Spatio-Temporal Kriging: Memory Issues

Dear RUsers,


I wastrying to perform spatio-temporal kriging using krigeST. But looks like I amhaving some memory issues. I get the error:


?Error:cannot allocate vector of size 8.9 GB?


I ampredicting using the following code:

predicted<- krigeST(values ~ 1, data, prediction.grd, v.model)

Here "data" has STSDF structure and "?prediction.grd" has STF structure. I fitted?separable, productsum and metric models in different times.
I was workingwith the yearly data of around 4600 stations (these are water level monitoringwells). They have yearly data for 19 years. I am trying to interpolate it using 0.5 mile by 0.5 mile pixel sizes(or grid size). I want to cover a total area of around 55000 square miles. But code stopped working and that error message came out.


Then I kept all those 4600 stations, but chopped my area of prediction to around1600 square miles (over a county) as I want to capture the overall effects ofall stations while predicting. But this did not work as well yielding the same error as above.?

I am afraid, R is inverting a huge matrix to get the weights each time.
However,when I work with small number of stations (around 500) and predicting over amuch smaller area of around 1600 square miles (over a county), the code worksnicely and gives accurate results using different fitted models. Sometimes I have to use a coarser pixel sizes to reduce myprediction locations. But I don't think this an accurate option.


Can anyonehelp me so that I can use all the 4600 stations with finer pixel sizes? I canpredict over small areas and then merge them. But again I have to keep all mystations while doing that. Is going parallel an option (but I really want toavoid it)? Please let me know if you used nearest neighbors with anisotropy.
I can share my code if you want.


Thanks!


Abdullah??PhD StudentTexas Tech University