These univariate probabilistic forecasts are useful for decisions that depend on a single weather variable; however, one generally cannot use them to derive proper probabilistic information for weather impacts dependent on combinations of variables, locations or times. For example, choosing a good time for a picnic involves the joint probability of a warm temperature, no precipitation and light winds occurring simultaneously. Similarly, the likelihood of a crop freezing, and thus the need for mitigation measures, depends on multi-hour temperature sequences. And forecasting whether energy use in a region will exceed a critical threshold involves the spatial and temporal patterns of temperatures and winds, among other factors.

For this reason, we have introduced a PFP capability that provides a large number of possible weather scenarios, or “prototypes.” Prototypes can be thought of as analogous to individual NWP ensemble members, but in this case, they are constructed to remove systematic error and to collectively span the statistically calibrated forecast probability distributions derived from the multi-model ensemble. Prototypes are computed based on the Ensemble Copula Coupling - Quantile (ECC-Q) methodology, which ensures that the set of prototype values for a given weather variable, location and lead hour comprise an equally-spaced set of forecast PDF percentiles. Thus, the prototypes can be thought of as “equally likely;” that is, each prototype has an equal chance of verifying for each weather variable, location and lead hour. Because they are derived from a canonical set of input NWP forecasts, PFP prototypes generally represent coherent scenarios across space, time, and weather variables. As such, the prototypes can be used as input to weather impact models that traditionally have been driven by deterministic forecasts. If the impact model is run on each prototype, then a probability distribution of outcomes can be derived from the suite of results. For example, the probability of “picnic weather” can be estimated from 100 prototype forecasts of temperature, precipitation, and wind speed by counting the number of prototypes for which all three variables fall within the desired ranges. This approach has a significant advantage over running the impact model using inputs directly from an ensemble of NWP inputs since it draws from a set of statistically calibrated inputs of equal likelihood. It does have shortcomings as well, and these will also be discussed. Nevertheless, we argue that PFP prototypes provide a valuable new tool for forecasting uncertainty in weather-dependent outcomes and thus helping to optimize a broad range of decisions.