6.2 Observations used for model initialization

period is used to judge observation quality. For example, Hollingsworth et al. (1986) describe how the operational ECMWF data-assimilation system can be used to monitor observation quality. This automated and economical approach to the QC process allows suspect instruments to be identified and corrective action taken, without routinely visiting and inspecting every instrument.

6.2.4 Other observation processing

Whether winds are observed and reported in terms of the individual components or as speed and direction, the measurements may need to be converted to the model wind components. This is because the model u that is defined to be parallel to the grid-point rows, and the model v that is parallel to the grid-point columns, generally differ from the geocentric u and v that are defined relative to latitude and longitude lines. For every vertical column of grid points (the same i, j coordinate), the mathematical transformation will be slightly different. This necessity may be most easy to accidentally overlook when the model coordinates are Cartesian, and the grid-point rows and columns are approximately oriented east–west and north–south.
            Software that interpolates (analyzes) observations to a model grid operates in the framework
of the model’s horizontal coordinate system. Thus, because observation locations are typically defined in terms of latitude and longitude coordinates, there needs to be a transformation to the horizontal coordinates of the model, if it is x–y and not latitude–longitude based.
            Lastly, the units of the observations may need to be transformed to those employed by the model. For example, wind speeds are often reported in knots, but models generally use the meter–kilogram–second (mks) system. And it is common for humidity observations to require conversion as well.

6.2.5 Metadata

Metadata (also called meta-knowledge) accompany the observations themselves, and provide information necessary for their use. Essential types of metadata include the file structure, data format (e.g., NetCDF), the variable (e.g., wind speed), the units (e.g., mks), and the time and three-dimensional-spatial coordinates of the observation. Optional, but useful, information includes the instrument type, the date of the most-recent calibration, and a photo of the instrument site and surroundings. The concept of metadata also applies to model-generated data as well, although the relevant information will obviously be different.
          Conventions have been established for the format of metadata. For example, the NetCDF (Network Common Data Format) Climate and Forecast (CF) Metadata Convention is a welldocumented standard for observational and forecast metadata, which is designed to promote the processing and sharing of files created with the NetCDF Application Programmer Interface [NetCDF API]. The CF conventions generalize and extend the convention of the Cooperative Ocean/Atmosphere Research Data Service, a NOAA/university cooperative group





Model initialization
whose goal is the sharing and distribution of global atmospheric and oceanographic research
data sets.


6.2.6 Targeted or adaptive observations

Economic and other constraints limit the number of observations that are made of the atmosphere, and thus it is reasonable to want to obtain observations from locations where they will have the largest positive impact on model-forecast accuracy, for a particular prevailing weather situation. Methods have been developed to satisfy this need, where the measurements are referred to as adaptive or targeted observations. However, it is clearly not economically feasible to deploy mobile observation platforms on a day-to-day basis. But, there are high-impact weather events, such as hurricanes or severe extratropical cyclones, for which special aircraft observations are made. If the aircraft can be routed so as to provide observations from locations for which the forecast skill is very sensitive to the accuracy of the initial conditions, the procedure can save lives. The routine use of targeted aircraft observations may become more common with the continued development of
unmanned aerial vehicles.

      Various strategies for observation targeting have been evaluated as part of the following
field programs.

• Fronts and Atlantic Storm Tracks EXperiment (FASTEX; Emanuel and Langland 1998;
   Bergot 1999, 2001; Bishop and Toth 1999; Joly et al. 1999; and Bergot and Doerenbecher
   2002)
• NORth Pacific EXperiment (NORPEX, Langland et al. 1999, Majumdar et al. 2002a)
• Atlantic THORPEX (The Hemispheric Observing-system Research and Predictability
   EXperiment) Observing System Test (Langland 2005)
• Annual US NWS Winter Storm Reconnaissance (WSR) programs (Szunyogh et al.
   2000, 2002; Majumdar et al. 2002b)

       The following notational framework for viewing the adaptive-observation problem is provided by Berliner et al. (1999), Majumdar et al. (2006), and others. Let Xi , Xa, and Xv represent n dimensional vectors that define the state of the atmosphere at times ti , ta, and tv , respectively, in terms of the grid-point values of variables or spectral coefficients. The initial time, ti, is when the decision must be made, based on Xi information, about the types and locations of special observations to be collected at time ta (the targeted observation time, and the analysis (initial) time of the operational forecast), where the objective is to optimize the statistical properties of a forecast Xv at the verification time tv . Within the interval ta − ti, the observing platforms need to travel to the target locations so that observations can be made at ta for use in initializing the forecast. The time
interval ta − ti is chosen based on logistical considerations associated with planning the surveillance mission, launching the aircraft, and getting the aircraft to the necessary locations to make the observations. The data set Xa is the result of assimilating standard observations and the special targeted observations, and is used as the initial conditions for the forecast.