What causes extreme storms to vary from year to year?  How will they change in the future?

Climate extremes are caused by specific weather phenomena, such as tropical cyclones and atmospheric rivers.  Observations and climate model simulations can help answer these questions, but it is first necessary to isolate these specific weather phenomena.  This requires sifting through large amounts of data–from terabytes to exabytes–to search for these storms. On top of that, there are numerous sources of uncertainty that must be considered in order to inform robust scientific conclusions, including observational uncertainty, model uncertainty, and even uncertainty in the detection of weather events.

The CASCADE team is combining expertise in atmospheric science, computational science, and statistics to develop tools and workflows that enable efficient and robust exploration of specific weather phenomena in massive datasets.  A key feature of the CASCADE research portfolio in this area is the development and implementation of innovative new algorithms for detecting weather phenomena, at hourly timescales, in datasets that involve hundreds of years of simulation and hundreds of unique simulations.  These algorithms explicitly include advanced forms of uncertainty quantification at their core. The team is advancing multiple types of approaches, including computer vision (machine learning), Bayesian statistics, and theoretical fluid dynamics.

The computational and statistical tools that CASCADE produces are designed explicitly around the team’s scientific needs, but they are developed with broader use in mind.  All tools produced by the CASCADE team are publicly and freely available for research use, and all are deployed using modern, community-oriented development practices (e.g., using github).  Current products in the CASCADE portfolio include:

• TECA: the Toolkit for Extreme Climate Analysis
• climextRemes
• fastKDE