Human estimates of warning uncertainty: Numerical and verbal descriptions

Abstract

The uncertainty of weather warnings is mostly expressed only in textual form (e.g., “thunderstorms are possible tomorrow afternoon”). Thus linguistic uncertainty might be added to the numerical uncertainty of the warnings. Two questions arise: can human forecasters estimate the uncertainty and how well is this done in verbal terms. Subjective and statistical forecasts of the probability of the occurrence of severe weather events for the city of Berlin were verified. Human estimates of the probability for the occurrence of thunderstorms and wind gusts > 14 m/s in Berlin were found to be reliable and possess significant skill in comparison to the statistical reference forecast. Additionally, the verbal description of warning uncertainty in an operational textual warning report was classified and objectively verified. Results indicate that forecasters actually are aware of the inherent uncertainty, yet express this by means of a multitude of verbal terms. In order to improve the communication and reduce confusions arising from linguistic uncertainty inherent to severe weather information, forecasts should thus contain few and well defined verbal phrases expressing forecast uncertainty. Relating numerical to verbal descriptions of uncertainty revealed a fundamentally different usage of wording when comparing warnings of thunderstorms and wind gusts >14 m/s, with “stronger” wording used in case of thunderstorms. This might indicate that risk information rather than probability information is communicated to the users of the considered warning information.