Weird interpolation "bug" - Arc Geostatistical kriging, lower bound > upper bound
I'm using the wizard in Geostatistical Analyst, ArcMap (10.0 I think) in order to run some simple kriging. Output is set to Prediction.
I have two separate inputs, which have points at the exact same locations, but one has values that are either equal or less than the other. Mostly the output makes sense. However, in a few locations the 'lower bound' dataset results in a krigged surface that interpolates higher values than the 'upper bound' dataset. I've tried messing with the Transformation Type and other basic variables, and some have the problem worse than others, but in all cases there are a few exceptions where the lower bound output exceeds the upper bound.
I've checked and re-checked through the input data carefully - it's not a problem there.
For the most part I am trying to use the default auto-populated parameters, but I do change the lag size (to ~1/6th the default) and I change the Model type from Stable to Exponential. I don't see how either of those could cause this.
What might be causing this? If I want to guarantee the lower bound data results in lower krigged values in all locations, is there anything I can in order to force that?
أكثر...
I'm using the wizard in Geostatistical Analyst, ArcMap (10.0 I think) in order to run some simple kriging. Output is set to Prediction.
I have two separate inputs, which have points at the exact same locations, but one has values that are either equal or less than the other. Mostly the output makes sense. However, in a few locations the 'lower bound' dataset results in a krigged surface that interpolates higher values than the 'upper bound' dataset. I've tried messing with the Transformation Type and other basic variables, and some have the problem worse than others, but in all cases there are a few exceptions where the lower bound output exceeds the upper bound.
I've checked and re-checked through the input data carefully - it's not a problem there.
For the most part I am trying to use the default auto-populated parameters, but I do change the lag size (to ~1/6th the default) and I change the Model type from Stable to Exponential. I don't see how either of those could cause this.
What might be causing this? If I want to guarantee the lower bound data results in lower krigged values in all locations, is there anything I can in order to force that?
أكثر...