Probabilistic Hazard Detection using a Time-lagged Ensemble

Operational forecasters are being provided today with an increasing range of probabilistic hazard detection tools, however many of these tools lack sufficient calibration or objectively demonstrated skill and reliability over a long period. Here we demonstrate a method for predicting heavy precipitation over the eastern United States at short (0-18-h) lead times, using an time-lagged ensemble of forecasts from the High-Resolution Rapid-Refresh model. Various sensitivity experiments are performed in order to determine the optimal number of ensemble members, neighborhood size, member weighting and latency. Forecasts are calibrated using the Stage-IV precipitation analysis, with the goal of developing a system capable of correcting biases from a relatively short training dataset. The technique is evaluated using a large archive of model forecasts spanning multiple seasons. Future work will involve testing with other types of short-range ensembles (e.g., multi-core or single suite, stochastic physics), and expanding the tool to additional hazards, including but not limited to: winter precipitation, severe convective storms, high wind speeds, icing and low visibility.