We present a prototype human-computer interface designed to assist military forecasters in the evaluation of uncertainty in numerical model forecasts. We refer to it as the uncertainty monitor. The modern military weather forecaster relies heavily on numerical models (Joslyn, et al., submitted) to produce their forecasts. The reliability of the information provided by the models is variable, however. Therefore, the forecaster must also consider the amount of uncertainty inherent in the information provided by the models. The steps involved in conducting a thorough evaluation of model uncertainty are time consuming. The forecaster must determine the answers to the following questions: How accurate have the models been over the past few days? How do the model initializations compare to the observational data? How uncertain are the current model predictions? Because military forecasting is often done under time pressure, the amount of uncertainty evaluation that can be done is often limited. In addition, the optimum means to convey this information to the end user is not well understood (Sauter, 2003). Nevertheless, the level of uncertainty in a forecast can be a crucial factor in tactical decisions.

The uncertainty monitor was developed to streamline the process of uncertainty evaluation. This effort is a component of a Department of Defense Multi-disciplinary University Research Initiative (MURI) on statistical and cognitive approaches to visualizing uncertainty in mesoscale meteorology. We began with a cognitive task analysis (CTA) of operational forecasters producing the Terminal Aerodrome Forecast (TAF). The CTA revealed that Navy forecasters are concerned about model uncertainty and have specific techniques for evaluating it. We designed a prototype human-computer interface that, while compatible with their cognitive process, also provides tools and visualizations that speed and enhance the forecasters’ ability to rapidly assess model uncertainty. The interface, viewable via a web browser, provides real-time information on global and mesoscale model performance, composite satellite pictures overlaid with model analysis and observational data, and a variety of probabilistic forecast products, which are derived from the University of Washington’s Short-Range Ensemble Forecasts (SREF) system.