Probabilistic, Ensemble-Based, 2D Flood Forecasting Using Torrent

We’ve become accustomed to speaking in the language of probability when talking about precipitation forecasts ⏤ a 30% chance of rain or a 50% exceedance probability ⏤ yet computational challenges have generally prevented the translation of these probabilistic precipitation forecasts into fully-2D probabilistic flood forecasts. Torrent is a Lagrangian rivulet-based model of overland flooding that is 4-7 orders of magnitude more computationally efficient than the solution of the shallow water equations. These efficiency gains now enable the full, probabilistic assessment of flood depth and inundation area projections incorporating uncertainties in both precipitation forecasts and those inherent in Lagrangian flood simulations. This advancement allows researchers to capture and quantify uncertainty in future flood risks under a variety of climate projections as well as to probabilistically assess distributions of flood risks preceding hurricane landfalls and atmospheric-river events. This will allow climate researchers, weather forecasters, and first responders to more accurately identify communities at risk even under significant levels of uncertainty.