Researchers usually assume that all spots on a weather map are equally chaotic, meaning that small uncertainties in initial conditions lead to unpredictably different results. A realistic model of Earth’s atmosphere must be high dimensional—there are a great many degrees of freedom—which further complicates the art of forecasting. Now, a multidisciplinary team of scientists at the University of Maryland has shown that, locally, the finite-time dynamics of the atmosphere is often low dimensional. They developed a statistic called the bred vector (BV) dimension that characterizes the differences between a model atmospheric state (the initial condition) and several perturbations of that state that evolve for a finite time. Using real “ensemble forecasts” from the National Weather Service, they mapped the atmosphere’s BV dimension globally, as shown in the figure, where red is low dimensionality and blue is high. The team says that the low BV-dimension regions might be particularly important for obtaining accurate weather forecasts; atmospheric data obtained there will tend to evolve only along the subspace spanned by the few dominant bred vectors. (D. J. Patil et al. , Phys. Rev. Lett. 86 , 5878, 2001 https://doi.org/10.1103/PhysRevLett.86.5878 )
