As global temperatures rise, tropical cyclones (TCs) are projected to generate increasingly intense precipitation, yet traditional climate models with coarse resolution fail to capture small-scale convective processes critical to storm dynamics. This presentation will discuss findings from multiple studies utilizing high-resolution idealized Weather Research and Forecasting (WRF) model simulations to examine the impacts of sea surface temperature (SST) warming on TC precipitation, vertical mass flux (MF), and three-dimensional convective storm structures. Results show that SST warming significantly enhances inner core (IC) precipitation and TC intensity through increased deep convection, hydrometeor mixing ratios, elevated melting levels, deeper vertical circulations, and stronger mean upward velocities, while precipitation changes in outer rainbands (OR) are comparatively modest. The role of MF is further explored, with findings indicating a robust quasi-linear relationship between upward MF and TC intensity in IC regions, while the weaker relationship in OR regions is driven by increased lower-atmosphere evaporation and altered precipitation processes.
Building upon these results, research in the Rasmussen group connects advancements in machine learning (ML) applications as a computationally efficient tool to bridge the gap between global climate models (GCMs) and convection-permitting simulations. Simple pixel-based neural networks demonstrate high accuracy in predicting convective storm frequency from large-scale environmental variables, successfully capturing the spatial distribution, diurnal cycle and orographic convection with only environmental input data. Model performance declines when fewer input features are used or specific regions are excluded, underscoring the role of diverse physical mechanisms in convective activity. These results show that machine learning, specifically a pixel based model that predicts each grid point independently rather than relying on CNN style spatial structure, can efficiently represent convection and enable scientific discovery while remaining independent of resolution, generalizing to unseen environments, and linking environmental conditions to convective features across diverse model grids and climate scenarios.
"
|