Kriging vs Spline with Barriers

2045
6
05-23-2019 09:47 AM
JenifferSoto_Perez
New Contributor II
Hi, I'm Jeniffer Soto from the Biogeochemistry group at USGS. We want to create a prediction surface for different water quality parameters along the Sacramento-San Joaquin Delta. Predictions are currently being completed using Kriging/Cokriging (we determine which is the best method using the summary report), and Spline with barriers. These prediction results will be extracted using the "Extract values to points" tool from the raster output to make temporal comparisons. We are also interested in generating a vector file for presentation purposes.

Our primary concern at the moment is which interpolation method is best suitable for our needs. For example, Kriging works best for showcasing purposes since we have the option of export as a vector, but some of the surface predictions are not accurate because it interpolates across the entire surface. On the other hand, we can set boundaries using Spline with barriers, so the interpolation occurs only within the river channels. The problem with this method is that the output is only a raster file. I can easily convert the prediction layer to vector, but the image will still be pixelated when you zoom in (this maps will eventually end up in a story map). I need a vector file so I can clip it to my delta polygon. Also, the Spline with barriers tool does not allow for much flexibility like the searching neighborhood, applying transformations to the data, and it does not give you a cross-validation summary report when the interpolation is complete. Again, I can calculate this myself, but it becomes too time consuming because we are generating over 100 maps.

If we want more information regarding background processes and algorithms applied for this interpolation methods, should I open an Esri case? We are interested in the possibility of setting barriers with Kriging or scripting Spline with barriers, so it produces a vector layer instead of having to convert from raster to polygon. We are not familiar with software engineering processes for Esri products, so we would like to have an understanding of what's possible with these tools. Let me know if you need additional information or have any suggestions.
0 Kudos
6 Replies
SteveLynch
Esri Regular Contributor

Kernel Interpolation with Barriers is also an interpolation tool that allows you to specify the output as a GA layer, a feature class or raster. Like kriging it also allows you to do cross validation to decide which interpolator and parameters to pick.

JenifferSoto_Perez
New Contributor II

Hi Steve,

I apologize for the late response. Thanks for the reminding me about Kernel with barriers, I have also tried Diffusion with barriers. The problem with these methods is the processing time. When the default properties are left unchanged, the surface does not cover the entire area (image attached). To fix this, I increase the bandwidth but then the wizard takes over 20 minutes to finish, most times even more. 

Diffusion Interpolation with barriers do not cover delta extent with default settings.

Diffusion Interpolation with increased bandwidth takes too long to run.

Kernel with barriers do not show prediction surface with default settings. Also, takes too long to run with increased bandwidth.

0 Kudos
SteveKopp
Esri Contributor

You can often improve performance when using barriers if you simplify the geometries as much as possible. For a first pass I'd try the SimplifyPolygon / SimplifyLine tool with POINT_REMOVE  option and a tolerance of 1 ft. See if these features still look good enough for you and haven't changed the input points used in a way you don't like (use the largest value you are comfortable with), then test to see how the performance improves.

JenifferSoto_Perez
New Contributor II

Hi Steve,

I actually tried the simplify tool with your settings, but the processing time is still pretty slow (~10 - 20 minutes). The polygon I'm using has very complicated geometry and I don't think I could get it fast enough for it to be efficient in my workflow. I have decided to stay with Spline with Barriers and using Focal Statistics along with the Contour tool to produce a smooth vector file. The problem I'm encountering now is a weird "halo" effect along the edge of the polygon, like a gradient in some cases. I wonder if is an artifact of Spline with Barriers. 

0 Kudos
SteveLynch
Esri Regular Contributor

Jennifer

Could you please post an image including;

  • the points and labels showing their values
  • legend for the raster
  • also show the barriers

and have you applied a mask?

Thanks

Steve Lynch

0 Kudos
JenifferSoto_Perez
New Contributor II

Here are the images requested:

I have not applied a mask.

0 Kudos