| Neural Networks as an Alternative to Statistical Modeling in the Semivariogram Analysis Portion of Krigging Procedures
ABSTRACT
The purpose of this study is to create and evaluate a hybrid interpolation procedure based on an extension of ordinary krigging procedures. This hybrid will replace the statistical model used in the semivariogram analysis portion of traditional krigging with an artificial neural network model. This hybrid approach should have the functionality of the popular krigging method combined with the superior performance of a neural network in recognizing patterns in noisy datasets. Previous studies have shown that when compared to traditional statistical techniques, artificial neural networks (ANNs) have superior capabilities in recognizing patterns and discovering relationships in noisy sample data. However, traditional krigging procedures rely upon standard statistical techniques to identify the relationship between lag distances and the semivariances between sample points, even though this relationship is often hidden in a very noisy dataset. ANNs provide an obvious and appealing alternative to these traditional statistical techniques. This study will develop interpolation models based both on traditional krigging procedures and upon a hybrid krigging/ANN approach for a standard forest inventory dataset. The absolute and relative accuracies of these interpolation procedures will be compared, and the strengths and weaknesses of the two approaches will be identified. |