Applying Machine Learning to Associate Precipitation Extremes with Synoptic-Scale Weather Events

Abstract:

Extreme precipitation events continue to have wide-ranging impacts across the United States. Rainfall associated with atmospheric rivers (ARs) and tropical cyclones (TCs) can cause devastating damage on communities and ecosystems. Precipitation originating from fronts and mesoscale convective systems (MCSs) can also be extreme, and these systems are often associated with severe thunderstorms that can produce large hail and damaging winds. Machine learning-based detection algorithms can help with the automated classification of synoptic weather features such as ARs, TCs, MCSs, and frontal systems. Here we use new and existing machine learning algorithms (e.g., Biard and Kunkel 2019, Prabhat et al. 2020) to identify these types of systems in observations and climate model output over the contiguous United States. We use high resolution Community Earth System Model (CESM) simulations to compare detection results from the machine learning models with standard tracking algorithms such as the TempestExtremes feature tracking software (Ullrich and Zarzycki 2017). We further compare results using CESM simulations with present-day and future climate forcing, to study how these events might change and evolve with climate change. We then associate these features with precipitation extremes to better understand the intersection of extreme precipitation and meteorological events.