_window.gif)
Cross-validation analysis is a means for evaluating effective parameters for kriging and IDW interpolations. Two types of validation are provided by GS+: Cross-validation and Jackknife analyses. In cross-validation analysis each measured point in the spatial domain is individually removed from the domain and its value estimated as though it were never there. Then the point is replaced and the next point is removed and estimated, and so on. In this way a graph can be constructed of estimated vs. actual values for each sample location in the domain.
In Jackknife analysis, estimates are compared against measured values for a set of locations different from those used as input data. Before performing jackknife analysis you must specify the jackknife data in a worksheet that appears when you press the Define command.
Cross validation and jackknife analysis are not available for Conditional Simulation.
Each point on the cross-validation and jackknife graph represents a location in the input data set for which an actual and estimated value are available. Information about individual points is provided at the bottom of the screen; points are displayed by placing the cursor on them. In the case above, the cursor was placed on the point representing record 100, as noted at the bottom of the window. By right-clicking on the graph you can list the data for all points.
The regression coefficient described at the bottom of the graph represents a measure of the goodness of fit for the least-squares model describing the linear regression equation. A perfect 1:1 fit would have a regression coefficient (slope) of 1.00 and the best-fit line (the solid line in the graph above) would coincide with the dotted 45-degree line on the graph. The standard error (SE = 0.162, above) refers to the standard error of the regression coefficient; the r2 value is the proportion of variation explained by the best-fit line (in this case 37.9%; it is the square of the correlation coefficient); and the y-intercept of the best-fit line is also provided. The SE Prediction term is defined as SD x (1 - r2)0.5 , where SD = standard deviation of the actual data (the data graphed on the y-axis).
If you have chosen to transform the input data without a backtransformation (see the Data Summary window), then the Estimated Z values will appear very different from Actual Z values regardless of the integrity of the interpolation system.
You may Print, Copy, Edit, Export, or List graph values for either graph using the menu commands of the main GS+ window, or via a right-click menu.