Cross validation showed that triangulation and ordinary kriging were the two best estimation methods for drop-test data. Because replicate drops were made, an analysis of variance was performed to determine whether differences in line lengths due to the firefighting chemical and flow rate were significant. Also, cross validation helped to determine whether changing the grid spacing improved the accuracy of the results. Either triangulation or ordinary kriging are the recommended interpolation methods. If the grid spacing is changed, cross validation can be performed again to see which of the two methods is superior.

Replicate drops should be made whenever investigators need to know whether differences in line length are due to changes in factor levels or whether they are just a reflection of the inherent variability in the test. An analysis of variance can determine how much variability is due to changes in factor levels versus the variability inherent in the experiment. Many sources of variability are associated with drop testing. For instance, variability exists in how we measure wind, height, speed, flow rate, and volume. There may also be unknown variation in retardant cloud formation and deposition. The variance associated with predicting gpc values must also be considered. For more information on calculating the prediction variance of a triangulated gpc value, see appendix A.

The investigation into the sampling scheme reveals that increasing the spacing reduces the accuracy of the estimates. This fact must be weighed against the added time and cost of tighter spacing. While going from a 20-foot spacing to a 40-foot spacing is probably too large an increase, the cups could be spaced a little farther apart in the downrange direction without losing much information. In the crossrange direction, the present 10-foot spacing is recommended. A 5-foot spacing wouldn't give that much more accuracy, but it would cost much more in time and money. The appendix examines the predictive capabilities of 20- and 30-foot spacings.

Overall, drop testing gives us a relatively good idea of the performance of an airtanker in a controlled setting. Drop tests would be even more accurate if a permanent grid could be set up. This would allow greater consistency in the experiment.

Because gpc values from a drop test are used to calculate line lengths, it is important to remember that the gpc values are simply estimates. Specifications based on these estimates should probably be expressed in a range that reflects the variability around the estimate.