Hurricanes are one of deadliest and costliest natural hazards humans face. Accurate predictions of hurricanes therefore have enormous economic value. Unfortunately, today's predictions continue to have significant errors. Through ensemble based data assimilation and sensitivity experiments of Hurricane Katrina (2005) and a tropical disturbance closely preceding Hurricane Alex (2004), our results highlight the extreme sensitivity and thus limited predictability that can be present in short-term forecasts of tropical cyclones, especially their initiation and intensity. Strong sensitivity exists in simulations with a grid spacing comparable to that of global models currently used for numerical guidance (using parameterized convection) and with a convective-permitting grid spacing of 1-5 km.

However, the predictability of hurricanes varies under different flow regimes and under different stages of evolution, which is controlled by differences and uncertainties in the large-scale environment and internal dynamics. For example, the extreme sensitivity of the 2004 case only occurs with the disturbance on the Gulf side of the Florida Peninsula (which in nature did not develop into a tropical cyclone) and not with the disturbance that eventually led to Alex. This occurs despite both systems being disturbed with similar initial perturbations and residing in nearly the same large-scale environment of weak synoptic forcing (at 00Z 30 July 2004). Hurricane Katrina, on the other hand, has large forecast uncertainties in both track and intensity forecasts initialized before its landing in Florida Peninsula. Much less uncertainty is found for its forecast initialized after moving into the Gulf of Mexico but the intensity forecast can be very limited by improper hurricane initialization. Through assimilating both ground-based and airborne radar and sounding observations obtained during RAINEX into a cloud-resolving NWP model, the ensemble-based data assimilation is shown to be capable of “hot-starting” a major hurricane near its full strength which has great potential for hurricane intensity forecasts.

The strong flow-dependent sensitivity exemplifies the inherent uncertainties in hurricane intensity prediction in which moist convection and large-scale environment are the keys that limit the predictability. The current study implies that the predictability of tropical cyclones is strongly limited at all temporal and spatial scales. These inherent uncertainties in hurricane forecasts illustrate the need for event-dependent probabilistic hurricane forecasts and risk assessment. Inherent uncertainties in hurricane forecasts also illustrate the need for raising public awareness on the inherent uncertainties of hurricane forecasts and for designing appropriate policies for disaster planning, risk management, and vulnerability reduction.