Francisco J. Doblas-Reyes
ICREA and Institut Catala de Ciencies del Clima
Dr Trueta 203, 08005 Barcelona, Spain
Magdalena A. Balmaseda, Antje Weisheimer, Tim N. Palmer
ECMWF, Shinfield Park
RG2 9AX Reading, UK
Three ten-year ensemble decadal forecast experiments were performed with the ECMWF
coupled forecast system using an initialization strategy common in seasonal forecasting where
both the ocean and the atmosphere use analysis data. One initializes the ocean using data from
an ocean-only simulation forced with data from an atmospheric re-analysis. The other two
take the ocean initial conditions from an ocean-only run that assimilates subsurface
observations. This is the first time that such a set of experiments was carried out. The coupled
model drifts from the realistic initial conditions towards the model climate. The small drift
differences found between the experiments suggest that in the tropics they are due to
differences in latent heat and outgoing top of the atmosphere radiation in the first three
forecast years. In the extratropics, the initial-condition differences between the experiments
persist during the simulations, associated to a persistent difference in latent heat flux between
the experiments initialized with and without ocean data assimilation. In spite of the drift, the
decadal predictions show that the system is able to predict the interannual variability of the
observed global and regional mean air temperature anomalies up to several years in the future.
No significant forecast quality benefit of the assimilation of ocean observations is found over
the extratropics, although a negative impact of the assimilation of incorrect XBT profiles is
hinted at. The results illustrate the importance of reducing the important model drift,
increasing the sample size typically adopted in decadal forecasting and reducing the ocean
analysis uncertainty.