WATER RESOURCES RESEARCH, VOL. 27,NO. 3, PAGES 327-339, MARCH1991 Trends in Stream Quality in the Continental UnitedStates, 1978-1987 DENNIS P. LETTENMAIER, ERICR. HOOPER, COLIN WAGONER, ANDKATHLEEN B. FARIS Department of CivilEngineering, Universit3' of Washington, Seattle The nonparametric seasonal Kendall's test, and a related multivariate test were used to analyze 403 National Stream Quality Accounting Network (NASQAN) stations for trend for the period !978-1987. Nominal sampling frequencies were quarterly orbimonthly. For all groups and individual constituents, trends were detected foronly a minority of stations atthe relatively liberal 10% significance level. The groups withthegreatest percentage of trends were common ions {,mostly upward) and nutrients and suspended sediment (mostly upward fortotal nitrogen and mostly downward fortotal phosphorus). In thepH andalkalinity group, the trends were dominantly upward; while thismight be the result of reductions inatmospheric deposition, this hypothesis isclouded bythe existence ofuptrends in sulfate at many of thesame stations. Thetrace metal analyses showed that theconstituents with thegreatest numberof trends (dominantly downward) were arsenic and cadmium; most of the stations with downtrends were in the major population areas of the east andmidwest. Additional exploratory analyses of possible relationships between trends andlanduseandpopulation did not give strong evidence of possible causation. INTRODUCTION Since passage of the Federal Clean Water Act of 1972, well over $100 billion of federal and local funds have been expended on municipalwastewatertreatmentplantsthrough the Construction Grants Program. Comparable expenditures of private funds have been made to meet the effluent guidelines for industrial discharges mandated by the Act and its 1977 amendments. In addition, large amounts of public and private monies have been expended to protect streams and rivers from nonpoint source pollution, such as urban stormwater, and changes in agricultural and other land use practices (e.g., forestry and mining) that affect water quality. The present condition of the nation's streams and rivers, and indicationsof how that condition might be changing,are therefore of considerable interest. The assessmentof long-term water quality changesis a challenging problem. Water quality data tend to be poorly behaved statistically, that is, they do not usually follow convenient probability distributionssuchas the well-known normal and lognorma! distributions onwhich many classichi statistical methods are based (see, for instance, Gilbert [1987]).These qualitiesmake the application of parametric statisticalmethods (that is, methods that make use of the numerical values of the observations directly, and which therefore require knowledge of the probabilitydistribution from which the data were drawn) inadvisable. Another complication is that the observation record of some water quality constituents, especially traceelements, may include high proportions of nondetectable values. Also, water qual- ity observations tend not to be collected uniformly in time, andsome water qualityconstituents maybe closely related to streamflow.In addition, the concentrations of somewater qualityconstituents exhibit strongseasonal variations. Overthe last 10 years,methods have been developed or adapted for the assessment of trendsin water qualitytime series that recognize these characteristics of the data. Hirsch et al. [1982] suggested an adaptation of the Kendall nonpa- rametric test for detection of trends in seasonallyvarying Copyfight 1991 by the American Geophysical Union. Papernumber 90WR02140. 0043- ! 397/91/90WR-02140505.00 water quality time series. The method essentially computes Kendall statistics for each season individually, then sums over seasons. For this reason, it is often referred to as the seasonal Kendall's test. Hirsch et al. also suggested that, for those constituents that are highly streamflow-dependent, the confounding effects of streamflow variations be removed by analyzing the residualsfrom a flow-concentrationrelation- ship for trend, rather than the raw data. Hirsch et al. hypothesizedtwo general classes of flow-concentration re- lationships: a dilution effect, for which concentration de- creases with flow, and an enhancement effect, for which concentration increases with flow. Several alternative math- ematical forms were suggested for each class of flow- concentration relationship. Hirsch and Slack [1984] sug- gested a generalization of Hirsch et al.'s nonparametric seasonal Kendall's method that allows for serial correlation in the water quality constituents (that is, correlation between the observationsin subsequentseasons).Lettenmaier [!988] suggested a multivariate extension to the seasonal Kendall's methodthat allows simultaneous analysis for trend of groups of water quality constituents. Several investigatorshave previously attempted national assessments of water quality trends. One of the first efforts was by Wolman [ 1971],who was frustrated by the absence of a comprehensive and consistent long-term observation record. In part because of Wolman's experience, two na- tional water quality data collection programs, the U.S. Environmental Protection Agency's National Water Quality Surveillance System (NWQSS) and the U.S. Geological Survey's National Stream Quality Accounting Network (NASQAN) were initiated in the early 1970s.Both networks employed fixed stations, with periodic observationstrategies (initially monthly). Most NWQSS stations were located in pairs upstream and downstream of major water quality influences (such as cities), whereas NASQAN stationswere generally located at the downstreammost stream gaging stationin a U.S. Water ResourcesCouncil AccountingUnit (of which there are about 500 in the continental United States). NWQSS was discontinuedin the early 1980s,but a number of NWQSS stations were absorbed into NASQAN. NASQAN, which in 1981 had in excess of 500 stations sampled monthly for common ions, trace metals, nutrients, 327