Abstract
The main goal of this work is to quantify the predictability of tropical cyclone (TC) intensity. Motivated by the lack of improvement in TC intensity prediction, a systematic study on the intrinsic predictability of TC intensity is conducted using a set of five high-resolution, cloud-resolving realistic model ensembles. The ensembles are generated with a stochastic kinetic-energy backscatter (SKEBS) method. Error growth is addressed by imposing stochastic perturbations with various spatial scales on the TC and its environment. The SKEBS ensembles feature convective-scale, mesoscale and synopticscale perturbations to better understand the growth of scale-dependent errors and their impact on TC uncertainty and predictability. TC intensity predictability is determined by computing the error magnitude associated with each component of the Fourierdecomposed TC wind fields at forecast times up to 7 days. It is found that forecast errors grow rapidly and saturate within 6-12 h on small scales (~ 30 km) in all five ensembles, independent of perturbation scale. Errors grow relatively slower on scales corresponding to rainbands (200-500 km), limiting the predictability of these features to 1-5 days. The predictability limit of rainbands strongly depends on perturbation scale, indicating that error downscaling is more detrimental than the upscale spread of small-scale errors. In long-lived TCs, the storm-scale circulation (i.e., the mean TC vortex and wavenumber-1 asymmetry) is resistant to upscale error propagation and remains predictable for at least 7 days. Uncertainty of the storm-scale circulation is only significant when the mean vortex is perturbed dircectly, demonstrating that TC intensity uncertainty and predictability is mainly affected by large-scale errors. This suggests that the predictability of the stormscale circulation is predominately controlled by the large-scale environment. A novel TC wind speed climatology based on 15 years of aircraft observations is created to investigate the predictability of various TC wind speed metrics using an information theory approach. Consistent with the results from the error growth approach, we show that the storm-scale wind field is predictable for more than 7 days. However, the wind speed at the radius of maximum wind loses its predictability during a period of rapid intensification (RI), which agrees with the extreme uncertainty of the peak wind during RI in the SKEBS ensembles. The predictability of the wind speed at the radius of maximum winds “recovers” during the maximum intensity phase, demonstrating that TC intensity predictability is intimately related to the phase of TC evolution and distinct physical processes. Environmental and internal mechanisms associated with RI uncertainty were investigated to better understand RI predictability. Both environmental (e.g., vertical wind shear) and internal (e.g., inner-core inertial stability) parameters play a role in the uncertainty of RI timing, indicating that the predictability of RI is a complex problem. The environment, which has longer predictability, controls the general tendency for RI to occur. In contrast, the impact of small-scale processes implies that the exact timing of RI has a short intrinsic predictability limit. The ensemble members are divided into two groups depending on their RI onset time. A comparison of the dynamic and thermodynamic fields shows that the physical mechanisms associated with RI differ significantly between the two groups. In the early RI cases, mid-tropospheric radial inflow is strong, leading to a fast contraction of the radius of maximum wind, increasing inertial stability, and the development of an eyewall. In contrast, the late cases have a well-developed (but broader) eyewall before RI onset, and the radius of maximum wind contracts slowly. The development of an upper-level warm core accompanies RI in the late cases. These results indicate that RI is associated with different physical processes during distinct stages of the tropical cyclone lifecycle.