Studying flash floods is crucial in understanding and modeling risk to the built environment. While recent studies on flood-related hazards extensively focused on trends in extreme precipitation and riverine floods, work on analyzing the spatiotemporal trends in flash floods is limited. In this study, we present an analysis of the space-time trends in the frequency and duration of flash floods across the United States from 1996 to 2023 using flash flood event information reported in the NOAA Storm Events Database. Both supervised and unsupervised machine learning techniques are usedon area-normalized annual flash flood counts aggregated at the weather forecast office spatial unit to uncover the trends. Preliminary results suggest variability (up to 22%) is dominated by a monotonic increase or decrease in flash flood frequency in the Northeast, Midwest, and Southern United States. Meanwhile, the Greater New York Area and Puerto Rico regions have the highest overall annual flash flood occurrence risk per unit area. A hierarchical Bayesian modeling framework is proposed to infer anthropogenic and climate-explanatory variables. The resulting improved understanding of the variability and trends in flash floods should be of immense value to public planners, businesses, and insurance-based risk management agencies.