Re: [ferret_users] Re: handling data

[Ryo Furue <furue@xxxxxxxxxx>]

Ansley,

> As always we encourage contributions of Ferret scripts, Python tools
> that might be incorporated using PyFerret, and also general ideas
> for handling this sort of data. So if you're thinking about
> non-gridded data, please lend your ideas.

I don't have good ideas but I'd like to share my experience
in plotting station data using Ferret as a novice of handling
station data.

I had to reproduce a depth-distance contour plot of oxygen
along a ship track from the same original station dataset
as the published figure I wanted to reproduce used.

You can skip the DETAILS below.

>>> BEGIN DETAILS >>>
Using a programming language, I pre-processed the
data to produce arrays of oxgen F(d,n) and depth dep(d,n),
where d is horizontal distance and n is the bottle number.
Reading the arrays into Ferret, I mapped F onto a depth
axis using  ZAXREPLACE(F, dep, z[gz=my_z_axis]), which
you can now plot using the CONTOUR or FILL command.
<<< END DETAILS <<<

Comparing with the original figure, I noticed differences
in "policies" of interpolation and extrapolation between
Ferret and whatever the plotting program the original author
used.

1) Vertical extrapolation.
Since shallowest bottle samples are typically from 2--10 m
depths, data values were typically missing at the
surface (z=0) grid point in my Ferret plot.  In contrast,
the original plot had NO missing data.  That must have meant
vertical extrapolation toward the surface for the original
plot.

2) Horizontal interpolation.
At one station bottle samples were missing for
the upper some 200 meters.  My Ferret plot of course
treated the portion as missing, but the original plot
filled the gap.  That must have meant a liberal
horizontal interpolation.

3) Horizontal extrapolation.
The last station is at a shallow sea near the coast
and so the last bottle sample was near the bottom ~ 500m.
Below 500m, my Ferret plot was white.  But, the original
plot extrapolated oxygen values to under the ground!

4) Smoothness of contour lines.
Ferret contour lines are very faithful to the original
gridded data and tend to be ziggy where the original
data is noisy.  But, the original plot had a very smooth,
liberal contour lines.  That suggests that the plotting
tool the original author used uses a smooth vertical
interpolation (like spline).

All in all, I got the impression that the plotting tool
the original author used has a very liberal policy for
interpolation and extrapolation.  This difference may
be from the "culture" of hydrography.  You know, in old
days, observationalists drew contours by hand for very
sparse dataset, when they had to interpolate and
extrapolate "subjectively".

The last point (4) is interesting to me. Since observed
data always have error, it makes some sense to move the
contour lines around a bit within error, assuming some
smoothness in the field we are plotting.  Of course,
there is always the danger of producing spurious values
by doing this.  (So, I can imagine you can devise an
"optimal contour line"-generating algorithm that weighs
both observational error and smoothness requirement.)

Anyway, this is what I learned as a person who is totally
unfamiliar with handling station data.

Ryo