Lawrence Livermore National Laboratory

NARAC models are extensively tested and evaluated against exact mathematical solutions to the model equations, controlled laboratory and field experiments, and real-world releases. These tests ensure that NARAC models meet the following criteria:

  • Physically realistic equations and parameterizations are used in the models
  • Model equations are solved correctly and numerical methods are sufficiently accurate
  • Necessary input data are available to drive the models, including meteorological, geographical, and material/release properties
  • Models are accurate enough to reproduce data from tracer experiments to real-world incidents
  • Models are fast and robust enough to be used for emergency response applications
  • Software meets DOE Software Quality Assurance standards

Model Verification Using Analytic Solutions

Analytic mathematical solutions to model equations are used to test algorithms against known, exact results to ensure that numerical methods have been correctly implemented as shown in the example figure above.

Model Validation and Evaluation Using Field Study Data

NARAC uses experimental data from tracer gas and explosive dispersal releases to evaluate the accuracy of its models on local, regional, and continental scales (see example in figure below). NARAC models have performed in the top tier of models in international model evaluation and inter-comparison studies such as post-accident modeling of the Chernobyl accident, and the European Tracer Experiment (see Publications for additional information on NARAC model evaluation studies). Comparisons with data show that NARAC model predicted values are typically within a factor of 2 of measured values for simpler cases (relatively flat terrain and steady-state meteorological conditions) and within a factor of 5–10 of measured values for more complex conditions (e.g., heterogenous terrain, time-varying meteorology, or complicated emissions). Factor of 2 agreement means that the ratio of observed to predict values are between 1/2 and 2. Even in complex conditions, predicted peak air and ground contamination values are typically within a factor of 2 of measured values for the same downwind distance, but not necessarily the same exact location.

The two panels above show the results of NARAC wind field and dispersion model predictions for a tracer gas release experiment near the Diablo Canyon Nuclear Power Plant on the central coast of California in 1986. Sulfur Hexafluoride (SF6) tracer gas released at the plant was measured at air sampling stations 1-40 km downwind. The left panel shows the LODI model predicted tracer particle positions and the ADAPT model wind field 10 meters above ground level at 11:45 PDT on 8/31/1986. The right panel compares the LODI-predictions (color contours) and the observed values (marked by "+") of 1-hr average SF6 surface air concentrations overlaid on model terrain contours (100 m contour interval).

Real-World Incident Response

Real-world incidents provide opportunities to test models under complex conditions. NARAC has responded to numerous events since it's creation in 1979. Examples include the Chernobyl nuclear power plant accident, industrial fires and accidents, the Algeciras Spain Cesium release, the Tokaimura Japan criticality accident, and the Fukushima Dai-Ichi Nuclear Power Plant accident.


Nasstrom JS, Sugiyama G, Baskett RL, Larsen SC, Bradley MM, 2007: The NARAC modeling and decision support system for radiological and nuclear emergency preparedness and response. Int. J. Emergency Management, 4, 524-550.

Foster KT, Sugiyama G, Nasstrom JS, Leone, Jr. JM, Chan ST, and Bowen BM, 2000: The use of an operational model evaluation system for model intercomparison, Int. J. Environment and Pollution, 14, 77-88.