[R-sig-Geo] strange variance patterns in spatiotemporal kriging

[Rawling, Geoffrey]

Hello list members

I have encountered some strange map patterns of variance when using
spatiotemporal kriging in R - gstat on water level data. I have
measurements from wells in a basin over a period of 40 years. Most
wells have 10 or fewer measurements. But 4 wells in one cluster have
from 70 to 180 measurements.

The kriged water level surfaces look fine. But the variance (shown as
std error on linked maps) is sensitive to the two parameters nmax and
buffernmax in krigeST. Below are three dropbox links to maps titled as
to the settings of these parameters in krigeST. The triangular region
of high variance persists around the cluster of wells with many
measurements (east of Silver City on the maps) in all cases.

Is there any theoretical reason for the irregular variance patterns?
And the persistence of the triangular pattern? Just an artifact of the
computation algorithm...if that is the case how best to explain it?

Many thanks for any insight

https://www.dropbox.com/s/nro53i84k3cce2x/nmax_20_buffnmax_1.jpeg?dl=0

https://www.dropbox.com/s/8k4zll7aqv3xh1k/nmax_100_buffnmax_2.jpeg?dl=0

https://www.dropbox.com/s/czg4ix70a4wzouq/nmax_200_buffnmax_2.jpeg?dl=0

-- 
*********************************************

Geoffrey Rawling

Senior Field Geologist
NM Bureau of Geology and Mineral Resources

Geoffrey.Rawling using nmt.edu

1015 Tijeras Avenue NW, Suite 200
Albuquerque, NM, 87102

505-366-2535 phone
505-999-8307 cell