Information-based potentialclimate predictability
Youmin Tang
University of Northern British Columbia, Canada
Potential Predictability Signal-to-Noise Ratio (SNR)
Rowell, D. P. (1998), Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations, J. Clim., 11, 109–120.
Peng, P., A. Kumar, W. Wang (2009), An analysis of seasonal predictability in coupled model forecasts, Clim. Dyn., 36, 637-648.
Mutual Info. MI
.
])()(
),(ln[),(
pvp
vpvpMI
Relative Entropy ( Gaussian)
.
Yang, D. Tang, Y and Zhang, Y and Yang X, 2011: JGR-Atmosphere.Yang, D. Tang, Y and Zhang, Y and Yang X, 2011: JGR-Atmosphere,
Multiple Model Ensemble
ENSEMBLES project stream-2 Hindcasts (1-tier forecast) http://ensembles.ecmwf.int/thredds/ensembles/stream2/seasonal/atmospheric/monthly.html
UKMO, ECMWF, MF, CMCC_INGV, IFM_GEOMAR
MI-based potential predictability and its difference from SNR-based measure
For 2-seasons prediction (the calendar season is the target time of prediction, such as MAM meaning the prediction starting from Nov.
The most predictable pattern of the NA TAS at the lead of one season for different seasons (prediction target)
Predictable Component Analysis ( PrCA)
PI Maximum ; SNR Maximum
The time series of PrCA mode 1 and mode 2
The SST patterns associated with the PrCA mode 1 and mode 2 of TAS.
Correlation RMSE
The predicted time series of the first PrCA mode against the observation counterpart.
Same as above but for the prediction of the first principal component.
Conclusions
Signal-to-noise ratio (SNR) is a special case of information-based predictability measure. When the ensemble spread changes with initial condition, SNR often underestimates the potential predictability.
The most predicable component of the Northern America climate (temperature) is the interannual mode and a long-term trend (the global warming). The most predicable interannual variability is highly related to the ENSO.