[CF-metadata] date and time

From: Nan Galbraith <ngalbraith>
Date: Fri, 24 Oct 2008 20:37:24 -0300

For the case where there are multiple variables with different sensor times
in a NetCDF file, the 'sensor time' standard name somehow needs to be
tied to the data variable produced by that sensor - can this be a compound
term that includes the standard name of another variable?
>> > time_from_device: time reported by some device (data source)
>>
> Somehow this seems a bit unsatisfying to me. It doesn't really suggest that
> this time might not be true time.
I'm not sure that it's been determined that the sensor time is *not* the
true
time.
> The two cases you outline seem pretty similar to me, but maybe I'm
> missing something.
In one case, the sensor_time can be used to get a more exact time for
a particular measurement. The time variable (in the CF sense) provides
a 'good enough' t-axis, and allows you to combine data from different
sensors on a single time base; the param_sensor_time would preserve a set
of more exact sensor times for specific measurements, possibly one for
each of several sensors.

We routinely include sensors with different sample schemes on a single
logger, so that one sensor spot-samples at the start of the minute and
another takes its measurement sometime later... etc. We also telemeter
subsampled and/or averaged data where different variables can be from
different times within the record interval. It's not feasible to
represent this
in a standard way in CF, as far as I know.

The second use of sensor_time would document sensor clock drift, and
would mainly be interesting to us as a data provider, or to someone who
is initiating commands to a sensor in the field using that sensor's clock.

As a data user, though, it would be fairly important to know which of these
types of sensor_time was which - I'd want to use one, but not the other.
> As a general observation, the question 'how are you going to use this'
> reflects an assumption that the only (or primary) use of the netCDF
> files and the standard names is purposeful and anticipatory. In fact,
> especially if you are dealing with observation data, both can be
> strictly documentary; and of course the data can be used in unintended
> ways. So it shouldn't matter how the data will be used; I just want
> to know how to document what I have.
The standard names are only useful to the extent that their meanings are
agreed upon, so how you're going to use this term is not an unreasonable
question.
>> I agree that it would be useful to have a standard name to indicate a
>> sensor time, but it would be a good idea to discuss the purpose a
>> little
>> more. Is it going to be used to indicate a clock drift, i.e. that
>> the time
>> in 'DateTime' (or, in CF, time) is an absolute, or at least a
>> 'corrected'
>> time, and this sensor_time is an indication of clock drift in an
>> instrument?
>> Or would it be used for data where there's a nominal time base and
>> multiple
>> sensors that actually have a separate, valid sample time? Does one
>> standard
>> name cover both of those cases?

Regards - Nan
Received on Fri Oct 24 2008 - 17:37:24 BST