Special issue: Spatial and spatio-temporal models for interpolating climatic and meteorological data

Dear R-sig-geo,

is anyone interested in working together on a submission for the below 
mentioned special issue of SPASTA?

The situation is as follows:
We have about 3.5 yrs worth of temperature and humidity data from Mt. 
Kilimanjaro at hourly resolution. In addition we have numerous landscape 
data derived from a high-resolution DEM (30m) plus NDVI images at the 
same spatial resolution.

So far, we have conducted a (though not in the strict sense) 
spatio-temporal interpolation study on a monthly basis for the above 
mentioned data focussing on machine learning / data mining algorithms. 
As a reference we also used ordinary kriging (through the 'automap' 
package). The results of this exercise look quite promising (see link 
below for a figure showing the RMSE of predictions - observations for 
250 repeated random subsampling runs). There are 6 machine learning 
algorithms that perform significantly better than our reference kriging 
runs.

https://www.dropbox.com/s/rlvvxxbr355hi44/ML_results_temperature_monthly.pdf

Here's the caveat:
The kriging done for this exercise can hardly be considered optimal. We 
simply used the autoKrige function.

Therfore, I would like to find one (or more) person(s) with profound 
knowledge of 'classical' spatio-temporal interpolation methods to 
provide an exhaustive comparison of the most promising machine learning 
algorithms and optimised classical approaches (e.g. various kriging 
flavours, IDW, GWR etc).

So, if anyone from this list might be interested, please write me and 
I'll be happy to provide more detailed information on both the results 
of the presented figure as well as the intended comparison study for the 
special issue (or another journal - this merely seems a very appropriate 
opportunity to come forward).

Regards,
Tim

Special issue: Spatial and spatio-temporal models for interpolating climatic and meteorological data

Thread (2 messages)