1 Nuclear Utility Meteorological Data Users Group (NUMUG) Walter D. Bach, Jr., Ph.D. Science and Technology Corporation Samuel P. Williamson Federal Coordinator for Meteorological Services and Supporting Research Margaret R. McCalla Office of the Federal Coordinator for Meteorological Services and Supporting Research 28 June 2011 Oak Brook, IL Uncertainty in Atmospheric Dispersion Modeling

2 The Report Purpose User’s Needs from User’s Perspective Model Capabilities to Meet User’s Needs Modeling and Measurement Research Needs R&D Strategy to Meet User’s Needs Recommendations

3 A R&D Strategy to Meet User Needs COORDINATED AGENCY SUPPORT & FUNDING GOALS MEASUREMENT CAPABILITIES MODEL EVAL STANDARDS ATD TEST BEDS BRIDGE THE SCALE GAP CAPTURE AND USE EXISTING DATA SETS LOCAL/ REGIONAL MEASUREMENT SITING Routinely Quantify Uncertainty Interpret Uncertainty OBJECTIVES

4 Why Emphasize Uncertainty? Critical to making intelligent life/death decisions Ubiquitous (permeates all aspects of life) Poorly characterized in modeling efforts Measure of the robustness of the modeling/observing system. Reducing uncertainty requires knowledge of current uncertainty Believability of our products

5 ATD Modeling Uncertainty The total model uncertainty is measured by the variance in the predicted and the observed quantity over a large number of events that have similar properties (an ensemble).

6 Total Model Variance Let C o = C oa +  C o and C p = C pa +  C p Square of model bias Total model variance Error variance of observations Stochastic uncertainty Error variance of input data

7 Contributions to Model Uncertainty Internal: Numerical approximations, modeling errors, and treatment of dynamical processes External: Data errors in execution and evaluation, model parameterizations, and initial & boundary conditions Stochastic:Natural variability of the atmosphere (turbulence)

8 What to Do? Capture and Use Existing Data Sets Basis of present ATD knowledge Accessible Establish Test Beds All-weather conditions (24/7 operation) Daily test and evaluations Feedback among operational modeling and users Develop Measurement Technologies Meteorological variables – T, RH, winds, turbulence Tracer materials – cheap, harmless, remote, in situ Conduct controlled experiments Physical modeling – flow channels, wind tunnels, convection tanks Computational modeling – DNS, CFD, LES

9 Recommendations Use existing facilities to begin routinely quantifying model uncertainty at NUMUG sites. Establish a local benchmark for model uncertainty Expand observational network and capabilities Initiate and nurture interaction between modelers and model product users to incorporate and understand uncertainty Evaluate improvements to model and observation networks using uncertainty benchmarks.

10 Questions? Comments?