Statistical analyses of the most recent 40 yr of hurricane track (1956-95)
are presented, leading to a version of the North Atlantic climatology and p
ersistence (CLIPER) model that exhibits much smaller forecast biases but si
milar forecast errors compared to the previously used version. Changes to t
he model involve the inclusion of more accurate historical tropical cyclone
track data and a simpler derivation of the regression equations. Nonlinear
systems analysis shows that the predictability timescale in which the aver
age errors increase by a factor e is approximately 2.5 days in the Atlantic
basin, which is larger than that found by similar methods near Australia.
This suggests that 5-day tropical cyclone track forecasts may have some ben
efit, and therefore a version of CLIPER extended to 5 days to be used as a
baseline to measure this skill is needed.