Project 6: Kriging Using The Geostatistical Wizard

Preliminaries

To use the Geostatistical Analyst, look along the main menu listing and choose Analysis. Then, along the tool ribbon, click on the Geostatistical Wizard toolbar icon. The dialogue window will appear.

'Simple' kriging

We use the Geostatistical Wizard to run kriging analyses. 

1. The Geostatistical Wizard: Choose Method and Input Data dialog (Figure 6.8).
   Begin by selecting Kriging/Cokriging from the list. Next, specify the Input Data (your random spot heights layer) and the Data field (Height) that you are interpolating. Then click Next >
   

   Figure 6.8: The Choose Method and Data Input screen of the Geostatistical Wizard: Kriging/CoKriging user interface.

   Credit: Kessler
2. The Geostatistical Wizard - Kriging dialog
   The next step allows you to select the exact form of kriging you plan to perform (Figure 6.9). You should specify a Simple Kriging method and Prediction output type. Under the Transformation type, select None. Note that you can explore the possibilities for trend removal later, once you have worked through the steps of the wizard. If you do so, you will be presented with an additional window in the wizard, which allows you to inspect the trend surface you are using. For now, just click Next.
   

   Figure 6.9: The Kriging screen of the Geostatistical Wizard: Kriging/CoKriging user interface.

   Credit: Kessler
3. The Geostatistical Wizard - Semivariogram/Covariance Modeling dialog (Figure 6.10).
   
   Here you should see the semivariogram plot on the left by selecting the Function Type. A goal with kriging is to choose a model that "fits" the scatter of points. There are several models from which to choose. These models are found under the pull-down menu located to the right of the Model #1 option. Choosing different models automatically updates the "fit" (or blue line) in the semivariogram. You can experiment with choosing different models. Note that you may not immediately see a difference in the fit of any given model to the scatter of points. 
   
   There is also an option here to specify assuming Anisotropy (True/False) in constructing the semivariogram (you may want to refer back to the readings in the text on understanding this idea better). This will be an important concept later, when it comes to the deliverables for this part of the project.
   
   When you are finished experimenting, click on Next.
   

   Figure 6.10: The Semivariogram/Covariance Modeling screen of the Geostatistical Wizard: Kriging/CoKriging user interface.

   Credit: Kessler
4. The Geostatistical Wizard - Searching Neighborhood dialog (Figure 6.11).
   
   Here you can get previews of what the interpolated surface will look like given the currently selected parameters. Also, specify how many neighbors to include using the Neighborhood Type options. You can specify how many Neighbors to Include in the local estimates and also how they are distributed around the location to be estimated using the Sector Type options. The various 'pie slice' or sector options define several regions around the estimation location. These sectors are each required to contain the defined number of neighboring control points, as specified by the minimum and maximum neighbors options.
   
   Together, these options enable a wide range of neighborhood types to be applied, and you should feel free to experiment. Often, you will find that it takes dramatic changes to the neighborhood specification to make much noticeable difference to the outcome surface.
   

   Figure 6.11: The Searching Neighborhood screen of the Geostatistical Wizard: Kriging/CoKriging user interface.

   Credit: Kessler
5. After this stage, click Next and then Finish. (We won't be going into the Cross Validation screen in the Wizard, although you can explore it, if you like!) As a suggestion, the different parameters listed on this screen give an assessment of the overall error in the kriging. You can compare one or more of the values on this screen between the Anisotropy option set to True and False. The Geostatistical Analyst makes a new layer and adds it to the map. You may find it helpful to right-click on it and adjust its Symbology so that only contours are displayed when comparing this layer to the 'correct' result (i.e., the pa_topo layer).

Things to do...

The above steps have walked you through the rather involved process of creating an interpolated map by kriging. What you should do now is simply stated, but may take a while experimenting with various settings.