Hi,
I encountered some problems into the Geostatistical Analyst processing tools.
First of all, since IDW and RBF are exact interpolator, shouldn't they provide errors all equal to zero in correspondence of the training points? this holds also for the previous version of the software, but the errors are not zero.
Secondly, in the previous version of ArcGIS it was possible, before choosing an interpolation model, to specify a training dataset and a validation one; then, continuing in the steps of choosing the model parameters and looking at the results, it was also possible to save the complete report in a .dbf file of both the results on training points (it was called "CrossValidationResults") and those on test points (it was called "ValidationResults").
Now, in the ArcGIS 10, it's not possible to select a test set of points, and moreover it's not possible to save any report regarding the results on the training points: infact, selecting "Export Result Table", the only way is to save a shapefile, but the associated .dbf file is completely empty (it has just the name of columns: "Measured", "Predicted", "Error" and so on, but there isn't any value). So it's possible just to know mean and RMS, but not to export the complete report.
Regarding the test points, I know it's possible to right-click on the generated surface and select "Validation/Prediction": in this way, selecting the test set of points, it's possible to save a shapefile and fortunately also to open its associated .dbf file with all the values I need. By comparing these procedures with those of the previous version, I discovered that:
- in the previous version, choosing the same interpolation model (with the same parameters) the report on the test set of points were different, depending if we generated it from the interpolation menu or by right-clicking on the generated surface (the numbers were similar, but not the same!);
- in this version, as said before, I think that only the second way is possible (right-click on the generated surface, and saving of the report): but, for the same models (and same parameters), the results of the report on test points were different from the ones of the previous version!
Do you know some explanation about this? do you know if it is possible also to save the entire report about training points?
Thank you very much,
Marco
