(NAWQA) (Gilliom et al., 2000; see Appendix A). That report describes the design and implementation strategy for Cycle II investigations in NAWQA study units. In this chapter, the committee considers the extent to which the NAWQA Cycle II design supports the three themes and six related objectives concerning the determination of trends in water quality and the association of spatial or temporal variations in water quality with urbanization and agricultural practices. In a larger sense, this chapter serves as a bridge between the statistical inference focus on water quality status assessments discussed in Chapter 3 and the scientific inference focus on understanding factors and processes that affect water quality in Chapter 5. To provide continuity between Chapters 3 and 5, this chapter includes an examination of statistical issues in trend analysis and causal inference in nonexperimental studies.
One cannot begin a discussion of trend assessment without first describing the design of water quality networks. It is only through carefully controlled water quality monitoring that trend detection is possible. Carefully controlled water quality monitoring includes attention to numerous issues including, but not limited to, site selection, determination of monitoring locations, sampling frequency and protocols, field operations, data reporting, laboratory protocols, and so on. Next, general water quality monitoring network design issues are discussed, including the development of concise monitoring objectives, the use of statistical methods for optimal design of water quality networks, and the accurate estimation of pollutant loads. When trends are the focus of a monitoring program, other important water quality monitoring design issues include the frequency and location of sampling, the length of the data series, the possible collection of collateral information that might be used to fill in missing data or augment the data series, and trend detection methods.
A sensible approach to the design of water quality monitoring networks is essential to the detection and evaluation of water quality trends—the topic of the third section of this chapter. Ideally, a water quality data series collected for trend analysis would include samples evenly spaced in time (e.g., monthly) that exhibit temporal independence and minimal measurement error. In reality, water quality data typically exhibit nonnormal distributions, seasonality, missing values, values below detection limit, changes in analytical detection methods, changes in sampling frequency and location, and serial correlation. Thus, from a practical standpoint, effective trend detection and assessment requires methods to deal with these features of real environmental data. The field of environmental statistics has emerged over the past few decades in an effort to address these complexities. Trend detection requires a rigorous background in environmental statistics. In this regard, many fundamental contributions to the field of environmental statistics have been developed and described by USGS researchers (e.g., Helsel and Hirsch, 1992).
The issue of causal inference (or scientific understanding) based on observational data is discussed in the next section of this chapter. The conventional view