Using linear discriminant analysis (LDA) of bulk lake sediment geochemical data to reconstruct lateglacial climate changes in the South Carpathian Mountains

A multi-proxy approach is widely used for reconstructing climatic change in alpine lake sediments. This study applied bulk sediment geochemistry to reconstruct lateglacial and early Holocene climatic change in a glacial lake (Lake Brazi, 1740 m a.s.l.) in the Retezat Mts. (South Carpathians, Romania). The lowermost 1 m part of a 4.9-m long sediment core, covering the period between 9950 and 15,750 cal BP, was used for high resolution bulk analysis of major elements (Al2O3, SiO2, TiO2, CaO, MgO, K2O, Na2O, Fe2O3, MnO, SO3). Linear discriminant analysis (LDA) was used to compare a priori classified main chemical groups. Subsamples from the core were priory ordered to warm and cold groups respectively, according to their age and evidence of cold and warm events in the record, as suggested by proxy correlation with the lateglacial event stratigraphy of North Greenland Ice Core Project (NGRIP). The discriminant function was calculated using concentrations of Al2O3, TiO2, CaO, MgO, K2O, Na2O, Fe2O3, and MnO after log ratio transformation. Loss-on-ignition, silicon and sulphur concentrations were not used for the discriminant analysis, but regarded as comparison proxies for checking up the validity of outputs. Sediments deposited during cold and warm events were separated significantly by the LDA function; 85.1% of the originally grouped cases were correctly classified under the LDA analysis. The calculated discriminant scores indicated four cold and four warm events. Sediments ordered into the warm group contained larger amounts of organic matter and sulphur, while sediment samples of the cold group were characterized by the overall predominance of major oxides bound into inorganic silicates. The discriminant scores showed strong correlation with the NGRIP delta O-18 data (r = 0.8135) and with the pollen percentage sum of trees and shrubs (r = 0.9460). Loss-on-ignition showed a somehow weaker, but still conclusive linear relationship with the discriminant scores (r = 0.7505). Discriminant analyses of bulk sediment major oxide chemical data may be a useful tool to identify the impact of climatic events upon the nature and composition of materials delivered to a lake basin. (C) 2012 Elsevier Ltd and INQUA. All rights reserved.