An Analysis of Atlantic Tropical Cyclone Forecast Errors, 1970–1979

Abstract

Official Atlantic tropical cyclone track forecasts issued by the National Hurricane Center over the decade of the 1970's are examined in this study. The purpose is to isolate any significant biases, trends or other statistical properties of the forecast errors. In this analysis, we consider (i) storms Initially located in the easterlies and (ii) storms initially in the westerlies. Latitude 24.5°N approximately delimits the two zones. The mean forecast error (MFE) on group (i) storms is shown to be substantially less than group (ii). However, it also is shown that ?skill,? defined as the ability to improve on a forecast based only on climatology and persistence, is greater for storms embedded in the westerlies. An examination of the year-to-year variation in MFE, after these errors have been adjusted according to attendant year-to-year variation in average storm latitude and translational speed, fails to disclose any upward or downward trend in the ability to forecast tropical cyclone motion over the decade of the 1970's. This finding is evaluated in relation to an earlier study, which describes a 10?12% decline in MFE over the 8-year period 1959?66. To examine biases and dispersion patterns of the errors, probability ellipses associated with the bivariate normal distribution were fitted to the classes of forecast errors described above. Two coordinate systems were used: one system was oriented in the conventional zonal/meridional sense and the other was rotated in accordance with storm emotion. Small right-of-track and slow biases were identified. Overall, these biases were found to be small, although a number of sizable regional biases were identified. The study concludes with an evaluation of temporal and spatial aspects of landfall error, i.e., the distance between forecast and observed point of tropical cyclone landfall measured along the coast.