Hi Simon,
I'm not sure that a "moment" dimension would be very helpful for a lot
of typical applications.
To give the specific example I'm working with, we have a wave model
writing files with up to 20 variables which are functions of (time,
longitude, latitude). Of these, 9 are wave parameters derived from the
spectra computed in the model, the rest are other environmental
variables (water depth, currents, winds, etc.). Only one of those 9 is a
mean period (the -1 moment in this case). Introducing a dimension which
would be of length one, and only applying to one of our 20 variables,
would seem an unnecessarily cumbersome way to determine which, if any,
of the variables is appropriate to compare with data from another
source. It would be much easier to do that from the standard name and/or
other attributes that apply to all the variables in the dataset.
Regards,
Richard
On 17/11/2006 10:01 a.m., Simon Wood wrote:
> Hi Richard,
>
> Would it be appropriate to consider the moment as a dimension? That way
> you could just define 3 standard names for the wave periods:
>
> sea_surface_wave_period;s
> sea_surface_wind_wave_period;s
> sea_surface_swell_wave_period;s
>
> and use a 'tm' dimension for the different means, taking values -1, 0,
> 1, 2 etc.
>
> Obviously not all tm values would have to be present in any particular
> file (so you don't always have to provide values for every moment if not
> required). You would probably make it the left most dimension (in CDL
> notation) so it is least tightly bound (based on the fact that the
> original proposal was to have them as separate datasets -- but maybe
> thats not what you would actually want? maybe right most would be
> preferable?).
>
> regards,
>
> Simon Wood
>
>
> Richard Gorman wrote:
>> Hi,
>>
>> I'm also coming to grips with CF-compliance for wave data and model
>> products, so think I should comment.
>>
>> The various quantities Heinz and Beate describe are all derived from
>> the directional spectrum. In the most general case (and explicitly so
>> inside a spectral wave model) this is a function S(t,x,y,f,theta) of
>> five dimensions:
>> t = time,
>> x,y = spatial coordinates (e.g. longitude & latitude)
>> f = frequency
>> theta = direction
>>
>> S has the standard name
>> "sea_surface_wave_directional_variance_spectral_density"
>>
>> Integrating over direction:
>> S1 = integral[S d(theta)], with the standard name
>> "sea_surface_wave_variance_spectral_density"
>>
>> The parameters of interest can be derived from moments of the spectra,
>> i.e. if
>> M(n) = integral[S1*f^n df]
>>
>> then the various definitions of mean period are
>> Tm-1 = M(-1)/M(0)
>> Tm1 = M(0)/M(1)
>> Tm2 = sqrt(M(0)/M(2))
>>
>> Also:
>> Hm0 = 4*sqrt(M(0)) = "sea_surface_wave_significant_height"
>> while "mean direction" and "directional spread" come from various
>> directional moments of S, e.g.
>> integral[[S cos(theta) df] d(theta)]
>>
>> Now the problem is how to describe these variables where the frequency
>> and directional dimensions are not necessarily explicitly present:
>> i.e. our data are Hm0(t,x,y), etc. These may come from a model which
>> has derived them from spectra, or from satellite data where the
>> spectrum hasn't been used explicitly at all.
>>
>> If we accept that these quantities are well defined in terms of these
>> accepted definitions, we could just give them standard names as Heinz
>> and Beate propose (except I'd suggest
>> >> for the periods from the moment of order 2:
>> >> sea_surface_wave_tm2_period;s
>> >> sea_surface_wind_tm2_peak_period;s
>> >> sea_surface_swell_tm2_peak_period;s
>> should read:
>> sea_surface_wave_tm2_period;s
>> sea_surface_wind_wave_tm2_period;s
>> sea_surface_swell_wave_tm2_period;s
>> for consistency).
>>
>> But in some cases it could be helpful to spell out how these are
>> derived. For example, ideally the frequency range is 0 to infinity,
>> but in practice a limited frequency range might have been used,
>> especially if we are distinguishing swell and wind waves. Then we
>> could give explicit frequency bounds, and use a cell method
>> description, e.g. for Tm1:
>>
>> standard_name = "sea_surface_wave_tm1_period"
>> cell_method = "sea_surface_wave_variance_spectral_density: normalised
>> inverse of first frequency moment" or similar.
>>
>> Unfortunately it's rather wordy, and that's one of the simpler ones:
>> directional spread would be quite messy to define in words!
>>
>> Regards,
>> Richard
>>
>> On 12/11/2006 2:59 a.m., Jonathan Gregory wrote:
>>> Dear Heinz and Beate
>>>
>>>> we propose new standard_names for serveral variables concerning wave
>>>> periods deduced from the one dimensional frequency wave spectrum:
>>>
>>> These concepts look to me like different statistical methods for
>>> characterising
>>> the spectrum, and as such, can they be expressed by cell_methods? For
>>> instance,
>>> if we define a standard name of
>>> probability_density_function_of_sea_surface_wind_wave_period
>>> we can express the "peak" period (I presume this is the mode of the
>>> pdf - is
>>> that right?) with a standard name of sea_surface_wind_wave_period and a
>>> cell_methods of
>>> "probability_density_function_of_sea_surface_wind_wave_period: maximum"
>>>
>>> I imagine that something like that could done for the others, but I
>>> am not
>>> clear what the "mean" is over, or what "tm" means. "spread" also
>>> sounds like
>>> a statistic of some kind, that would need definition, or is it perhaps a
>>> standard deviation?
>>>
>>> Cheers
>>>
>>> Jonathan
>>> _______________________________________________
>>> CF-metadata mailing list
>>> CF-metadata at cgd.ucar.edu
>>> http://www.cgd.ucar.edu/mailman/listinfo/cf-metadata
>>>
>>
>
--
=========================================================
Richard Gorman
National Institute of Water and Atmospheric Research
PO Box 11-115, Hamilton, 3251, New Zealand
Tel: +64 7 856 1736 Mob: 021 074 7490 Fax: +64 7 856 0151
Email: r.gorman at niwa.co.nz Web: http://www.niwa.co.nz
=========================================================
Received on Sun Nov 19 2006 - 17:03:40 GMT
