Hi all,
in the context of long time series, the word "time stamp" also was used differently:
Imagine you are living some time ago and are running a
mechanically recording thermometer scratching a curve into wax
paper. Every month you put in a new roll of that wax paper and put
the previous one into the cubicle . To indentify it, you make a
"time stamp" on it just marking the day when you took it out. So
in some data sets the time stamp marks the end of a measuring
period.
People producing data, especially time series, are often living
in a scientific environment with different conventions like those
using the data. But all are using the same words and are not aware
of possible sources of misunderstanding. Even in times of netcdf
with many meta data attached to the data, the very meaning of the
numbers may be unclear and can be even lost over time. So it is
very important, to have some communication between these different
worlds.
I remember that the data set in question her is related to era-5. So its worth to read the era-5 description in the COPERNICUS portal.
https://climate.copernicus.eu/climate-reanalysis
A mirror is here:
https://rda.ucar.edu/#!lfd?nb=y&b=proj&v=ECWMF%20ERA5%20Reanalysis
If this does not answer the question, it's definitely worth to
ask there for support. I did this some time ago for a similar
problem with ncep and got friendly and helpful answers.
Greetings,
Martin
Hi Alexander and Ferret fans,
On Wed, Feb 26, 2020 at 7:15 AM Alexander Audet <alexander.c.audet@xxxxxxxxx> wrote:
As far as I can tell (link) although the monthly means are timestamped at the first of the month and seem to be centered there, they actually represent the average value of the month instead 1/2 the last and 1/2 the named month. I am still confused about this, so I could be wrong.
If this netCDF file was created by some respectable organization, you may want to file a bug report. This error affects a lot of users. The best netCDF configuration for a monthly-mean dataset is, I think, such that
- Each monthly mean is defined at the center of the month.
- The "time_bnds" coordinate variable correctly specifies the beginning and end of each month ( http://cfconventions.org/cf-conventions/cf-conventions.html ).
Having said all that, it all depends on how accurate you want your seasonal means to be. For many use cases, the simple (unweighted) average may be fine. In that case, the 3-point boxcar average (is that @SBX ?) on the L axis will do.
Regards,Ryo