Automated parameter tuning applied to sea ice in a global climate model

Lettie A Roach,Michael J Mineter,Simon F B Tett,Kuniko Yamazaki,Cameron D Rae

doi:10.1007/s00382-017-3581-5

Abstract

This study investigates the hypothesis that a significant portion of spread in climate model projections of sea ice is due to poorly-constrained model parameters. New automated methods for optimization are applied to historical sea ice in a global coupled climate model (HadCM3) in order to calculate the combination of parameters required to reduce the difference between simulation and observations to within the range of model noise. The optimized parameters result in a simulated sea-ice time series which is more consistent with Arctic observations throughout the satellite record (1980-present), particularly in the September minimum, than the standard configuration of HadCM3. Divergence from observed Antarctic trends and mean regional sea ice distribution reflects broader structural uncertainty in the climate model. We also find that the optimized parameters do not cause adverse effects on the model climatology. This simple approach provides evidence for the contribution of parameter uncertainty to spread in sea ice extent trends and could be customized to investigate uncertainties in other climate variables.

Highlights

Lack of consistency between observed sea ice extent trends and those simulated by complex climate models has garnered significant public and scientific attention in recent years
The proximity of the final parameter configurations suggests that both experiments have located the same stationary point and provides evidence that the optimization is robust under changes in spinup period
This paper reports on the application of automated parameter tuning to uncertainty in climate model simulations of sea ice

Summary

Introduction

Lack of consistency between observed sea ice extent trends and those simulated by complex climate models has garnered significant public and scientific attention in recent years. Simulations from the latest Coupled Model Intercomparison Project (CMIP5) continue to exhibit various biases in their mean state, during winter months (Stroeve et al 2012; Turner et al 2013), as well as in the magnitude and even sign of hemispheric trends (Flato et al 2013) Such model biases in recent historical sea ice can broadly be attributed to two sources of uncertainty: firstly, internal variability arising due to the chaotic nature of climate, and secondly, structural uncertainty arising from incomplete knowledge of and inability to represent the complete physics of the climate system (Hawkins and Sutton 2009). Sea ice parameter uncertainty can be larger than that expected from natural variability (Ridley et al 2007) and

Methods

Results

Conclusion