Project

General

Profile

Gridbox (sellonlatbox) Selection and Resampling Issue

Added by Nathanael Harwood about 7 years ago

Hi - I have used two ways to (gridbox) subset the same data, and retrieved two different results. I cut down ECMWF Era-Interim data to my chosen coordinates on ECMWF's website, and again on CDO using sellonlatbox chained with fldmean to get a spatial average over the gridbox. Same data, same coordinates, mean variance of .02 when I put them into R dataframes and use the all.equal () command.

You can even see the difference by eye on Panoply (below).

Now, I get that because my data is 0.75 x 0.75 resolution, choosing coordinates that are not an integer multiple of that is a problem (70-80N, 30-105E). ECMWF suggested that the difference probably lay in the CDO code for sellonlatbox, and that some resampling is occurring.

Does anyone know how to solve this? I can cheat for now (degrade to 1x1 res on ECMWF website), but I want to be able to chain these operators (-sellonlatbox -fldmean) in the future for spatial averages, and I'll use this a lot.

Thanks for reading,

NH

comparison 1 - ECMWF.PNG (227 KB) comparison 1 - ECMWF.PNG ECMWF Gridbox (website)
comparison 2 - CDO.PNG (218 KB) comparison 2 - CDO.PNG CDO Gridbox (sellonlatbox)

Replies (2)

RE: Gridbox (sellonlatbox) Selection and Resampling Issue - Added by Ralf Mueller about 7 years ago

hi!

for real help I need input file and the commands. all I see, is that the image created from the website stuff has a little bit large area.

sellonlatbox does not to any change to the data. My recommondation is: create a grid for you region of interrest and use remapnn. or use the bounds of your ECMWF output grid as values for sollonlatbox.

hth
ralf

RE: Gridbox (sellonlatbox) Selection and Resampling Issue - Added by Michael Böttinger about 7 years ago

Hi Nathanael,

First, sellonlatbox is supposed to only consider only those grid cells with the grid
center inside the lon/lat box. Can this be the reason for the different area?

Second, I suspect that R doesn't know about area weights when the field mean is computed.
cdo fldmean accounts for the different sizes of grid cells depending on the latitude
when computing the field mean.

Cheers,
Michael

    (1-2/2)