| Home > Research > Ifsdocs > ASSIMILATION > |
DATA ASSIMILATION
IFS documentation Front Page
Table of contents
CHAPTER 1 Incremental
formulation of 3D/4D variational assimilation-an overview
CHAPTER 2 3D variational assimilation
CHAPTER 3 4D variational assimilation
CHAPTER 4 Background term
CHAPTER 5 Conventional observational
constraints
CHAPTER 6 Satellite observational
constraints
CHAPTER 7 Background, analysis
and forecast errors
CHAPTER 8 Gravity-wave control
CHAPTER 9 Data partitioning (OBSORT)
CHAPTER 10 Observation screening
CHAPTER 11 Analysis of snow
CHAPTER 12 Land surface analysis
CHAPTER 13 SST and sea-ice analysis
CHAPTER 14 Reduced-rank Kalman filter
Previous Section
7.7 Calculation of forecast error variances
The analysis errors are inflated according to the error growth model of Savijärvi (1995) to provide estimates of short term forecast error. This is done by a call to ESTSIG. There is also an option to advect the background errors for vorticity as if they were a passive tracer. The advection is performed by ADVSIGA.
The error growth model is
|
(7.15) |
Here,
represents growth due to model errors, 
represents the
exponential growth rate of small errors, and 
represents
the standard deviation of saturated forecast errors.The saturation standard deviations are calculated as
times the standard
deviation of each field. The standard deviations have been calculated for
each month from the re-analysis dataset. ESTSIG reads these climatological
error fields from file `stdev_of_climate' by calling READGRIB, and interpolates
them in the horizontal and vertical using SUHIFCE and SUVIFCE. The climatological
errors may also be artificially increased in the tropics under the control
of LFACHRO. This is the default, and is recommended in preference
to using LFACHR, since it means that the forecast errors that are
archived, and are used to screen observations, are closer to those used
to formulate the background cost function. If climate standard deviations
are not available for any field, they are estimated as 10 times the global
mean background error for the field.The growth due to model error is set to 0.1 times the global mean background error per day. The exponential growth rate,
, is set to 0.4
per day.The error growth model is integrated for a period of NFGFCLEN hours. The integration is done analytically using the expression given by Savijärvi (1995). Two precautions are taken in integrating the error growth model. First, negative analysis error variances are set to zero. Second, the growth rate due to model error is limited to a sensible value with respect to the saturation errors. This was found to be necessary to prevent numerical problems when calculating specific humidity errors for the upper levels of the model.
ESTSIG overwrites the contents of ANEBUF with the estimated variances of forecast error. The variances are converted to standard deviations and written out by WRITESD.
Next Section
Previous Section
 |
20.03.2002 |
 |
© ECMWF |
