Create a test dataset(s) with range of grid types/ challenging features

Reported by @AnsleyManke on 11 Jun 2010 19:57 UTC Captured from an email...

Putting this in the component LASTest, though it belongs to a number of topics including adddXML and products.

We have a need for an LAS test dataset -- much in the spirit that the Ferret benchmark depends upon several test datasets. Roland agreed to be thinking about the range of datasets needed. It occurred to me that some of that knowledge resides with you as the creator of most of the products. So this email is to suggest that you might be the person most easily able to get the ball rolling -- to create a reasonably comprehensive initial test dataset that Roland and others can see and add to.

Here's a start on characterizing a single dataset that might, alone, test a broad range of issues. In fact embedding the range of characters of the variables into a single dataset, makes the test even more demanding -- both from a UI standpoint, and from a multi-variable selection standpoint. The LAS test dataset should contain

* variables on combinations of irregular and irregular axes (X, Y, Z and T)
* variables that are 4d, 3d in each view, 2d in each view and 1d in each view
* variable pairs that are staggered in X, Y, and Z in the manner that MOM model output is as well as non-staggered pairs (for vector testing)
* very high resolution localized in XY and also global in X
* combinations of very high resolution (n=8000 on an axis) and very low resolution (n=1, 2, or 3 on an axis) (must exercise automatic striding, among other things)
* variables that should be on centered contour levels and palettes and variables that should not
* a variable with units of "number of observations" and similar heuristic choices that might be assumed to effect FILL versus SHADE defaults (see note below (*))
* tests of global and variable attributes that are relevant to LAS and Ferret -- e.g. scale/offset
* manufacture values for the variables that can help in gleaning diagnostic information from looking at plots.  For example, values like v = i + 10*j + 100*k + 1000*l might serve us well
* --- also --
* probably a curvilinear version of this dataset distinct from the initial rectilinear one
* probably another for sigma vertical coordinates

(*) We need increasingly to find ways to clue LAS into automating the special treatments that some variables deserve -- vector pairs, centered palettes, SHADE v FILL, enumerated versus continuous variables. Presumably this will be through a combination of heuristics ("TAUX" and "TAUY " as a vector pair) and explicit attributes that might be added to the dataset via attributes (e.g. a new attribute such as semantic_clues = "centered, scattered, enumerated" that could be added to the variable via ncML). Over time the LAS test dataset(s) should exercise the new levels of automation. AddXML will get exercised by these datasets, as well.

Migrated-From: http://dunkel.pmel.noaa.gov/trac/las/ticket/812

NOAA-PMEL / LAS

Create a test dataset(s) with range of grid types/ challenging features #818