Page History

Atmospheric Model Data Sources

The best analysis is that which allows the IFS models  to subsequently produce forecasts that verify nearest to the actual evolution.  The analysis is not necessarily true to the observations in every respect, though of course the analysis processes (4D-Var and LDAS) try to assimilate them to best effect.  For the purposes of the ensemble, the analysis process also tries to quantify the uncertainty of our estimate of the initial state.  Advanced analysis procedures have to be used to assimilate non-conventional observations. 

...

The observations used for the analysis of the atmosphere are available at both synoptic and asynoptic hours and can be divided roughly into direct observations and remote-sensing observations. 

Direct or Ground-based observations 

These consist of observations from:

...

• be biased or be reported at incorrect locations or times.  For most observations there are dynamic bias correction mechanisms (e.g. bias correction used on one day may differ from bias correction used the previous day). Incorrect locations are generally flagged out by the monitoring.  Any duplicate observations are filtered out.
• not be representative of the grid resolution or height - perhaps lying in one corner of a grid box or at a different altitude.  Some observations are excluded using certain land/sea mask criteria.  For the other observations, the observation errors include implicitly a representativeness error.
  
• be of non-uniform quality – some radiosondes have good quality, others less so; and absolute calibration can vary with age of the instrument.  For most data types we try to use adapted observation errors.  Radiosondes are also identified by instrumentation type, which facilitates adaptations to the expected data quality.
  

• have insufficient detail - in radiosonde messages, perhaps only significant levels are reported (as in the old style TAC message format) instead of full resolution data (as in the new style BUFR format being gradually introduced around the world).

   

Indirect or Satellite-based observations

These are achieved in two different ways:

• passive technologies sense natural radiation emitted by the earth and atmosphere or solar radiation reflected, refracted or retransmitted by the earth and atmosphere.
  
  • Contamination of atmospheric signals over land and over coastlines can be a problem, although increasingly new ways are being found to utilise such data. For example with cycle 45r1 introduced in June 2018 ECMWF began assimilating non-surface-sensitive infra-red channel data over land, and all sky micro-wave sounding data over coasts.
    
• active technologies emit radiation and sense how much is transmitted, reflected or scattered back.  For example:
  • the GPS radio occultation satellite-to-satellite signal is very sensitive to the temperature and humidity structure of the atmosphere particularly to the sharp moisture and temperature gradients beneath the boundary layer inversion,
  • Scatterometers derive surface-wind vectors from back-scattered radar signals from sea-surface ripples,
  • the Advanced Scatterometer (ASCAT) also enables soil moisture pseudo-observations that observe subsurface/subcanopy sub-surface and sub-canopy climate-related features such as water content of sub-canopy and continental surfaces.
    

...

Satellite data is vital for an effective analysis and the use satellite observations is increasing rapidly.

Fig2.4.-1: Pie chart showing the proportion of data types used by the IFC assimilation.  ATMS predominate. Ground-based observations constitute a relatively small proportion.

...

Users need to be aware of potential problems with the forecast due to deficiencies in coverage of data or conflict of observations with background fields.  Users should inspect:

• the data coverage (use obs drop down menu to select observation type) to show if and where there are deficiencies in data availability. 
• the analysis increments and analysis increment charts to show the location of significant differences between observed and background values.

...