with and without declustering, are compared. Overall modeling parameters were simplified and held constant to isolate the effect of hierarchical ordering in the simulations. The analysis is based on the Jura dataset (Goovaerts, 1997). This dataset comprises spatially distributed point samples with geochemical variables, land-use information, and rock classes. In this study only the rock variable is considered. This variable is composed of five mutually exclusive categories defined over a local coordinate system. Although sequentially numbered, these categories do not imply any chronological or geological ordering. Five new columns were created in the data table, each representing one rock class, labeled Rock_1 to Rock_5. Each indicator takes value 1 when the original Rock variable matches the corresponding class, and 0 otherwise (Table 1). Figure 1a shows a point map of the sample locations colored by their Rock values. As observed, the sampling exhibits an incipient spatial pattern, but with some areas more densely sampled than others. Such clustering may lead to biased estimates of global proportions if sample proportions are used directly. To obtain more representative global proportions, a nearest-neighbor interpolation is applied to mitigate the effects of sample clustering. Figure 1b presents the nearest neighbor interpolation map obtained using a search radius of 500 m. Figure 2 compares the original global proportions derived directly from the sample data (a) with those obtained via nearestneighbor interpolation (b). Table 1 – First and last records from the dataset used in this study. Sample number X (m) Y (m) Rock Rock_1 Rock_2 Rock_3 Rock_4 Rock_5 1 2386 3077 3 0 0 1 0 0 2 2544 1972 2 0 1 0 0 0 … … … … … … … … … 258 3859 4022 3 0 0 1 0 0 259 2593 3312 3 0 0 1 0 0
RkJQdWJsaXNoZXIy MTM0Mzk2