The concept of predictability limits in meteorology dates back to Edward Lorenz in the 1960’s, when he first shows that infinitesimally small perturbations to the initial conditions (ICs) would eventually diverge, as result of the chaotic nature of the atmosphere, to entirely different solutions. For, the past ~60 years, meteorologists have sought to quantify the absolute, or intrinsic, predictability limit of weather, as well as our current, or practical, predictability limits. The schematic below from Melhauser and Zhang (2012) depicts a scenario on the left where the truth lies far from the bifurcation point, such that further reducing the IC uncertainty improves the practical predictability. However, in the scenario depicted to the right, the truth lies near the bifurcation point and further reducing the IC uncertainty, yields little improvement in predictability, suggesting this scenario is limited by the intrinsic predictability.

Atmospheric predictability in general results from both uncertainty in ICs, as well as model dynamics and physics uncertainties. My research has sought to examine the impact of both on tropical cyclone (TC) forecast uncertainty.

One such example comes from work I completed examining Hurricane Joaquin (2015), which provided operational forecasters with significant challenges due to large uncertainties in both track and intensity forecasts. Using a convection allowing 60-member ensemble initialized from a cycling EnKF, I examined the regions of ICs that contributed most to the track and intensity forecasts by isolating the IC differences within the large-scale environment and the tropical cyclone itself (Nystrom et al. 2018). When IC differences to the TC itself are removed (Rcore), this ensemble is the only one whose track spread approaches that of CNTL. These results suggest that Joaquin’s large track forecast spread likely occurs regardless of perturbations to the initial storm’s inner core region. On the other hand, when IC uncertainty to the storm itself are removed the intensity spread is significantly reduced relative to CNTL, however intensity spread still exists and results from differences in environmental wind shear, relative humidity and track differences. When IC differences to the environment are removed (Renv), the track spread is significantly reduced relative to CNTL, further suggesting that IC differences in the environment are necessary for Joaquin’s track divergence. Conversely, even without environmental IC uncertainty the intensity spread from Renv is nearly identical to CNTL, suggesting that if the practical predictability of Joaquin’s intensification is to be improved IC uncertainty to the storm itself must be reduced.In summary, for Hurricane Joaquin, IC differences for the TC itself, specifically the initial intensity, contributed most to the intensity forecast uncertainty, while IC uncertainty to the near storm environmental steering flow contributed most to the track forecast uncertainty.