[CF-metadata] daily maximum of running 8-hour means

From: Jonathan Gregory <j.m.gregory>
Date: Sun, 23 Oct 2011 18:29:47 -0600

Dear Martin and Philip

Thank you for your use-cases showing the need for standardised metadata for
daily maxima of 8-hour running means. I expect that similar cases will arise
in other areas and actually it is quite surprising they haven't already, I
think.

What you want to describe involves two cell_methods processing operations: (1)
Calculate an 8-hour running mean for consecutive hours, (2) Calculate a
maximum of these within days. We could record this explicitly and completely
if we had different time coordinates for the two operations. For the first
operation, the cell_method is a mean, and we have time coordinates at hourly
intervals, each with bounds specifying an 8-hour period. The cell bounds
overlap, but that's no problem. E.g. they could be [0:00, 8:00], [1:00, 9:00],
[2:00, 10:00], ... (just the time, omitting the date). For the second
operation, the cell_method is a maximum, the bounds indicate periods of days,
spaced at daily intervals i.e. not overlapping.

One possibility would be to extend CF to allow you to preserve the original
time axis to which the first operation applied, even though it is no longer
provides the time coordinates once the second operation has been carried
out. For instance, we could allow:

  float air_quality(time);
    air_quality:cell_methods="old_time: mean time: maximum";
  double old_time;
    old_time:units="hours since 2011-10-23 0:00";
    old_time:bounds="old_time_bounds";
  double old_time_bounds(old_time,two);
  double time(time);
    time:units="hours since 2011-10-23 0:00";
    time:bounds="time_bounds";
  double time_bounds(time,two);

  old_time=0.5, 1.5, 2.5, ...;
  old_time_bounds=-3.5,4.5, -2.5,5,5, -1.5,6.5, ...;
  time=12, 36, 60, ...
  time_bounds=0,24, 24,48, 48,72, ...;

This is not currently legal because old_time is not a dimension of air_quality,
but it seems a reasonable extension to me.

For climatological time, we have a similar issue of multiple time processing,
and in that case we do not find it necessary to keep old time coordinates.
Instead, we use a special interpretation for the time bounds (CF sect 7.4).
However, this doesn't seem to be easy to adapt for the general case, because
of the overlapping periods. Climatological time doesn't allow that.

Another possibility would be to define new kinds of standardised metadata,
following Philip's categories, to describe the superseded time coordinate e.g.
cell_methods="time: mean (period: 8 hours offset: -15.5 hours) time: maximum
(interval: 1 hour)". The second entry is standard, and records that hourly
values were input to the calculation of the daily maximum. The first entry
records that those values were themselves means calculated over periods of 8
hours, and the first such period began 15.5 hours before the first time
coordinate (12 - 15.5 = -3.5, the lower bound of the first old_time cell).

I am sure there are plenty of other possibilities, but I wonder what you think
of either of these as a start?

Best wishes

Jonathan
Received on Sun Oct 23 2011 - 18:29:47 BST