PREDICTIONS OF DISPERSION AND DEPOSITION OF FALLOUT …

Paper

PREDICTIONS OF DISPERSION AND DEPOSITION OFFALLOUT FROM NUCLEAR TESTING USING THE

NOAA-HYSPLIT METEOROLOGICAL MODEL

Brian E. Moroz,* Harold L. Beck,† Andre Bouville,* and Steven L. Simon*

Abstract—The NOAA Hybrid Single-Particle Lagrangian In-tegrated Trajectory Model (HYSPLIT) was evaluated as aresearch tool to simulate the dispersion and deposition ofradioactive fallout from nuclear tests. Model-based estimatesof fallout can be valuable for use in the reconstruction of pastexposures from nuclear testing, particularly where little his-torical fallout monitoring data are available. The ability tomake reliable predictions about fallout deposition could alsohave significant importance for nuclear events in the future.We evaluated the accuracy of the HYSPLIT-predicted geo-graphic patterns of deposition by comparing those predictionsagainst known deposition patterns following specific nucleartests with an emphasis on nuclear weapons tests conducted inthe Marshall Islands. We evaluated the ability of the computercode to quantitatively predict the proportion of fallout parti-cles of specific sizes deposited at specific locations as well astheir time of transport. In our simulations of fallout from pastnuclear tests, historical meteorological data were used from areanalysis conducted jointly by the National Centers forEnvironmental Prediction (NCEP) and the National Center forAtmospheric Research (NCAR). We used a systematic ap-proach in testing the HYSPLIT model by simulating therelease of a range of particle sizes from a range of altitudes andevaluating the number and location of particles deposited. Ourfindings suggest that the quantity and quality of meteorologi-cal data are the most important factors for accurate falloutpredictions and that, when satisfactory meteorological inputdata are used, HYSPLIT can produce relatively accuratedeposition patterns and fallout arrival times. Furthermore,when no other measurement data are available, HYSPLIT canbe used to indicate whether or not fallout might have occurredat a given location and provide, at minimum, crude quantita-tive estimates of the magnitude of the deposited activity. Avariety of simulations of the deposition of fallout fromatmospheric nuclear tests conducted in the Marshall Islands(mid-Pacific), at the Nevada Test Site (U.S.), and at theSemipalatinsk Nuclear Test Site (Kazakhstan) were per-formed. The results of the Marshall Islands simulations

were used in a limited fashion to support the dose recon-struction described in companion papers within this volume.Health Phys. 99(2):252–269; 2010

Key words: Marshall Islands; nuclear weapons; fallout; mod-eling, meteorological

INTRODUCTION

COMPUTER MODELS have been both influential and benefi-cial in predicting fallout dispersion and deposition. Thesemodels have been used historically for such diverse tasksas producing quick fallout estimates necessary for imme-diate health assessments, extending exposure estimatesdownwind beyond ground-based measurements in retro-spective dose and risk assessments (Cederwall and Peter-son 1990; Hoecker and Machta 1990), and projectingpotential physical damage, including atmospheric effectssuch as smoke production from regional nuclear conflictsand individual acts of nuclear terrorism (Toon et al.2007). Computer codes used for such purposes weredeveloped and applied by the scientific and defensecommunities as early as the 1960’s (Rowland 1994).

Modeling the transport and deposition of particlesreleased from a nuclear weapons test is both a complexand highly uncertain exercise. This is true even when themeteorological data used in the simulation are accurate.Furthermore, in order to simulate the deposition densityof specific radionuclides or total radioactivity, a model isrequired for the spatial distribution of radionuclides inthe initial debris cloud as well as the distribution ofactivity as a function of particle size. The most compu-tationally burdensome factors in performing the simula-tions are the large size of the debris cloud and, therefore,the large number of particles and particle sizes that areneeded to conduct a realistic fallout simulation over longdistances. An additional difficulty is presented whenmodeling wet removal processes. Both in-cloud andbelow-cloud wet removal processes may be of greatimportance to accurately simulating deposition whenprecipitation occurred downwind. The data available

* Division of Cancer Epidemiology and Genetics, National Insti-tutes of Health, National Cancer Institute, Bethesda, MD, 20892;† New York, NY.

For correspondence contact: Steven L. Simon, National CancerInstitute, National Institutes of Health, 6120 Executive Blvd., Be-thesda, MD 20892, or email at [email protected].

(Manuscript accepted 22 June 2009)0017-9078/10/0Copyright © 2010 Health Physics Society

DOI: 10.1097/HP.0b013e3181b43697

252

from most meteorological archives are generally insuffi-cient for accurately modeling these processes (Draxlerand Hess 1997, 1998). Therefore, simplifying assump-tions are usually incorporated into wet removal algo-rithms, leading to predictions with low reliability.

Given the present national security concerns, there isa need for the scientific and defense communities to beaware of the capabilities of available atmospheric trans-port models and their possible application to predictfallout in the case of future events. This paper discussesone particle transport and dispersion model, the HybridSingle-Particle Integrated Trajectory (HYSPLIT) model,and describes how the model was tested and evaluatedfor the purpose of reconstructing fallout resulting fromnuclear testing. Our main evaluation was of U.S. nucleartests conducted in the Marshall Islands (MI). In additionto the Marshall Islands nuclear tests, two other falloutevents were simulated to test the model: the 1953Upshot-Knothole Harry test at the Nevada Test Site(NTS) and the first Soviet nuclear test conducted at theSemipalatinsk Test Site in 1949. In a latter section of thispaper, we discuss one application of the HYSPLITmodel: our use of the model to support deposition anddose estimates in the Marshall Islands reported in com-panion papers (Beck et al. 2010; Bouville et al. 2010;Simon et al. 2010a, 2010b). Based on the test simulationsand the application mentioned above, the potential use ofHYSPLIT predictions for both past and future falloutevents is discussed.

MATERIALS AND METHODS

The HYSPLIT modelHYSPLIT (Draxler and Hess 1997, 1998; Draxler

1999) was developed and is maintained by the NationalOceanic and Atmospheric Administration Air ResourcesLaboratory (NOAA ARL). The HYSPLIT model com-putes the dispersion and deposition of particles originat-ing from single or multiple source locations upon asimultaneous release. In this paper, HYSPLIT was usedto simulate the advection and dispersion of particles in aradioactive debris cloud over the time period of severalhours to several days following the nuclear test. Here werecognize that HYSPLIT was not developed as a predic-tive fallout model; it makes no attempt to simulate thedynamics of the debris cloud prior to stabilization, nordoes it simulate the radioactivity associated with aparticular particle size. However, by assuming the debriscloud is stabilized, and assuming reasonable distributionof particles and particle sizes within the stabilized cloud,multiple simulations of the transport of particles releasedat various altitudes, for a range of particle sizes, can becombined to approximate the total fallout deposited from

a debris cloud as the particles from each altitude aretransported downwind.

Meteorological input data used by HYSPLIT aregridded meteorological data fields generated and ar-chived from other meteorological models, although it ispossible to perform simulations with user-defined winddata to a limited extent (Draxler and Hess 1997, 1998;Draxler 1999). Because it is common for differentmeteorological models to use different vertical coordinatesystems, HYSPLIT linearly interpolates the meteorolog-ical data at each horizontal grid point in the meteorolog-ical input data to an internal sub-grid containing aterrain-following coordinate system where all heights areexpressed relative to mean sea level (Draxler and Hess1997, 1998). The default vertical resolution for HYSPLITdefines the model’s lowest level (level 1) at approxi-mately 10 m and level 2, which is considered the surfacelayer, at approximately 75 m above ground level (AGL).Vertical resolution continually decreases away from theground surface following a quadratic form. It is possibleto modify the model’s internal vertical resolution bymodifying the internal parameters corresponding to themodel’s internal height index (Draxler and Hess 1997).In contrast, the horizontal resolution applied by themodel is equivalent to the horizontal resolution of themeteorological input data.

The spacing between the grid points of the inputdata influences the accuracy of model computations. Asa general rule, the grid resolution should, at minimum,correspond to the scale and the purpose of the simulation.Data gridded at a coarse resolution may yield less preciseresults than desired. In contrast, finely gridded data canimprove model results, assuming that the meteorologicalinput data are accurate. The small amount of griddedmeteorological data at fine resolutions in the tropic zonesof the Pacific during the period of U.S. nuclear testing inthe Marshall Islands was one limiting factor to our work.

HYSPLIT offers several particle or puff modelingapproaches which are discussed in Draxler and Hess(1997, 1998). We used the three-dimensional particlemodel to compute the advection and dispersion of thedebris cloud. Advection is computed independently, orprior to, the dispersion calculation. The dispersion rate isdependent upon the vertical diffusivity profile, windshear, and the horizontal deformation of the wind field(Draxler and Hess 1997).

The particle advection algorithm has two primarysteps. After linearly interpolating the velocity vectors (u,v, w) to the current particle position and time, a displace-ment calculation yields a first-guess position by integrat-ing the velocity component at the current position andtime over the duration of the time-step. The final positionis then calculated by averaging the velocity components

253Dispersion and deposition of fallout from nuclear testing ● B. E. MOROZ ET AL.

at the two successive particle positions, integrating overthe duration of the time-step, and then adding thedisplacement to the initial position of the particle. Theintegration time-step can vary from 1 to 60 min but isbound by a user-specified advection distance per time-step to limit the advection to less than one grid point pertime-step.

Dispersion is computed after the advection compu-tation; however, it is necessary for the model to firstcompute stability and mixing coefficients. Stability andmixing are estimated from the meteorological input data.Heat and momentum fluxes, if they are present in themeteorological data, are used to compute the stability,otherwise temperature and wind data at each grid pointare used to estimate it. Vertical mixing within theboundary layer is computed as an average at eachhorizontal grid point based upon flux data. Above theboundary layer, vertical mixing is estimated from thewind and temperature profiles. Horizontal mixing iscomputed using the deformations in the wind field and isadjusted based on the size of the meteorological grid.

To realistically simulate the dispersive nature of theatmosphere, a random turbulent component is incorpo-rated into the dispersion calculation by adding the turbu-lent component to the mean velocity obtained from themeteorological input data at each time-step. This turbu-lent component is a Gaussian based pseudo-randomlygenerated number resulting from the product of theGaussian random number and the standard deviation ofthe computed turbulent velocity of the velocity vector(Draxler and Hess 1997). The Gaussian random numberis generated using a variation of the linear congruentialmethod, Xi�1 � (aXi � c) mod m, where the element iindicates the position of the random number within thesequence. When the parameters a, c, and m are chosencorrectly, generators of this class can ensure a nonrepeat-ing sequence on the scale of 109. It should be noted thatthough the HYSPLIT model incorporates a randomturbulence element, the model is not stochastic becausethe same random sequence is generated with each invo-cation of the model, meaning that the model results forany single simulation will always be the same assumingthe simulation parameters are not changed. This can bealtered by simply modifying the model’s random numberalgorithm to apply a different seed value with eachinvocation.

Several dry deposition options are available to themodel user. In our case, dry deposition was simulatedunder the assumption that the deposition velocity for allparticles was equivalent to the gravitational settlingvelocity. For local fallout from weapons tests, this is areasonable approximation since most of the radioactivityis found on particles of diameter greater than 5 �m

(Heidt et al. 1953; Crocker et al. 1965; Ibrahim et al.2010). Other HYSPLIT options include implicitly spec-ifying a dry deposition velocity or using the resistancemethod (Draxler and Hess 1997). In our simulations,gravitational settling was computed by the model basedon particle diameter, a fixed particle density of 2.5 gcm�3, and a fixed spherical particle shape. The computedsettling velocity is applied to the vertical position of theparticle at each time-step. Particles are subject to drydeposition removal processes upon entering the model’ssurface layer. The model computes dry deposition usingone of two options: either removing a fraction of theparticle’s mass over successive time-steps until the massbecomes zero, or computing the probability that a parti-cle will deposit all of its mass during a single time-step(Draxler and Hess 1997). In our simulations the deposi-tion probability option was used.

Wet deposition processes impose difficulties inmeteorological computer models. The difficulty stemsfrom the simplified assumptions incorporated into wetdeposition models coupled with a general lack of reliableprecipitation observations in the meteorological inputdata (Draxler and Hess 1997). Both in-cloud (rainout)and below-cloud (washout) wet deposition are estimatedin the HYSPLIT model by defining the fraction of totalpollutant mass within and below the cloud layer andapplying an estimated deposition rate. The extent of thecloud layer is defined using relative humidity (RH) inthe meteorological profile at each horizontal grid point.The cloud top and bottom are, by default, defined at 60%and 80% RH, respectively. In the case of rainout, a wetdeposition velocity is calculated as the product of theprecipitation rate at the grid point and a pollutant-specificscavenging ratio. The scavenging ratio is based on theamount of pollutant (g L�1) in the air within the cloud tothat in the rain (g L�1) measured on the ground at the gridpoint (Draxler 1999). The wet deposition velocity is thenapplied to the fraction of pollutant mass within the cloudlayer. Below-cloud removal is defined using only a rateconstant (s�1) and is independent of precipitation rate(Draxler and Hess 1997). The rate constant is applied tothe fraction of pollutant that is below the cloud bottom.In our simulations, the model’s default values for thein-cloud scavenging ratio (3.2 � 105 L per L) and thebelow-cloud rate constant (5.0 � 10�5 s�1) for wetdeposition processes were used.

The total deposition over a time-step is the sum ofthe removal amounts resuting from each process. Thetotal pollutant mass is then reduced by the computedremoval fraction (Draxler and Hess 1997).

Several verification examples demonstrating theapplicability and accuracy of HYSPLIT computations inthe areas of particle advection, dispersion, and deposition

254 Health Physics August 2010, Volume 99, Number 2

are provided in Draxler and Hess (1998). Draxler andHess (1998) discussed a HYSPLIT simulation of therelease from the Chernobyl reactor accident which tookplace in the former Soviet Union in 1986. Resultingdeposition contours and peaks were compared againstthose reported in Klug et al. (1992). It was found thatcontour patterns were reasonably consistent and theHYSPLIT-computed deposition peaks were numerically�10% higher than the reported values in the best case.Two additional studies which used HYSPLIT as aresearch tool for modeling fallout processes includeKinser (2001) and Swanberg and Hoffert (2001), whosimulated releases of 137Cs from the Chernobyl reactoraccident to investigate model-predicted wet depositionand resuspension as a source of 137Cs in Europe.

Meteorological input dataHigh spatial density meteorological data are either

not available or are very sparse during the 1950’s formany areas including the mid-Pacific Ocean where theMarshall Islands are located, the NTS, and the Semipal-atinsk Test Site. For that reason, meteorological inputdata from a reanalysis data set were used. The reanalysiswas conducted as a collaborative effort between theNational Centers for Environmental Prediction (NCEP)and the National Center for Atmospheric Research(NCAR) with the purpose of recovering historicalweather observations from many sources, providingquality control of the data, and compiling the results so asto fulfill the research needs for climate monitoring andprediction. The NCEP/NCAR database covers the period1948–2008 and provides forecasts at regular temporalintervals of four times daily (though, during the period1948–1957, forecasts are provided eight times daily) andat a spatial resolution of 2.5 degrees.

The most important observational data incorporatedinto the reanalysis data set for the purposes of our studyare the upper-air wind data. Upper-air wind data areprimarily derived from the world’s upper-air rawinsondenetwork, although a considerable amount of aircraftreconnaissance data was also incorporated into theNCAR/NCEP reanalysis. The U.S. Air Force prepared aglobal collection of aircraft data that covered the timeperiod 1948–1970. Aircraft data were also contributedby the University of Hawaii from locations in the tropicsfor the time period 1960–1973.

Early rawinsonde coverage in the U.S. is fairlycomplete, though less data are available for other parts ofthe world. Good coverage in the U.S. began in 1948while periods of good coverage in China, India, andRussia cannot be found earlier than the 1950’s and1960’s. Wind profiles from several operational weatherstations in the Marshall Islands were available in addition

to upper-air wind measurements taken at test site loca-tions by the U.S. Army (DNA 1979), although thesupporting literature does not unequivocally report thatthose observations were incorporated into the NCAR/NCEP reanalysis.

Precipitation observations in the reanalysis data setare also important, but appear to be much less reliablethan upper-air wind data. In the NCAR/NCEP reanalysismodel, observed precipitation data do not play a directpart in the reanalysis output. Precipitation data are solelydriven by the reanalysis model and can exhibit regionalbiases (Kalnay et al. 1996).

NUCLEAR TEST SIMULATIONS ANDMODEL EVALUATIONS

General methodsThe HYSPLIT model was tested using a systematic

approach for simulating the transport and deposition ofradioactive particles resulting from a nuclear weaponstest. A range of particle sizes were released from a rangeof altitudes at a single source location and tracked overtime. The proportion of deposited particles and thelocation of deposition were then evaluated by comparingmodel-predicted deposition patterns and time of falloutarrival against known deposition patterns and best avail-able estimates.

In order to simulate the deposition density of aspecific radionuclide, such as 137Cs, a crude model wasdeveloped to qualitatively relate HYSPLIT particle dep-osition to 137Cs deposition as a function of fission yield,debris cloud size, and stabilization altitude. This modelassumes a lognormal distribution of activity as a functionof particle diameter based on data from the NTS (Izrael2002). In order to simulate 137Cs ground depositiondensity (Bq m�2) for comparison with the 137Cs deposi-tion densities in the Marshall Islands reported by Beck etal. (2010), the 137Cs particle-size distribution was modi-fied to reflect the fact that 137Cs tends to be depleted onlarger particles. Much of the 137Cs is formed after theheavier particles have already deposited due to 137Cshaving a gaseous precursor, 137Xe (Beck et al. 2010).Here we assume that �80% of the total 137Cs activity isfound on particles less than 50 �m in diameter. Thisassumption is consistent with data and models for a coralsurface shot, as reported by Freiling et al. (1965). Theapportionment of the 137Cs on particle sizes less than 50�m (i.e., 80%) was chosen based on experience andjudgment since little actual data are available from theliterature. The particle-size distributions for 137Cs activityused for the Marshall Island simulations are shown inTable 1. In order to estimate the sensitivity of thesimulated deposition density of 137Cs, two other particle-size distributions were tested in selected simulations. The


two alternative distributions, also shown in Table 1,varied the fraction of 137Cs on particles greater than 50�m slightly from the distribution labeled MI.

The distribution of activity and of the number ofparticles within the stem and the assumed spherical headof the nuclear debris cloud were based on assumptionsfrom previous publications on meteorological modelingof nuclear debris clouds. Here, we assumed that 12% ofthe activity was deposited in the stem as derived fromdata on NTS nuclear test debris clouds (Hoecker andMachta 1990; NCI 1997). The remaining 88% of theactivity was assumed to be distributed homogenouslythroughout the head of the cloud as well as the number ofparticles in each size fraction. Another simplifying as-sumption made for the simulations was to release allparticles from the vertical axis through the center of thespherical head of the debris cloud.

The total amount of 137Cs in the debris cloud wascalculated from the estimated fission yield of each test.Note that the fission yields used for normalization areonly estimates, since the fission yields for U.S. thermo-nuclear tests remain classified. The 137Cs activity for eachparticle tracked by the model was based on the total 137Csproduced and its apportionment among the total numberof particles using the distributions of activity as afunction of particle diameter (Table 1).

The debris cloud model used in these simulations isacknowledged to be crude; hence, the fallout estimatesare subject to a large degree of uncertainty. Moresophisticated simulations of the distribution of activity inthe cloud and on various sized particles have been done

in other studies (see for example, Cederwall and Peterson1990). However, since the particle-size and activitydistributions can vary significantly with the particularconditions of a test such as height of burst, type of soil,and yield (see companion paper by Ibrahim et al. 2010),even simulations using more elaborate models (Ceder-wall and Peterson 1990) were forced to adjust theactivity, altitude, and particle size parameter estimatesfor each test to achieve even marginal agreement withactual measurements.

Once a particle is deposited in a HYSPLIT simula-tion, its location relative to points of interest has to bedetermined. For that purpose, we defined rectangularareas (termed “deposition domains”) in which the depo-sition density of particles and activity were determinedby counting the number of deposited particles stratifiedby release height and diameter. Each domain was defined bythe longitude and latitude of two points on opposingcorners. The coordinates of each deposited particle couldbe tested to determine if the particle had deposited withinthe domain bounded by the defined rectangle.

Marshall Islands nuclear testsThe HYSPLIT model was tested by simulating

fallout deposition from selected U.S. nuclear tests con-ducted at the Bikini and Enewetak test sites in theMarshall Islands. Model-predicted deposition patterns,density estimates, and fallout time of arrival were com-pared against patterns and values reported in the litera-ture (see Beck et al. 2010 for a listing of available data).

The large area of the Marshall Islands imposedsignificant computational constraints because it is neces-sary to simulate very large numbers of particles in orderto delineate spatial or temporal patterns with satisfactoryreliability. Using a three-dimensional particle dispersionmodel, the number of particles required for reasonablyprecise simulations of multi-day fallout dispersion froman entire debris cloud was too large to be practicallyfollowed in a single HYSPLIT simulation. Thus, smallersimulations were performed, each for a single releasealtitude and particle size. In each of these simulations,10,000 particles were released. The results of the simu-lations were then combined based on the assumed rela-tive fractions of total 137Cs activity released from variousportions of the debris cloud and the fraction of activityreleased from various altitudes as discussed above. Forthe individual altitude and particle size combinations,release heights were varied from ground level to thereported top of the radioactive debris cloud (DNA 1979)in 1,000 m increments. The particle sizes simulatedranged from 1 to 100 �m in 5 �m increments and from125 to 300 �m in 25 �m increments and were selectedbased on the typical range of particle sizes for weapons

Table 1. Estimated distribution of 137Cs activity on particles ofspecified diameters.

Median particlediameter, �m

(range)

Activity (%) of total

MI 137Csdistribution

Alternate 137Csdistribution #1

Alternate 137Csdistribution #2

5 (2.5−7.5) 12.5 16.6 23.610 (7.5−12.5) 11.0 11.6 11.715 (12.5−17.5) 10.0 9.3 7.520 (17.5−22.5) 9.0 8.1 5.325 (22.5−27.5) 8.0 6.9 3.930 (27.5−32.5) 7.0 6.2 3.035 (32.5−37.5) 6.5 5.4 2.440 (37.5−42.5) 6.0 5.0 2.045 (42.5−47.5) 5.5 4.4 1.650 (47.5−52.5) 5.0 3.7 1.455 (52.5−57.5) 4.5 3.3 1.260 (57.5−62.5) 4.0 2.9 1.065 (62.5−67.5) 3.5 2.6 0.970 (67.5−72.5) 2.5 2.3 0.875 (72.5−77.5) 1.8 2.0 0.780 (77.5−82.5) 1.2 1.8 0.685 (82.5−87.5) 0.9 1.5 0.690 (87.5−92.5) 0.7 1.4 0.595 (92.5−97.5) 0.3 1.2 0.5100� �0.1 �1.0 �1.0


debris reported in many publications (see Ibrahim et al.2010).

The fallout arrival time was assessed by calculatingthe time for various particle sizes released from eachaltitude to be deposited in specific deposition domainsused to define specific atolls in the Marshall Islands.Three separate deposition domain sizes were used toestimate average deposition density (Bq m�2) for theregion surrounding an atoll. As an approximation of theland area encompassed by an atoll, each atoll area wasapproximated by a deposition domain defined by themost extreme point of each atoll’s boundaries in north,south, east and west directions. To account for possibleprediction error resulting from inadequate meteorologi-cal data or due to the statistical limitations imposed bysimulating too few particles, a larger domain was alsodefined by increasing the size of the original domain byan additional 50% in both longitude and latitude. In thecase of very small atolls, increasing the domain sizeusing the method just described did not make a signifi-cant difference in the number of particles counted. Forthat reason, a third domain size, measuring 1 squaredegree, was also used in simulations. For large atollssuch as Kwajalein, the areas represented by the 1 squaredegree domain and the smaller rectangular areas were notgreatly different. By comparing the estimated depositiondensity averaged within each of the three differentdomains, we could assess the sensitivity of the estimateddeposition density to small spatial variations in particletrajectories, as well as the precision of the depositiondensity estimates since, for the smaller atolls, the depos-ited fraction of the 10,000 particle source term was oftenvery small. Thus, the precision of estimates based on thesmaller domains was often poor.

In addition to the primary particle size-activitydistribution (labeled MI in Table 1), two alternativedistributions were also used in the Castle Bravo simula-tion to investigate the sensitivity of the 137Cs depositiondensity estimates to the assumed size-activity distribu-tion. Because most of the atolls of interest were at largedistances (hundreds of kilometers) from the Bikini Atolltest site, the deposited particles were mostly less than 50�m in diameter; thus, the estimated 137Cs depositiondensities for most atolls were not highly sensitive to theassumed particle size-activity distribution.

The estimated deposition density of 137Cs at anyatoll will be sensitive to the assumed spatial distributionof activity in the cloud, particularly, the spatial distribu-tion of the smaller particles that carried most of the 137Cs.However, as discussed earlier, the actual spatial distribu-tion of activity in the cloud probably varies with thelocation, yield, and conditions of the test. For this reason,errors in the exact meteorology tended to have a much

larger impact on the estimated deposition at a specificatoll than did assumptions inherent in the debris cloudmodel, resulting in (1) simulations failing to predictfallout at an atoll when it was known to actually haveoccurred, and (2) predicting fallout arrival times thatwere much later than the reported or assumed arrivaltime.

In order to estimate the relative impact of theHYSPLIT wet deposition model on the predicted falloutdeposition, each simulation was run twice, once withprecipitation processing enabled and once with precipi-tation processing disabled.

General results. Predicting deposition density (Bqm�2), usually of 137Cs, was of potential use in assessmentsof radiation doses in the Marshall Islands (Beck et al. 2010;Bouville et al. 2010; Simon et al. 2010a, 2010b). How-ever, evaluating the accuracy of the simulations of deposi-tion density for the Marshall Islands tests was hindered bythe absence of a large and consistent set of empirical groundmeasurements of radioactivity following individual nucleartests as well as accurate meteorological data. The reliabilityof trajectories in close proximity to the test site could beinferred from comparisons between observed wind data atthe test site, and the initial wind speed and direction used bythe model. The HYSPLIT-interpolated wind data resultingfrom the meteorological input data sets used in simulationswere compared with the actual wind speed and direction atmany different altitudes reported from measurements at thetest sites (DNA 1979). Direct observations of wind speedand direction as a function of altitude downwind from thetest sites were not available except for a few tests, and inthose cases, the observations were available at only onelocation downwind. For these reasons, no other systematiccomparisons of wind speed and direction could be made. Ingeneral, the agreement between HYSPLIT simulated dep-osition and available measurement data tended to be muchbetter when the initial wind speed and direction from themeteorological reanalysis data (interpolated by HYSPLIT)were similar to that measured at the test atoll at the time ofthe test.

Because our assumptions used to estimate activityper particle were crude, and because the meteorologicalinput data had limited accuracy, the HYSPLIT simula-tions could only yield estimates of deposition densitywith significant uncertainty. Even determining whetheror not any fallout had even occurred was often difficultgiven the generally coarse resolution of the meteorolog-ical input data. Our findings indicate that even a rela-tively small error in wind direction at any altitude belowthe particle release height can result in errors in themagnitude of deposition density downwind. Further-more, even when the actual meteorological conditions at


the test site were in agreement with the meteorologicalinput data used by the model, there is no guarantee thatthe model will predict correct depositions because themeteorology downwind can also be in error.

Significant differences in predicted fallout werefrequently observed when precipitation processing wasenabled as compared to when precipitation processingwas disabled. As shown in Table 2, the differences indeposition were occasionally as great as a factor of three,although generally much less. Often, the amount offallout was reduced at a given atoll when precipitationprocessing was enabled, apparently due to cloud deple-tion at upwind locations. At other times, predicteddeposition was higher, suggesting local rainout or wash-out. The limitations in the HYSPLIT wet depositionmodel (in common with other meteorological models),and the lack of precipitation data on an event-specificbasis, may have contributed, in some cases, to thequantitative differences seen between the measured fall-out deposition and model simulated estimates at somelocations at some times.

Based on comparisons for tests where significantmonitoring data were available, predicted depositionsusing the HYSPLIT three-dimensional particle modelcoupled with NCEP/NCAR reanalysis meteorologicaldata agree with measured 137Cs densities to within afactor of ten. This was almost always true when theinitial wind speed and direction of the meteorologicalinput data agreed reasonably well with that of the actualwind data at the test site.

Inadequate meteorological data were generally thelimiting factor in the HYSPLIT model’s ability to predictaccurate arrival times of fallout for Marshall Islandstests, although high-quality data on actual arrival timeswere also often lacking due to few measurements havingbeen made and some inconsistencies between the avail-able measurements (Beck et al. 2010). These conditionsmade comparing model-based estimates with measurementsa difficult exercise. A comparison of model-predicted fall-out arrival times with reported best estimates (Beck et al.2010) is provided in Table 3.

As discussed, the HYSPLIT model uses a simplerainout and washout model that may not adequatelysimulate such complex processes, particularly for therelatively large amounts of debris in nuclear test clouds.For this reason and because of the normal high frequencyof precipitation events in the Marshall Islands, most ofwhich would not have been recorded in archival meteo-rological data sets, particularly in the southern atolls(Beck et al. 2010) where rainfall is the highest, theHYSPLIT simulations may not have predicted someactual deposition events. Moreover, it is possible thatsome of the differences in the HYSPLIT-predicted fall-out arrival times as compared to the generally earlierarrival times observed could be a result of falloutdeposited as a result of precipitation scavenging fromunrecorded precipitation events.

Test-specific results. The results of simulations offallout from five nuclear tests in the Marshall Islands arediscussed here. Simulation results were compared againstexisting measurement data when possible, although an-ecdotal reports of fallout from test participants wereconsidered as well. Further details on the characteristicsand dates of the tests are given in an appendix of acompanion paper (Beck et al. 2010).

Greenhouse Dog was a pure fission device whichwas detonated on Enewetak Atoll on 7 (GMT) April1951. No radiological survey data are available for theDog test. However, the HYSPLIT simulations suggestthat small amounts of fallout could have occurred atseveral atolls in the Marshall Islands includingUjelang (Fig. 1), Wotho, Kwajalein, and Utrik. Model-predicted wind speed and direction agree fairly wellwith the observed values reported in DNA (1979)(Table 4). Dog is an example where HYSPLIT simu-lations were particularly valuable because there are nohistorical monitoring data.

Greenhouse Item was a fusion device detonated onEnewetak Atoll on 24 (GMT) May 1951. Similar to thesituation for Dog, no radiological survey data exist forthe Item test. The HYSPLIT simulation suggests therewas significant fallout at Ujelang Atoll (Fig. 1). Because

Table 2. Comparison of HYPSLIT simulations of 137Cs depositiondensity (Bq m�2) with and without wet deposition for selected testsand atolls.

Atoll/Test

Wet depositiondisabled

Wet depositionenabled

Atolldomain

1 degreedomain

Atolldomain

1 degreedomain

Fir (11 May 1958)a

Kili 30 22 15 22Ebon 30 30 40 22Mejit 78 100 110 100

Flathead (11 June 1956)a

Namorik 340 300 140 300Wotho 5,200 3,300 4,100 3,000

Nectar (13 May 1954)a

Namorik 40 160 85 160Kili 74 40 48 40Jabat 130 85 180 110Lib 850 520 560 520Lae 440 740 560 780

Dog (7 April 1951)a

Kwajalein 5.2 3.3 9.3 6.7Ujelang 24 110 60 190Lae 0 2.6 3.0 5.6

a All dates GMT.


the wind speed and direction used in the HYSPLITsimulation are in good agreement with data collected atthe test site (Table 5), we assumed that significant falloutmost likely did occur at Ujelang. A comparison of thetotal estimated 137Cs deposition from tests after 1951 withsoil sample data (Beck et al. 2010) also suggested thatthere was significant fallout at Ujelang Atoll from thistest. The combination of empirical evidence and sim-ulation supports the supposition that the missingfallout had to occur from either the Item or Dog test,or possibly both.

Castle Bravo, the largest U.S. test ever and the testthat resulted in the largest individual exposures (Simon1997), was a thermonuclear device detonated on 28(GMT) February 1954 with a reported yield of 15 Mt. Aconsiderable amount of monitoring data is available forthe northern Marshall Islands for the Bravo test includingdata from fixed-wing aircraft as well as ground surveysof many atolls (Beck et al. 2010; Breslin and Cassidy1955). In addition, gummed film (GF) data were col-lected at Kwajalein and Majuro Atolls; automatic con-tinuous exposure rate monitors were also in operation atMajuro and Ujelang (Beck et al. 2010; Breslin andCassidy 1955) during the Bravo test. Unfortunately, noGF data were available for Kwajalein covering the daysimmediately following the test; however, later dataindicate that there was significant fallout several daysafterwards, suggesting that additional fallout may haveoccurred at other atolls after the air monitoring hadceased. Since some of the atolls were surveyed only onceby air at about 2 d after the test (Breslin and Cassidy1955), there is also a strong possibility that additionalfallout occurred after the surveys, particularly at Kwaja-lein, Wotho, and atolls south of Kwajalein. Daily GFmeasurements at Kwajalein and Majuro (Beck et al.2010) often indicated that fallout persisted for many daysafter the air surveys were completed. The HYSPLITsimulations for Bravo proved useful in helping to inter-polate fallout deposition over some of the southern atollswhere the airplane monitoring data were either sparse orsuspect because of known instrument problems. The

model predicted significant fallout at Ujae and Laecontrary to survey data, although estimates at most otheratolls agree within an order of magnitude (Fig. 2). Theagreement between HYSPLIT and DNA (1979) winddata at the detonation site was poor at some altitudes(Table 6) suggesting a reason for the slight shift in theHYSPLIT-predicted fallout pattern compared to the ob-served pattern of deposition.

The Redwing Flathead test was detonated at BikiniAtoll on 11 (GMT) June 1956 with a reported totalexplosive yield of 356 kt of which, according to unoffi-cial reports, was �73% from fission. The pattern ofHYSPLIT-predicted fallout for the Flathead test wasgenerally consistent with the few available monitoringdata. However, HYSPLIT-predicted 137Cs deposition wasabout a factor of three higher than the GF measurementfor Kwajalein and about a factor of twenty higher thanthe survey data for Wotho.

Test Fir, in the Hardtack I series, was a thermonu-clear test which was detonated on Bikini Atoll on 11(GMT) May 1958. Based on monitoring data at Utrik,Ujelang, Wotho, and Rongelap Atolls, as well as GF dataat Kwajalein Atoll, only very light fallout occurred in theMarshall Islands from any of the 35 tests in 1958,including Fir. HYSPLIT simulations of other 1958 testsindicated that the Fir test was probably the most signif-icant contributor to regional deposition during 1958.Model predictions for the Fir test indicated that most ofthe fallout occurred in areas to the east of the test site andthat little fallout occurred south of Kwajalein, consistentwith the available monitoring data. The estimated 137Csdeposition agreed with 137Cs deposition density monitor-ing data within a factor of three at Ujelang and Utrik andwithin a factor of 5 to 10 at the three other sites (Wotho,Kwajalein, and Rongelap).

Summary of Marshall Islands simulations. Theavailability of high quality three-dimensional and tem-poral meteorological data is a key factor for success inpredicting the arrival time and location of fallout depo-sition. The comparisons presented here indicate the

Table 3. Comparisons of time of arrival (TOA) predicted by HYSPLIT with estimates based on measurements.

TestTest date(GMT) Test site

Distance totest site (km)

TOA: bestestimate (h)

TOA: HYSPLITestimate (h) % difference

Bravo 28 Feb 1954 Rongelap 180 5.6 4.5 �20Ujelang 521 18 40 122Majuro 836 48 102 113

Romeo 26 Mar 1954 Rongelap 180 �30 97 223Kwajalein 426 100 104 4Majuro 836 100 118 18

Yankee 4 May 1954 Rongelap 180 �30 5 �83Kwajalein 426 35 154 340Utrik 486 30 105 250


importance of accurate upper-air wind data which largelyinfluence the trajectory of the radioactive cap cloud thatcontains over 80% of the radioactivity. There were,however, only a few weather stations in the Marshall

Islands during the years of nuclear testing (1946–1958)and they were located at significant distances from thenuclear weapons test sites. Thus, there were limitedmeteorological data collected that were directly relevant.

Fig. 1. Top Panel: HYSPLIT-predicted fallout pattern near Ujelang Atoll resulting from the Dog test. Bottom Panel:HYSPLIT-predicted fallout pattern at Ujelang Atoll resulting from the Item test.


Other sources of weather observations from this regionare largely conjectural, but may have included datacollected from passing ships at sea and aircraft. Compar-isons with the actual wind data from the test site (DNA1979) and the model-predicted wind data at the test siteresulted in only sporadic agreement between the two.Furthermore, Kistler and Kalnay (2000) indicated thatupper-air rawinsonde observations were inconsistent andvery few in the tropics during this time period resultingin reanalysis forecasts of poor quality. For these variousreasons, the meteorological input data used for HYSPLITsimulations, solely based on the NCAR/NCEP reanalysismodel, did not often reproduce downwind meteorologi-cal conditions accurately enough to predict trajectories ofthe radioactive debris clouds with strong certainty. De-spite these limitations, the simulations proved useful forindicating which tests could have impacted the MarshallIslands and which likely did not, particularly for yearswith no actual monitoring data (Beck et al. 2010).Simulations were also useful for interpolating betweenactual monitoring data for atolls that were not surveyed.

Nevada Test Site: Upshot-Knothole HarryUpshot-Knothole Harry was a 32 kt fission device

detonated on 19 May 1953 at the NTS. The Harry testwas simulated using the HYSPLIT model with thepurpose of comparing deposition patterns and falloutarrival times to published data (Beck et al. 1990; Beckand Anspaugh 1991) as well as with the meteorologicalmodeling results of Cederwall and Peterson (1990).

Particle sizes and release heights for the Harrysimulation closely followed those selected by Cederwalland Peterson (1990). Trajectory endpoint calculationsand fallout pattern plots produced by Cederwall andPeterson (1990) indicated two diverging air masses atapproximately H�10 h downwind. At lower altitudes,particles were shown to deposit at latitudes between thenorthern border of New Mexico and Denver, CO, whileparticles aloft deposited further to the south at latitudesbetween Cedar City, UT, and Albuquerque, NM.Using this information, two separate simulations wereperformed, each modeling the respective bottom andtop halves of the debris cloud. The particle sizes usedin the simulation ranged from 5 to 1,000 �m indiameter (Table 7).

Deposition parameters in our Harry simulationsdiffered from those used by Cederwall and Peterson(1990). Cederwall and Peterson (1990) chose a fixeddeposition velocity of 0.005 m s�1 to represent a range ofvalues for various radionuclides and modeled only wash-out, not rainout, assuming that precipitation removedonly airborne material below the radioactive debriscloud. As stated previously, in applying HYSPLIT, thedeposition velocity was assumed to be only attributed togravitational settling. Also, both below-cloud and in-cloud wet deposition processes were simulated. It shouldbe noted that the model used by Cederwall and Peterson(1990) incorporated a washout coefficient dependent onprecipitation rate; in contrast, in the HYSPLIT model,the rainout coefficient is dependent upon the precipita-tion rate and the washout coefficient is independent of it.

Results. The fallout patterns and fallout arrivaltimes resulting from the Harry simulation agree reason-ably well with those reported by Cederwall and Peterson(1990), but there are some noticeable differences. Fig. 3shows that at H�12 h, the HYSPLIT-predicted debriscloud had entered Colorado and New Mexico and byH�18 h the debris cloud had reached Denver in the northand crossed into New Mexico much further south. Thefallout pattern produced by HYSPLIT appears to agreewith the estimated centerline of the plume produced byCederwall and Peterson (1990). However, the patternsdisagree in some locations. Cederwall and Peterson’spattern is broader north to south. Furthermore, the

Table 4. Comparison of wind speed and direction at time ofdetonation at the Enewetak test site for the 7 April 1951 (GMT)Dog test.

Altitude(m)

DNA (1979) HYSPLIT

Wind speed(km h�1)

Wind direction(deg)


Wind direction(deg)

1,524 48 80 45 813,048 35 80 27 574,572 38 70 21 356,096 35 30 19 227,620 19 300 19 1349,144 50 280 23 298

10,668 47 230 37 26612,192 53 220 45 25913,716 42 280 45 26215,240 35 310 32 27516,764 50 340 19 288

Table 5. Comparison of wind speed and direction at time ofdetonation at the Enewetak test site for the 24 May 1951 (GMT)Item test.

Altitude(m)

DNA (1979) HYSPLIT


Wind direction(deg)


Wind direction(deg)

1,524 26 90 21 1033,048 8 90 16 1144,572 14 260 11 506,096 14 290 13 27,620 19 250 19 1479,144 16 360 23 352

10,668 14 250 18 31812,192 13 280 13 28013,716 — — 11 27815,240 — — 11 31016,764 — — 11 345


deposition estimates do not agree well quantitatively. Ingeneral, HYSPLIT indicated that the more significantdeposition occurred slightly north of the location re-ported in Cedarwall and Peterson (1990). For example,the HYSPLIT simulation indicated little fallout at St.George, UT, a location in which significant fallout is

known to have occurred (Anspaugh and Church 1986;Beck and Anspaugh 1991). This difference is likely dueto disagreements between the wind data at downwindgrid locations used in our simulations as compared to thesimulations of Cederwall and Peterson (1990). Cederwalland Peterson (1990) also reported that adjustments tocertain fallout parameters in their model were needed tocreate an agreement between simulation results andmeasurement data.

Semipalatinsk Nuclear Test Site: Test #1The first Soviet nuclear detonation took place on 29

August 1949 with a yield of 22 kt. This detonation was asurface burst at the Semipalatinsk Test Site in Kazakh-stan. The test is believed to have been identical inconstruction to the first U.S. nuclear test, Trinity (Rhodes1986), conducted in New Mexico in 1945. The maximumcloud height was �9 km. At the time of the test, there

Fig. 2. Ratio of predicted 137Cs deposition density (Bq m�2) from simulations using the NOAA-HYSPLIT model anddeposition density (Bq m�2) inferred from available measurement data from the Bravo test.

Table 6. Comparison of wind speed and direction at time ofdetonation at the Bikini test site for the 28 February 1954 (GMT)Bravo test.

Altitude(m)

DNA (1979) HYSPLIT


Wind direction(deg)


Wind direction(deg)

1,524 16 100 27 1363,048 10 310 16 3034,572 24 290 19 2826,096 24 380 32 2647,620 35 260 47 2539,144 48 250 58 250

10,668 64 240 66 25812,192 64 230 71 26513,716 84 250 69 26415,240 58 250 56 28516,764 29 200 42 30818,288 — — 24 33121,336 — — 10 12424,384 — — 23 7427,432 — — 45 9230,480 — — 72 96

Table 7. Particle size distribution used for the Harry simulation.

Particle size range (�m) Increment in range (�m)

5 to 50 560 to 100 10

125 to 300 25350 to 700 50800 to 1,000 100


were strong northeasterly winds estimated at 47–60 kmh�1 with almost no wind shear (Shoikhet et al. 1998;Imanaka et al. 2005).

HYSPLIT simulations were performed to comparepredicted estimates of 137Cs deposition density withpublished estimates of the fallout pattern and the spatialdistribution of 137Cs and 239,240Pu near the village ofDolon, Kazakhstan. Fallout was reported to have reachedDolon, approximately 118 km northeast of ground zero,at roughly H�2 h (Yamamoto et al. 2008; Gordeev et al.2002). Soil samples were collected in 2005 byYamamoto et al. (2008) at 21 locations along a lineapproximately perpendicular to the supposed centerlineof the plume. Their analyses of 137Cs and 239,240Pusuggested that (1) the spatial distribution of 137Cs and239,240Pu is roughly Gaussian in shape and perpendicularto the axis of the fallout trajectory with maximumslocated near the supposed axis-center, and (2) the widthof the fallout pattern near Dolon was approximately8–10 km (Yamamoto et al. 2008).

To calculate the fallout deposition at locationsdownwind, particle releases at varying altitudes weresimulated using HYSPLIT. The total number of particlesin the simulated debris cloud was apportioned using theestimated total 137Cs activity, particle size, and spatialdistribution model described previously for the MarshallIslands simulations as well as the alternate distributions

given in Table 1. The release heights in the simulationranged from 450 m AGL to the reported maximum cloudheight, �9 km, and are shown in Table 8. The cloudbottom, estimated at 2.7 km, was based on the reportedcloud dimensions of the Trinity test. The assumedparticle sizes varied from 5 �m up to 300 �m, in 5 �mincrements, depending on the activity distribution, andthe simulation was carried out to 5 h post-detonation. Thetotal number of particles tracked in different simulationsvaried from 1 � 107 to 2.5 � 107.

Results. The calculated 137Cs deposition pattern forthe first Soviet nuclear detonation, using the MI 137Csactivity-size distribution from Table 1, is shown in Fig. 4.The simulation data were gridded at a resolution of �4.3km2. Sites A, B, and C in Fig. 4 correspond to the

Fig. 3. HYSPLIT fallout pattern resulting from simulations of the Upshot-Knothole Harry test. The solid black lineindicates the estimated centerline of the radioactive cloud as simulated by Cederwall and Peterson (1990). Dashed linesdelineate the HYSPLIT predictions of the geographic boundary of the fallout pattern at the time noted.

Table 8. Particle release heights and fraction of total 137Cs activitycorresponding to each release height for the first test at theSemipalatinsk Nuclear Test Site.

Release height(m AGL)

Height increment(m AGL)

Fraction of total137Cs activity

450 to 2,700 450 0.122,875 to 3,750 175 0.103,925 to 4,800 175 0.165,150 to 6,900 350 0.367,075 to 7,950 175 0.168,125 to 9,000 175 0.10


Yamamoto et al. (2008) sampling locations. Location Dis the grid cell with the highest HYSPLIT-predicteddeposition density at �118 km downwind. Sites A and Care the respective northern and southern-most sites listedin the Yamamoto et al. (2008) study. The predicted

fallout arrival time at Dolon is reasonably consistent withthat reported by Gordeev et al. (2002) and Yamamoto etal. (2008) at approximately 2 to 3 h, but the spatialdistribution of 137Cs is much wider than that inferred bythe measurements of Yamamoto et al. (2008), and the

Fig. 4. HYSPLIT predictions of 137Cs deposition pattern near the Semipalatinsk Test Site in Kazakhstan following event1 (29 August 1949) using: MI particle-size distribution (Top Panel), alternate particle-size distribution #1 (Center Panel),and alternate particle-size distribution #2 (Bottom Panel), all described in Table 1. The key in each panel comparessimulation results and measurements reported by Yamamoto et al. (2008).


direction of the 137Cs deposition pattern is significantlyfurther to the north than the Yamamoto et al. (2008) data.However, the deposition patterns resulting from thealternative 137Cs activity-size distributions given in Table1 (Fig. 4) shift the HYSPLIT fallout pattern closer to thatreported in Yamamoto et al. (2008) and clearly indicatethat the predicted 137Cs deposition is very sensitive to theestimated fraction of 137Cs on particles greater than 50�m. As discussed earlier, and shown in Table 1, thedistribution used for the Marshall Islands simulationsassumed only about 20% of the 137Cs activity on particlesgreater than 50 �m, while the alternate distributionsassume a larger fraction of 137Cs on particles of diametergreater than 50 �m. As shown in Fig. 5, where theactivity on various size groups of particles is plottedseparately, the HYSPLIT simulation indicates that mostof the particles depositing in the vicinity of Dolon, andparticularly along the axis of the fallout pattern, to begreater than 50 �m in diameter, while the particlesfurther from the centerline were generally less than 50�m. However, even assuming a greater fraction of 137Csactivity on large particles, the HYSPLIT pattern stilldeviates from the axis of the Yamamoto data and is muchbroader, presumably reflecting the wind shear present in

the meteorological input data. HYSPLIT air mass trajec-tories (Fig. 6) clearly illustrate the significant wind shear,which is inconsistent with the assertion that wind shearwas minimal (Shoikhet et al. 1998; Imanaka et al. 2005).

Although the estimated peak 137Cs deposition den-sity predicted by HYSPLIT in the vicinity of Dolon(Table 9) is slightly closer to that measured byYamamoto et al. (2008) (corrected for decay) for thealternative particle-size distributions than for the MIdistribution, the HYSPLIT maximum deposition densitynear Dolon is still much lower than the maximummeasured by Yamamoto et al. (2008) (corrected fordecay). However, this is to be expected since, as a resultof the predicted wind shear, the HYSPLIT fallout isspread over a wider area compared to the Yamamoto etal. (2008) soil data and, thus, is diluted.

We surmise the shift of the HYSPLIT pattern to thenorth, compared to the measurements of Yamamoto et al.(2008), to be a result of wind shear in our meteorologicaldata and other limitations of those data. However, it mayalso partly reflect the fact that the 137Cs activity-sizemodel used in HYSPLIT simulations is too crude andmay not be apportioned to give enough of the total 137Csactivity on the larger particles that deposit closer to the

Fig. 5. HYSPLIT-predicted fallout patterns of different particle sizes at the Semipalatinsk Test Site in Kazakhstanfollowing event 1 (29 August 1949).


centerline of the Yamamoto et al. (2008) pattern. Thiscould account for a broadening of our predictions. Itshould also be noted, however, that the Yamamoto et al.(2008) soil sample data were obtained almost a halfcentury after the test and may not accurately reflect theoriginal deposition pattern because of weathering andredistribution. The Yamamoto measurement data exhibitsignificant scatter and many of the samples were takenover bare soil where those processes could have beenparticularly important. This is particularly true for thesamples taken in Dolon itself. Thus, the true width of theoriginal 137Cs deposition pattern may lie somewherebetween that predicted by the contemporary soil mea-surements and that predicted by the HYSPLIT model.

The HYSPLIT simulations reflect, at least, the sameorder of magnitude of the peak 137Cs deposition density

in the vicinity of Dolon, taking into account the dilutionof the HYSPLIT maximum deposition density as a resultof the additional dispersion of the fallout about the axis.They also illustrate the impact of fractionation on therelative deposition density of volatile nuclides such as137Cs (i.e., deposition of small particles) compared torefractory elements such as 239,240Pu (i.e., deposition oflarge particles). As illustrated in Yamamoto et al. (2008),the soil data clearly show a different dispersion (patternwidth) about the fallout pattern centerline for 239,240Pu asopposed to 137Cs, as expected since the 239,240Pu is mostlyon large particles. The pattern of particle size and 137Csdeposition indicated by the HYSPLIT model is qualita-tively consistent with that expected from highly fraction-ated local fallout (see companion paper by Ibrahim et al.2010).

Fig. 6. HYPSLIT air mass trajectories illustrating wind shear derived from archival meteorological data that wereinconsistent with reported actual weather conditions at the Semipalatinsk Test Site at time of detonation. Symbols areplotted in 1 h intervals. The wind shear resulted in different sized particles depositing in the vicinity of Dolon over awider area (Fig. 5) than indicated by retrospective soil sample analyses.

Table 9. Comparison of HYSPLIT predicted peak 137Cs deposition density using three different particle sizedistributions with decay corrected measurements of Yamamoto et al. (2008).

DistributionHYSPLIT-predicted

maximum (kBq m�2)

HYSPLIT prediction atlocation of Yamamoto et al.

(2008) axis (kBq m�2)

Measurements of 137Csfrom Yamamoto et al.

(2008) at axis (kBq m�2)Ratio: Yamamotodata to HYSPLIT

MI (from Table 1) 2.2 0.5 12−16 24 to 32Alternate #1 3.4 1.6 12−16 7 to 10Alternate #2 1.4 0.8 12−16 15 to 20


APPLICATION OF HYSPLIT TO MARSHALLISLANDS FALLOUT ASSESSMENT

HYSPLIT model simulations were used to supportthe analysis of deposition of fallout in the MarshallIslands and related dose assessments (Beck et al. 2010;Bouville et al. 2010; Simon et al. 2010a, 2010b). Thisapplication is described below.

Simulations of air mass trajectories and depositionpatterns were used to help make 137Cs deposition densityestimates for specific locations in the Marshall Islandsfor tests in which fallout monitoring data were eithersparse or nonexistent. The results were used to assist ininterpolations of deposition at atolls where no monitoringdata were available using measurements of deposition atnearby atolls (Beck et al. 2010).

Despite the uncertainty found after testing the modelunder conditions of questionable input data and limitedmeasurement data, the model-based deposition estimateswere useful in estimating fallout deposited in the Mar-shall Islands (Beck et al. 2010) in certain specific cases.For example, the HYSPLIT predictions were the onlysource of information on fallout deposited and oftensupported anecdotal reports of significant fallout prior to1952 when there were no monitoring data. This isparticularly true at Ujelang Atoll where anecdotal reportsindicated fallout resulting from the 1951 GreenhouseDog and Item tests but no actual measurements werereported. Additionally, the HYSPLIT model simulationswere used to support interpolations of deposition andtime of arrival, and to fill out the estimated depositionpatterns from the 1956 and 1958 tests for which only afew (4 to 6) atolls were monitored. For example, for the1956 Flathead test, model predictions were used inconjunction with GF measurements at Kwajalein andsurvey measurements at Ujelang, Wotho, Rongelap, andUtrik, to estimate fallout at atolls south and east ofKwajalein where no actual measurements were made.Only very low levels of fallout deposition were predictedin other areas. Similarly, for the 1958 Fir test, theHYSPLIT simulations were used to aid in estimating therelatively low levels of deposition at atolls not moni-tored. As discussed in Beck et al. (2010), a high uncer-tainty estimate (a probability distribution functionwith a geometric standard deviation of 3.0) wasapplied to the HYSPLIT-based deposition densityestimates.

Although uncertain, the HYSPLIT results had rela-tively little impact on estimates of total fallout in theMarshall Islands presented in Beck et al. (2010) becausethe model-predicted fallout estimates used were almostalways small compared to fallout levels from tests withmonitoring data. This was the case, for example, for the

1956 and 1958 tests compared to the 1954 Castle tests.For this reason, the uncertainty contribution of theHYSPLIT simulations to the overall uncertainty of theestimated external and internal doses (Bouville et al.2010; Simon et al. 2010a) was also small.

CONCLUSION

A well-established meteorological model, HYSPLIT,was tested for its ability to predict dispersion and depositionof nuclear test-related fallout at varying distances down-wind. The model was evaluated by comparing model-predicted deposition patterns and arrival times againstmeasured deposition density. Particles of varying sizes werereleased from a range of starting heights to represent astabilized radioactive debris cloud. Deposition domainswere defined to track the locations of deposited particles soas to test the model’s ability to predict downwind deposi-tions of differently sized particles at specific locations inagreement with known patterns.

Because of the general limited availability ofground-based radiological measurement data, it is verydifficult to separate the relative contributions of differentfactors to the overall predictive ability of HYSPLIT.Results from our simulations suggest that the accuracyand spatial resolution of the meteorological input data isone of the most important factors in modeling falloutwith the HYSPLIT model since the advection and dis-persion calculations directly depend on the meteorolog-ical data. The simplification of physical processes andthe particle distribution that is assumed in the debriscloud model, as well as in the wet deposition modelimplemented in HYSPLIT, all may have had an impacton the quantitative predictions of fallout at a particularatoll. When relatively accurate wind data were used,however, we confirmed that the model-predicted depo-sition is reasonably consistent with available groundmeasurement data.

In our simulations, meteorological reanalysis datawere used. Although methods in data assimilation andreanalysis have greatly improved, reanalyses prior to thegeophysical year (1957–1958) still suffer from a lack ofsatisfactory observations. However, the HYSPLIT modelwas able to predict reasonably accurate fallout arrivaltimes for simulations in which the meteorological reanal-ysis data were consistent with observed data at severalaltitudes within the cap of the stabilized debris cloud atthe test site. Under those conditions, model-predictedarrival times were often within several hours of thosereported. Conversely, when the reanalysis data did notagree with the local observed wind measurements, falloutarrival times and depositions deviated, largely in somecases, from reported values in the literature.


The type of meteorological data provided by mostreanalysis models makes it difficult to model wet depo-sition. Atmospheric models may incorporate simplifyingassumptions for wet deposition, as does the HYSPLITmodel, or may disregard the wet removal process alto-gether. Wet deposition can be a large contributor tofallout under conditions of precipitation and can oftenlead to localized pockets of elevated radioactivity. Thus,to accurately model fallout through computer simulation,a refined wet deposition model would be needed inconjunction with accurate precipitation data.

The assessment of fallout deposition in the MarshallIslands discussed in a companion paper (Beck et al.2010) is a good example of retrospective modeling of thedispersion and deposition of fallout using the HYSPLITmodel. HYSPLIT predictions of fallout deposition can beused for supplementing existing ground-based falloutmeasurement data, particularly when no ground-basedfallout measurement data are available. In such cases,HYSPLIT can be used to indicate whether or not falloutmight have occurred at a particular location and provide,at minimum, crude estimates of the magnitude of thedeposited activity. This, in itself, can be a very valuableasset for the reconstruction of past fallout events.

Acknowledgments—This work was supported by the Intra-Agency agree-ment between the National Institute of Allergy and Infectious Diseases andthe National Cancer Institute, NIAID agreement #Y2-Al-5077 and NCIagreement #Y3-CO-5117. The authors are indebted to Roland Draxler andBarbara Stunder of the NOAA Air Resources Laboratory in Silver Spring,MD, for their willingness and generosity in providing instruction andinformation about the operation and theory of the HYSPLIT model(http://ready.arl.noaa.gov/HYSPLIT.php).

REFERENCES

Anspaugh L, Church B. Historical estimates of external expo-sure and collective external exposure from testing at theNevada Test Site through Hardtack II, 1958. Health Phys52:35–51; 1986.

Beck HL, Anspaugh LR. Development of the county database:estimates of exposure rates and times of arrival of fallout inthe ORERP Phase-2 area. Washington, DC: U.S. Depart-ment of Energy; DOE/NV-320; 1991.

Beck HL, Helfer IK, Bouville A, Dreicer M. Estimates offallout in the western U.S. from Nevada weapons testingbased on gummed-film monitoring data. Health Phys59:565–570; 1990.

Beck HL, Bouville A, Moroz BE, Simon SL. Fallout depositionin the Marshall Islands from Bikini and Enewetak nuclearweapons tests. Health Phys 99:124–142; 2010.

Bouville A, Beck HL, Simon SL. Doses from external radiationto Marshall Islanders from Bikini and Enewetak nuclearweapons tests. Health Phys 99:143–156; 2010.

Breslin AJ, Cassidy ME. Radioactive debris from OperationCastle, islands of the mid-Pacific. Washington, DC: U.S.Atomic Energy Commission; 1955.

Cederwall RT, Peterson KR. Meteorological modeling ofarrival and deposition of fallout at intermediate distancesdownwind of the Nevada Test Site. Health Phys 59:593–601; 1990.

Crocker G, O’Connor J, Freiling E. Physical and radiochemicalproperties of fallout particles. San Francisco, CA: U.S.Naval Radiological Defense Laboratory; Report USNRDL-TR-899; 1965.

Defense Nuclear Agency. Compilation of local fallout datafrom test detonations 1946–1962 extracted from DASA1251. Washington, DC: DNA; DNA 1251-2-EX; 1979.

Draxler RR. HYSPLIT4 user’s guide. Silver Spring, MD:National Oceanic and Atmospheric Administration; ERLARL-230; 1999.

Draxler RR, Hess GD. Description of the hysplit4 modelingsystem. Silver Spring, MD: National Oceanic and Atmo-spheric Administration; ERL ARL-224; 1997.

Draxler RR, Hess GD. An overview of the hysplit4 modelingsystem for trajectories, dispersions, and deposition. Austra-lian Meteorological Magazine 47:295–308; 1998.

Freiling EC, Crocker GR, Adams CE. Nuclear debris forma-tion. In: Radioactive fallout from nuclear weapons tests.Proceedings of an USAEC Conference. Washington, DC:U.S. Atomic Energy Commission; 1965: 1–41.

Gordeev K, Vasilenko I, Lebedev A, Bouville A, LuckyanovN, Simon SL, Stepanov Y, Shinkarev S. Fallout fromnuclear tests: dosimetry in Kazakhstan. Radiat Environmen-tal Biophys 41:61–67; 2002.

Heidt WB, Schuert EA, Perkins WW, Stetson RL. Nature,intensity, and distribution of fallout from MIKE shot. SanFrancisco, CA: U.S. Naval Radiological Defense Labora-tory; Report WT-615; 1953.

Hoecker WH, Machta L. Meteorological modeling of radioio-dine transport and deposition within the continental UnitedStates. Health Phys 59:603–619; 1990.

Ibrahim SA, Simon SL, Bouville A, Melo D, Beck HL.Alimentary tract absorption (f1 values) for radionuclides inlocal and regional fallout nuclear tests. Health Phys 99:233–251; 2010.

Imanaka T, Fukutani S, Yamamoto M, Sakaguchi A, Hoshi M.Width and center-axis location of the radioactive plume thatpassed over Dolon and nearby villages on the occasion ofthe first USSR A-bomb test in 1949. Radiat Res 46:395–399; 2005.

Izrael YA. Radioactive fallout after nuclear explosions andaccidents. In: Baxter MS, ed. Radioactivity in the environ-ment (vol. 3). Oxford, UK: Elsevier Science Ltd.; 2002.

Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D,Gandin L, Iredell M, Saha S, White G, Wollen J, Zhu Y,Leetmaa A, Reynolds R, Chelliah M, Ebisuzaki W, HigginsW, Janowiak J, Mo KC, Ropelewski C, Wang J, Jenne R,Joseph D. The NCEP/NCAR 40-year reanalysis project.Bulletin American Meteorological Society 77:437–471;1996.

Kinser AM. Simulating wet deposition of radiocesium from theChernobyl accident. Columbus, OH: Air Force Institute ofTechnology Wright-Patterson Air Force Base; 2001. Thesis.

Kistler R, Kalnay E. The NCEP/NCAR reanalysis prior to1958. In: Proceedings of the second World Climate Re-search Programme international conference on reanalysis.London: World Meteorological Organization; 2000: 27–35.

Klug W, Graziani G, Gridppa G, Pierce D, Tassone C.Evaluation of long range atmospheric transport modelsusing envrionmental radioactivity data from the Chernobylaccident (The ATMES report). Essex, England: ElsevierScience Publishers; 1992.


National Cancer Institute. Estimated exposures and thyroiddoses received by the American people from Iodine-131 infallout following Nevada atmospheric nuclear bomb tests: areport from the National Cancer Institute. Washington, DC:Department of Health and Human Services; 1997.

Rhodes R. The making of the atomic bomb. New York: Simonand Schuster, Inc.; 1986.

Rowland R. Fallout computer codes: a bibliographic perspec-tive. Washington, DC: U.S. Defense Nuclear Agency;DASIAC-SR-93-022; 1994.

Shoikhet YN, Kiselev VI, Loborev VM, Sudakov VV, AlgazinAI, Demin VF, Lagutin AA. The 29 August 1949 nucleartest. Radioactive impact on the Altai region population.Barnaul, Russian Federation: Institute of Regional Medico-Ecological Problems (IRMEP); 1998.

Simon SL. A brief history of people and events related toatomic weapons testing in the Marshall Islands. Health Phys73:5–20; 1997.

Simon SL, Bouville A, Land CE, Beck HL. Radiation doses andcancer risks in the Marshall Islands associated with exposure toradioactive fallout from Bikini and Enewetak nuclear weaponstests: summary. Health Phys 99:105–123; 2010a.

Simon SL, Bouville A, Melo D, Beck HL, Weinstock RM.Acute and chronic intakes of fallout radionuclides byMarshallese from nuclear weapons testing at Bikini andEnewetak and related internal radiation doses. Health Phys99:157–200; 2010b.

Swanberg EL, Hoffert SG. Using atmospheric 137Cs measure-ments and HYSPLIT to confirm Chernobyl as a source of137Cs in Europe. In: 23rd Seismic Research Review: World-wide Monitoring of Nuclear Explosions. Los Alamos, NM:Los Alamos National Laboratory; 2001: 64–70.

Toon OB, Turco RP, Robock A, Bardeen C, Oman L,Stenchikov GL. Atmospheric effects and societal conse-quences of regional scale nuclear conflicts and acts ofindividual nuclear terrorism. Atmospheric Chem Phys7:1973–2002; 2007.

Yamamoto M, Tomita J, Sakaguchi A, Imanaka T, Fukutani S,Endo S, Tanaka K, Hoshi M, Busev BI, Apsalikov AN.Spatial distribution of soil contamination by 137Cs and 239,

240Pu in the village of Dolon near the Semipalatinsk NuclearTest Site: new information on traces of the radioactiveplume from the 29 August 1949 nuclear test. Health Phys94:328–337; 2008.

f f


PREDICTIONS OF DISPERSION AND DEPOSITION OF FALLOUT …

Documents

Transcript of PREDICTIONS OF DISPERSION AND DEPOSITION OF FALLOUT …