THE ECMWF PERSPECTIVE
377
essentially the same, though they do differ slightly as the analysis for the
monthly system starts from the last analysis for the seasonal system but then
is accelerated to real time using the surface forcing from the atmospheric
analyses and assimilating what ocean data are available.
Because model error is significant at longer forecast range, it is necessary
to take this into account when preparing forecast products. This is done by
calculating the model drift and climate over a calibration period which is
currently 15 years but will be longer in future. The forecast products are
anomalies relative to the model climate. Hindcasts over the period 1987 -
2004 are possible, largely because ocean forcing fields are available from the
reanalyses ERA-15, ERA-40. These atmospheric reanalyses are not without
their problems, as shown in the paper, but mark a significant improvement
over previously available products, such as the analyses carried out at the
time. It is hoped that atmospheric reanalyses will continue to be performed,
perhaps every 5 to 10 years, and covering perhaps the last 50 years or so.
These are major undertakings using advances in model development and
assimilation techniques and some recovery of old data. They should be
viewed as part of the effort to produce monthly, seasonal and even multiannual forecasts. Ocean reanalyses spanning the period of the atmospheric
reanalyses are also made on a routine basis. These are currently done every
year or so but as model resolution increases and/or assimilation techniques
become more sophisticated, they will be undertaken less frequently.
Currently, several ocean groups are active in this area, in contrast to
meteorology where only two or three are performed.
At present it is unclear how best to initialize the coupled model.
Initialising the atmosphere and ocean separately, as is done currently, may
not be the best. One would like to do a more coupled assimilation but this is
some way off at present. It is not straight forward because of the disparate
time scales of the atmosphere and ocean.
The ocean observing system has advanced rapidly over the last decade.
In the tropics it is clear that there are systematic differences between the
ocean model state and the observations. This could be because of deficient
forcing fields, ocean model physics or the way the surface fluxes are
transferred from atmosphere to ocean. ECMWF uses a wave model as part
of all its forecast systems, but this model is not fully tied into the ocean as
yet. Most other groups using coupled models do not even include a wave
model and do not pass the fluxes through the wave field. The assimilation
system itself may contribute to the bias increments noted here. Finally, we
are just beginning to evaluate the observing system and full use is not yet
made of all the data available. For example, salinity from the ARGO floats
is only just beginning to be used (Haines et al 2005). The combination of an
expanding ocean observing system, new strategies for assimilation, different
monthly forecast has to be available within one day of real-time, whereas
the seasonal forecasts start 11 days behind real-time. The analysis for the
377
essentially the same, though they do differ slightly as the analysis for the
monthly system starts from the last analysis for the seasonal system but then
is accelerated to real time using the surface forcing from the atmospheric
analyses and assimilating what ocean data are available.
Because model error is significant at longer forecast range, it is necessary
to take this into account when preparing forecast products. This is done by
calculating the model drift and climate over a calibration period which is
currently 15 years but will be longer in future. The forecast products are
anomalies relative to the model climate. Hindcasts over the period 1987 -
2004 are possible, largely because ocean forcing fields are available from the
reanalyses ERA-15, ERA-40. These atmospheric reanalyses are not without
their problems, as shown in the paper, but mark a significant improvement
over previously available products, such as the analyses carried out at the
time. It is hoped that atmospheric reanalyses will continue to be performed,
perhaps every 5 to 10 years, and covering perhaps the last 50 years or so.
These are major undertakings using advances in model development and
assimilation techniques and some recovery of old data. They should be
viewed as part of the effort to produce monthly, seasonal and even multiannual forecasts. Ocean reanalyses spanning the period of the atmospheric
reanalyses are also made on a routine basis. These are currently done every
year or so but as model resolution increases and/or assimilation techniques
become more sophisticated, they will be undertaken less frequently.
Currently, several ocean groups are active in this area, in contrast to
meteorology where only two or three are performed.
At present it is unclear how best to initialize the coupled model.
Initialising the atmosphere and ocean separately, as is done currently, may
not be the best. One would like to do a more coupled assimilation but this is
some way off at present. It is not straight forward because of the disparate
time scales of the atmosphere and ocean.
The ocean observing system has advanced rapidly over the last decade.
In the tropics it is clear that there are systematic differences between the
ocean model state and the observations. This could be because of deficient
forcing fields, ocean model physics or the way the surface fluxes are
transferred from atmosphere to ocean. ECMWF uses a wave model as part
of all its forecast systems, but this model is not fully tied into the ocean as
yet. Most other groups using coupled models do not even include a wave
model and do not pass the fluxes through the wave field. The assimilation
system itself may contribute to the bias increments noted here. Finally, we
are just beginning to evaluate the observing system and full use is not yet
made of all the data available. For example, salinity from the ARGO floats
is only just beginning to be used (Haines et al 2005). The combination of an
expanding ocean observing system, new strategies for assimilation, different
monthly forecast has to be available within one day of real-time, whereas
the seasonal forecasts start 11 days behind real-time. The analysis for the