Some New Techniques for Hurricane Forecast Verification

Verification information helps forecasters find models that improve on current hurricane forecasting capability. Many new metrics are being used and researched for assessment of tropical cyclone forecasts, some very simple and others more complex.

For deterministic forecast models, average errors in track and intensity are examined by operational forecasters and verification systems. Some supplemental metrics have been introduced for recent assessments to examine both the frequency and magnitude of forecasting errors. Additionally, methods for making comparisons between a single model and several operational models have been added to improve the efficiency and understandability of verification results. These measures are simple and understandable, in the hopes of finding acceptance and use by the operational community. A rank histogram of errors compares a candidate model with a set of operational standards. Comparisons of two models are made using a frequency of superior performance for differences deemed practically significant. The traditional comparison of mean errors is enhanced by reporting p-values for the significance of the differences and making note of the practical significance of the typical error magnitudes. The increasing frequency of model forecasts raises the issue of consistency. Forecast changes and autocorrelations are calculated for both forecast intensities and errors to assess the consistency of model forecasts through time.

Examples of these analyses and some proposed improvements for future assessments will be presented.