Application of adjoint-derived sensitivity gradients to tropical cyclone intensification

With properly defined response functions, adjoint models can provide information about the dynamical sensitivity of specific aspects of a modeled tropical cyclone's (TC) evolution to the model forecast trajectory. The output of adjoint models, coupled with statistical measures of forecast trajectory uncertainty, can be used to define regions for targeted observations to improve subsequent forecasts of a TC's evolution. However, lack of rigorous dynamical interpretation of adjoint-derived sensitivity gradients makes it difficult to apply these sensitivities to the problem of targeted observing.

The NOGAPS model and its adjoint are used to demonstrate the inappropriateness of various response functions typically used to define TC intensity. Perturbations to initial conditions in regions of high sensitivity are used to facilitate dynamical interpretation of the sensitivity gradients, and several response functions used to define TC intensity are tested for their utility. Several response functions based on low-level vorticity and kinetic energy in the region of the TC are shown to be sensitive to perturbations to the location of the TC and its surrounding environment within the response function box, yielding a change to the response function without significantly affecting the intensity of the TC itself. Response functions describing specific processes related to TC intensification (e.g. vertical motion leading to the stretching of the air column) are shown to be more successful, but these response functions constrain sensitivity gradients to only focus on the one aspect of TC intensification described by the response function.