Stochastic simulation for the comparison of methods for assessing equivalence and data leveling in geochemical mapping

Biases between geochemical survey datasets and procedures for dataset leveling are issues that have been discussed for a long time. Methods proposed in the literature are generally based on common sense, but their performance... [ view full abstract ]

Biases between geochemical survey datasets and procedures for dataset leveling are issues that have been discussed for a long time. Methods proposed in the literature are generally based on common sense, but their performance are not discussed with statistics and indicators figures. Here we propose a general procedure for randomly generating geochemical datasets and assessing the performance of methods applied in geochemical mapping by computer simulation. This procedure is illustrated by the performance comparison between two methods for testing the equivalence between geochemical datasets and for data leveling. The first compared method is based on the Kolmogorov-Smirnov test and quantile regression. The second method is based on BLS (Bivariate Least Squares) regression. Each simulation run consists in randomly generating a reference dataset and a biased dataset and applying the methods to test the equivalence between these two datasets and to level the bias between datasets. After completing 10 000 simulations, we compute several performance indicators aimed at answering the following questions: are the methods efficient to detect the bias introduced into the biased datasets? Are both methods efficient to remove the bias? What is the best method depending on the bias? The results of this study show that computer simulation is a useful tool to identify the particular benefits and limitations of each method, and provide practical recommendations for prudent use of these methods in geochemical mapping.