In recent years, it has become clear that it will be difficult or impossible to develop deterministic forecasts of convective activity on the time and space scales required by decision makers such as air traffic managers. Current capabilities in forecasting convection generally are based on extrapolation techniques in the very short term (0-2 hours) and numerical model predictions in the longer term (2-12 hours). However, even in the very short term, uncertainties associated with initiation, growth, decay, and movement of storms and storm systems are very difficult to capture or anticipate with precision. Thus, these forecasts are an ideal candidate for probabilistic forecasts in which the uncertainty associated with the development and movement of the systems can be captured and expressed directly, and much effort is currently being focused in this direction.

But how should a probabilistic forecast of convection be formulated? How should it be interpreted? And how should it be evaluated? These issues are of critical concern as we head down the road toward spatial probabilistic forecasts of a variety of phenomena. One approach is to equate the probability values to an expected coverage in the region of a grid point. Then temporal and spatial scaling issues associated with the forecasts can be taken into account directly by adjusting the size of the area of interest for computing coverage. Another approach might involve developing an ensemble of possible scenarios characterizing future convective activity that can be evaluated to determine the probabilities of occurrence of particular types of scenarios.

These issues are considered in this paper, with particular attention to the national scale convective forecasts produced by the National Convective Weather Forecast (NCWF). Experimental probabilistic forecasts produced by the NCWF are evaluated to help understand the appropriate interpretation of these forecasts as well as to determine how well the forecasts perform at different spatial and temporal scales.