Sensitivity Analysis of a Moist 1D Eulerian Cloud Model Using Automatic Differentiation

Abstract

An automatic differentiation tool (ADIFOR) is applied to a warm-rain, time-dependent 1D cloud model to study the influence of input parameter variability, including that associated with the initial state as well as physical and computational parameters, on the dynamical evolution of a deep convective storm. Storm dynamics are found to be controlled principally by changes in model initial states below 2 km; once perturbed, each grid variable in the model plays its own unique role in determining the dynamical evolution of the storm. Among all model-dependent variables, the low-level temperature field has the greatest impact on precipitation, followed by the water vapor field. Mass field perturbations inserted at upper levels induce prominent oscillations in the wind field, whereas a comparable wind perturbation has a negligible effect on the thermodynamic field. However, the wind field does influence the precipitation in a more complex way than does the thermodynamic field, principally via changes in time evolution. The simulated storm responds to variations in three physical parameters (the autoconversion/accretion rate, cloud radius, and lateral eddy exchange coefficient) largely as expected, with the relative importance of each, quantified via a relative sensitivity analysis, being a strong function of the particular stage in the storm?s life cycle.