Time-mean output files are named for the end of the averaging interval #554

A time-averaged output file is named for the instant the average was finalized, and its time value is that instant, so the file holding the mean for December of year 0 is named 0001-01 and carries time = 0001-01-01. A reader that maps files to months by name, as ncclimo and MPAS-Analysis do, takes it for January.

What happens

writeStream in IOStream.cpp uses the current clock time both for the file name (FileTime = ModelClock->getCurrentTime(), line 2529 at d086d49) and for time (OutTime(0) = ElapsedTimeR8, line 2544). Both are right for a snapshot. For a stream written by an analysis group with a ReductionPeriod, the averaging interval ends at that instant, so GlobalStats with ReductionPeriod: [1Month] and Filename: global_stats.$Y-$M.nc on develop names its December mean 0001-01.

Observed on a #481 build, whose IOStream.cpp has the same two lines: with StartTime: 0000-12-01_00:00:00, the first monthly MOC file is moc_1MonthTimeStats.0001-01.nc with time = 2678400 seconds since the start, i.e. 0001-01-01, holding the average over 0000-12-01 to 0001-01-01.

What was expected

MPAS-Ocean's timeSeriesStatsMonthly names a monthly mean for the start of the month it averages (…timeSeriesStatsMonthly.0001-01-01.nc is January of year 1), and CF asks for time to be representative of the interval, in practice its midpoint, with time_bnds giving the ends. For a stream whose fields are time means: name the file for the start of the interval, the previous ring time of the stream's alarm, and set time to the midpoint. #481 adds time_bnds to such streams from that same previous ring time, so once it is in, both follow from values already at hand.

Until then, Polaris will read the file named Y-M as the mean of the month before, so a run over calendar year 1 is read from the files 0001-02 through 0002-01.

How to reproduce

Any analysis group with ReductionPeriod: [1Month] and a $Y-$M file name; compare the file name with the interval it covers. The run above is at /lcrc/group/e3sm/ac.xylar/polaris_1.1/chrysalis/test_20260907/qu240-moc-rerun-20260909/monthly_mean_run on Chrysalis, with omega.yml beside it; ncdump -v time moc_1MonthTimeStats.0001-01.nc.

Posted by Claude Code on @xylar's behalf. The testing, analysis and wording above are AI-authored; please check them accordingly.