Enhancing Precipitation Predictions with the Cloud-Associated Parameterizations Testbed using Artificial Intelligence

Project Overview

This project applied advanced machine learning / artificial intelligence (ML/AI) techniques to address DOE's mission needs to improve predictions of precipitation and extremes of DOE's Energy Exascale Earth System Model (E3SM) over the US. There are two objectives in the proposed study: 1) Conduct phenomenon-based model diagnosis to better understand the underlying problems in E3SM model physics critical to the prediction of precipitation and its extremes; and 2) Apply ML/AI techniques to improve model physics: ML equation discovery to better improve current deep convection scheme by improving the representation of organized convection. The outcome of the project on storm-analysis and ML models provides an innovative capability for E3SM to strengthen predictive capabilities for wide-range energy, environmental, and national security applications, as well as provide actionable information for resource allocations, energy infrastructure planning and policy making.

Mission Impact

The outcome of this project provides a historical storm dataset for studying the impact heavy-precipitating storm systems on infrastructure. This information can provide solutions to ensure infrastructure resilience in the future. On the science side, the approach can further provide a new way to identify model problems in precipitation and can improve model predictions of these storm systems in the future. The outcome of this project also provides a faster, online, feature tracking capability for storm systems for high-resolution model simulations, which can rapidly provide storm information for infrastructure needs. Both of which expand Lawrence Livermore National Laboratory's (LLNL) capability in better understanding extreme precipitation from heavy-precipitating storms, and enhance Laboratory's Core Competency in Earth and atmospheric science. This project also clearly addresses the needs on Mission Focus Area on Energy Security and Climate Resilience.

Publications, Presentations, and Patents

Galea, Daniel, Hsi-Yen Ma, Wen-Ying Wu, 2023 Deep learning image segmentation for atmospheric rivers. Artificial Intelligence for the Earth Systems, accepted pending minor revision.

Wen-Ying Wu, Hsi-Yen Ma, David Lafferty, Zhe Feng, Paul Ullrich, Qi Tang, Chris Golaz, Daniel Galea, and Hsiang-He Lee "Assessment of Storm-Associated Precipitation and its Extremes using Observations and Short-range Climate Model Hindcasts" (Presentation, Gordon Research Conference for Radiation and Climate. Lewiston, Maine, July 23-28, 2023).

Hsi-Yen Ma, Wen-Ying Wu, David Lafferty, Zhe Feng, Paul Ullrich, Qi Tang, Chris Golaz, Daniel Galea, and Hsiang-He Lee, "Analysis of storm-associated precipitation and extremes in observations and climate model short-range hindcasts" (Presentation, The 2023 Joint CFMIP-GASS meeting on cloud, precipitation, circulation and climate sensitivity, Paris, France, July 10-13, 2023).

Wen-Ying Wu, Hsi-Yen Ma, David Lafferty, Qi Tang, Chris Golaz, "Contributions of storm-associated precipitation and its extreme using observations" (Presentation, American Geophysical Union Fall Meeting 2022, Chicago, Illinois, December 12-16, 2022).

David Lafferty, Hsi-Yen Ma and Wen-Ying Wu, "Atmospheric feature tracking and associated precipitation extremes" (Presentation, Lawrence Livermore National Laboratory Climate and Weather Seminar Series. Livermore, California, August 10, 2022).

Daigo Kobayashi, Hsi-Yen Ma, Wen-Ying Wu and David Lafferty, "Improving climate prediction through machine learning" (Presentation, Lawrence Livermore National Laboratory Climate and Weather Seminar Series. Livermore, California, August 10, 2022).