Using nudged simulations to investigate parameter sensitivities in CAM5

Abstract:

There are still many tunable parameters in the physical parameterizations used by global climate models. This compromises their skill in predicting future climate change. To quantify the sensitivity of modeled climate to these parameters, sufficiently long simulations are usually required to reduce the effect of natural variability. However, such long simulations are too expensive for a high-resolution model, especially when many simulations are needed to identify the parametric uncertainty. As an alternative approach, nudging (also called Newtonian relaxation) constrains large-scale meteorology and can distinguish the target signal from the natural noise in relatively short simulations. Here, we test whether this method biases the sensitivity of simulations to model parameters. We nudged the model towards a common baseline simulation, for both winds and temperature or winds alone. We found that constraining both winds and temperature suppresses the interaction between clouds and meteorological fields, inhibiting the response of moisture convergence and precipitation to changes in parameter values. Instead, nudging winds alone can capture the climate response to parameter perturbations more effectively. In addition, nudging simulations can reduce the computational cost by more than a factor of six.