conclusions:
This investigation tested the accuracy and precision of interpolation methods available in the ArcGIS 9.3 Geostatistical Analyst Wizard. The results were compared to a previous spatial statistics study by Flores-Garnica (2001) that utilized the interpolation methods available in ArcView 3.1, GS+ and S-Plus statistical software to predict forest fire fuel loads. The statistical results from the 1-hour and 10-hour fuel groups were evaluated as examples of the method.
Based upon the mean square of the errors (MSE) criterion of Flores-Garnica (2001) this study finds close agreement with the conclusions of the previous investigation. The relative precision of the interpolation methods is confirmed. This is important because interpolation is an iterative process, therefore different software application utilizing different algorithms, may result in slightly different results (e.g. Pebesma 2003; Davis and Ierapetritou 2008).
Although Ordinary Kriging is regarded as the best (i.e. most precise) unbiased linear estimator, problems can arise when the data distribution does not satisfy the tests for normality or otherwise shows a biased distribution (Mardia 1980). In this case there is strong evidence for a biased fuel load distribution based upon changes in elevation. This causes the accuracy of the Ordinary and Universal Kriging methods to fall short, in spite of their higher precision (Tables 7-8). Inverse Distance Weighted interpolation with a 2nd order polynomial solution provides the most consistently accurate estimates for both 1 Hour and 10 Hour fuel loads, and requires no data transformation. Cokriging with elevation to account for the biased fuel distribution provides the next most accurate estimate for the 1 Hour fuels. These results are also consistent with the findings of Flores-Garnica (2001).
Continuing investigations should consider applying the Fisher’s F-Ratio test to each fuel type to confirm spatial dependence of the data. The Geostatistical Analyst Wizard allow for data transformations and corrections for anisotropy, but specific options must be investigated further in order to select the best option(s). This approach should improve the results of both Ordinary Kriging and Cokriging.
