CERFACS CERFACS Scientific Activity Report Jan. 2010 – Dec. 2011

CLIMATE VARIABILITY AND PREDICTABILITY
Additional experiments have also been carried out and published. Additional members have been added
leading to a total of 10 for each forecast date. 2 additional dates per decade have been also integrated,
namely 1959, 1964, 1969, ..., 2004, as a supplement to the mandatory 1960, 1965, 1970, ..., 2005 suite.
Finally, a fake 2010-pinatubo eruption has been introduced in an additional set of forecasts for the 2005-
2035 timeslot. The decadal forecast activity at Cerfacs is also part of the ongoing EC-FP7 COMBINE
project coordinated by MPI, the GICC EPIDOM project started at the beginning of 2011 and coordinated
by Cerfacs and the ANR CECILE coordinated by BRGM. The two main preliminary results are listed
below.
2.2.1 Ocean initialization
In order to minimize the initial shocks and drifts during the forecasts, initial conditions are obtained from
a coupled simulation of CNRM-CM5 integrated from 1958 to present where the sole ocean is nudged in
temperature and salinity towards the NEMOVAR-COMBINE (Balmaseda et al. 2010) ocean reanalysis.
Within this simulation, the other components (atmosphere, sea-ice, continents etc) are thus freely coupled
and do not contain any observational data. The NEMOVAR-COMBINE reanalysis has been produced using
a multivariate 3D-var data assimilation method (Weaver et al. 2005, Daget et al. 2009) applied to the NEMO
model running on the same ORCA1 configuration as CNRM-CM5, and with the physical parameterizations
as close as possible. The surface temperature and salinity are restored towards NEMOVAR using a flux
derivative term while the subsurface ocean nudging is based on a 3D Newtonian damping in temperature
and salinity (see Madec et al. 2008 for details) which is function of depth and space. No 3D damping is
applied between the surface and the thermocline thus leaving the mixed layer free to evolve ; the damping
is equal to 10 days below the thermocline and to 360 days for the deep ocean (below 800m depth). There is
no nudging near the coasts to avoid spurious effects on boundary currents, neither within the tropical band.
The latter choice tends to account for the fact that tropical subsurface ocean and tropical surface wind should
be consistent to minimize the excitation of tropical Kelvin and Rossby waves when one lets the model go
at the beginning of the forecast. We performed two sets of experiments where nudging is excluded either
within 1 o N-1 o S (hereafter GLOB) or within 15 o N-15 o S (hereafter EXTROP). We show that GLOB has
a much stronger initial shock than EXTROP ; it is characterized by a systematic excitation of an El Niño
events in the Pacific leading, by teleconnection, to enhanced drift in the entire Northern Hemisphere. We
found that such a disruption affects the mean drift of the model up to about 4 year leadtime stressing out
the importance of the initialization step (Sanchez et al 2011).
2.2.2 Skill score
Decadal predictability arises from two sources : anomalous states assimilated in the initial conditions of
the climate system (here the NEMOVAR ocean, -first kind predictability) and so-called external forcings
linked to natural (volcanoes and solar activities) events or anthropogenic influences (e.g. GHGs increase,
aerosols emission, land use etc. -second kind of predictability). Skill scores have been computed for global
temperature as a function of leadtime. Data have been debiased following the CMIP5 linear protocol. Very
high positive scores (from 0.93 to 0.98) are found whatever the leadtime. At global scale, the predictive skill
is mostly explained by the warming trend and the interannual cooling events associated with the external
forcings, respectively the increase of GHG concentrations and the volcanic aerosols that are prescribed in
the model (Figure 2.1).
118 Jan. 2010 – Dec. 2011