northern Hemisphere, both zonally as well as vertically, after 25-30 days (see Figs. 2-4 in Stohl et al., 2002).In this well-mixed case the different temporal shape in the emissions of 137 Cs and 133 Xe gets less important.Figure 4a shows that τ a = 14.3 days (C.I. 13.0-15.9days) and τ ω = 12.9 days (C.I. 12.1-13.7 days) are ob-tained.These results are not significantly different from the removal times obtained when using the complete measurement data set.
For the second sensitivity test we took into account the fact that emissions were not from a single emission time but varied over a few days (e.g.Stohl et al., 2012a).Using the Lagrangian particle dispersion model FLEXPART (Stohl et al., 1998(Stohl et al., , 2005) ) we estimated the most probable emission time and the age of the radioactive plume when it reached each measurement station.This was done by using simulated sensitivities from the dispersion model.These sensitivities were calculated from the same model simulations that were used and described in the source term estimations by Stohl et al. (2012a).The sensitivity of the modeled concentrations to the emissions and time-varying source terms of 137 Cs (Stohl et al., 2012a) was calculated for each measurement sample.This determined the modeled emission contributions to each measurement sample as a function of emission time.Only measurement data which the model clearly associated with the emissions from FD-NPP were used.Here, the effective emission time for a given measurement sample was considered to be the time with the highest emission contribution of 137 Cs.Before fitting the exponential model, time was counted relative to the effective emission time of every measurement sample.The resulting removal times (Fig. 4b) of τ a = 13.5 days (C.I. 12.5-14.7)and τ ω = 11.9 days (C.I. 11.2-12.7 days) are again not significantly different from the results when using a single reference time for the emissions.
Using a different method, Yang and Guo (2012) found atmospheric residence times of 131 I ranging 7-33 days with an average of 12 ± 3 days from precipitation samples taken in Southern United States.This is shorter than the 131 I removal times estimated with our method for the Southern U.S. stations.However, the measurements, methods and time periods which were used are not directly comparable for the two approaches.Also atmospheric transport is not taken into account in the method using precipitation samples, but is considered in our approach using the 133 Xe signal.Furthermore, there are complicating factors related to 131 I as it coexists in both particulate and gaseous forms.
The longer removal time for 131 I than for 137 Cs is expected since 131 I has a large gaseous fraction, and the gas-particle conversion leads to the formation of new particulate iodine from gaseous iodine during the transport.Irrespective of the lifetime of gaseous 131 I, on-going particulate 131 I generation can only lead to longer removal times for aerosol-bound 131 I. Masson et al. (2011) show that the gaseous 131 I fraction (∼ 70 %) that was released at the FD-NPP stays more or less on the same level during the two weeks long transport time to Europe.Also Uematsu et al. (1988) suggest that the gasto-particle conversion time for 131 I typically requires about 2-3 weeks to occur.However, the time scale of our analysis is considerably longer (∼ 80 days) which means that the gas-to-particle conversion will act as a source for the particulate fraction of 131 I and will increase the removal times for the aerosol-bound 131 I. Therefore, the removal time derived from the 131 I data can only be considered as an upper estimate for the AM aerosol lifetime.Due to these complicating factors related to 131 I, most of the further discussion on the application for aerosol lifetime focuses on the results obtained for 137 Cs which is found only in particulate form.
To evaluate the initial source emissions and the exponential model fit used to derive the aerosol lifetimes we estimated the initial 137 Cs/ 133 Xe ratio at the reference time from the exponential model fitted to the data by extrapolating the fitted curve to time zero (the intercept value).From the boxmodel, the initial 137 Cs/ 133 Xe ratio is 7.2 × 10 −5 and the method using the direct measurements gives an initial ratio of around 2 × 10 −5 .These initial ratios are up to 2 order of magnitude lower than the ratio of the total releases of 137 Cs and 133 Xe found by Stohl et al. (2012a) using an inversion technique (36.6 × 10 15 Bq /15.3 × 10 18 Bq 137 Cs/ 133 Xe, giving an emission ratio of 2.4 × 10 −3 ).Their total 137 Cs emissions have an estimated uncertainty of about 50 %, while 133 Xe emissions should be accurate to within 20 %.For 131 I we obtain an initial 131 I/ 133 Xe ratio of about 1 × 10 −3 from our two methods.Compared to the ratio of the total release estimates of Chino et al. (2011) and updated by Terada et al. (2012) for 131 I, and Stohl et al. (2012a) for 133 Xe, of 1 × 10 −2 (150 × 10 15 Bq/15.3 × 10 18 Bq 131 I/ 133 Xe), our estimate is roughly one order of magnitude lower.
The fact that our radionuclide ratios extrapolated to the time of the initial release are so much lower than those reported for the emission ratios (even considering uncertainties in the emission ratios), suggests that this extrapolation is not valid.There are indeed indications in our data that the initial removal rates of 137 Cs and 131 I were higher than the removal rates encountered in the well-mixed situation afterwards.Figures 3c-d and, especially, Fig. 4b show that most of the first few data points deviate strongly upward from the exponential model curve.Unfortunately, the initial phase of plume dispersion was not sampled by the CTBTO network and it is therefore not possible to derive removal rates or lifetimes for the first few days after emission.However, there are some potential reasons for higher initial removal rates, namely the fact that there was strong precipitation co-located with the plume during the period of the highest emissions.This caused strong scavenging of the plume immediately after its emission (Stohl et al., 2012a).Also the fact that the initial plume was close to the ground facilitated effective dry deposition.High deposition rates over Japan were also indicated by Katata et al. (2012a, b).Furthermore, hot particles (particles that carry very high radioactivity, e.g., fragments of the nuclear fuel) were present in the FD-NPP plume (Paatero et al., 2012).Hot particles can be much larger than AM aerosols (Paatero et al., 2010) and deposit much more quickly, e.g., by gravitational settling.Thus, our derived AM aerosol lifetimes must be considered valid for a well-mixed background AM aerosol, whereas the lifetime of a fresh aerosol directly emitted from the ground may be substantially shorter.
The major uncertainty factor with regard to our removal time estimate is the possibility of additional releases of radionuclides long after our assumed reference time.These additional releases can be either direct late emissions from FD-NPP or indirect releases by resuspension of deposited radionuclides.First we discuss the possibility that our results are influenced by resuspension.It has been seen that resuspension was important after the Chernobyl accident (Garland and Pomeroy, 1994), and monitoring data from Japan suggest resuspension occurred also there after the FD-NPP accident (Stohl et al., 2012a).However, the data also show that the 137 Cs concentrations in Japan are lower by two to three orders of magnitude in between major plume passages, suggesting that resuspension was quite limited.Applying a 137 Cs resuspension rate γ = 1 × 10 −9 s −1 , typical for summer conditions in central Russia (Makhon'ko, 1979as cited by Gavrilov et al., 1995), and accounting for the fact that only 20 % of the FD-NPP emissions were deposited over land (Stohl et al., 2012a), we estimate that a fraction of 1.7 × 10 −4 of the originally emitted 137 Cs could be resuspended in a 10day period.For a 10-day removal time, we find that a fraction of 3.3 × 10 −4 of the originally emitted 137 Cs would still be suspended in the atmosphere after 80 days.Thus, even for the shortest removal times obtained in our study and at the very end of our study period, resuspension could account for only about half of the 137 Cs mass suspended in the atmosphere.For the moist climate of Japan, which received 90 % of the FD-NPP fall-out over land as estimated by numerical simulations (Stohl et al., 2012a), the resuspension rate is likely to be lower than the value used above.Even more importantly, long-range transport of resuspended 137 Cs to our monitoring  3c, but using measurement data only after 1 April 2011, the exponential model yields estimates of τ a = 14.3 and τ ω = 12.9 days.(b) Using plume age calculations based on FLEX-PART model output yields τ a = 13.5 and τ ω = 11.9 days.Notice that the age here is not referenced to the time of the earthquake but to the actual emission time most relevant for each sample.
stations is highly unlikely because of the larger particle sizes typical of resuspended material.In the surroundings of Chernobyl, 137 Cs activity size distributions with median diameters of 5-10 µm were measured for resuspended material (Garger et al., 1998).Such particles have a very short residence time in the atmosphere compared to AM aerosols and cannot be transported far from the source.Finally, there was also no indication of an elevated background level of 137 Cs after July, compared to the levels before the FD-NPP accident, at any of the stations.This implies that resuspension, which occurs over much longer time scales than aerosol deposition (Maxwell and Anspaugh, 2011) and should still be observable after July if it was important during the period of our study, did not impact the measurements used in this study.
Secondly, we discuss if direct late emissions affect our results.One study (Stohl et al., 2012a) found 137 Cs emissions in late March to be about two orders of magnitude smaller than during the first week, and emissions were decreasing further in April, but there is a possibility that 137 Cs emissions had not ceased completely in April.We have carefully screened the measurement data (including also Japanese data as presented in Stohl et al., 2012a) for any evidence for a late emission pulse from FD-NPP and found one possible event in late May seen at the station Ussuriysk that would be large enough to affect our results.We have excluded these data and ended the period of our study on 25 May to avoid impacts of this event on other stations.However, late emissions on the order of only 1 % of the maximum emission in the early phase could theoretically contribute to a large part to the measured activity concentrations in late May, and thus affect the estimated removal times.However, no other study has suggested strong late emissions of 137 Cs.Regarding 131 I, there is also no evidence for late emissions from FD-NPP.
Lastly, we compare our estimated removal times of the aerosol-bound radionuclides to aerosol lifetimes reported in other studies.Observation-based estimates of aerosol lifetimes are sparse, range from less than 4 days to more than a month (Moore et al., 1973;Cambray et al., 1987;Papastefanou, 2006;Williams et al., 2002;Paris et al., 2009;Schmale et al., 2011), and are associated with substantial uncertainties.According to the existing aerosol models the average residence times of AM aerosol in the atmosphere are on the order of 3-7 days (Chin et al., 1996;Feichter et al., 1996;Stier et al., 2005;Berglen et al., 2004;Liu et al., 2005;Bourgeois and Bey, 2011;Chung and Seinfeld, 2002;Koch and Hansen, 2005;Textor et al., 2006).The aerosol lifetimes vary regionally and are generally longest in dry or cold regions (Koch et al., 1996).Modeled lifetimes longer than 10 days were obtained primarily for aerosols originating from the stratosphere (Koch et al., 1996).
Our removal times for 137 Cs and associated AM aerosol lifetimes range from 10 to 14 days and are compatible with the much larger range of aerosol lifetimes given in previous observation-based studies, but they are not consistent with the 3-7 days annual global averages obtained from the aerosol models.The difference cannot be explained by the fact that our study extended only over 80 days and covered only the Northern Hemisphere.The emissions were exposed to extratropical cyclones and experienced strong lifting in the North Pacific storm track (Stohl et al., 2012a), a region where and during a time of the year when storm activity is considerably enhanced.This should have caused stronger-thanaverage rather than weaker-than-average wet scavenging of aerosols.
One possible explanation for the discrepancy between our estimated removal times and those reported for aerosol models is the different aerosol distribution.Our estimated removal times should be representative for AM aerosol which is already reasonably well mixed in the atmosphere (background aerosol), whereas aerosol models report lifetimes for aerosols which are either emitted at the surface, or formed mainly in the boundary layer.Indeed, our comparison to the reported emission ratios for the FD-NPP accident suggests that aerosol removal rates must have been much larger (and, thus lifetimes shorter) shortly after the emission, compared to the later period.However, the comparison between our results and modelled aerosol lifetimes may still indicate that the lifetimes in the aerosol models are too short.This certainly deserves further clarification and investigation, for example by running the major aerosol models directly against the FD-NPP accident case, or at least specifically for a wellmixed aerosol that is more comparable to the FD-NPP accident case than the AM aerosol tracers for which lifetimes are normally reported.

Conclusions
The removal times of the aerosol-borne radionuclides 137 Cs and 131 I have been quantified by using a global set of measurements which recorded the activity concentrations following the release of both nuclides from the FD-NPP accident in March 2011.The radioactive noble gas 133 Xe also released during the accident served as a tracer of the atmospheric transport.The main findings from this study are summarized as follows: -A removal time for 137 Cs of 10.0-13.9days and for 131 I of 17.1-24.2days was estimated from the decrease over time of the ratios 137 Cs/ 133 Xe and 131 I/ 133 Xe, respectively.The removal times can serve as estimates of accumulation-mode (AM) aerosol lifetimes since the radionuclides attach to AM aerosols and trace their fate during transport and removal.
-The longer removal times for 131 I were affected by the gas-to-particle exchange that occurs during transport while no such effect influences 137 Cs which thus gives a better estimate of AM aerosol lifetimes.Lifetimes derived with 131 I must be considered as upper estimates.
-The removal rates must have been higher (and, thus, removal times shorter) during the initial phase of the plume transport that was not captured by our measurements.This can be seen by too low aerosol/noble gas radionuclide ratios obtained when extrapolating the exponential fit back to the time of the accident, compared to reported emission ratios.The same is suggested by an upward deviation of the first few measurement data points from the exponential model fit.Therefore, the estimated removal times are valid only for an aerosol reasonably well mixed in the troposphere (a background aerosol) and not for fresh aerosol directly emitted from the ground.
-Our results are highly sensitive to possible late emissions of radionuclides.However, there is no evidence for such late emissions, neither in our data nor in the existing literature on the FD-NPP accident.
-The effect of resuspension on the estimated removal times was likely negligible mainly due to the fact that resuspension is much smaller than the initial emission pulse and encompasses larger particles than AM aerosols.
-The estimated removal times are consistent with the large range of previous observation-based aerosol lifetimes, but they are about a factor of two higher than published model-based estimates on AM aerosol lifetimes in the atmosphere.The difference points towards a too quick removal of AM aerosol in models.
-Our study should serve as encouragement for aerosol modelers to run their models against the FD-NPP accident case.

Appendix A Statistical methods
From Eq. ( 3), the coefficients of the exponential model are ε (intercept) and 1/τ (slope).The 95 % confidence interval (C.I.) for the removal time τ (Table 1) is obtained by taking the inverse of the upper and lower C.I. limits of the slope.This inverse-conversion changes the distribution so that τ it is not longer normally distributed, thus τ is not the centre of the C.I. but can have a long tail towards longer τ (this is seen, e.g., at Ashland).The standard deviation for τ is not obtained due to the inverse-conversion.Therefore, only confidence intervals are reported throughout this paper, and not standard deviations.

Fig. 2 .
Fig. 2. Time-series of observed 137 Cs, 131 I and 133 Xe at the station Oahu.Shown are the observed activity concentrations (red squares), the observed concentrations with background subtracted (green plusses), and the decay-corrected enhancements over the background (blue diamonds).The time of the earthquake (11 March 2011) is indicated with the vertical dashed line.
et al.: Atmospheric removal times of aerosols emission time, so the time of the earthquake was chosen for convenience.

Fig. 3 .
Fig. 3. 137 Cs and 131 I decrease with time.(a) Total atmospheric burden of [ 137 Cs], [ 133 Xe] and the [ 137 Cs]/[ 133 Xe] ratio over time, from a multi-box model.The fit of exponential decay models to the data (dotted lines and text insets) with removal time τ b of 10.0 days, are also shown.The fraction of variance in the data explained by the exponential model is given by R 2 .(b) Same as (a), but for 131 I. (c) Measured 137 Cs/133 Xe ratios for 11 CTBTO-stations (colored marks) and the fit of an exponential decay model to the data (grey line) with removal time τ a of 13.9 days.ω (black triangles) represent the ratio of the activity concentrations for each day averaged over all stations and the black line is the exponential fit through these ratios with removal time τ ω of 12.6 days.(d) Same as (c), but for the 131 I/ 133 Xe ratios.The time-axes on all four figures are in days after the time of the earth quake (11 March at 05:46 UTC).

Fig. 4 .
Fig. 4.Sensitivity tests for removal times τ of 137 Cs.Measured 137 Cs/ 133 Xe ratios and fit of exponential models to the data (solid lines).(a) Same as Fig.3c, but using measurement data only after 1 April 2011, the exponential model yields estimates of τ a = 14.3 and τ ω = 12.9 days.(b) Using plume age calculations based on FLEX-PART model output yields τ a = 13.5 and τ ω = 11.9 days.Notice that the age here is not referenced to the time of the earthquake but to the actual emission time most relevant for each sample.