[R-sig-Geo] Co-Kriging with huge data

[Edzer Pebesma]

Kai Zhang wrote:
> Hi Edzer,
>
> Thanks for your reply. I am not sure if I get your point -- "Maybe a cross validation with your 17 temperature monitors can give you
> some idea on this". I am wondering if you mean sensitivity analysis of neighbor size using cross-validation. For example, based on 17 monitors, I could try the radius with 1 km, 2 km and so on using cross-validation and choose a neighbor size with smallest RMSE calculated from cross validation.
>   
Yes, that's what I meant.
> I appreciate your response and best regards,
> Kai
>
> ____________________________________________________
> Kai Zhang
> PhD Candidate, Environmental Health Sciences
> School of Public Health                        
> University of Michigan                         
> 109 S. Observatory                       
> Ann Arbor, MI 48109-2029         
> ____________________________________________________
>
> ======= 2009-12-14 02:49:22 You wrote =======
>
>   
>> Kai Zhang wrote:
>>     
>>> Dear all,
>>>
>>> I have temperature measurements at 17 monitors.  I have a secondary variable across the study area, which includes 230,000 points. I tried to conduct Co-Kring with this huge dataset and it appeared to be computationally intensive. Is there any way to run it except for running the program in a cluster?
>>>
>>> I appreciate your response and best regards,
>>> Kai
>>>   
>>>       
>> One alternative would be to use co-kriging and to limit the search
>> neighbourhood for the secondary variable. The question is how much.
>> Maybe a cross validation with your 17 temperature monitors can give you
>> some idea on this. If someone knows about existing literature on this,
>> I'd be happy to hear.
>>
>> Another possibility is to use universal kriging, with your secondary
>> variable treated as predictor. For this, you need the values of your
>> secondary variable to be known ath the 17 monitor sites, and the
>> prediction locations should coincide with (some of) the 230000 points.
>> It is not equivalent to the first approach, poorer in the sense that it
>> ignores the secondary variable at locations beyond the prediction
>> location, but assumes a linear relationship between the two as well.
>>
>> -- 
>>
>> Edzer Pebesma
>> Institute for Geoinformatics (ifgi), University of Münster 
>> Weseler Straße 253, 48151 Münster, Germany. Phone: +49 251 
>> 8333081, Fax: +49 251 8339763  http://ifgi.uni-muenster.de 
>> http://www.52north.org/geostatistics      e.pebesma at wwu.de
>>
>>     
>
>
> ��i��'���ɨh��&

-- 
Edzer Pebesma
Institute for Geoinformatics (ifgi), University of Münster 
Weseler Straße 253, 48151 Münster, Germany. Phone: +49 251 
8333081, Fax: +49 251 8339763  http://ifgi.uni-muenster.de 
http://www.52north.org/geostatistics      e.pebesma at wwu.de