| Home > Newsevents > Training > Rcourse_notes > DATA_ASSIMILATION > OBS_AND_DIAG_TOOLS > |
Observations and diagnostic
tools for data assimilation:
October 1998
| Table of contents | ||||||
|
Table of contents >>
Next Section >>
1 . Observation preprocessing
1.1 The incoming observations
The observations arrive at ECMWF through GTS (Global Telecommunications System) and are stored in a decoded format in the RDB (Report Data Base). Prior to the data assimilation the observations are extracted from the data base. These data have already undergone some rudimentary quality control, e.g. a check for the observation format and position, for the climatological and hydrostatic limits as well as for the internal and temporal consistency, respectively. Then an observation file suitable for assimilation is created in an observation preprocessing module. This entails format conversions, change of some observed variables, like calculation of relative humidity from dry and wet bulb temperatures, as well as assignment of observation error statistics. The resulting file contains all the observational information from the data window (currently six hours) and is an input for the IFS (Integrated Forecast System). The observation screening then selects the best quality and unique observations. In 3D-Var closeness to the middle of the data window is preferred as the background is not interpolated to the exact time of the observation whereas in 4D-Var the screening can be performed hourly. Unlike the OI, the 3D/4D-Var data assimilation is global and therefore no separate data selection for analysis boxes is needed (OI analysis involves a matrix inversion of the size of the number of observations and therefore the analysis equation is solved separately for smaller areas where the number of observations is sufficiently small).
1.2 Bias correction
The feedback files are extensively used for monitoring the performance of the observing and assimilation systems. One use is to determine the bias corrections for some observing systems, currently for temp temperature observations, TOVS radiances and scatterometer (scatt) winds.
Bias correction, in general, is a very difficult task as there is no fixed reference point with respect to which the bias should be corrected. If one removes, for instance, all the bias between the model background field and the TOVS radiances, there is a risk that part of the removed bias actually originates from the forecast model rather than from the observing system. In this case, the true effect of the bias removal is that the observations will actually enforce the model bias in the subsequent assimilations. Due to the risks involved, often a policy of "conservative bias correction" has been adopted, i.e. removing for instance only a half of the bias appearing in the observations.
The biases change in time due to changes in observing and assimilation systems and therefore the bias correction has to be updated from time to time. An update to the bias correction coefficients for TOVS radiances is performed once a month on the past 2 to 4 weeks of radiance background departure statistics. The bias correction is calculated with an off-line code using feedback files as input. The coefficients are substituted to input observations at the preprocessing stage.
Training Course Notes Front Page >>
Table of contents >>
Next Section >>
 |
07.06.2002 |
 |
© ECMWF |
