An integrated measure of tropical cyclone track forecast performance

A new measure is proposed to evaluate the performance of Tropical Cyclone (TC) track forecasts. This measure, called the “Track Forecast Integral Deviation” (TFID) is based on the idea that a good TC forecast has small track errors at both early and later lead times. The TFID is the mean of two sub-scores, calculated for latitude and longitude errors, and defined as the average value of the mean absolute error and mean absolute deviation of relative errors from the mean relative error along a track, across lead times. The TFID provides a clear measure of forecast performance: the smaller the TFID, the more accurate the forecast track. A perfect forecast has zero TFID. The proposed approach has the benefit of reflecting the accuracy of the whole track instead of just one position. In an experimental application, TFID was calculated for track forecasts from the ECMWF-IFS during 2010–2012. A comparison of TFID results with standard position error (PE) values indicates that TFID is a good supplement to the PE in discriminating good or bad track forecasts; typically, some forecasts have small PE but large TFID, or vice versa. Analyses of forecast subsets for the example forecasts also suggest that TFID can provide additional insights regarding TC forecast performance, including performance as a function of TC strength and wind shear.