Add variables to MPAS_Stream_type for CF-compliance

[mark-petersen]

I would like to propose that we add variables to the MPAS_Stream_type so that the time can be CF-compliant. The specific problem is that a variable Time_bnds is required that defines the start and end time points of each measurement (see CF-compliant requirements), but we write multiple output files with different time bounds, so we can not just request a general Time_bnds variable in a stream that is specific to that stream.

My proposal is to add a new flag at the top of a stream:

<stream name="timeSeriesStatsMonthlyOutput"
        TimeInFile="standard" (default) or 
        TimeInFile="CFCompliant" 

where standard retains the current behavior, and the new option, CFCompliant, writes the time variables that we need, like Time_bnds and the variable Time. A core can then access a particular stream and set the values for these variables.

I think the framework alterations would be:

1. Add variables to MPAS_Stream_type in mpas_io_streams_types.inc, for example

logical :: writeCFCompliantTime = .false.
real :: Time_bnds(2)

2. In xml_stream_parser.c, read in the flag TimeInFile and pass information through subroutine stream_mgr_create_stream_c, say with iTimeInFile.
3. In mpas_stream_manager.F, subroutine stream_mgr_create_stream_c, set the writeCFCompliantTime logical for that stream.
4. In mpas_io_streams.F, subroutine MPAS_writeStream, if writeCFCompliantTime is true, write additional variables to file, like Time_bnds.

[xylar]

I think Time would also be needed, since this could be different for different streams.

[xylar]

@mark-petersen, I renamed time_bnds to Time_bnds everywhere, just FYI.

[mark-petersen]

@mgduda my main question for you is if it sounds acceptable to add the new flag to a stream header,

        TimeInFile="standard" (default) or 
        TimeInFile="CFCompliant" 

where the second option will add the output for these CF-Compliant variables. If so, I will proceed with the project. We can discuss more details in the actual PR, when we can see the code and example output.

[xylar]

@mark-petersen and @mgduda, maybe this is worthy of a quick Developer's telecon, maybe next Monday?

[mgduda]

@mark-petersen My first thought is that the MPAS_Stream_type (used by the mpas_io_streams module) isn't quite the right place to handle this. My counter-proposal would be to handle this at the mpas_stream_manager level, which has more information about, e.g., the times at which streams are written. Regardless of where this is handled, how would the stream determine whether fields represented instantaneous or time-averaged values?

I agree with @xylar that this is a topic with enough depth that a telecon might be warranted.

[mark-petersen]

For our reference, I experimented with these changes using two methods.
https://github.com/mark-petersen/MPAS-Model/tree/cf_compliant_time_using_streams_types
adds the changes described in this PR (#665 (comment))

https://github.com/mark-petersen/MPAS-Model/tree/cf_compliant_time_using_io_streams
tries to add time_bnds one level higher in mpas_io_streams.F