Optimizing Percentile Thresholds and Buffer Distances for Stream-Sediment Geochemical Anomaly Delineation: A Spatial-Efficiency Analysis from the Zhongdian Porphyry-Skarn Metallogenic Belt, Southwestern China

We jointly evaluated percentile threshold and buffer distance for stream-sediment geochemical anomaly mapping in the Zhongdian porphyry-skarn belt of northwestern Yunnan. The dataset comprises 4307 geochemical cells on a 500 m lattice and 13 mineralized targets. Seven elements and three ratios were tested at P85, P90, and P95 using six buffer radii from 250 to 2000 m. For each of the 180 configurations, buffers were dissolved, clipped to a fixed 1076.75 km² boundary, and measured directly. Performance was assessed by target capture and spatial efficiency, with Wilson intervals, 1000 spatial-block bootstrap replicates, blocked validation, and 999 max-statistic spatial permutations. Cu-P95 at 250 m gave the highest full-data efficiency (SE = 11.63), capturing 6 of 13 targets within 3.97% of the area. Pb-P95, Ag-P95, and Cu/(Pb + Zn)-P95 at the same radius ranked next. Seven configurations exceeded the search-corrected 5% familywise threshold of SE = 5.87. Capture favored a broader setting: Cu-P95 at 750 m captured 12 of 13 targets within 18.35% of the area, whereas extending the buffer to 1000 m added footprint without adding capture. The resulting eight-configuration efficiency frontier separates selective and higher-recall operating rules. Spatially blocked selection recovered P95 and 250 m as the modal setting, but global out-of-fold capture was 23.1% (95% CI: 8.2% - 50.3%), showing that the full-data maximum overstates transferable performance. The selected 250 m and 750 m configurations were finally converted into prospectivity-target footprints. The workflow is reproducible, but the numerical settings are specific to the gridded Zhongdian dataset and should be re-optimized locally.