METEOROLOGICAL DATA PERSPECTIVE
357
the assimilation scheme, others manipulate ensembles of forecasts. This
concept is currently operational in the US and is called the Winter Storm
Reconnaissance Program. The National Centers for Environmental
Prediction uses the dispersion of the ensemble of forecasts run routinely and
a set of pre-defined flight plans to evaluate which of the flight scenario
would bring the maximum reduction in the dispersion of the forecasts. This
flight scenario leads to a designated aircraft flying in the area and dropping
dropsondes at regular intervals to provide additional observations.
Majumdar et al (2002) provide a detailed comparison of various targeting
techniques.
Apart from being used for selecting additional observations, targeting
observing systems can be extended to other applications such as controlling
the sampling rate of satellite sensors or the timing and location of mobile
upper-air soundings. Targeting can also be used to determine which
observations are to be discarded, i.e., to conduct effective thinning of the
observations. This capability will become increasingly important, given the
very large numbers of observations that will be available from nextgeneration satellites. Among the tools which can be useful for targeting,
being able to quantify the impact of any observation on the analysis and the
subsequent forecast is crucial. Such a tool has been developed in particular
using the adjoint of the various operations involved (analysis step and model
forecast) by Baker and Daley (2000) and Doerenbecher and Bergot
(2001).This sensitivity to observations is illustrated in Figure 8. The
“forward” step consists in the analysis represented by the Kalman gain
matrix K and the forecast model M. From the background x b and the
observations y, it creates the analysis x a and the subsequent forecast x f . The
“adjoint” step consists in the adjoint of the forecast model M
T followed by
the adjoint of the analysis process K
T
. It allows to compute the sensitivity of
any aspect of the forecast with respect to the observations and/or the
background.
This sort of tools initially developed mainly for targeting can also allow
to compute the sensitivity of the analysis or forecast to various satellite
channels in an adaptive manner (Fourrié and Rabier, 2004).
Beyond these targeting issues, NWP is also expected to progress (within
THORPEX and also independently) in flow-dependent specification of
various parameters used in assimilation. The major of these parameters is the
background error covariance matrix B, where a lot of work is ongoing to
incorporate more flow-dependence in the statistics through 4D-Var or
ensemble methods mainly. The link to the observations is then the
estimation of background errors in observation space (HBH T ) to perform
first-guess check (Andersson et al, 2000). There are also interesting
developments in the context of flow-dependent tolerances for outlier
observations for an adaptive buddy check (Dee et al, 2001).
sounder channels (eg Fourrié et al, 2002) and can also be used to select
357
the assimilation scheme, others manipulate ensembles of forecasts. This
concept is currently operational in the US and is called the Winter Storm
Reconnaissance Program. The National Centers for Environmental
Prediction uses the dispersion of the ensemble of forecasts run routinely and
a set of pre-defined flight plans to evaluate which of the flight scenario
would bring the maximum reduction in the dispersion of the forecasts. This
flight scenario leads to a designated aircraft flying in the area and dropping
dropsondes at regular intervals to provide additional observations.
Majumdar et al (2002) provide a detailed comparison of various targeting
techniques.
Apart from being used for selecting additional observations, targeting
observing systems can be extended to other applications such as controlling
the sampling rate of satellite sensors or the timing and location of mobile
upper-air soundings. Targeting can also be used to determine which
observations are to be discarded, i.e., to conduct effective thinning of the
observations. This capability will become increasingly important, given the
very large numbers of observations that will be available from nextgeneration satellites. Among the tools which can be useful for targeting,
being able to quantify the impact of any observation on the analysis and the
subsequent forecast is crucial. Such a tool has been developed in particular
using the adjoint of the various operations involved (analysis step and model
forecast) by Baker and Daley (2000) and Doerenbecher and Bergot
(2001).This sensitivity to observations is illustrated in Figure 8. The
“forward” step consists in the analysis represented by the Kalman gain
matrix K and the forecast model M. From the background x b and the
observations y, it creates the analysis x a and the subsequent forecast x f . The
“adjoint” step consists in the adjoint of the forecast model M
T followed by
the adjoint of the analysis process K
T
. It allows to compute the sensitivity of
any aspect of the forecast with respect to the observations and/or the
background.
This sort of tools initially developed mainly for targeting can also allow
to compute the sensitivity of the analysis or forecast to various satellite
channels in an adaptive manner (Fourrié and Rabier, 2004).
Beyond these targeting issues, NWP is also expected to progress (within
THORPEX and also independently) in flow-dependent specification of
various parameters used in assimilation. The major of these parameters is the
background error covariance matrix B, where a lot of work is ongoing to
incorporate more flow-dependence in the statistics through 4D-Var or
ensemble methods mainly. The link to the observations is then the
estimation of background errors in observation space (HBH T ) to perform
first-guess check (Andersson et al, 2000). There are also interesting
developments in the context of flow-dependent tolerances for outlier
observations for an adaptive buddy check (Dee et al, 2001).
sounder channels (eg Fourrié et al, 2002) and can also be used to select