improved edge behavior of sparse data

[karlmsmith]

Reported by steven.c.hankin on 4 Jun 2010 21:21 UTC The attached figure is a particularly bad example of a common problem -- where fill-over-shade produces some undesirable effects at the plot edges. But this one looks really "wrong". Why are the square corners of the red SHADED squares at 37.50 and 37.75N showing through at all? These cells are separated by a valid data point from the right hand edge of the available data? Shouldn't the FILL completely hide the SHADE on those cells? Is it possible that the same color levels are not being used on both the SHADE and the FILL?

An related by aside suggestion: Datasets that have sparse data should routinely be configured so that the default graphics style is a simple SHADE plot

Migrated-From: http://dunkel.pmel.noaa.gov/trac/las/ticket/798

[karlmsmith]

Attachment from steven.c.hankin on 4 Jun 2010 21:21 UTC

[karlmsmith]

Attachment from @AnsleyManke on 8 Jun 2010 22:55 UTC

[karlmsmith]

Attachment from @AnsleyManke on 8 Jun 2010 22:56 UTC

[karlmsmith]

Attachment from @AnsleyManke on 8 Jun 2010 23:02 UTC

[karlmsmith]

Comment by @AnsleyManke on 8 Jun 2010 23:01 UTC The region chosen is not sparse, but is at the edge of good data, and has a strong gradient which is causing the yellow line on the fill plot. I've attached separate SHADE and FILL plots.

The /LEVELS=v qualifier for variance-based color levels is part of the problem. With linear levels it looks like the last attachement

yes? use "http://edac-dap.northerngulfinstitute.org/thredds/dodsC/ncom/region1/ncom_glb_regp01_2010060100.nc"
yes? shade/x=-76.33:-75.63/y=37.44:37.86/k=1/l=1/lev=v water_temp
yes? fill/lev/x=-76.33:-75.63/y=37.44:37.86/k=1/l=1 water_temp

yes? shade/x=-76.33:-75.63/y=37.44:37.86/k=1/l=1/LEV=20 water_temp
yes? fill/over/lev/x=-76.33:-75.63/y=37.44:37.86/k=1/l=1 water_temp

[karlmsmith]

Comment by @AnsleyManke on 8 Jun 2010 23:08 UTC ''An related by aside suggestion: Datasets that have sparse data should routinely be configured so that the default graphics style is a simple SHADE plot''

The dataset as a whole is not sparse but has edges as all data do if they have missing values over land. Should there be logic so that if the plot involves less than xx points -- some small number -- then we do only a shade plot by default?

[karlmsmith]

Comment by steven.c.hankin on 8 Jun 2010 23:36 UTC ''Should there be logic so that if the plot involves less than xx points -- some small number -- then we do only a shade plot by default?''

==> I have not thought of sufficiently simple heuristic logic to detect sparse data. The fraction of total cells that have non-missing data is not a sufficiently reliable test.

See trac #800 for a related discussion of anticipating the need for RASTER style based upon units and variable names.

[karlmsmith]

Comment by steven.c.hankin on 8 Jun 2010 23:54 UTC Looking at the image examples, I'm still suspicious that the LAS plot (the first attachment) is showing a pathology that is deeper/worse than the final image (linear_lev.gif) -- Ferret commands, but not driven by LAS scripts). What concerns me is that the artifact in the LAS output where a triangle (FILL) appears over the top of a rectangle (SHADE) seems to be occurring a grid cell further back from the edge than we see in linear_lev.gif.

General conclusion here: RASTER-only plots are preferred whenever we are dealing with noisy data -- not just sparse data. I wonder if the LAS default should be changed to raster-only. (The plotting would be faster.) I know we've waffled on the choice of default in the past. Careful curatorship when configuring LAS configs is really the best answer.

[karlmsmith]

Modified by @AnsleyManke on 5 Jan 2011 23:37 UTC