7B.6
Initialized decadal prediction experiments using CCSM4

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 26 January 2011: 9:45 AM
Initialized decadal prediction experiments using CCSM4
609 (Washington State Convention Center)
Steve Yeager, NCAR, Boulder, CO; and G. Danabasoglu, J. Tribbia, J. Anderson, N. Collins, T. Hoar, J. Hurrell, K. Raeder, and H. Teng

Decadal-scale climate change is likely to be a major topic of the next Intergovernmental Panel on Climate Change assessment report (AR5), and yet generating skillful predictions of basin scale or regional climate change on decadal time scales remains a formidable challenge. At these time scales, the influence of particularly the ocean initial conditions on climate evolution exceeds that of the external forcing by greenhouse gases. In this study, we present results from initialized decadal prediction simulations using the Community Climate System Model version 4 (CCSM4). A suite of prediction ensembles starting in 1970, 1980, 1990, and 2000 are integrated to year 2005. These are initialized from forced ocean hindcast simulations which sample different choices of ocean salinity restoring timescale, an ad-hoc parameter setting which influences the strength of the Atlantic Meridional Overturning Circulation in the initial conditions. Another set of prediction ensembles are started at annual intervals from 2000 to 2005 and integrated to 2030. This set uses initial conditions from an ensemble of ocean reanalysis data obtained using data assimilation via the Data Assimilation Research Testbed (DART) in addition to the ones from the above ocean hindcast simulations.

The full-field initialization methodology is assessed in prediction simulations which overlap with the historical record, and the verisimilitude of the initialized predictions with that from uninitialized 20th century integrations is contrasted. In all prediction experiments, model drift in the Atlantic is the dominant decadal signal. Improved initial conditions do not mitigate the drift problem, although they do significantly reduce model bias relative to observations early on. We also explore analysis techniques for gleaning predictive skill from short-term projection experiments characterized by large decadal climate drift.