Donald Lucas | 17-ERD-045
Better characterization of meteorological uncertainty is needed to predict the effects of radiological and nuclear releases in the atmosphere. Multiphysics ensembles of atmospheric simulations can estimate uncertainty of physical processes, but they are often limited for computational reasons. Because weather uncertainty is important for emergency response applications and the conventional way of running multiphysics ensembles is infeasible, we developed new machine-learning-based methods that quickly estimate and identify key sources of meteorological uncertainty. Machine learning accelerates the analysis of weather uncertainty by iteratively learning about model physics options that affect plume predictions. We demonstrated the power of machine learning by creating a large multiphysics weather and dispersion ensemble dataset containing 1,200 simulations for two hypothetical radiological release scenarios. By training on less than 10 percent of the ensemble, we can quickly and accurately predict different outcomes, including the mass of radiological material deposited and spatial distribution and extent of contaminated areas.
The research capabilities developed under this project support several Lawrence Livermore National Laboratory mission and program areas in national security and energy and resource security, including consequence management, nuclear detection and monitoring, nuclear nonproliferation, and nuclear forensics. The most significant impact is the new capability to use machine learning to quickly estimate and incorporate weather uncertainty into atmospheric dispersion assessments conducted by the Laboratory's National Atmospheric Release Advisory Center (NARAC). In addition, research capabilities developed under this project can support other National Nuclear Security Administration (NNSA) programs, as well as the missions of the Defense Threat Reduction Agency (DTRA).
Publications, Presentations, and Patents
Gunawardena, N., et al. 2019a. "Rapid Prediction of Airborne Contamination Events." LLNL Research Slam, Livermore, CA, October 2019. LLNL-PRES-793708
——— 2019b. "Rapidly Emulating Spatial Dispersion Simulations." AGU Fall Meeting, San Francisco, CA, December 2019, LLNL-PRES-798206
——— 2020a. "Emulating and Interpreting Spatial Deposition Simulations." Conference on Data Analysis, Santa Fe, NM, February 2020. LLNL-POST-805224
Lucas, D. D., et al. 2018. "Using Machine Learning to Intelligently Select Members of Large Atmospheric Model Ensembles." AGU Fall Meeting, Washington, D.C., December 2018. LLNL-POST-763726
——— 2019. "Probabilistic Predictions and Uncertainty Estimation Using Adaptively Designed Ensembles for Radiological Plume Modeling." CTBTO Science and Technology Conference, Vienna, Austria, June 2019. LLNL-POST-774026.
Pallotta, G., et al. 2018. "Improving predictions of radiological surface contamination uncertainty via an ensemble of simulations." LLNL's First Data Science Institute Summer Workshop, Livermore, CA, August 2018. LLNL-PRES-754902
Schroeder, K. A., and D. D. Lucas, "Weather Uncertainty Total Deposition Novelty Detection." Data Science Summer Institute Poster Symposium, Livermore, CA, June 2019. LLNL-POST-782204
Simpson, M., et al. 2018. "Probabilistic Predictions and Uncertainty Estimation Using Adaptively Designed Ensembles for Radiological Plume Modeling." George Mason University Conference on Atmospheric Transport and Dispersion, Fairfax, VA, June 2018. LLNL-PRES-753230