ArcGIS Tutorial: geostatistical comparison model

Comparisons help determine how good or bad the model of the geostatistical layer was created relative to other models. To compare models, there must be two geostatistical layers (created using the ArcGIS geostatistical Analyst extension) for comparison. These two layers can be created using different interpolation methods, such as inverse distance weighting and ordinary kriging, or they can be created using the same method but with different parameters. In the first case, you will compare which method is best for the data, and in the second case you will examine the different input parameters for the same model when creating the output surface. To compare two models, right-click the name of one of the models in the table of contents and click Compare as follows:

　　

The Compare dialog box uses the cross-validation statistics that are described in performing cross-validation and validation. But it allows you to check the statistics and graphs side-by In general, the best model should have the standard mean value closest to 0, the minimum RMS prediction error, the average standard error closest to the error, and the standard RMS prediction error closest to 1.

A common approach is to create multiple surfaces, identify one of them as the best, and then determine whether it is the end result or will be passed to a larger model, such as a suitability model for housing location, to solve existing problems. Each surface can be systematically compared to other surfaces, removing the poorer of the two comparison surfaces until the remaining best two surfaces are compared to one another. It can be concluded that for this particular analysis, the better of the last two surfaces is the best possible surface.

　　

　　Considerations for comparing methods and models

There are two issues to consider when comparing the results of different methods and/or models: one is optimality and the other is validity.

For example, the RMS prediction error for a model may be smaller. It can therefore be concluded that it is the best model. However, when compared with another model, the RMS prediction error may be closer to the estimated average prediction standard error. This is a more efficient model, because when you make predictions at a point that does not contain data, you can only use the estimated standard error to assess the uncertainty of the forecast. You must also check whether the standard RMS is close to one. If the standard RMS is close to one, and the estimated average prediction standard error is close to the RMS prediction error based on cross-validation, you can determine that the model is appropriate. In the left-hand kriging model, the average averaging standard error is less than the right model, but the right-hand kriging model should be preferred, since it has a similar standard error of averaging. In addition, the left model has a very large standard RMS, which indicates that the model has serious problems.

In addition to the statistics provided in the Compare dialog box, prior information about the dataset and prior information from ESDA should be used when evaluating the best model.

　　Steps:

1. Right-click one of the geostatistical layers you want to compare in the ArcMap table of contents and click Compare.

2. In the Compare Object drop-down menu, click the Second layer in this comparison.

3. Click each tab to see the different comparisons.

4, click Close.

ArcGIS Tutorial: geostatistical comparison model