An Assessment of GCM Skill in Simulating Persistence across Multiple Time Scales

Abstract

limate change impact studies for water resource applications, such as the development of projections of reservoir yields or the assessment of likely frequency and amplitude of drought under a future climate, require that the year-to-year persistence in a range of hydrological variables such as catchment average rainfall be properly represented. This persistence is often attributable to low-frequency variability in the global sea surface temperature (SST) field and other large-scale climate variables through a complex sequence of teleconnections. To evaluate the capacity of general circulation models (GCMs) to accurately represent this low-frequency variability, a set of wavelet-based skill measures has been developed to compare GCM performance in representing interannual variability with the observed global SST data, as well as to assess the extent to which this variability is imparted in precipitation and surface pressure anomaly fields. A validation of the derived skill measures is performed using GCM precipitation as an input in a reservoir storage context, with the accuracy of reservoir storage estimates shown to be improved by using GCM outputs that correctly represent the observed low-frequency variability.Significant differences in the performance of different GCMs is demonstrated, suggesting that judicious selection of models is required if the climate impact assessment is sensitive to low-frequency variability. The two GCMs that were found to exhibit the most appropriate representation of global low-frequency variability for individual variables assessed were the Istituto Nazionale di Geofisica e Vulcanologia (INGV) ECHAM4 and L?Institut Pierre-Simon Laplace Coupled Model, version 4 (IPSL CM4); when considering all three variables, the Max Planck Institute (MPI) ECHAM5 performed well. Importantly, models that represented interannual variability well for SST also performed well for the other two variables, while models that performed poorly for SST also had consistently low skill across the remaining variables.