Kernel Interpolation with Barriers - unexplained results

[danielfenner]

I ran a Kernel Interpolation with Barriers using a stream buffer as my barrier.  Most of the results make visual sense to me, but there are results in tributaries of the mainstem river that I have no explanation for:



The data points I used to create the stream raster are included in the image above.  I've circled the incorrect results in blue.  These are at least the obvious ones, where instead of continuing to decrease in value (towards the red color direction) the values began to increase.

Can someone explain why this is occurring and what options are available to correct for this error?

Input variables are as follows: Output cell size 0.000833, Kernel function: Polynomials, Order polynomial: 1, Ridge parameter: 50, Output surface type: Prediction

Thanks!

Daniel

[EricKrause]

Can you send your data to ekrause@esri.com?  I can't tell what's going on just by the picture.  Thanks.

[danielfenner]

Can you send your data to ekrause@esri.com?  I can't tell what's going on just by the picture.  Thanks.

I just sent the data to your email....thanks!

[EricKrause]

Daniel and I have talked over email, and to sum up the conversation, the problem is that he's fitting a linear trend, and he doesn't have very many data points.

You can see in the two attached graphics that when the searching neighborhood moves away from the large value, the extrapolated predictions actually go up.  It's counter-intuitive that removing a large value causes predictions to increase, but you can get artifacts like this when you don't have many points.