Formatted Title
Application of Principal Component and Hierarchical Cluster Analysis to Delineate Hydrogeochemical Units at Fractured Rock Sites
Background/Objectives
The complexity of fractured bedrock sites makes development of conceptual site models (CSMs) and remediation strategies challenging. To address this, we developed Python scripts to apply principal component analysis (PCA) and hierarchical cluster analysis (HCA) to a large hydrological and geochemical data set from a complex fractured rock site with the goal of identifying groupings of wells that reflect hydrogeochemical units.
Approach/Activities
Data collected over many years at the site include chemical (volatile organic compounds [VOCs] and inorganic parameters) and hydraulic (transmissivity, storativity) characteristics. A phased approach was taken to implement PCA and HCA in which feature sets (e.g., chemical or hydraulic data) were first evaluated individually. Each feature set included data from between 65 and 169 individual wells. Once PCA and HCA were performed for each feature set individually, the results were reviewed to choose combinations of features from across data sets for further analysis to generate more holistic insights.
Each application of PCA and HCA followed the general steps of (1) calculating initial summary statistics to gain a better understanding of variability and distributions, (2) generating a Pearson correlation matrix to assess whether there are high correlations among variables supportive of reducing the features into a smaller number of components, (3) performing PCA, (4) visualizing the output of the PCA with three-dimensional (3D) charts of the factor loadings to determine which features most influenced each loading and two-dimensional (2D) and 3D scatter charts of the PCA-transformed values to identify potential groupings, (5) performing HCA, and (6) visualizing the output of the HCA together with the PCA by applying cluster labels as a color-scale to 2D and 3D scatter charts of PCA-transformed values.
Once the PCA and HCA results were used to assign wells to individual groups, the groupings were evaluated within the context of other analyses performed at the Site. Steps taken to integrate the PCA and cluster results within the larger CSM development effort include visualizing the results spatially using GIS and comparing the groupings to hydrogeochemical unit assignments based on professional judgment and field observations.
Results/Lessons Learned
Meaningful groupings identified by the PCA and HCA were apparent in 2D and 3D scatter charts of PCA transformed values. When labels assigned by the HCA were used as a color-scale on these charts, they generally aligned with the PCA results, providing multiple lines of evidence in support of the statistical groupings. The spatial distribution of cluster labels assigned during the VOC analysis led to a reevaluation of the hydrogeochemical units previously assigned to some of the monitoring wells. Analysis of geochemical parameters is in progress, and it is anticipated that results for geochemical parameters and combined analyses will be available to discuss at the time of the presentation.