Quantifying declustering-weight variability and risk across methods: Impacts on histograms and local uncertainty

Declustering methods were developed to reduce bias in statistical measures when samples are preferentially selected. In this case, the equal-weighted histogram of the variable of interest is not representative of the area of study. Declustering methods assign weights to samples so that the weighted histogram is more representative of the study area. These declustering weights depend on the chosen declustering method. Most declustering studies focus on comparing or developing new methods. However, these studies do not account for the variability in declustering weights across different methods. In this study, this variability is measured, and a risk measure related to declustering is proposed. The risk measure considers both the variability and the impact on the declustered histogram. The impact is measured by the distance from the global mean. The methods were illustrated in a case study. The results indicate that high declustering variability occurs in the transition zones between densely and sparsely sampled areas. High risk is related to high-grade samples in these transition zones. The study also investigates how the variability of the declustering weights affects local uncertainty measured by multi-Gaussian kriging. The study shows that the local uncertainty is more affected in the areas where the risk and variability related to declustering weights are higher.