The Puget Sound coastal region is located in the northwest corner of the contiguous U.S. and serves as the country’s second-largest estuary. It faces significant challenges of modeling extreme events due to its sharp mountain-to-coast topography and complex rural-to-urban development patterns. This abstract presents two research efforts aimed at addressing these challenges. The first study developed a novel regional weather system using a clustering model with a deep learning encoder based on daily meteorological data from a reanalysis product. The regional weather system successfully connected the influence of ENSO and MJO with regional precipitation and hydrological conditions, and identified two flood-inducing weather patterns: one causing excess precipitation and another causing both high precipitation and warm temperatures, which brings notable snowmelt contribution to flooding. The findings suggest new potential for understanding regional weather in the context of both large-scale climate variabilities and regional hydrological processes. The second study evaluates the Simple Cloud-Resolving E3SM Atmosphere Model (SCREAM) with regional refinement for simulating atmospheric rivers (ARs) that often produce heavy precipitation and strong winds and induce hazardous flooding when they make landfall over mountainous regions such as the U.S. west coast. Using two versions of SCREAM at a 3.125km regionally refined mesh, we simulated five AR events that made landfall and led to major flooding in the Puget Sound basin between 2006 and 2022. The New-SCREAM, which sets ice cloud fraction based on cell averaged ice mass mixing ratio, showed improved simulation of AR characteristics such as orographic precipitation, integrated vapor transport, and winds, underscoring the important role of improved cloud microphysics parameterization in modeling ARs.