Abstract：Forecasts of tropical cyclone (TC) tracks from six global models during 2010 and 2012 were assessed to study the current capability of track forecast guidance over the western North Pacific. To measure the performance of the global model forecasts, a series of statistical evaluations of track forecasts up to 120 h were carried out, including the mean, median, percentile distribution, regional distribution, relative position, correlation analysis, and binned analysis. Results showed that certain improvements have been made for the six global models in their prediction accuracy and stability in the past three years. Remarkably, stepped decreases in the values of each quantile were found at all lead time levels from 2010 to 2012 for NCEP-GFS. An analysis of the regional distribution of position errors showed that a high-latitude region, low-latitude region (which covers mostly the TC genesis region) and the South China Sea are the three main areas within which large errors tend to concentrate. The majority of the models show their own respective characteristics of systematic bias at each lead time, as established through the relative position analysis results. Only the results of NCEP-GFS and CMA-T639 did not show any obvious systematic bias in the three-year study period. Binned analyses indicated that the prediction accuracy and stability of most of the models were better for strong TCs than for weak TCs at short lead time levels. It was also found that the models tend to perform better for initially large TCs, or for those with weak vertical wind shear at lead times shorter than 48 h. The results demonstrate the heavy reliance of forecast errors upon the initial characteristics of a TC or its environmental conditions.
Guomin Chen, Hui Yu, Qing Cao et al., 2013: The Performance of Global Models in TC Track Forecasting over the Western North Pacific from 2010 to 2012. Tropical Cyclone Research and Review, 2(3), 149-158.