摘要：A new measure (the Track Forecast Integral Deviation; TFID) for the verification of tropical cyclone (TC) track forecasts is proposed, based on the mathematical consideration that a “good” forecast has a small distance to the observed track not only at zero-order but also at higher orders. The TFID is the mean of two sub-scores, which are respectively calculated for latitude and longitude and defined to be the average value of the mean absolute error and mean absolute deviation of relative errors from the mean relative error along a track. By definition, the smaller the TFID, the more accurate the forecast track. A perfect forecast has zero TFID. It is suggested that such a measure is superior to the widely-used position error (PE) in terms of reflecting the accuracy of the whole track instead of just one position. In an experimental application, TFID was calculated for the track forecasts from the ECMWF-IFS during 2010-2012. A comparison with PE showed that TFID can work as a good supplement to the PE in discriminating good or bad track forecasts, as there are generally some forecasts with small PE but large TFID, or vice versa. The binned characteristics of TFID and PE of ECMWF-IFS were also analyzed based on several traits of the TC or its environment at the initial time of the forecast. It was found that the model performs better for initially strong and large TCs, or those with weak vertical wind shear at lead times shorter than 48 h.
