Outlier Analysis

To improve the SHIPS and LGEM models, cases with large forecast cases are being analyzed in greater detail. Cases with both high and low forecast biases are considered. The outlier cases are identified by looking at all forecasts for individual storms.

Another way to identify large errors is to average the intensity errors for all the forecast times out to 120 h for each forecast. Before the errors are averaged, they are normalized by dividing by the error standard deviation for that time period. The resulting average is called the Time Averaged Normalized Intensity Error (TANIE). The error standard deviations used in TANIE are from the NHC official intensity forecasts from 2008-2013 for a combined Atlantic and east Pacific sample. The error standard deviations range from 12 kt at 12 hr to 35 kt at 120 hr. A TANIE value of 1 or greater indicates that the model forecast error was 1 standard deviation above the mean of the NHC official forecast error. The bias associated with the normalized intensity errors is also calculated to identify errors that are systematically too high or low on average throughout the 120 h forecast period.

TANIE Plots Storm by Storm Forecasts