Wiki Page for Ticket #636 (Review of Grid Properties in MindMaps)
See also ticket/657
Some thoughts prompted by a review of Marie-Pierre's document (see attached pdf) comparing grid properties defined in the mindmaps, and in the gridspec and CIM grid package data models.
Use of time-stepping information
For now (i.e. for CMIP5Q purposes) my inclination would be to restrict any grid properties to the usual 3 spatial dimensions and not try to factor in any additional time-stepping stuff into the CIM grid model. It may well be necessary to capture time-stepping information elsewhere in the CMIP5Q but I don't think the CIM Grid package is the right place to model it - at least not right now.
(Sub)Component-specific Grid Definitions
To solve the issue of grid properties being attached to various sub-components in the mindmaps we'd need to make grid definitions first-class objects with their own identity, just like models, simulations, platforms, and so on. I mooted this in my previous review of CMIP5Q grid artifacts back in Nov 2009 (see ticket #657). Bryan has, I believe, recently endorsed this approach. Whether or not it's feasible to engineer this change in time for CMIP5 is not a question I feel qualified to answer.
GridType / grid_descriptor enumeration
The values of this enumeration differ across the three conceptual models, although the values in the CIM grid package enumeration are the same as for gridspec only with the "_grid" suffix removed (oh, and a slightly different spelling for regular_lat_lon). Given that the current gridspec tools are spitting out the original gridspec enumeration values, I guess we should standardise on those, even though my preference would be to drop the "_grid" suffix. Why? Well, if one defined, say, a Dwelling entity with a property called 'type' to indicate the type of dwelling, then values of that property might be ("condo", "apartment", "bungalow", "manor_house"). It would be superfluous IMO to add a "_dwelling" suffix thus: ("condo_dwelling", "apartment_dwelling", "bungalow_dwelling", "manor_house_dwelling").
MeanZonalResolution and MeanMeridionalResolution
I think it would be useful to add definitions of these quantities to the mindmaps, which would then find their way through to the CMIP5Q. Is MeanZonalResolution asking for the mean spacing between grid points along a parallel of latitude, in which case it's referring to meridional spacing (i.e. in the X or longitude dimension), or the mean spacing between parallels of latitude (i.e. in the Y or latitude dimension). Conversely for MeanMeridionalResolution. How likely is it that these values will be known by metadata compilers?
SchemeType and SchemeMethod
These were new to me. I wasn't sure what the difference was between SchemeType? and grid_desriptor (in the gridspec model) and GridType? (in the CIM model): there seems to be some degree of overlap here. Are these 2 properties defined or described anywhere? From a point of ignorance then, I'd say that this low-level information may usefully be captured as part of the gridspec output but may not be essential for either the CMIP5Q or the CIM Grid package. Happy to be told otherwise though :-)
NumberOfZonalGridBoxes and NumberOfMeridionalGridBoxes
Not sure what these refer to, though I think they're related to the previous item. Definitions anywhere? Are they related in any way to the existing CIM grid properties nx, ny and nz, which refer to the number of grid points in those 3 dimensions? I was under the impression that the dimensions of 'metocean' datasets were typically expressed in either lat/long or X/Y. Will the use here of 'zonal' and 'meridional' be understood? Is it not possible to use 'latitudinal' and 'longitudinal' instead? I'm guessing that these properties make good sense in their particular context, one of the many that I'm not au fait with!
Phil Bentley, 22 Feb 2010