The current semi-arid coverage of one-third of the earth’s surface may increase in the future due to the effects of global warming. Many semi-arid regions are particularly affected by flash floods, caused mainly by convective storm systems. In the United States, flash floods kill more people annually than any other natural disaster, account for more than eighty percent of all flood-related deaths, and cause an average of one billion dollars of economic losses. Predicting these flash-floods is extremely difficult due to their short duration and the small geographic region over which they occur. An adequate predictive tool has to be based on a good geometric representation of the watershed and be driven by high-resolution, spatially distributed and accurate precipitation measurements. To this end, we developed a continuous-time version of the well established event-based rainfall-runoff model, KINEROS2. This is a spatially distributed kinematic wave model that represents the catchment as a cascade of overland flow planes and trapezoidal channel elements. The dynamic infiltration algorithm is particularly well suited to simulate hydrologic processes in semi-arid regions, such as Hortonian overland flow, transmission losses in ephemeral streams, etc. KINEROS2 has been restructured into discrete modules which can be configured to run in continuous-time mode for operational use. Prediction improvement involves model sensitivity analysis and calibration strategies. The aim of this project is to utilize a model that provides more accurate and reliable flood warnings, including probabilistic information, to aid in flash-flood forecasting, flood-related risk assessment and decision-making. Pilot applications to a few watersheds in the southwest US are presented.