Interseasonal Prediction of Western United States Snowpack with Deep Learning

Gemma Anderson | 19-ERD-032

Project Overview

Western U.S. snowpack provides over 75% of the annual water to the west. Accurate interseasonal snowpack forecasts are crucial for adequate water resource planning, yet are sorely lacking. In this project we developed a deep learning framework that downscales and corrects the inherent biases in the temperature and precipitation predictions from a dynamical forecast system, which can subsequently be used to improve both the accuracy and the resolution of predictions of western U.S. snowpack. Our deep learning bias correction and downscaling methods can be more broadly applied to make accurate and high-resolution predictions of other climate variables and regions. We also developed a deep learning probabilistic seasonal forecast system that was trained on pre-existing climate simulations. This can be used to make faster seasonal forecasts of snowpack and other climate variables and can also help to diagnose dynamical seasonal forecast systems that can allow for even further improvements to seasonal predictions in the future.

Mission Impact

Improving the accuracy and resolution of climate predictions can help to mitigate the threats from climate change to the security and resilience of the nation and its critical infrastructure. This research advances the Lawrence Livermore National Laboratory core competencies of earth and atmospheric sciences, and high-performance computing, simulation, and data science. This  research also supports efforts to develop science and technology tools and capabilities to meet future NNSA national security challenges.

Publications, Presentations, and Patents

B. Pan, G. J. Anderson, A. Goncalves, D. D. Lucas, C. J. W. Bonfils, & J. Lee. 2021. "Learning to correct climate projection biases," Journal of Advances in Modeling Earth Systems, 13, e2021MS002509. LLNL-JRNL-817982.

B. Pan, G. J. Anderson, A. Goncalves, D. D. Lucas, C. J. W. Bonfils, & J. Lee. 2021. Code Release: "RADA 1.0," https://github.com/panbaoxiang/RADA. LLNL-CODE-826222.

G. J. Anderson. 2021. "AI-enhanced Seasonal Forecasts for Electricity-Sector Climate Resilience." Electricity Subsector Coordinating Council (ESCC) Roundtable, Virtual. LLNL-PRES-826245.

G.J. Anderson. 2021. "Artificial Intelligence for Climate Impacts and Resilience." Climate Impacts Seminar Series, LLNL, Virtual. LLNL-PRES-827836.

B. Pan. 2020. "Improving Seasonal Forecast using Probabilistic Deep Learning." Climate and Weather Seminar Series. LLNL-PRES-814574.

B. Pan, G.J. Anderson, D.D. Lucas, A. Goncalves, C. Bonfils, J. Lee, Y. Tian. 2020. "Identifying and correcting climate projection biases using artificial intelligence." AGU Fall Meeting 2020, Virtual. LLNL-POST-817031.

G.J. Anderson, B. Pan, A. Goncalves, D.D. Lucas, C. Bonfils, J. Lee. 2020. "Probabilistic deep learning for seasonal forecasting." AGU Fall Meeting 2020, Virtual. LLNL-POST-816959.

A. Goncalves, G.J. Anderson, B. Pan, D.D. Lucas, J. Lee, C. Bonfils. 2020. "Joint inter-seasonal forecasts with deep multitask learning." AGU Fall Meeting 2020, Virtual. LLNL-PRES-816932.

A. Goncalves, G.J. Anderson, D.D. Lucas, J. Lee. 2019. "Inter-seasonal snowpack prediction with deep convolutional neural networks." AGU Fall Meeting 2019, San Francisco. LLNL-POST-798858.

G.J. Anderson. 2019. "Deep Learning for Climate Predictions." Data Science Institute Workshop, LLNL, Livermore. LLNL-PRES-781566.