The limited skill of dynamical tropical cyclone intensity models suggests that in addition to model error, there must be some significant loss of predictive skill because of inadequate knowledge of the initial state of the cyclone and its environment. This hypothesis is evaluated with ensemble forecasts of tropical cyclone intensity in a highly simplified situation, but utilizing an estimate of the analysis uncertainty of real storms. The results show divergent, and often times bimodal, intensity forecast probability density functions, due to the inherent predictability of the system in combination with the lack of sufficient observational constraints on the analysis uncertainty. The state-dependent sensitivity of the model forecast intensity to small differences in the magnitude of the environmental wind shear is especially striking in its tendency to create bifurcations in the ensemble intensity forecasts. If nature exhibits similarly bimodal behavior, it could explain part of the reason for the success of deterministic statistical forecasts relative to their dynamical counterparts in root-mean-square error verification. Ultimately, however, these preliminary results favor a probabilistic approach rather than a deterministic approach for producing tropical cyclone intensity forecasts of greatest utility to the user.