Insights into Hurricane Track and Intensity Verification Results through Temporally Homogeneous Samples

Average track errors from most prediction models for forecasts of hurricanes in the North Atlantic basically increase linearly out to 120 h. However, average absolute intensity errors increase somewhat linearly out to only about 60 h and then stay about the same through 120 h. Comparisons between models are only considered “fair” if the sample is “homogeneous,” i.e., all models in the group to be verified have a forecast at a particular forecast interval. This type of homogeneity is critical since even a 10-20% difference in sample size between comparisons can produce a noticeable difference in results. However, for the homogeneous sample, the number of cases diminishes substantially, by a factor of 2.5 or even ~7 between the first forecast interval (12 h) and the last interval (120 h). To investigate the impact of the reduction in the number of cases for later forecast intervals, verifications for various hurricane prediction models were compared for normal homogeneous samples and for temporally homogeneous samples, where temporally homogeneous samples are defined to have the same number of cases at all forecast intervals, i.e., only those forecasts are used which verify for every model for every forecast interval. The use of a temporally homogeneous sample examines the impact of the reduced number of cases at later forecast intervals in the search for the cause of the special shape of average absolute intensity errors. It is important to evaluate model results in as many ways as possible, and it is hoped that a better understanding of the nature of the forecast errors, can help elucidate needed areas for improvements in the prediction models.