GIST National Weather Service High-Resolution Upper-Air Dataset 
     
     1.0  General Description
     
          This is one of the upper air sounding datasets developed for the
     GEWEX Continental-scale International Project (GCIP) Integrated Systems
     Test (GIST) conducted from 01 April to 31 August 1994.  The GIST
     domain extends from approximately 91W to 107W longitude and 31N to
     40N latitude.  Included in this dataset are 13 National Weather Service
     (NWS) sounding stations in the GIST domain.  The soundings were
     typically released at 00 and 12 UTC, however,  several stations had
     additional launches on an as requested basis, particularly Norman OK
     (OUN).  The final dataset consists of 6-sec vertical resolution files.
          
     2.0  Detailed Data Description
          
     2.0.1     National Weather Service High-Resolution Sounding Algorithms
     
          The detailed description of NWS sounding collection and
     instrumentation is located in section 2.1 below
     
     2.1  Data Remarks
     
               The NWS soundings during GIST were comprised of two types: 
     Space Data radiosondes (Amarillo, El Paso, Midland, and Stephenville TX
     and Albuquerque NM) and VIZ radiosondes (Denver CO, Dodge City KS,
     Little Rock AR, Longview TX, Monett MO, Norman OK, Topeka KS,
     Albuquerque NM, and Stephenville TX).  Albuquerque NM and
     Stephenville TX both used Space Data radiosondes from 01 April to 31
     May 1994.  From 01 June to 31 August 1994 they used the VIZ
     radiosondes.   
     
               OFPS applied a correction to the Space Data radiosonde relative
     humidity calculations.  The NWS Micro-ART sounding system, in the
     resistance ratio and RH calculations, used, instead of the observed
     temperature, the observed temperature divided by 100.  The OFPS. applied
     correction was to rederive the resistance ratio, using the observed
     temperature divided by 100 and the observed RH, by iteration using
     the so termed "1A" and "1B" coefficients.  Now with the new calculated
     resistance ratio, the observed temperature and using the "1A" coefficients
     only, the new RH was calculated.
     
               OFPS also applied a correction to the VIZ radiosonde RH
     calculations.   Using the observed temperature and RH, the resistance ratio
     was rederived by iteration using both the "1A" and "1B" coefficients. 
     
          With the calculated resistance ratio, the observed temperature
     and using the "1A" coefficients only, the new RH was calculated.
     
               In both of the radiosonde types, the use of the raw 6-sec
     resolution elevation and azimuth angle data to derive the winds sometimes
     led to large oscillations in wind speed, due to the presence of
     oscillations in the elevation angle data, particularly at low elevation
     angles. The general approach to correct this problem was to remove the
     outlier radiosonde position data before computing the wind components.
     For both the azimuth and elevation angles from 360 sec to the end of the
     sounding, a ninth order polynomial was fit to the curve.  The residuals
     were calculated and compared to the observed values.  The outliers of the
     residuals were then removed.  

               Then to help correct the more extensive problems at low elevation
     angles within 10 degrees of the limiting angles (LA) some additional
     smoothing was applied.  If the elevation angle was between (LA + 7.5)
     and (LA + 10), the new elevation angle was computed with a 2 min linear
     fit.  If the elevation angle was between (LA + 5) and (LA + 7.5), the new
     elevation angle was computed with a 3 min linear fit.  If the elevation
     angle was less than (LA + 5), the new elevation angle was computed with
     a 4 min linear fit.  If the number of observations with low elevation 
     angles was greater than 20% of the total number of observations for the
     sounding no frequency smoothing occurred.
     
               Then, for the elevation angle only, a finite Fourier analysis
     was performed on the residuals.  Periods from 90-190 sec were removed and
     those below 30 sec were flattened.
     
               Finally, a 2 min second order polynomial was then fit to the
     position to derive the u and v wind components, except for the beginning
     and end minute (or 1.5 minutes if over 50 mb) which used a 3 min fit.  If
     there were less than 15% of the total number of points, not counting the
     beginning or end of the flight, on one side of the point for which the wind
     value was being computed, a linear fit was used.
          
     3.0  Quality Control Processing
          
               This dataset underwent a two-stage QC process.  First, the dataset
     underwent internal consistency checks.  This included two types of checks,
     "reasonable" limit checks on all parameters and rate-of-change checks on 
     temperature, pressure and ascension rate.  Second, each sounding was 
     visually examined to verify those parameters that are too variable for
     automatic checks (wind speed, wind direction and moisture).  This stage
     of the QC process also allows for a verification of the QC flags generated
     by the automatic checks.
          
     4.0  Reference
          
     Micro-ART Observation and Rework Programs Technical
          Document,  National Weather Service, National Oceanic and
          Atmospheric Administration, Washington, D.C., March 1991.
          
     Williams, S.F., C.G. Wade, and C. Morel, 1993:  A comparison of
          high resolution radiosonde winds:  6-second Micro-ART winds
          versus 10-second CLASS LORAN winds.  Preprints, Eighth
          Symposium on Meteorological Observations and Instrumentation,
          Anaheim, California, Amer. Meteor. Soc., 60-65.