Precipitation parameter sensitivity and optimization challenges in
climate models

Despite successful simulation of many large-scale climate variables,
climate models exhibit considerable bias in regional precipitation, and
very substantial differences in regional scale projections of
precipitation changes under global warming---currently a stumbling block
in regional assessment of human and ecosystem impacts. Systematic
examination of parameter sensitivity and optimization issues in a single
climate model can help to identify fundamental issues underlying this
regional hydroclimate sensitivity. Parameter-space exploration in the
International Centre for Theoretical Physics atmospheric general
circulation model forced with observed SST, or a coupled to an ocean
mixed layer suggests sufficiently smooth parameter dependence for
leading climate fields that simple metamodeling and multi-objective
optimization strategies for high dimensional design problems adapted
from the engineering literature can be highly efficient. This holds
despite nonlinearity sufficiently strong to reverse the curvature of RMS
objective functions for precipitation in one parameter direction. Optima
often occur at the limit of the feasible parameter range, identifying
key parameterization aspects warranting attention---here the interaction
of convection with free tropospheric water vapor. Analytic results help
to visualize the contributions to the optimization problem on a regional
basis.

http://www.atmos.ucla.edu/~neelin/