Re: eps and nscan in poisson_grid_fill

The code is a VERY simple sequential over-relaxation.
It iterates until the residuals from a previous pass at
all grid points are less than eps (always positive).
What is acceptable? That is a user decision.

The code for the function is at:
     http://www.cgd.ucar.edu/~shea/fill_msg_grid.f

Good luck

On 8/26/13 9:01 PM, Jiaxin Zhang wrote:
> Hi,
>
> Thank you for the information, Dennis.
> Just an additional question about eps. In the poisson_grid_fill link, the
> argument relc has a range of values, and I wonder if eps has also some
> range of values (minimum and maximum). Or maybe eps can take any
> value greater than 0?
> I would need to make sure about this before experimenting with different
> eps values in the code.
> Is there any clue about what makes and eps value work better than another
> one in the poisson_grid_fill function? How can I notice that one eps
> improves the output of poisson_grid_fill with respect to a different eps
> value?
>
> Thank you.
>
> Jiaxin
> 2013/8/25 Dennis Shea <shea@ucar.edu>
>
>> You have to experiment.
>>
>> The eps for geopotential height is certainly different than that
>> for (say) vorticity.
>>
>> Tha same eps should be used for each variable ... not 'year dependent'
>>
>>
>> On 8/23/13 9:49 PM, Jiaxin Zhang wrote:
>>
>>> Dear NCL users,
>>>
>>>
>>> I am using the function “poisson_grid_fill” for the first time and I have
>>> some questions about the arguments of that function. I guess that this
>>> function is only going to fill values in the grids with missing values,
>>> without modifying those grids with already existant values, right?
>>>
>>>
>>> Apparently, the “poisson_grid_fill” works very well with my data using the
>>> same default values for the arguments provided in example 1 in the web
>>> (see
>>> attached code for temperature).
>>>
>>>
>>>
>>> My questions are:
>>>
>>> - It seems that the argument eps is variable dependent according to the
>>> web. Do you think I can use the same eps value (for instance 1.e-2) for
>>> all
>>> my climate variables (temperature, rain, ...)? Or maybe is it necessary to
>>> modify this depending on the approximate ranges of values for the climate
>>> variables? I have included information about the ranges of my climate
>>> variables for 1 year at the end of the message in case this may help.
>>> Also,
>>> because I have climate values for different years, I guess I could use the
>>> same eps values for different years.
>>>
>>>
>>> - I have tried nscan=1500, and it takes some time to complete the grid
>>> filling for a netcdf with 2920 times. Do you think that when using a lower
>>> value than 1500 for nscan it is going to be quicker to complete this part?
>>> Or if it nscan is too low, perhaps then the results are not very reliable?
>>> What are your suggestions?
>>>
>>>
>>> Could anyone please have a look at the obtained plot with filled values
>>> for
>>> temperature from the script "test_grid_filling.ncl" and let me know your
>>> opinion? Do I need to modify something in the script? I am not completely
>>> sure about the most appropriate values for some of the arguments in the
>>> function. Can I use the same argument values for all my climate variables?
>>> With respect to the obtained results with this function, I am not
>>> interested in the climate values located in water grids.
>>>
>>>
>>>
>>> Thank you!
>>>
>>>
>>>
>>> Jiaxin
>>>
>>>
>>>
>>> -----------
>>>
>>> In case it helps, the ranges and means of climate values for 1 year in my
>>> netcdf files are, temperature (min=210, mean=282, max=325); longwave
>>> radiation (min=56, mean=298, max=478); shortwave radiation (min=0,
>>> mean=177, max=1119); specific humidity (min=1e-5, mean=0.0069,
>>> max=0.02969); rain (min=0, mean=2.32e-5, max=0.00565); pressure
>>> (min=56104,
>>> mean=95094, max=106001); wind (min=2.05e-5, mean=2.87, max=25.46).
>>>
>>>
>>>
>>> ______________________________**_________________
>>> ncl-talk mailing list
>>> List instructions, subscriber options, unsubscribe:
>>> http://mailman.ucar.edu/**mailman/listinfo/ncl-talk<http://mailman.ucar.edu/mailman/listinfo/ncl-talk>
>>>
>>>
>
>
>
> _______________________________________________
> ncl-talk mailing list
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>
_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Mon Aug 26 21:24:54 2013