THE ECMWF PERSPECTIVE
363
in performing the reanalyses means that it is not possible to use the latest
scheme, nor the highest resolution. So for ERA-40, the resolution was
T159.
3dvar does not try to fit the model trajectory throughout the data time
window. It shares with Optimal Interpolation (OI) the approach of
performing the analysis at discrete times. Account is taken of the time
however, in the sense that the observations are compared with the model
first guess at the time of the observation, rather than bunching all the
observations in a given data window to the central time. This procedure is
called FGAT (First Guess at Appropriate Time). Formally, if all assumptions
are the same, OI and 3d-var are equivalent. In practice this is never so.
The analyses from ERA-40 have been used as initial conditions for
medium-range weather hindcasts (hindcasts are made as if they were
forecasts but are made for past events). Results from these hindcasts have
been discussed by F. Rabier (this volume). The improvement of the
forecasts as the observing system changes can be clearly seen. The
improvements in forecast skill since the end of ERA-40 are also shown.
They can result partly from improved data assimilation procedures (4d-var cf
3d-var) and partly from improvements in the quantity and quality of data.
An interesting finding is that the skill of forecasts for the southern
hemisphere now rivals or exceeds that for the northern hemisphere, whereas
in the first decades of ERA-40 reanalyses, the southern hemisphere skill was
relatively poor compared to that of the northern hemisphere. Since the in
situ measurements for the southern hemisphere are sparse, the implication is
that the improvement results from increased satellite coverage. Fig 1, from
Kelly et al 2004, shows results of various Observing System Experiments
(OSE’s) which clearly indicate the importance of satellite data for the
southern hemisphere: when satellite data are withheld, the forecast skill is
much reduced. By contrast, there is a much smaller degradation in the
northern hemisphere. Other components of the observing system have a
smaller impact.
Although the use of reanalyses to provide initial conditions for hindcasts
is informative, it is not the sole or even primary purpose of performing a
reanalysis. The analyses are used for diagnostic work to understand physical
processes in the atmosphere. (The physical assumptions/simplifications that
went into the analysis are important here). The application of most use to
climate modelers is that they provide an estimate of the surface wind, heat
and fresh water fluxes. As we shall later see, these are crucial in developing
a monthly or seasonal forecast system, as they provide a means of dealing
with model error. For monthly forecasting, the atmospheric initial
conditions from the reanalysis are more important than the fluxes used to
generate the ocean analysis. As the forecast range increases this balance
shifts and for seasonal forecasts the fluxes are more important than the
atmospheric initial conditions, as they play a major role in determining the
ocean initial conditions.
363
in performing the reanalyses means that it is not possible to use the latest
scheme, nor the highest resolution. So for ERA-40, the resolution was
T159.
3dvar does not try to fit the model trajectory throughout the data time
window. It shares with Optimal Interpolation (OI) the approach of
performing the analysis at discrete times. Account is taken of the time
however, in the sense that the observations are compared with the model
first guess at the time of the observation, rather than bunching all the
observations in a given data window to the central time. This procedure is
called FGAT (First Guess at Appropriate Time). Formally, if all assumptions
are the same, OI and 3d-var are equivalent. In practice this is never so.
The analyses from ERA-40 have been used as initial conditions for
medium-range weather hindcasts (hindcasts are made as if they were
forecasts but are made for past events). Results from these hindcasts have
been discussed by F. Rabier (this volume). The improvement of the
forecasts as the observing system changes can be clearly seen. The
improvements in forecast skill since the end of ERA-40 are also shown.
They can result partly from improved data assimilation procedures (4d-var cf
3d-var) and partly from improvements in the quantity and quality of data.
An interesting finding is that the skill of forecasts for the southern
hemisphere now rivals or exceeds that for the northern hemisphere, whereas
in the first decades of ERA-40 reanalyses, the southern hemisphere skill was
relatively poor compared to that of the northern hemisphere. Since the in
situ measurements for the southern hemisphere are sparse, the implication is
that the improvement results from increased satellite coverage. Fig 1, from
Kelly et al 2004, shows results of various Observing System Experiments
(OSE’s) which clearly indicate the importance of satellite data for the
southern hemisphere: when satellite data are withheld, the forecast skill is
much reduced. By contrast, there is a much smaller degradation in the
northern hemisphere. Other components of the observing system have a
smaller impact.
Although the use of reanalyses to provide initial conditions for hindcasts
is informative, it is not the sole or even primary purpose of performing a
reanalysis. The analyses are used for diagnostic work to understand physical
processes in the atmosphere. (The physical assumptions/simplifications that
went into the analysis are important here). The application of most use to
climate modelers is that they provide an estimate of the surface wind, heat
and fresh water fluxes. As we shall later see, these are crucial in developing
a monthly or seasonal forecast system, as they provide a means of dealing
with model error. For monthly forecasting, the atmospheric initial
conditions from the reanalysis are more important than the fluxes used to
generate the ocean analysis. As the forecast range increases this balance
shifts and for seasonal forecasts the fluxes are more important than the
atmospheric initial conditions, as they play a major role in determining the
ocean initial conditions.