Learning Extreme Temperature Regimes (Papers Track)
Shirin Goshtasbpour (SDSC); Maxim Samarin (SDSC, ETH Zurich and EPFL); Michele Volpi (SDSC, ETH Zurich and EPFL)
Abstract
Recent changes in climate made previously predictable temperature and weather patterns increasingly unreliable, giving rise to increased volatility and extreme events such as prolonged heat waves, abrupt cold spells, and erratic temperature shifts. This growing unpredictability challenges the capabilities of physics-based climate models, particularly as low-return-rate temperature patterns become more common. In this paper, we present and explore a machine learning approach based on ClimODE to generate projections of future climate scenarios conditional on specific temperature quantiles. Our Uniform Quantile ClimODE (\textit{UQClimODE}) approach presents itself as a promising tool for capturing these atypical patterns, identifying localized impacts, and enabling proactive planning for climate adaptation and resilience under different scenarios.