376
DAVID ANDERSON ET AL
occasions when this is not so. There are good examples of this in weather
forecasting though we are not aware of this in ocean analyses for climate
forecasting. Evaluation of observing systems through their impact on
forecasts is standard practice in meteorology and should generally be so in
oceanography, even if it has not been so in the past.
It is important to realise that results from OSEs are dependent on the
analysis system used and on the weight given to the data. The results will be
application-dependent. We are interested mainly in seasonal forecasts; this
emphasizes the tropics over middle latitudes. For other forecast horizons, or
other objectives, different areas may be important and different conclusions
might be drawn.
Various experiments to assess the impact of the TAO and XBT networks
are given in Vidard et al 2005. To assess the importance of the observing
systems on forecasts, 200 six-month forecasts are made, spanning the
period 1993-2002 using the ocean analyses as initial conditions. The
importance of the equatorial moorings is demonstrated. The exciting new
aspect of the observing system is the development of the ARGO network.
Can this system replace the XBT network? To consider this, the impact that
can be derived from ARGO is discussed. The ARGO experiments cover
only the two years 2002-3, but show considerable impact even when the
array is only partially developed and optimum techniques for using the data
have not been developed (for example, salinity data from the ARGO floats
are not yet assimilated).
In addition to OSEs, one can conduct Observing System Simulation
Experiments (OSSEs) to assess the potential impact of a proposed observing
system, or to assess the relative merits of a given array design. It is difficult
to gauge the error characteristics of such observing systems and the
experience from meteorology is that results are often too optimistic because
errors or difficulties in using the data are underestimated.
6. Summary and conclusions
In this paper we have considered data assimilation methods and issues
arising, as related to operational analysis of the atmosphere and ocean. At
ECMWF, the primary purpose is, and has been, to provide initial conditions
for various forecasts made. These range from forecasts out to 10 days at
resolution of ~20km to seasonal forecasts at atmospheric resolution of
~200km. Forecasts for the monthly and seasonal timescales are made with
coupled atmosphere-ocean models and so require initial conditions for the
coupled system. In real-time, these are made by taking the atmospheric
analysis performed for the highest resolution forecast and truncating to the
appropriate resolution needed for the forecast. In the case of the ocean, they
are obtained by running the ocean analysis system, assimilating all in situ
thermal data and relaxing strongly to observed sea surface temperature. The
ocean analysis systems for the monthly and seasonal forecast systems are
DAVID ANDERSON ET AL
occasions when this is not so. There are good examples of this in weather
forecasting though we are not aware of this in ocean analyses for climate
forecasting. Evaluation of observing systems through their impact on
forecasts is standard practice in meteorology and should generally be so in
oceanography, even if it has not been so in the past.
It is important to realise that results from OSEs are dependent on the
analysis system used and on the weight given to the data. The results will be
application-dependent. We are interested mainly in seasonal forecasts; this
emphasizes the tropics over middle latitudes. For other forecast horizons, or
other objectives, different areas may be important and different conclusions
might be drawn.
Various experiments to assess the impact of the TAO and XBT networks
are given in Vidard et al 2005. To assess the importance of the observing
systems on forecasts, 200 six-month forecasts are made, spanning the
period 1993-2002 using the ocean analyses as initial conditions. The
importance of the equatorial moorings is demonstrated. The exciting new
aspect of the observing system is the development of the ARGO network.
Can this system replace the XBT network? To consider this, the impact that
can be derived from ARGO is discussed. The ARGO experiments cover
only the two years 2002-3, but show considerable impact even when the
array is only partially developed and optimum techniques for using the data
have not been developed (for example, salinity data from the ARGO floats
are not yet assimilated).
In addition to OSEs, one can conduct Observing System Simulation
Experiments (OSSEs) to assess the potential impact of a proposed observing
system, or to assess the relative merits of a given array design. It is difficult
to gauge the error characteristics of such observing systems and the
experience from meteorology is that results are often too optimistic because
errors or difficulties in using the data are underestimated.
6. Summary and conclusions
In this paper we have considered data assimilation methods and issues
arising, as related to operational analysis of the atmosphere and ocean. At
ECMWF, the primary purpose is, and has been, to provide initial conditions
for various forecasts made. These range from forecasts out to 10 days at
resolution of ~20km to seasonal forecasts at atmospheric resolution of
~200km. Forecasts for the monthly and seasonal timescales are made with
coupled atmosphere-ocean models and so require initial conditions for the
coupled system. In real-time, these are made by taking the atmospheric
analysis performed for the highest resolution forecast and truncating to the
appropriate resolution needed for the forecast. In the case of the ocean, they
are obtained by running the ocean analysis system, assimilating all in situ
thermal data and relaxing strongly to observed sea surface temperature. The
ocean analysis systems for the monthly and seasonal forecast systems are