Lab3: Ordinary Kriging

 

 

 

Geo597: Geostatistics

Spring 2009

 

Exercise

 

I. Adjust the search neighborhood

In the "Geostatistical Wizard: Step 3 of 4 - Searching Neighborhood" window:

(1) Select "Dataset 1" for "Dataset Selection". Choose "Neighbors" for "Preview Type". Examine the weight assignment among included neighbors in the ellipse. For example, if you see "3" in the red portion of the legend and "10%" next to it, it means that there are three red dots in the search neighborhood and each receives 10% weight.


(2) For "Neighbors to Include", select 3-4 neighbors and "Include at Least" 2. Then play with the "Sector Type". In the shape section, clear the "Default" box, then adjust the angle in "Angle", the range in the major direction of spatial continuity in "Major Semiaxis", and the range in the minor direction of spatial continuity in "Minor Semiaxis". Print the screen and complete assignment 1. -> Next.

 

II. Cross Validation
1. Examine cross validation results.
In the "Geostatistical Wizard: Step 4 of 4 - Cross Validation" window:

(1) In the top portion, select "Predicted" plot and print the screen. Switch to the "Standardized Error" plot and print the screen.


(2) Assess the unbiasedness of the estimation by examining the means: the "Mean" error, the "Root-mean-square" error, and the "Mean standardized" error. All values should be close to zero.

Root-Man-Square error = square root of the average of the squared errors

Mean Standardized error =  Mean [(absolute-error) / (standard-deviation-of-the-estimation-error)]


(3) Assess the variation of the error by examining the "Average Standard Error". The value should be small. The "Root-mean-square standardized" value should be close to 1.

Average Standard Error = standard-deviation-of-the-estimation-error

Root-Mean-Square Standardized error = (Root-Mean-Square) / (Average-Standard-Error)

If Root-Mean-Square < Average-Standard-Error, then the variability is over estimated, otherwise underestimated

 

2. Compare semivariogram models

Go back to steps in Lab 2, and select another semivariogram model to fit the sample semivariogram. Repeat steps in search neighborhood and cross validation. Print the "Predicted" and the "Standardized Error" plots corresponding to the new model, along with the associated summary statistics. Complete assignment 2. -> Finish.

 

III. Kriging

1. Kriging

Examine the "Summary" window. -> OK. Kriging result now should appear in ArcMap.

 

2. Examine the kriging estimation.


(1) "Display" the kriging layer. Right click the kriging layer name -> Properties -> Symbology. Check "Grid' in 'Show", then change color, value range, and the number of categories if  needed.


(2) Display both the kriging layer and the original sample data in "Display" view (lower left corner of the screen). Left click the name of the sample data layer, then hold and drag it above the name of the kriging layer. To match the color scheme of  the sample points to the kriging map, first check the classification scheme used by kriging. right click the kriging layer -> Properties->Symbology->Grid->Classify, then in the Classification section, it shows "10" classes and "Geometric Interval" method. Cancel this window, and use the same number of classes and the same method to classify the original sample points.

 

(3) Examine the consistence between the measured values and the estimated values. If the estimation is not satisfactory, go back to the Geostatistical Wizard and readjust the semivariogram modeling and search strategies. If satisfactory, print a layout map of  the kriging surface and the original sample data. Complete assignment 3.

 

 

Assignment (Due: April 16)

1. Print the search neighborhood for the kriging and briefly describe the number of neighbors used and the weight distribution among the neighbors.
2. Print cross validation results using two different semivariogram models.

3. Print the kriging result showing both the krigged map and the original sample values.