MRG Interactive Developments

Mesoscale Research Group, McGill/UAlbany

NARR Usage Notes

  • NARR Background
  • Converting NARR Grib Grids to GEMPAK
  • NARR Smoothing and Calculations

  • NARR Background

    The North American Regional Reanalysis (NARR) was generated at the National Centers for Environmental Prediction for the period 1979-2003. The NARR area covers all of North and Central America and much of the flanking ocean regions. The data produced by the NARR project is freely available for download. Interactive map plotting is also possible. An extensive amount of information is available from the NARR Homepage.

    Converting NARR Grib Grids to GEMPAK

    The MRG Interactive General GEMPAK Information and Products page contains information related to GEMPAK data conversions for the NARR dataset.

    Smoothing and Calculations

    The output from native E-grid of the NARR model (essentially NCEP's Eta model) is interpolated to a Northern Lambert Conformal Conic (NCEP Grid 221) projection before the data is archived. The reprojection from the native model grid to the 221 grid appears to generate a high level of noisiness in certain NARR archived fields. The u and v component wind fields are particularly problematic from a diagnostic perspective. An example of relative vorticity computed (using GEMPAK) from the raw u/v wind compontents on the NCEP 221 grid is available here (courtesy of Alan Srock) and clearly demonstrates that some level of data manipulation is required before diagnostics for the NARR fields can be accurately computed. The noisiness and spurious fine-scale stucuture of the vorticity plot makes use of the raw NARR data for diagnostic calculations inadvisable.

    Numerous smoothing and truncation methods were applied to the raw archived NARR grids in an effort to ascertain what kind of filtering was required for derived fields to be relatively noise-free. Spurious structures are particularly problematic in diagnostics requiring the use of horizontal derivatives or, worse yet, second derivatives of the raw fields. A series of qualitatives comparisions were made for smoothers including 5 point, 9 point and 25 point spatial filters; a Gaussian smoother; and a spectral trucation filter. The Gaussian smoother has been determined to provide the best results in a trade-off between smooth fields and unacceptable loss of valid small-scale structures (which are, after all, the whole point of using the NARR).

    The Gaussian filter (included with the standard GEMPAK distribution) is simply a moving-average filter with a normally-distributed set of filter weights applied over a window determined by the degree of the filtering. Essentially, the degree of filtering determines the standard deviation of the distribution used to calculate the filter weights; therefore larger degrees lead to more damping at all wavelengths (focused at shorter wavelengths) and a larger window area for the averaging. If a degree of 2 is used, then approx 1/e (33%) of the power of 2*dx waves will be retained (i.e. 67% eliminated). If a degree of 6 is used, then about 95% of 2*dx wave power is eliminated, along with 86% of 4*dx wave power and 67% of 6*dx wave power. With a grid spacing of about 32.5km (221 grid), this means that one would significantly (i.e. over 67%) decrease the amplitude of waves with wavelengths of 200km. Note that these damping values are for the ideal response function and that real damping values will in all cases be larger.

    The application of the Gaussian filter is simple, and should be undertaken on all raw fields before any calculations are performed. Although the noise problem is particularly noticable in the wind fields, spurious structures have also been found in the horizontal derivatives of the height, temperature and humidity fields. We therefore recommend that all fields be subjected to the Gaussian filter for consistency. The filter is simply applied using the GEMPAK gbdiag application. For each field (FIELD), set GFUNC=GWFS(FIELD,6) and run the gbdiag program. This applies the Gaussian filter with a weight of 6, implying that 95% of 2*dx noise is eliminated, as outlined in the previous paragraph. Although some useful high resolution information is also smoothed in this process, it is necessary for the production of realistic diagnostic fields. The result of a vorticity computation from filtered wind compontents (compare directly with the raw product shown above) is shown here.

    MRG Interactive is a supporter of open source initiatives and, Logo

    Questions or Comments? Contact our WebMaster.