Importance of internal variability for climate model assessment

Benchmarking climate model simulations against observations of the climate is core to the process of building realistic climate models and developing accurate future projections. However, in many cases, models do not match historical observations, particularly on regional scales. If there is a mismatch between modeled and observed climate features, should we necessarily conclude that our models are deficient? Using several illustrative examples, we emphasize that internal variability can easily lead to marked differences between the basic features of the model and observed climate, even when decades of model and observed data are available. This can appear as an apparent failure of models to capture regional trends or changes in global teleconnections, or simulation of extreme events. Despite a large body of literature on the impact of internal variability on climate, this acknowledgment has not yet penetrated many model evaluation activities, particularly for regional climate. We emphasize that using a single or small ensemble of simulations to conclude that a climate model is in error can lead to premature conclusions on model fidelity. A large ensemble of multidecadal simulations is therefore needed to properly sample internal climate variability in order to robustly identify model deficiencies and convincingly demonstrate progress between generations of climate models.