Water quality trends in the southeastern plains and piedmont ecoregions and the application of power analysis

Abstract

This study contributes to the growing body of literature on water quality monitoring and offers important observations on issues of scale and hypothesis testing. In order to understand the interaction between land use/land cover (LU/LC) and water quality trends and to implement landscape level management options, analysis at large spatial scales is required. The study recognizes that this type of broad scale analyses is confounded by biases and factors ranging from small scale to large scale. Such biases include potential lack of representativeness of specific sites within monitoring networks to biases and inherent assumptions of the ecoregional framework that was employed. The two components of this study are: 1) determination of trends in water quality at multiple spatial scales and assessment of the association between LU/LC changes and trends; and 2) demonstration of power analysis to determine adequacy of monitoring period and/or effect size (change in constituent values over time) for trend detection. It examines trends of six water quality constituents in two adjacent ecoregions at multiple over two time periods. Trends were not observed for most constituents in many areas except pH. This constituent showed increased values throughout the region, and this trend appeared to be affected by LU/LC changes. The power of a statistical test is rarely reported when water quality trends are cited. Using the software TRENDS, the requisite minimum monitoring duration, effect size for trend detection and stations that met a priori power were determined. Water quality assessment is confounded by the characteristic of the information, and current and future water quality monitoring efforts should focus more on addressing this issue by designing programs and expanding analysis of the database to meet research needs while satisfying compliance requirements. In order to meet criteria for rigorous statistical tests of spatial and temporal patterns, monitoring design should include longer, uninterrupted recording periods and more strategically sampled sites that best represent the landscape. Reporting power results should be included to strengthen observations and analyses. Analyses that fail to detect trends should be viewed with caution since the data may not be robust enough to support the null hypothesis.