An Evaluation Metric for Intraseasonal Variability and its Application to CMIP3 Twentieth-Century Simulations

Abstract

The intraseasonal variability (ISV) is an intermittent phenomenon with variable perturbation patterns. To assess the robustness of the simulated ISV in climate models, it is thus interesting to consider the distribution of perturbation patterns rather than only one average pattern. To inspect this distribution, the authors first introduce a distance that measures the similarity between two patterns. The reproducibility (realism) of the simulated intraseasonal patterns is then defined as the distribution of distances between each pattern and the average simulated (observed) pattern. A good reproducibility is required to analyze the physical source of the simulated disturbances. The realism distribution is required to estimate the proportion of simulated events that have a perturbation pattern similar to observed patterns. The median value of this realism distribution is introduced as an ISV metric. The reproducibility and realism distributions are used to evaluate boreal summer ISV of precipitations over the Indian Ocean for 19 phase 3 of the Coupled Model Intercomparison Project (CMIP3) models. The 19 models are classified in increasing ISV metric order. In agreement with previous studies, the four best ISV metrics are obtained for models having a convective closure totally or partly based on the moisture convergence. Models with high metric values (poorly realistic) tend to give (i) poorly reproducible intraseasonal patterns, (ii) rainfall perturbations poorly organized at large scales, (iii) small day-to-day variability with overly red temporal spectra, and (iv) less accurate summer monsoon rainfall distribution. This confirms that the ISV is an important link in the seamless system that connects weather and climate.