Re: Preventing extrapolation

Chad Herman wrote:
> Hi everyone,
>
> I've spent all of yesterday investigating this issue and haven't found
> any satisfactory solution. I have a file containing land surface
> temperature anomalies. The data is represented on an equal-area grid
> with 8,000 grid boxes, so I have three arrays: one for longitude,
> latitude and anomaly.
>
> My task is to interpolate it into 5x5 resolution. I can't use any
> interpolation functions that require the data present on a regular 2D
> grid. I've used CSSGRID and NATGRID and they interpolate fine, but they
> interpolate too much. It's completely infilling the oceans as well as
> large land surfaces which must be missing. I've looked into the two
> aforementioned interpolation routines and can't seem to find a way to
> completely stop them from interpolating into data absent regions.
>
> For NATGRID, I used nnsetp("ext", 0) and nnsetp("nul",
> anomaly@_FillValue) which at least stopped excessive interpolation in
> the Arctic. I tried to use bin_avg to find grid boxes where there is no
> data whatsoever and use the results to mask out the interpolated grid.
> While this solves the ocean data problem, it creates a new problem by
> masking out too many grid boxes on land. If there are a couple of grid
> boxes with non-missing data surrounding a grid box or two with no data,
> it seems quite reasonable to interpolate into the missing region. But if
> there is no data there, the interpolated values get masked.
>
> Any suggestions?
>
> Thanks,
> Chad
Chad

Suggestion: You could build your mask from a separate topography
dataset, say, ETOPO or Smith and Sandwell, at the same resolution/grid
of your interpolated data.

http://iridl.ldeo.columbia.edu/SOURCES/.Sandwell/.seafloor/.height/
http://iridl.ldeo.columbia.edu/SOURCES/.Sandwell/.seafloor/.height/figviewer.html?map.url=X+Y+fig-+colors+-fig
http://www.ngdc.noaa.gov/mgg/global/global.html

Would this help?

Gus Correa
_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Mon Jan 24 16:06:47 2011