You are here

Making Sense of the Multi-Model Decadal Prediction Experiments from CMIP5

Vertical Tabs


Session Description: 

Decadal prediction, defined as initialized climate model experiments to predict the regional time-evolving statistics of climate over the next 10 to 30 years, has elicited a lot of interest in both the climate science and policy communities. However, decadal prediction is a new area of climate science, and in recognition of this fact a seminal 2008 AGCI session formulated a first-ever experimental design to address the science issues involved with decadal prediction that became incorporated in the Coupled Model Intercomparison Project phase 5 (CMIP5). Modeling groups from around the world are currently running these decadal prediction experiments, with the intention that they will be assessed as part of the IPCC AR5.

Since decadal prediction is new for the climate science community, most of the decadal experiments for CMIP5 are hindcasts designed to quantify expected skill of the predictions. These are 10-year hindcasts for initial states starting in 1960 and performed every five years, and two 30-year hindcasts for initial states of 1960 and 1980. There are two predictions for the initial state of 2005, one for 10 years and another for 30 years. There are a number of other optional experiments that some modeling groups may also perform.

As the modeling groups begin to make the outputs of these decadal hindcasts/predictions available for analysis in late 2010 and early 2011, there will be a bewildering array of results that will have to be synthesized in some way prior to their inclusion in the IPCC AR5. This is because IPCC lead authors are supposed to assess published literature, and not perform analyses.

Therefore, it is proposed to convene an AGCI session to make sense of the decadal hindcasts/predictions in terms of evaluation metrics, skill quantification, and summary figures that communicate the synthesis of the multi-model results. The product of the session will be a journal article that can then be assessed as part of the AR5. Participants will be asked to prepare some preliminary diagnostics in advance of the meeting. This will be the first time a synthesis will be attempted of a multi-model dataset of decadal hindcasts/predictions, and will be crucial in order to make a vital contribution to the IPCC AR5.