Sensitivity analysis of a 3D convective storm: Implications for variational data assimilation and forecast error

In this study a nonhydrostatic 3D cloud model, along with an automatic diff erentiation tool, is used to investigate the sensitivity of a supercell sto rm to prescribed errors (perturbations) in the water vapor field. The evolu tion of individual storms is strongly influenced by these perturbations, th ough the specific impact depends upon their location in time and space. Gen erally, perturbations in the rain region above cloud base have the largest impact on storm dynamics, especially for subsequent storms. while perturbat ions in the ambient environment above cloud base influence mostly the main storm. Although perturbations in the subcloud layer have a relatively small impact on upper-level storm structure, they do impact the low-level struct ure, especially during the period immediately following insertion. Sensitivities are also examined in the context of variational data assimila tion and forecast error. For perturbations added inside the active storm, t he cost function, which is prescribed to measure the discrepancy between fo recast and observations for all variables over rime and space, is found to be most sensitive to temperature, followed by pressure and water vapor. Thi s implies that the quality of variational data assimilation can be affected by the inaccuracy of observing or retrieving those quantities. It is also noted that, at least for the case studied here, the pressure held has the l argest influence on forecast error immediately after the errors are inserte d, while the temperature field does so over a longer time period.