File:Ambiguity in ensemble forecasting- evolution, estimate validation and value (IA ambiguityinensem1094510467).pdf

An ensemble prediction system (EPS) generates flow-dependent estimates of uncertainty (i.e., random error due to analysis and model errors) associated with a numerical weather prediction model to provide information critical to optimal decision making. Ambiguity, or uncertainty in the prediction of forecast uncertainty, arises due to EPS deficiencies, including finite sampling and inadequate representation of the sources of forecast uncertainty. An EPS based on a low-order dynamical system was used to investigate the behavior of ambiguity, validate two practical estimation methods against a theoretical (impractical) technique, and apply ambiguity in decision making. Ambiguity generally decreased with increasing lead time and was found to depend strongly on ensemble forecast variance and the variability of ensemble mean error. The practical estimation techniques provided reasonably accurate ambiguity estimates, although they were too low at early lead times. The theoretical ambiguity estimate added significant value when combining ambiguity with forecast uncertainty to provide a single normative decision input. Additionally, value added to secondary user criteria (e.g., minimizing repeat false alarms), was explored using the practical estimations. Repeat false alarms were significantly reduced while maintaining primary value by using ambiguity information to selectively reverse normative decisions to take protective action, which effectively redistributed negative outcomes.

Subjects: Value.; Ensemble Forecast; Ambiguity; Uncertainty; Ensemble-of-Ensemble; Calibrated Error Sampling; Randomly Calibrated Resampling; Optimal Decision Making; Cost-Loss; Uncertainty-Folding; Secondary Criteria; Lorenz '96; Ensemble Prediction Systems;