[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 20040218: contour plot aesthetics



Michael,

The SKIP parameter has 2 components, the first being the
skip factor for contours, the second being the skip factor
for vector/grid point i;j respectively- eg
SKIP = contour_skip/i_skip;j_skip

SKIP thins the resultant calculation on output, for the graphic
rendering, not the input data.

To subsample yor 40km grid to an 80km input data set for calculation, use
IJSKIP=1. IJSKIP is a different parameter than SKIP. For more information
on these parameters, see the help and release notes.

Also, you can apply a 5 or 9 point smoother on the display to remove the
higher frequency noise in such a field as: GFUNC = sm5s ( div.......
Note that your function showed a smoothing of the input to the qvector
calculation, but not on the output of divergence.

Steve Chiswell

On Thu, 19 Feb 2004, Michael Evans wrote:

> Steve,
>
> Thanks for the reply.  You've certainly given me some things to think about.
> I'm using a 40 km grid, and the reason that I am using skip is that a reviewer
> told me that Q vector convergence is best to calculate with a larger input
> grid spacing than 40 km.
>
> My understanding of the skip variable, is that the first parameter thins the
> input grid, so that setting skip=1 changes a 40 km grid to an 80 km grid,
> setting skip=2, changes a 40 km grid to a 120 km grid, ect.  In other words, I
> am removing data before I do any calculations. Meanwhile, setting the second
> parameter should thin the output grid, but not the input grid, which would
> give the results that you showed in your examples (for example, setting
> skip=0/2;2 would not thin the input grid, but would plot only every third
> point of the calculation resulting doing the calculation on the full data set
> (the 40 km grid))..  Is this right?
>
> I found that using skip=0 gave me lots of noise and bullseyes, while skip=4
> gives me a much easier to interpret pattern, with just a few main maxima and
> minima.  The downside seems to be that maybe I am setting skip too high, to
> the point where I am compromising the accuracy of my result. Maybe I need to
> try a smaller value for skip (1 or 2), and try setting contur=3?
>
> Steve Chiswell wrote:
>
> > Mike,
> >
> > You can subscribe to the gembud email list at:
> >
> > http://www/mailinglist/mailing-list-form.html
> >
> > Also, our general GEMPAK support address is
> > address@hidden.
> >
> > Since you did not mention what the grid spacing of the data set you
> > are looking at, my first question is why do you want to use SKIP
> > to thin the data set? The GEMPAK help states:
> >
> > SKIP is a variable which determines the contour points or plot
> >       points to skip.  Input is entered as:
> >
> >         skip_contour / skip_plot_x ; skip_plot_y
> >
> >       The defaults for skip_contour and skip_plot are 0.
> >
> >       Skip_contour thins the input grid before computing the
> >       contours to be drawn.efore contouring.
> >
> > When you change the GAREA, you will change the grid points
> > being displayed, and therefore which points are skipped.
> > If you are skipping a relative maxima or minima, then you will notice
> > a shift in location of the feature. If you reduce the data points being
> > used, and your data set is a highly varying field (eg, your divergence
> > field will sharply change from + to - values) then you will effect the
> > max/min locations. A smoothly varying field of 500mb hght would be less
> > sensitive and therefore SKIP can be more useful.
> >
> > If your grid is larger than the number of grid points you can view
> > within your area, then you would want to use IJSKIP to subset your grid
> > rather than SKIP. If you want to smooth the shape of the contours,
> > then look at the CONTUR parameter. Using SKIP will speed up the display
> > process since you are considering fewer grid points in the contour
> > algorithm, but you should consider how noisy the resultant data field
> > is, and do not under sample the data in highly varying quantities.
> >
> > For example, using the ETA 212 grid, today at F000, the quantity above
> > veries from -19**-13 to 34**-13 over the Sierras. Here is a GDLIST
> > output of a section of a row:
> > SCALE = 13
> > ROW 59
> > -1.80
> > -0.47
> > 0.89
> > -2.48
> > -0.68
> > -2.66
> > -3.40
> > 14.34
> > 19.46
> > -11.62
> > -13.53
> > 0.60
> > 0.40
> > -1.40
> > -0.50
> > -0.95
> > 1.87
> > -0.51
> > -2.99
> > 0.30
> >
> > If you skip every other point, EG skip=1, values remainin are:
> > -1.8, .89, -.68, -3.4, 19.46, -13.53, .4, -.5, 1.87, -2.99
> > you will still have both the greatest value (19.46) and the least
> > (-13.53) in your list. This is still useful.
> > If you set SKIP=2, values remaining are:
> > -1.8, -2.48, -3.4,  -11.62,  .4,  -.95,  -2.99
> > then you will miss both 14.34 and 19.46. In fact, you will lose
> > almost all positive values- and the minimum will shift to
> > the -11.62 location instead of the -13.53 location.
> >
> > The fact that the field varies from 19.46 to -11.62 in adjacent
> > grid points illustrates how sensitive the field is (some q vector
> > references can be found from Fred Sanders and/or Mel Shapiro).
> > Also beware of using auto SCALE=999 since in boring conditions,
> > you might have lots of contours, but the SCALE several orders
> > of magnitude greater (If you are generating a time series, you
> > most likely want to fix the SCALE appropiately). This also changes the
> > perception/look of the plot.
> >
> > Steve Chiswell
> >
> >
> >
> > On Wed, 2004-02-18 at 13:17, Michael Evans wrote:
> > > Steve and Robbert,
> > >
> > > I have a question about GEMPAK and was told by Dave Novak at NWS eastern
> > > region headquarters that you two are often helpful when it comes to
> > > answering questions about gempak on a gempak mailing list.
> > >
> > > My question: I am looking at divergence of Q vectors using gempak, using
> > > the following:
> > >
> > > GLEVEL=500:700
> > > GFUNC=div(sm5v(qvec(ldf(hght),vlav(geo)))).
> > >
> > > I have been varying the SKIP and GAREA parameters, and have been seeing
> > > some fairly large differences in my resulting Q vector convergence
> > > patterns.  I have mostly used SKIP=4, occasionally SKIP=3, and I have
> > > been specifying my GAREA by listing a station, for example: GAREA=avp,
> > > GAREA=avp-, GAREA=abe ect.  Obviously I would like to see not much
> > > change in my div Q result when I change these parameters, but I have
> > > been seeing some big variations.  For example, changing GAREA from avp
> > > to mdt (keeping everything else the same, including SKIP=4) has been
> > > enough to move my convergence maximum from southwest WV, to northern
> > > Virginia.
> > >
> > > So, I guess what my question boils down to is: is there a recommended
> > > size for my GAREA, and recommended value for SKIP to use when
> > > calculating Q vector divergence.  The calculation seems to be very
> > > sensitive to how those two parameters are set.
> > >
> > > Any help appreciated!
> > >
> > > - Mike
>