Hurricane predictability: are there simple linear invariants within these complex nonlinear dynamical systems?

Abstract

Forecasting the tracks of hurricanes is a problem of immense importance. It is a major scientific exercise in solving the complicated set of mathematical equations that govern the behavior of atmospheric flow in general and hurricanes in particular. Moreover, hurricanes rank as the most devastating of all natural phenomena, in terms of loss of life and destruction of property. Hitherto, unlike many other atmospheric and oceanic systems, hurricanes have defied rapid advances in prediction of their motion, and progress has been of the order of a mere one percent or so reduction per annum in mean 48 hour forecast position errors over the past two decades. A research program aimed at estimating inherent and actual mean absolute forecast position errors, has produced an apparent paradox. Despite the fact that the equations governing hurricane motion are a complex, coupled, nonlinear set of dynamical equations, there is very strong evidence for the existence of an underlying simple, linear, invariant behavior. The original aim of the research program was to determine the lower limits of mean hurricane forecast position errors and to quantify them out to 72 hour leadtime. The appearance of the paradox meant that the focus shifted first to examining and explaining the paradox. Attention then turns to showing that the mean forecasts errors are still a very large 40 to 50 percent lower than the mean position errors currently being achieved in practice by state-of-the-art models numerical weather prediction (NWP) models.