[CF-metadata] Is there a convention defining day offsets to use for monthly average time series?

From: Steve Hankin <Steven.C.Hankin>
Date: Mon, 08 Aug 2011 15:40:51 -0700

Hi Jim,

[My gosh, two responses arrived as I typed this. A hot topic, clearly.]

The short answer depend upon what are you trying to *do* with values,
since conceptually a monthly average doesn't have a point location in
time -- it represents a range. When plotting, for example, it is
equally correct to plot a step function that changes value at each month
boundary, or a line plot with points centered on each month. Depends
what you are trying to communicate. (Using the 15th or 16th is at best
a crude estimate of the center point.)

You'll find extensive CF email threads about the use of integer values
to represent monthly time increments. Myself, I am in the school of
thought that says a "month" is not a valid unit of time, since it does
not represent a fixed amount of time (January and February months being
unequal quantities). This school says you should encode your CF files
using a meaningful unit such as "DAYS since ...". Yes it is more up
front work than using an integer for month number, but it saves endless
confusions after the fact (per your question here). Others argue that
because use of a month number is such common practice, then CF must
support it.

Climatological monthly averages are even messier, of course. Ferret
users, for example, most commonly weigh climatological February as the
long term average length of the month (28.2485 days or something like
that). A non-integer number of days, of course, plays havoc with the
diurnal cycle.

     - Steve

========================

On 8/8/2011 2:43 PM, Jim Biard wrote:
> Hi.
>
> I have a time series of monthly averaged values. I have an
> integer-valued time coordinate variable and an associated time_bounds
> variable. Is it correct to use the 15th of February and the 16th of
> all the other months for my time centers, or should I use the 16th of
> every month?
>
> Also, should I do anything differently if my data are climatological
> monthly averages (say, over 30 years of data)? And, in this case,
> should the time coordinate values be day numbers from the beginning of
> the 30-year time interval, the end of the time interval, or something
> else entirely?
>
> Grace and peace,
>
> Jim Biard
>
Received on Mon Aug 08 2011 - 16:40:51 BST