Water-quality data analysis protocol development
Date
1990
Authors
Harcum, Jonathan Brooks, author
Loftis, Jim C., advisor
Ward, Robert C., advisor
Hirsch, Robert M., committee member
Salas, Jose, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Several agencies have developed networks to routinely monitor water quantity and quality in an attempt to assess society's influence on the environment, including the impacts of modern agriculture. Data from these networks are often plagued with attributes that inhibit analysis and interpretation. As more and more emphasis and public pressure is placed upon demonstrating environmental results, it is increasingly necessary that a consistent protocol for analyzing data from water quality monitoring networks be developed. Common data record attributes which inhibit data analysis include distribution applicability, variance heterogeneity, seasonality, serial correlation, extreme events, censoring, erroneous observations, small sample size, missing values, different sampling frequencies, multiple observations and measurement uncertainty. Each data record attribute is described in this study. In establishing a protocol to analyze water quality data, the handling of censored data and detection of trends in the presence of serial correlation and missing data are particularly difficult to quantify. This study focuses on these issues of protocol development. Seventeen procedures are evaluated for estimating the mean, median, standard deviation and interquartile range from data sets with singly and multiply censored observations. The results from this evaluation support previous investigations. In addition, the "no censoring” rule was found superior to methods which used censored observations for estimation of the mean, median and standard deviation. This study also compared the use of the Mann-Kendall tau test (and variations) for evaluating monotonic trends in water quality data. The Seasonal Kendall (Mann-Kendall) tau test should be used for data records with no serial correlation and five or less (ten or more) years of record. An ideal test for short data records which have serial correlation was not found in this study. The Seasonal Kendall tau test with serial correlation correction should be used for data sets of at least ten years of record and serial correlation. Furthermore, if monthly data sets have on the order of 40 to 50 percent missing values, monthly data should be collapsed to quarterly data by computing seasonal means or medians.
Description
Rights Access
Subject
Water quality -- Measurement
Water quality management
Agriculture -- Environmental aspects
Environmental monitoring
Time-series analysis