A sensitivity-enhanced simulation approach for Community Climate System Model

Abstract

A global sea-ice modeling component of the Community Climate System Model was augumented with automatic differentiation (AD) technology. The numerical experiments were run with two problem sets of different grid sizes. Rigid ice regions with high viscous properties cause computational difficulty in the propagation of AD-based derivative computation. Pre-tuning step was required to obtain successful convergence behavior. Various thermodynamic and dynamic parameters were selected for multivariate sensitivity analysis. The major parameters controlling the sea-ice thickness/volume computation were ice and snow densitives, albedo parameters, thermal conductivities, and emissivity constant. Especially, the ice and snow albedo parameters are found to have stronger effect during melting seasons. This high seasonal variability of the thermodynamic parameters underlines the importance of the multivariate sensitivity approach in global sea-ice modeling studies. © Springer-Verlag Berlin Heidelberg 2006.