Re: [ferret_users] dealing with monty-mean time series

[Jaison Kurian <jaison@xxxxxxxxxxxxxxxxxx>]

Hi Sang-ki,
            I have few suggestions ...

1. Suppose your 'original' dataset is in grib format. In this case
     you have to define and assign a time axis while reading that
     data on to a Ferret session. In this case you can just open
     the NetCDF dataset and use its time axis.

2. Suppose your 'original' dataset is in NetCDF format. To regrid
     this data to the time axis of the second dataset, you can use
     @ASN regridding, which should take minimum time since it does
     not involve any interpolation :

       let err = sst_cdc - sst_org[gt=sst_cdc@ASN] 

     You can even save this new regridded variable to a NetCDF file
       for future use (and you can delete the original file only if 
       the new time axis is suitable for all your applications)

3. If the 'original' dataset is in NetCDF format, you can even use
     a descriptor file consisting of the time details & path of 
     the netcdf file/files so that it have time variables 
     (like T0, S_START, S_END, S_DELTA) corresponds
     to the second dataset. In this case Ferret will not bother 
     about the time axis/details present in the NetCDF file and
     hence will be the faster method. 

     You can get details about descriptor file and tools for 
     generating it from the Ferret Mail Archive.

     This will be an ideal solution because the time being spend
     for creating a descriptor file will be worth....you can 
     access the same dataset on different time axis without 
     costing any extra disk space.

Hope this helps...

Jaison


On Tue, 19 Sep 2006, Sang-ki Lee wrote:

> Hi,
> 
> I have a general question about dealing with monthly-averaged
> time series data from different sources.
> 
> NCEP1 reanalysis data is a good example. The original (grib format)
> data from NCEP have the time index at the beginning of each month,
> like 01/JAN/1949, 01/FEB/1949, and etc. One has to be very careful,
> because this time index means "initial_time" of the monthly average.
> For instance, t=01/JAN/1949 means time average from 01/JAN/1949 to
> 31/JAN/1949. If one download the CDC-derived NCEP1 reanalysis data
> (netcdf format), sometimes the time index is relocated to the
> middle of each month.
> 
> If I want to mix the original NCEP1 data with the cdc-derived
> data, one thing that I should not do is to regrid the original
> NCEP1 data to the time index of CDC-derived data in the following
> way:
> 
> ex) err = sst_cdc - sst_org[gt=sst_cdc]
> 
> This is like mixing orange and apple. In my case, I do the
> following:
> 
> let err = sst_cdc - sst_org[gt=month_reg@asn]
> or
> let err = sst_cdc[l=1:660] - sst_org[l=1:660]
> 
> However, I find such regriding processes time-consuming and
> unnecessary. And, it is also very easy to make mistake. I always
> get ambiguous time axis warning when I do this way, but I guess
> that is ok.
> 
> It will be very convenient if there is a way to tell Ferret
> to ignore the day, hour and second. So whenever Ferret see JAN/1949,
> for instance, it automatically understand that the date is in the
> middle of JAN/1949.
> 
> I would welcome any other tips for dealing with monthly-mean time
> series data.
> 
> Sang-ki
> 
>