Adopting Model Uncertainties for Tropical Cyclone Intensity Prediction

Tropical cyclone (TC) intensity prediction remains highly uncertain, despite the current efforts in improving the performance of numerical prediction models. This uncertainty has been attributed to many factors, one of them being the poor representations of physical processes within the models. Particularly, TC intensity predictions are sensitive to the choice of the physical parameterizations that represent small-scale processes that would otherwise not be resolved by the models, such as cloud microphysics, planetary boundary layer processes and turbulence. In order to better understand which set of parameterizations should be used to improve TC intensity forecasts, the Generic Inversion by Transfer Function Analysis (GITFA) is introduced in this study. The method produces a joint probability density function (PDF) of inverse estimation solution for a selected set of parameters given the forecast model and observations with their associated errors. This PDF in the parameter space is non-Gaussian for the nonlinear models and provides information about likelihood of the joint values of the parameters that would result in the model forecast within a given range of the uncertainty in the observation space. The PDF of the inverse estimate defines the optimal, mutually correlated ensemble of parameter values.

In this study, two physical parameterizations from an axisymmetric model were perturbed to produce different idealized TCs simulations. Results from those simulations were used to form the transfer functions for GITFA to obtain the inverse solutions. Preliminary results show that when the observation is a point within the TC field, such as the maximum wind speed, the optimal range of parameters is poorly constrained. On the other hand, when an entire kinematic field is observed, the optimal parameters can be constrained to a subset of joint range of values. The results suggest that an ensemble of physical parameterizations should be employed to improve TC intensity forecasts and that the ensemble could be optimally derived using inverse solutions constrained by the observations that contain information about the vortex structure.