Application of partial least squares regression for sample-specific lithium geochemical background estimation and anomaly identification: a case study from Northeastern Hunan Province, China

Lithium (Li) geochemical background concentrations exhibit pronounced lithological dependence, challenging reliable resource prospecting in heterogeneous terrains. Accurate, spatially resolved background estimation is essential for robust anomaly detection. This study applies partial least squares regression (PLSR) to a 1:200,000-scale stream sediment dataset from northeastern Hunan, China, developing a site-specific Li background model. Nine lithologically sensitive elements (Al 2 O 3 , Fe 2 O 3 , MgO, CaO, K 2 O, Sr, Ti, V, Y) were selected as predictors based on geochemical coherence with Li and rock-type responsiveness. Leave-one-out cross-validation (LOOCV) was used for component selection and sample-wise cross-validation, and spatial block cross-validation was further conducted to assess model applicability. The resulting regression equation in clr-transformed space is: Li ( b a c k g r o u n d ) = 3.83750 - 0.17707 × S r - 0.35090 × T i - 0.17547 × V - 0.32998 × Y + 0.49467 × Al 2 O 3 - 0.08215 × C a O - 0.30069 × Fe 2 O 3 + 0.10333 × K 2 O + 0.25553 × MgO . Using this model, Li background values were computed for all samples in clr-transformed space. Li anomalies were identified by comparing observed clr-transformed Li values with the upper prediction limit of the PLSR-derived background estimate, yielding 115 anomalous samples and enabling spatial delineation of geochemical anomalies. Performance was evaluated against two benchmarks: a deep autoencoder and the conventional [mean + 2 standard deviations] criterion. Results demonstrate that PLSR effectively decouples anomalous signals from lithological background trends. These sample-level estimates substantially reduce false anomalies in high-background zones and enhance sensitivity to weak mineralization in low-background settings. Notably, in areas with significant background variation, the PLSR method produces Li anomalies that are more geologically consistent than those identified by the DAE and the [mean + 2 standard deviations] method. This work establishes a reproducible, interpretable, and geologically grounded framework for Li exploration in lithologically complex regions, applicable to other pathfinder elements under strong lithological control