Daughney, C.J. 2007 Multivariate statistical methods for assessment of groundwater chemistry between the Waingawa and Waiohine Rivers, Wairarapa Valley GNS Science report 2007/19 37 p.
Abstract: This investigation explored the use of multivariate statistical methods to provide insight into the groundwater chemistry in the Wairarapa Valley, covering the area between and near the Waingawa and Waiohine Rivers. Prior to this investigation, Greater Wellington Regional Council (GWRC) had defined seven preliminary hydrostratigraphic units within this study area. An independent comparison to the GWRC conceptual hydrostratigraphy was provided by hierarchical cluster analysis (HCA), which was used to re-categorise the monitoring wells based on major ion concentrations without any consideration of well location, depth, or assumed hydrostratigraphic unit. Two major hydrochemical categories and nine subcategories were defined by HCA, and these were generally consistent with the GWRC conceptual hydrostratigraphy. However, HCA revealed some cases where a well’s hydrochemistry was inconsistent with the expectation for its assumed hydrostratigraphic unit. Thus discriminant analysis (DA) was used to predict the likelihood that each well taps into each GWRC hydrostratigraphic unit, on the basis of major ion chemistry and well depth. The results of DA were also generally consistent with the GWRC conceptual hydrostratigraphy, with the DA prediction matching the assumed hydrostratigraphic unit for 75% of the monitoring wells (n = 99). Most of the wells for which the DA prediction did not agree with the GWRC unit assignment were clustered in three parts of the study area: 1) near the confluence of Managatere and Waiohine Rivers, where DA suggested that older sediments are closer to the surface than the GWRC conceptualisation implies; 2) along a line roughly parallel to the axis of the Tararua Ranges, where DA suggested that fan gravels are thinner than the GWRC conceptualisation implies, or absent altogether; and 3) for shallow sites in the Parkvale sub-basin, where DA performed poorly where hydrochemistry is controlled more by local land use than by regional hydrostragraphy. Overall, this investigation has shown that multivariate statistical methods can be valuable for the development and validation of a conceptual hydrogeological model. (auth)