[R-sig-Geo] Spacetime regression
Hi all, I want to do a spatio-temporal regression on a quite large dataset. I have 100 k records. These correspond to measurements taken at 3000 locations, approximately every half year. The geographic area is all of the Netherlands (240 x 300 km). Is spatio-temporal kriging advisable for a dataset that is so large? When I make the sample space-time variogram (with variogramST), it automatically chooses a time-lag difference of about 2 days. This is much too small to be meaningful for my data; half-year periods would be interesting. Is there a way to tell this to the variogramST function? Thanks in advance, Roelof Coster [[alternative HTML version deleted]] ___ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Re: [R-sig-Geo] Spacetime regression
Thanks for the answers. I have cleaned up the data by rounding the dates to half years, and now the variogram turns out beautiful. Now I can proceed to kriging and see what I get. Best regards, Roelof Coster 2014-02-13 17:54 GMT+01:00 ldec...@comcast.net: when confronted with a large dataset i usually begin by trying methods out on subsets, e.g. start with 2^n for n = 8 and work your way up to 2^17... you might find convergence to the answer to your questions before analyzing the full dataset. Lee De Cola -- *From: *Roelof Coster roelofcos...@gmail.com *To: *r-sig-geo@r-project.org *Sent: *Thursday, February 13, 2014 7:30:04 AM *Subject: *[R-sig-Geo] Spacetime regression Hi all, I want to do a spatio-temporal regression on a quite large dataset. I have 100 k records. These correspond to measurements taken at 3000 locations, approximately every half year. The geographic area is all of the Netherlands (240 x 300 km). Is spatio-temporal kriging advisable for a dataset that is so large? When I make the sample space-time variogram (with variogramST), it automatically chooses a time-lag difference of about 2 days. This is much too small to be meaningful for my data; half-year periods would be interesting. Is there a way to tell this to the variogramST function? Thanks in advance, Roelof Coster [[alternative HTML version deleted]] ___ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo [[alternative HTML version deleted]] ___ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
[R-sig-Geo] Local spacetime kriging
Hi, The krigeST function has a 'nmax' parameter that sets the maximum number of neighbouring observations to be used in the prediction. However, the help for this function states that it does not support kriging in a local neighbourhood. So, is this just a mistake in the help? Also, in case kriging in a neighbourhood is possible, how does the function determine this neighbourhood? How is distance in space compared to distance in time? Thanks! Roelof Coster [[alternative HTML version deleted]] ___ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
[R-sig-Geo] Predicted kriging variances don't match errors in cross-validation
Dear list, I'm working on a local (nmax=100) space-time kriging model. I did a cross-validation in which I made my model predict the values at 2000 randomly selected data points, based on the rest of the observations. The results are quite good (the average error is very small, the errors are symmetrical and the spread is not too large). However, I don't see any correlation between the squared errors and the predicted variances. The Kendall's tau correlation coefficient between the two is even slightly negative. I would expect larger squared errors, on average, for data points in which the predicted variances are large. Should I consider this as a sign that my model is incorrect? Best regards, Roelof Coster [[alternative HTML version deleted]] ___ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Re: [R-sig-Geo] Spatio-Temporal Kriging: Memory Issues
Hello Karim, You might want to use local kriging, as in: predicted<- krigeST(values ~ 1, data, prediction.grd, v.model, nmax = 500, stAni = 1) The nmax parameter is the number of observations that will be used for each location. The stAni is a parameter controlling whether observations that are 'spatially' or 'temporally' near will be preferred. Local kriging may also be theoretically better because it only assumes local stationarity rather than the existence of a constant mean for the entire domain. Hope this helps! Roelof [[alternative HTML version deleted]] ___ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
[R-sig-Geo] Geostatistics on geometric network
Dear list members, I'm working with electric measurements that were taken on pipelines. These are spatio-temporal data whose spatial domain is not Euclidean, because the pipelines form a geometrical network. Has any work been done before to study this kind of data? Best regards, Roelof Coster [[alternative HTML version deleted]] ___ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
