geostatistical interpolation questions

Question

Hello all,

I'm new to geostatistical analyst and am trying to use the method of kriging for an academic project. I was wondering if anyone might be able to guide me on how to use the geostatistical analyst for the data that I have.

I have water potential and stomatal conductance data from 10 plants at 3 different sites in the Mojave. This data was taken on 5 different days over the course of two months following the addition of a pulse of water. I would like to take this data and spatially interpolate the water potential values and stomatal conductance values for each sample day to get a series of predictive and progressive surfaces. I know the sample size is pretty small and the prediction is going to be made over a very small scale as well.

I've made some preliminary attempts at analyzing this data already. I found the kriging to be rather complicated in terms of determining all the proper settings, and the model was not coming out as I expected that it should (perhaps due to small sample size?). I also tried local polynomial interpolation (LPI) and the deterministic method of spline with barriers. The LPI method seems to go along more with what I might expect to see from my data, but I'm not sure what the best method is to use.

Any help or suggestions would be greatly appreciated!

Eric6346 · Answer

Kernel Interpolation with Barriers calculates distances as the shortest distance around the barrier.  If two points can be connected with a straight line and the line does not cross a barrier, the distance will not be altered.  If a barrier blocks the line connecting two points, the distance between them will be altered to the shortest path around the barrier.  In the case of polygon barriers, it may not be possible to go "around" a barrier.  In that case, the points do not influence each other.

AimeeRoach · Answer

If you only have 10 points for each day, we do not recommend kriging.  No interperpolator is going to be very effective with so few points, but if I had to choose one, I would go with Kernel Interpolation with Barriers.  Just don't supply a barrier.

Thank you for the response. I've been working on trying Kernel Interpolation with Barriers with my data set. I have a question regarding this method though, perhaps you or someone can answer. I would appreciate any suggestions. 
I do actually have a barrier: the channels around which the plants that have been measured are located. However, the barrier limitation in the kernel method sounds like it would limit the information provided by values on opposite sides of the channel. For instance, plants on opposite sides of the channel won't be analyzed together. I.e., if there are no samples, say, within 1m on one side, we won't have a good estimate at that distance on that particular side. Does this method still use all the distances regardless of side, but then place the barrier after interpolation? 
Since my sample size is low, if the barrier is placed and each side is analyzed individually, this would reduce the sample size further and perhaps not be an ideal method.

Eric6346 · Answer

If you only have 10 points for each day, we do not recommend kriging.  No interperpolator is going to be very effective with so few points, but if I had to choose one, I would go with Kernel Interpolation with Barriers.  Just don't supply a barrier.