Skip to content

nlme: spatial autocorrelation on a sphere

3 messages · David Winsemius, Dan Bebber, Spencer Graves

David Winsemius
#
On Oct 1, 2012, at 8:34 AM, Spencer Graves wrote:

            

      
      
      
    

        
Please accept my apologies for attempting to be cute and I also apologize to Mr Bebber for reading his request far too quickly. I was trying to use (C)RAN (T)ask (V)iews as a verb.

            

      
      
      
    

        
I don't think it warrants being enshrined. I might also  have written:  require(sos); findFn("metric spherical latitude longitude")  might produce. In this instance that approach found the ramps::corRSpher function, which has a "haversine" metric, much  more quickly than did my subsequent efforts with RSeek, which I conducted after I realized that Bebber had already made a good faith effort at identifying resources.

  
    

      
      
    

  


  


    

    

      
      

        


  

        

      Dan Bebber
    

    

      
      #
    

  


  

      
Please be assured, I really did "RTFM" (i.e. Pinheiro & Bates) as well as all the online sources, and have accessed the codes of the corStruct functions, but they do not reveal how the "metric" argument is used. Therefore, I can't reprogram to include "haversine" distance, as seen in the ramps package.

Smith et al. (2008) write "The spatial correlation structures in nlme are not directly used because they do not allow great circle distance, which is very commonly needed for spatial data" so there may be a case for adding this functionality to nlme.

If anyone knows a way in to the relevant code, please let me know.

Thanks
Dan

Reference
Smith BJ, Yan J & Cowles MK 2008. Unified Geostatistical Modeling for Data Fusion and Spatial Heteroskedasticity with R Package ramps. Journal of Statistical Software 25(10)

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Spencer Graves
    

    

      
      #
    

  


  

      
Hi, Dan:

        
On 10/1/2012 8:07 PM, Dan Bebber wrote:

            

      
      
      
    

        
Thanks for the clarification and for your persistence in this 
issue.  Others clearly have been asking for an enhancement of this nature.


       I assume you've checked and confirmed that "ramps::corRSpher", as 
suggested by David Winsemius, will NOT do what you want.  If so, I 
suggest you subscribe to "R-sig-mixed-models" (per r-project.org -> 
"Mailing Lists"), and post your question, citing Smith et al., to there. 
  When you do, I suggest you include cc:  Douglas Bates 
<bates at stat.wisc.edu>.


       I'm sorry I can't help more, but I hope you will soon find a 
solution to this problem.


       Best Wishes,
       Spencer