Analysis

The Analysis window summarises the general accuracy of the block model using a leave-one-out cross-validation analysis. This iteratively removes each borehole reading one at a time and estimates the result at its location using nearest neighbour, inverse distance weighting and kriging. The leave-one-out prediction is then compared to what the true value was at each point and the cumulative percentage error is plotted on a probability of non-exceedance plot. A distribution of the parameter values across the borehole readings is also provided to assist in understanding the cross-validation output.

Select the domains to be analysed, and the results within these domains will be automatically plotted.

An estimation of the accuracy of the results for each block can be made based on both the leave-one-out cross validation (e.g. in the example above, > 90% of the model is within a 10% error for RQD, while only 50% of the model is within a 10% error for log Q') and the relative confidence rating. Use these measures to evaluate the level of 'trustworthiness' of the model's predictions, and to decide what the most accurate interpolation approach is likely to be.