topic Re: Co-kriging model in ArcGIS GeoStatistical Analyst Questions

Co-kriging model

PaulineScheelbeek — Tue, 22 Oct 2013 09:10:00 GMT

Dear all,

I am a junior researcher and not fully experience in Geostatistical Analyst, so I hope someone can help me with the following.

I have water salinity data (approximately 100 samples) and I would like to create a prediction map of water salinity in the study area. I used an universal kriging model for this, however there are a few variables that I would like to add to the model, such as "distance to closest river" (already calculated / continuous variable), time point in the dry season (early-mid-late dry season/ categorical variable), and 3 other variables

I believe I have to use a cokriging model for this, however I have never used it and am experiencing problems in determining what parameters to use.

Is there anyone very experienced with these models that I could contact?

Thanks a lot,

Charlotte

Re: Co-kriging model

EricKrause — Tue, 22 Oct 2013 15:14:54 GMT

Fitting a cokriging model isn't very different than fitting a kriging model, but it will take more time and more work.

On the first page of the Geostatistical Wizard, kriging lets you input up to four datasets. The first dataset is the one you will interpolate, and the other three are cokriging datasets. In kriging, you fit a single semivariogram. In cokriging, you will need to fit semivariograms for each dataset and cross-covariance curves between each pair. At the top of the semivariogram screen, you'll see something that says "Var1 - Var1". That means you're fitting the semivariogram for the primary dataset. "Var2 - Var2" is the semivariogram for the second dataset. "Var1 - Var2" is the cross-covariance curve for the first and second dataset. With more than two datasets, there will be more combinations. You'll need to simultaneously fit all of these semivariograms and covariance curves, which can be difficult. I would definitely recommend using the Optimize Model button to get you started.

Re: Co-kriging model

PaulineScheelbeek — Wed, 23 Oct 2013 10:16:17 GMT

Dear Eric,

Many thanks for your reply! I tried some different models (thanks so much for the "Optimize Model" tip, that really helped me getting started!, but now think I might just make kriging 3 models (1 for each early - mid - late dry season), rather than a cokriging model adjusting for time of the season.

I was wondering however how you can change the window of prediction. I would like to cut the prediction area off along the borders of the main rivers. Any tips on how to do that?

Again, many thanks! Really appreciated that people reply on this forum!!!

Charlotte

Re: Co-kriging model

EricKrause — Wed, 23 Oct 2013 19:48:21 GMT

You cannot clip geostatistical layers. However, if you convert your geostatistical layer to a raster with GA Layer to Grid geoprocessing tool, you can clip the raster. GA Layer to Grid also supports the Mask environmental setting, so you can provide your area of interest as either a feature or a raster.

Re: Co-kriging model

SteveLynch — Wed, 23 Oct 2013 20:57:28 GMT

Exercise 1 will guide you through the process of blanking out part of the geostatistical layer.

This comes in handy when you are preparing a graphic to use in document.

Steve

Re: Co-kriging model

PaulineScheelbeek — Thu, 24 Oct 2013 11:56:05 GMT

Thanks a lot! That worked well Eric!
And also thanks for the tip Steve, good explanation on the settings of all parameters!

Charlotte