A practitioner’s guide for exploring water quality patterns using principal components analysis and Procrustes
To design sustainable water quality monitoring programs, practitioners must choose meaningful variables, justify the temporal and spatial extent of measurements, and demonstrate that program objectives are successfully achieved after implementation. Consequently, data must be analyzed across several variables and often from multiple sites and seasons. Multivariate techniques such as ordination are common throughout the water quality literature, but methods vary widely and could benefit from greater standardization. We have found little clear guidance and open source code for efficiently conducting ordination to explore water quality patterns. Practitioners unfamiliar with techniques such as principal components analysis (PCA) are faced with a steep learning curve to summarize expansive data sets in periodic reports and manuscripts. Here, we present a seven-step framework for conducting PCA and associated tests. The last step is dedicated to conducting Procrustes analysis, a valuable but rarely used test within the water quality field that describes the degree of concordance between separate multivariate data matrices and provides residual values for similar points across each matrix. We illustrate the utility of these tools using three increasingly complex water quality case studies in US parklands. The case studies demonstrate how PCA and Procrustes analysis answer common applied monitoring questions such as (1) do data from separate monitoring locations describe similar water quality regimes, and (2) what time periods exhibit the greatest water quality regime variability? We provide data sets and annotated R code for recreating case study results and as a base for crafting new code for similar monitoring applications.
Sergeant, C. J., E. N. Starkey, K. K. Bartz, M. H. Wilson, and F. J. Mueter. 2016. A practitioner’s guide for exploring water quality patterns using Principal Components Analysis and Procrustes. Environmental Monitoring and Assessment 188(4):1-15.