NUCLEAR REACTOR and radioactive waste management operations occur at the Idaho National Laboratory (INL), a large (2,300 km2) U.S. Department of Energy (U.S. DOE) national laboratory in southeastern Idaho (Fig. 1). Emergency response drills are sometimes performed to assess hypothetical, accidental, airborne releases of radioactive material from some of these facilities. The Field Resource Division (FRD) of the U.S. National Oceanic and Atmospheric Administration’s (NOAA) Atmospheric Research Laboratory has primary responsibility for assessing these airborne releases, using an INL version of the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model called HYRad (NOAA-FRD 2015). HYRad uses the HYSPLIT dispersion model (Draxler and Hess 1998 ; Stein et al. 2015) for basic transport and dispersion but has modifications to evaluate radioactive material, use the local INL mesoscale meteorological monitoring network (Mesonet) data, and provide graphical output specific to INL releases.
The State of Idaho Department of Environmental Quality INL Oversight Program (OP) provides independent oversight and assessment of these potential releases and performs other monitoring activities around INL. Since all models have inherent uncertainty, OP decided to use a different model to provide a completely independent assessment and to increase overall confidence in the range of INL emergency response predictions and actions. OP chose the RASCAL model (Ramsdell et al. 2012), which is the primary reactor-emergency response code used by the U.S. Nuclear Regulatory Commission (NRC). In order to have a better understanding of the RASCAL model for evaluating INL releases, OP decided to compare RASCAL model (version 4.3.1-DEQ) predictions to those from HYRad (version 2.0) for typical meteorological scenarios at INL. Although model comparisons such as these cannot tell us which model is more accurate, they can provide important information on how they perform under different conditions, which model is more conservative, and what are some of the more important factors that analysts should be aware of when using them for emergency response operations. RASCAL output was also compared to two sets of sulfur hexafluoride (SF6) tracer data measured during FRD’s Project Sagebrush (PSB) experiment at INL in October 2013 (Finn et al. 2015). This comparison does provide information on model accuracy, although only for the meteorological conditions and downwind distances evaluated in the experiment.
U.S. NRC RASCAL MODEL
The RASCAL code is a model used by the Protective Measures Team in the U.S. NRC’s Operations Center for making independent dose and consequence projections during radiological incidents and emergencies. RASCAL was developed by U.S. NRC over 25 years ago to provide a tool for rapid assessment of an incident or accident at U.S. NRC-licensed facilities but can be used for other nuclear facilities such as those at INL. RASCAL uses both a straight-line Gaussian plume model for close-in distances where wind directions are more likely to be constant (maximum distance of 16 km) and a Lagrangian trajectory, Gaussian puff model to account for changing wind fields and complex terrain effects over longer plume transport times (out to a maximum distance of 160 km). The maximum distance these Gaussian models are best applied at is constrained by their use of dispersion parameters (sigmas), which are based on historical experimental tracer measurements. RASCAL’s puff model is the preferred Gaussian model for INL because it accounts for changing wind directions over the long transport times to INL site boundary locations (20–50 km) where emergency response actions need to be assessed. Both the plume and puff models in RASCAL are simpler than the more complex, three-dimensional, particle model available in HYRad-HYSPLIT. RASCAL has been configured by OP for INL applications to use short-term (15 min) surface meteorological data (wind direction, wind speed, and stability class) from multiple INL Mesonet stations. These meteorological observations are preprocessed in RASCAL to interpolate over the model output grid and to adjust the wind field for topographic influences. RASCAL has a comprehensive, built-in, radionuclide inventory with a decay-chain processor and models multipathway doses, including inhalation and external exposure to radionuclides in the plume (cloud shine) and deposited on the ground (4 d ground shine dose). The dispersion models in RASCAL have undergone significant model verification and validation testing of long-term (monthly and longer) predictions with generally good (factor of two) results (Rood et al. 1999 ; Molenkamp et al. 2004 ; Ramsdell et al. 2012). However, no known previous study has compared RASCAL short-term (hourly) exposure predictions with those from HYSPLIT-HYRad or similar short-term tracer measurement data.
NOAA-FRD HYRad MODEL
HYRad (NOAA-FRD 2015) uses the HYSPLIT Lagrangian particle model (Draxler and Hess 1998 ; Stein et al. 2015) to calculate individual radionuclide air concentrations, deposition, cumulative concentration or deposition, and radiological doses. HYSPLIT is one of the most extensively used atmospheric transport and dispersion models in the atmospheric sciences community and was used to evaluate airborne releases of radioactive material from the Fukushima Daiichi nuclear power plant accident in March 2011. The model simulation method is a hybrid between the Lagrangian approach (using a moving frame of reference for the advection and diffusion calculations as the air parcels move from their initial location) and the Eulerian methodology (using a fixed three-dimensional grid as a frame of reference to compute pollutant air concentrations). Like RASCAL, HYRad has the capability to use short-term meteorological observations from INL’s local Mesonet network. Because HYSPLIT’s particle model does not rely on experimentally based dispersion parameters (like RASCAL’s puff model), it can be effectively used across regional distances (hundreds of kilometers) and can easily handle the 20–50 km site boundary distances that occur at INL.
METHODS—RASCAL-HYRad MODELING COMPARISON
Release scenarios and input parameters
For the RASCAL-HYRad modeling comparisons, a 37 GBq s−1 release of 131I over 1 h (1.3 × 1014 Bq) was assumed to occur from a 50 m release height at the Advanced Test Reactor (ATR) at INL. Two meteorological scenarios were evaluated: (1) a typical daytime [summer afternoon (PM)] release with southwesterly winds, moderately high wind speeds, and neutral stability conditions and (2) a nighttime [early morning (AM)] release with variable winds, low wind speeds, and stable conditions (Table 1).
The RASCAL model used 15 min surface observations over a 6 h period starting 2 h prior to the release (Table 1). Data from seven INL Mesonet towers in the general downwind area from the release at ATR were used. The HYRad runs used the same short-term Mesonet data.
Both models were run with a 4 h plume transport time to allow adequate time for plume transport to INL site boundary locations (20–50 km). Time-integrated 4 h air concentrations and total deposition were selected for output because these are the primary output for both models, and all radiological doses are directly proportional to these values over the plume exposure time. RASCAL time-integrated concentration output is in units of μCi s cm−3 while the HYRad output is in equivalent units of Ci s m−3. These fixed model output units can be converted to MBq s m−3 by multiplying by 3.7 × 104. RASCAL total deposition output is in fixed units of μCi m−2 while the HYRad output is in fixed units of Ci m−2. These units can be multiplied by 37 and 3.7 × 107, respectively, to convert them to kBq m−2.
HYRad was run with a 0.02° × 0.02° (~2.3 km) output grid node spacing, but output from this model is limited to order-of-magnitude isopleth plots. The RASCAL puff model was run for maximum distances of 16 km, 40 km, and 80 km which gives output values at corresponding grid node spacings of 0.8 km, 2 km, and 4 km. Because the output grids from the two models do not align precisely and because the HYRad output is limited to order-of-magnitude isopleths, comparison of the two models’ output at specific locations had to be estimated. This was done by overlapping the RASCAL grid output over the HYRad isopleth plots, then visually determining the corresponding RASCAL grid node values at the maximum range of each HYRad isopleth (roughly corresponding to plume centerline). The RASCAL straight-line Gaussian plume model was also run for maximum distances of 3 km (nighttime release) to 8 km (daytime release) to provide higher resolution of close-in concentrations.
METHODS—RASCAL MODEL COMPARISONS TO PSB TRACER DATA
RASCAL was also compared to SF6 tracer measurements collected during NOAA’s Project Sagebrush 1 experiment (PSB1) conducted by FRD in October 2013 at INL (Finn et al. 2015). Five tests were conducted during PSB1, all during the daytime with conditions ranging from near neutral with higher wind speeds to unstable with low wind speeds. Two of the five 2 h intensive operational period tracer release tests (IOP‐2 and IOP‐3) were selected for comparison to RASCAL model output because these tests occurred when wind directions best corresponded with sampler locations. Each IOP consisted of a continuous 2.5 h SF6 release with consecutive 10 min average bag sampling at numerous downwind locations over the last 2 h of the tracer release period. Meteorological conditions were unstable, mostly sunny, with light-to-moderate southwest winds for IOP‐2 and neutral, mostly sunny, with moderately strong southwest winds for IOP‐3. Bag sampling was done on four arcs of almost 90° each ranging in distance from 200 m to 3,200 m from the source. During each test, SF6 tracer was released at a steady-state nominal rate of 10 g s−1 at 1.5 m above ground level (AGL) and sampled over a 2 h period by 12 sequential 10 min bag samplers at each sampling location.
The SF6 data was processed for comparison to RASCAL output by identifying the sampler locations with good data (quality control flags = 0 or 2), averaging the 12 sequential 10 min bag samples at each location (giving a 2 h average sample at that location), then taking the average of the highest four values along each arc to represent maximum plume centerline concentrations. These 2 h average concentrations (g m−3) were then converted to 2 h time-integrated concentrations (g s m−3) by multiplying by 7,200 s (2 h)−1 for comparison to RASCAL integrated concentration output.
Since the tracer measurements were all made at relatively close-in distances (maximum of 3,200 m from release), only the RASCAL straight-line Gaussian plume model (vs. the puff model) was run for comparison. RASCAL documentation (Ramsdell et al. 2012) states that plume model output should be used for receptors near the release point because of errors associated with puff transport at these close-in distances.
Meteorological data over 5 min intervals from the GRID3 10 m INL Mesonet tower (200 m from the tracer release point) were provided by NOAA and averaged to 15 min, giving sixteen 15‐min observations for input to RASCAL over the 2 h sampling period. Although the SF6 was released at a height of 1.5 m AGL, the RASCAL releases were assumed to occur at 10 m height since that is the minimum height available in the model. The RASCAL release rate was set to 370 GBq s−1 131I over a 2 h period to correspond with the tracer release rate of 10 g s−1 SF6 over the 2 h measurement time. Time-integrated (2 h) 131I concentration was selected for output. Although 131I is a depositing species and SF6 is an inert gas that does not deposit, previous RASCAL assessments (Ramsdell et al. 2012) have shown there is an insignificant amount of 131I plume depletion (less than 4%) that would occur over the maximum PSB experiment distance of 3.2 km.
RESULTS AND DISCUSSION—RASCAL-HYRad MODEL COMPARISONS
Figs. 2 and 3 show the models’ integrated concentration footprints for the neutral stability, moderate wind speed, daytime meteorological scenario [7/17/17 1900-2300 mountain standard time (MST) plume transport time]. Total deposition output (not shown) showed very similar patterns for both models.
Figs. 4 and 5 show the maximum plume centerline modeling results vs. downwind distance for the neutral, moderate wind speed, meteorological scenario. Determination of the HYRad results vs. distance was made by visually examining the order-of-magnitude isopleth values given in the model’s graphical output.
Neutral, moderate wind speed case discussion
Figs. 2 and 3 (plume shape and direction):
- Both models predicted similar northeasterly plume paths representative of what would be expected from moderate southwest winds.
- RASCAL puff transport was in a more northeasterly straight line, while the HYRad particle model predicted a plume trajectory slightly curving north-northeasterly from 20–60 km.
- RASCAL puff transport extended well beyond 80 km over the 4 h plume transport time, while the HYRad plume traveled out a maximum of 70 km over the same time.
Figs. 4 and 5 (maximum plume centerline concentration/deposition values):
- RASCAL Gaussian plume concentrations from 2 to 8 km essentially bracket the HYRad results at 2 km, although the RASCAL concentrations at distances of less than 5 km are generally higher than HYRad’s (up to six times greater). This may be influenced by the fact that the HYRad particle model uniformly averages concentrations vertically over a 0–50 m AGL layer, while RASCAL calculates the Gaussian concentration at a point near the ground.
- RASCAL puff air concentrations slightly underpredict HYRad concentrations at distances of less than about 40 km, after which the HYRad concentrations drop below the RASCAL concentrations. This could again be influenced by the different way the models average air concentrations vertically (HYRad in a uniform 50 m vertical layer and RASCAL at a point on the ground). The decrease in HYRad air concentrations relative to RASCAL’s beyond about 40 km (Fig. 4) may be influenced by the relatively high deposition-velocity value currently hard-wired into HYRad for iodines (0.01 m s−1) and the resulting higher plume depletion that would occur at these far-field distances.
- RASCAL total deposition is generally lower than HYRad’s (by up to an order of magnitude), likely influenced by the lower deposition-velocity values from RASCAL’s resistance model (Ramsdell et al. 2012). This difference becomes insignificant at distances of greater than 50 km, possibly due to fact that HYRad’s plume is more depleted.
- Air concentration outputs for both models are very close at 40 km, and deposition outputs are similar at 50 km. The similarity in the results at these distances provides confidence that, at least under typical daytime conditions, both models can provide similar predictions at typical INL site boundary distances (20–50 km).
Figs. 6 and 7 show the models’ integrated concentration footprints for the stable, low wind speed, nighttime [early morning (AM)] meteorological scenario (7/17/2017 0100–0500 MST plume transport time). Total deposition output (not shown) showed very similar patterns.
Figs. 8 and 9 show maximum plume centerline modeling results vs. downwind distance for the stable, low wind speed, nighttime [early morning (AM)] meteorological scenario. Interpolation of the HYRad results was limited to the order-of-magnitude isopleth values given in the model’s graphical output.
Stable, low wind speed case discussion
Figs. 6 and 7 (plume shape and direction):
- Both models predicted similar northeasterly plume paths which slow (due to low wind speeds) and curve to the east at around 16 km. The RASCAL plume appears to reach the east sector while the HYRAD plume reaches only the east-northeast sector (i.e., remains farther north).
- The RASCAL plume seems to travel slightly farther than the HYRad plume, but neither model predicts the plume will reach the INL site boundary (50 km) over the 4 h transport time.
Figs. 8 and 9 (maximum plume centerline concentration/deposition values):
- All of the models (RASCAL plume, RASCAL puff, and HYRad particle) predict very similar integrated air concentrations at 1.6 to 3 km. The RASCAL puff model shows the pronounced concentration peak characteristic of an elevated Gaussian-distributed plume that finally reaches the ground. This peak is not apparent in the HYRad results possibly because it averages the concentration vertically over a 0 to 50 m AGL layer at all distances.
- Beyond about 3 km, the two models diverge rapidly with the RASCAL air concentrations decreasing at a much slower rate than HYRad’s. At the outer edges of the plume, the RASCAL concentrations are about two orders of magnitude higher than HYRad’s.
- Total deposition shows the same relative rates of change with distance except that the RASCAL values are generally much lower than those from HYRad during most of the plume transport distance (≤10 km). After this distance, the HYRad deposition values decrease well below those from RASCAL’s. These relationships are likely influenced by HYRad’s higher assumed deposition-velocity value (0.01 m s−1), which gives higher deposition rates early on until the plume is more depleted at greater distances.
RESULTS AND DISCUSSION—RASCAL MODEL COMPARISONS WITH PSB TRACER DATA
Fig. 10 shows the RASCAL integrated air concentrations compared to those measured in the PSB tracer study (IOP-2 and IOP-3) out to the maximum tracer study range of 3.2 km. When the tracer results are averaged across the highest four valid samples at a given distance, the RASCAL predicted-to-observed (P/O) ratios range from 0.8 to 1.8 for IOP‐2 (unstable conditions) and 0.4 to 0.9 for IOP‐3 (neutral conditions) (Table 2). Historically, a factor of two (0.5–2) P/O ratio has been considered good model performance. Given this, RASCAL can be said to have performed very well at these near-field distances under the given meteorological conditions. However, these straight-line Gaussian plume model results do not necessarily suggest similar model performance at the much larger INL site boundary distances (20–50 km) and plume transport times, where the probability of changing meteorological conditions greatly increases.
These results suggest that RASCAL is a good independent assessment tool for OP modeling of short-term (hours) accident releases at INL relative to the more complex HYRad model. Under typical daytime conditions, RASCAL will likely provide results that are within a factor of two from HYRad’s at typical INL site boundary distances (20–50 km). This is comparable to what other, more comprehensive, RASCAL model comparison studies (Rood et al. 1999 ; Molenkamp et al. 2004 ; Ramsdell et al. 2012) have found, even though these studies used much longer averaging times (e.g., 30 d), which generally provide much better comparisons. For more difficult modeling conditions (nighttime, low variable wind speeds), RASCAL results will likely provide a conservative upper bound (by up to two orders of magnitude) to those provided by HYRad, which will provide OP a conservative range of emergency response actions that might be considered under those conditions. In addition, RASCAL’s very good performance at predicting 2 h average tracer measurements (P/O: 0.9–1.4 at 3.2 km) provides additional confidence in its accuracy even though these results may not apply under all meteorological conditions or at INL site boundary distances farther downwind.
This study also helped OP to better understand the capabilities and limitations of each model. Both models seem to predict very similar plume trajectories under the two very different meteorological cases, which is not surprising given that both models used the same short-term meteorological data. This suggests that horizontal mixing and advection are similar between the two models. The two orders of magnitude discrepancy between the models’ predictions under the nighttime, low wind speed conditions then suggests significant differences in how the models simulate vertical mixing during transport and how they calculate receptor concentrations. These differences include model diffusion algorithms (HYRad particle vs. RASCAL puff), how the models determine mixing heights, and the different way the models calculate receptor concentrations—HYRad’s volume averages over 0–50 m AGL while RASCAL calculates them for points near the ground. This latter difference would be more important for near-field results and less important for large INL site boundary distances (20–50 km) where the plume will be more mixed. Other important differences between the models as currently configured include the calculated rates of deposition and the resulting plume depletion of activity for receptors that are at INL site boundary distances. For depositing radionuclides like 131I, RASCAL uses a resistance model for calculating deposition velocities that are a function of meteorological and surface conditions (Table 4‐3, Ramsdell et al. 2012). This is a more refined methodology that generally gives significantly lower deposition-velocity values (by up to a factor of seven) than the fixed, upper-bound value of 0.01 m s−1 that is currently hard-wired into HYRad for iodine isotopes. It is important to recognize that conservatively high deposition rates also result in greater plume depletion rates and less activity in the plume at far-field distances (such as those at the INL site boundary). This is important when assessing short-term (emergency response) dose, where greater than 90% of the total effective dose will be from the inhalation exposure pathway and less than 5% is from ground surface pathways. Finally, these comparisons demonstrated that RASCAL’s output grid provides better resolution of results compared to HYRad’s order-of-magnitude isopleths, which may be important when OP needs to determine emergency response actions at specific downwind communities.
Since the accuracy of both these models for INL applications cannot be quantified by simple model comparisons, it may be prudent to consider both models’ results in emergency response decision making if a release occurs under what we know are difficult modeling conditions (low, variable wind speeds often occurring at night). However, under typical daytime conditions with moderate-to-high wind speeds and relatively uniform wind fields, either model will probably suffice.
The author gratefully acknowledges the NOAA Air Resources Laboratory (ARL), particularly the ARL Field Research Division in Idaho, for the provision of the HYSPLIT transport and dispersion model together with a special model interface for radiological applications.
Draxler RR, Hess GD. An overview of the HYSPLIT_4 modeling system for trajectories, dispersion, and deposition. Aust Meteor Mag 47:295–308; 1998.
Finn D, Clawson KL, Eckman RM, Carter RG, Rich JD, Strong TW, Beard SA, Reese BR, Davis D, Liu H, Russel E, Gao Z, Brooks S. National Oceanic and Atmospheric Administration Air Resources Laboratory. Project Sagebrush Phase 1. NOAA Technical Memorandum OAR ARL‐268; 2015. DOI: 10.7289/V5VX0DHV.
Molenkamp CR, Bixler NE, Morrow CW, Ramsdell JV Jr, Mitchell JA. Comparison of average transport and dispersion among a Gaussian, a two-dimensional, and a three-dimensional model. Washington, DC: U.S. Nuclear Regulatory Commission; NUREG/CR-6853; 2004. 2012. Available at https://www.nrc.gov/reading-rm/doc-collections/nuregs/contract/cr6853/
. Accessed 9 April 2018.
National Oceanic and Atmospheric Administration—Field Research Division (NOAA-FRD). HYSPLIT transport and diffusion model for radiological applications (HYRad) user’s guide vol. I, operator manual, version 2.0. Idaho Falls, ID: NOAA FRD; 2015.
Rood AS, Killough GG, Till JE. Evaluation of atmospheric transport models for use in Phase II of the historical public exposures studies at the Rocky Flats Plant. Risk Analysis 19:559–576; 1999.
Stein AF, Draxler RR, Rolph GD, Stunder JB, Cohen MD, Ngan F. NOAA’s HYSPLIT atmospheric transport and dispersion modeling system. Bull Am Meteor Soc 96:2059–2077; 2015. Available at https://doi.org/10.1175/BAMS-D‐14‐00110.1
. Accessed 9 April 2018.
Keywords:© 2018 by the Health Physics Society
accident analysis; emergency planning; emissions, atmospheric; modeling, dose assessment