A LISA-DBSCAN framework for screening multielement geochemical hotspots and raw-profile similarity domains

Accurate screening of soil trace-element anomalies is needed for risk control in geochemically heterogeneous regions, where spatially discrete hotspots may retain partly redundant multielement profiles. We developed a data-driven framework integrating local indicators of spatial association (LISA), density-based spatial clustering of applications with noise (DBSCAN), and cosine angle homology auditing. The framework identifies multielement hotspots, defines operational high-high candidates, delineates spatially continuous composite halos, and screens intercluster compositional redundancy using a 15° cosine angle criterion. Applied to 1330 soil samples and 14 target trace elements from Jieyang, the workflow identified 89 candidate multielement sites, which DBSCAN classified into five spatial hotspots. Cluster 1 remained consistently separated across raw-profile and scale-adjusted representations, whereas Clusters 2-5 formed a connected similarity domain only in the raw concentration space and showed metric-dependent internal differentiation after scale adjustment. This contrast indicates that apparent compositional redundancy among spatially separated hotspots can depend on elemental scaling and similarity representation. The framework flags potential compositional redundancy in fragmented hotspot maps and supports targeted field investigation of candidate anomaly signals in heterogeneous soil systems.