From The Changing Illinois Environment: Critical Trends, Volume 2: Water Resources , Technical Report of the Critical Trends Assessment Project
During the last decade increased awareness and concern for the environment have changed human lifestyles and contributed to changes in industrial and commercial activities. Governments at the local, state, and federal levels are interested in assessing the impact of environmental policies that address the problem of externality and target entities that pollute. Water quality data collection began in the early 1970s, and a number of locations in Illinois now have long-term data. Consequently, it is now possible to evaluate trends in water quality and determine whether increased environmental awareness in conjunction with public policy for pollution abatement has contributed to a general improvement in water quality. This chapter describes a method for identifying trends in parameters that are generally used in water quality studies. The data sources and limitations are described briefly. Finally, the results of the analysis of water quality data are illustrated, followed by suggestions for future research.
DATA SOURCES AND LIMITATIONS
Data on water quality parameters for streams and lakes in Illinois are available in STORET, a database maintained by the Illinois Environmental Protection Agency (IEPA). To analyze and determine water quality trends in Illinois streams and rivers, data collected at water quality stations in the Ambient Water Quality Monitoring Network (AWQMN) were used (21ILAMB in STORET). The database contains observations for the period 1971 to 1991 at 204 stations in 15 different river basins in Illinois (figure 1). Not all parameters are recorded at all stations in the network. For example, phosphorus concentrations are measured at all 204 stations, while arsenic concentrations are available at only 67 stations in the AWQMN. Water quality data are also measured at Illinois lakes. To determine water quality trends in Illinois lakes, data collected at 650 lake water quality stations in the AWQMN were used in this study (21LLAKE in STORET) and their locations are shown in figure 2. The locations of the major Illinois drainage basins included in the IEPA database are shown in figure 3.
Water quality measurement procedures for parameters such as metals indicate whether the metal was detected and whether the reading was below a certain threshold value. This threshold value varies for different parameters. With lead, for example, the measurements indicate if the concentration was <5 parts per billion (ppb), or <50 ppb, or > 50 ppb. There is no acceptable procedure to convert such measurements to unique concentration values. Where exact concentrations are not available, the observations are first grouped by categories such as low, medium, and high, based on the frequency distribution of the observations and water quality standards established by the IEPA for that parameter. The percentage of observations that fall in each group is then computed for each parameter for each year. For example, for arsenic for 1985, 25 per-cent of the observations lie in the first group (0-1 ppb), and the remaining 75 percent lie in the second group (>1 ppb). Data for the other four metals--cadmium, chromium, lead, and mercury--and for chloride and phenolics were grouped into three categories. The percentage values in each category and the time were used instead of actual concentration values to determine temporal trends.
When exact concentration values are available, the data groupings are not needed. The mean values of all the observations in a particular year for each parameter are averaged--first for the entire state and then for each basin. The average concentration for each parameter is used to determine temporal trends.
Detection of temporal trends in water quality consists of the following five steps. Some assumptions needed for this test are explained later.
1. State the null hypothesis for the test. For this analysis, the null hypothesis is that the water quality concentration data have no temporal trend. 2. Calculate an appropriate test statistic to evaluate the time-series data. 3. Interpret the test statistic by comparing it with its known probability distribution. 4. If the test statistic falls within preselected values, then accept the null hypothesis. 5. If the test statistic does not fall within preselected values, then reject the null hypothesis. In this case we can conclude that the time-series data exhibit a statistically significant trend.
The limits are calculated for a preselected level of probability, denoted by a. Usually a value of a = 0.1 is selected. This corresponds to a confidence level of 10 percent, which is not a very high level of significance. An a value of 0.01, or a confidence level of 1 percent, however, indicates a very high level of significance. To detect trends at various levels of significance, three different values were selected for a, 0.1, 0.05, and 0.01, which correspond to confidence levels of 10, 5, and 1 percent statistical significance, respectively. The significance level is not very meaningful, however, when there are an inadequate number of observations. In this study, the significance level is not reported in cases with less than eight years of data.
One common test for trend is based on ordinary least- squares regression of the water quality variable as a function of time. The assumptions necessary to statistically determine the presence of trends begin with a dependent variable that is normally, independently, and identically distributed. Water quality data series frequently violate these assumptions, however, making the statistical test invalid.
Distribution-free tests that use the relative rankings of the data series rather than their magnitude do not require that the data conform to certain probability distributions. The distribution-free test employed in this study is the Kendall tau (Kendall, 1970). This procedure does not require the data series to be normally distributed, only that it is independently and identically distributed. The Kendall test does not provide a mea-sure of the magnitude of the trend, but only indicates whether or not a trend is present. This procedure has been used in many studies of water quality data. Some recent examples include a study of trends for total phosphorus (Smith et al., 1982) and the determination of surface water quality trends in Virginia (Zipper et al., 1992).
Kendall's Tau-b Association Measure
Kendall's tau-b is a nonparametric measure of association between two continuous variables. This rank measure uses the rank or order of the observations, rather than the actual value of the variables. It is based on concordance and discordance; that is, the degree to which values of two variables respectively vary together for pairs of observations. It has a range of -1 to 1.
To calculate the Kendall correlation, the observations are first ranked in terms of values of the first variable, and then in terms of values of the second variable. The number of interchanges that occur in the position of the first variable is noted and used to compute the Kendall tau-b. A correction is made for pairs that are tied.
Data Groups and Average Concentrations
Water quality measurement procedures for parameters such as metals only indicate whether the metal was detected and whether it was below a certain threshold value. This threshold value varies for different parameters. For example, in the case of lead, the measurements indicate if concentrations were <5 ppb, or <50 ppb, or > 50 ppb. The concentrations of arsenic, cadmium, chromium, lead, and mercury were all very low, and some values were reported as less than the detection limit. Other values reported as greater than the detection limit were still less than the limit of quantitation (Keith et al., 1983). Where exact concentrations were not available, the observations in such cases were first grouped in low, medium, and high categories, based on the frequency distribution of the observations and water quality standards established by the IEPA for that parameter. The percentage of observations that fall in each group is then computed for each parameter for each year. Kendall correlations were calculated between the percentage values in each category and the time needed to identify any significant change over time in the percentage of observations in that category.
When exact concentration values are available these data groupings are not needed. The mean values of all the observations in a particular year for each parameter are averaged for the entire state and for each basin. Then Kendall correlation tests are performed for each parameter between the "year" and the "average concentration." This procedure was used to determine time trends for all parameters for the entire state of Illinois as well as for each basin.
ANALYSIS OF STREAMWATER QUALITY IN ILLINOIS
The water quality time-series data underwent the Kendall analysis on an IBM 6000 RISC workstation using the Statistical Analysis System (SASc). A summary of the streamwater quality data used for the trend analysis is shown in table 1. The results of the Kendall correlation analyses of selected water quality parameters for streams averaged over the entire state of Illinois are given in table 2.
In general for all the metals, a decreasing trend was found in the levels of concentration. For chromium, however, the tau-b values were positive for the lower group and negative for the higher group, showing some movement in the observations from the higher group to the lower group. Although the highest groups for cad- mium, lead, and mercury showed a significant negative trend, the rest of the correlation values were low and insignificant.
The tau-b value of -0.63 or the >10 group for cadmium means that 60 percent of the pairs of observations were concordant. This indicates a moderate declining trend in cadmium concentrations over time. This association was also highly significant, and the probability of its being a null relationship was less than 1 percent.
Only the lowest value group in chloride exhibited a significant negative trend. The results show average and highly significant trends for chemical oxygen demand (COD) at 0.50 and for phosphorus at 0.46. COD concentration levels decreased over time, while phosphorus displayed a positive trend. There is also some evidence of an increasing trend, albeit small, for dissolved oxygen (DO).
The data for nitrate nitrogen (NO2+NO3) show a highly significant positive trend, indicating increasing levels of concentration over time. There was no evidence of any temporal trends for total dissolved solids (TDS). Significant positive trends for nitrate nitrogen were detected in the Rock, Kaskaskia, and Illinois River basins, with the correlations varying in magnitude from 0.4 to 0.5. In contrast, the Ohio River basin displayed a significant negative trend of -0.41 for this parameter, indicating that concentrations had decreased over time.
No significant temporal trends could be detected for water quality parameters classified as pesticides. The number of nonzero average values for all the years for all parameters in this category, for the state or for any basin, ranged between 2 and 5. Annual water quality concentrations for selected parameters in streams are plotted as a function of time in figures 4-7.
ANALYSIS OF STREAMWATER QUALITY BY RIVER BASINS
The results of the Kendall correlation analyses for selected streamwater quality parameters for 14 basins in Illinois are given in tabular form in Ramamurthy (1993) and are summarized below.
Big Muddy basin: A positive trend for the low group (0.36) and a negative trend for the middle group (-0.60) for chromium implies that concentration levels are declining over time. The mean COD concentration has also decreased during this period. Des Plaines River basin: Significant trends were observed for the low and middle groups of chromium: a negative trend (-0.68) for the low group and a positive trend for the middle group (0.58). This means that chromium levels have increased in this basin over time, in contrast to the Big Muddy basin. The high group of cadmium shows a strong negative trend (-0.87). A strong negative trend (-0.79) was also observed for ammonia nitrogen. Fox River basin: Chromium concentrations are increasing over time, similar to the trend in the Des Plaines basin. A weak negative trend was found for ammonia nitrogen in this basin. Kankakee River basin: Chromium levels showed a strong increase from the low group to the middle group. Kaskaskia River basin: Some evidence indicates a moderate positive trend for phosphorus concentrations and a strong negative trend for COD. These trends are similar to the average trends observed for the state. The middle group for chromium exhibits a moderate negative trend (-0.42). Mississippi River, north basin: No significant trends were observed for any of the parameters analyzed. Mississippi River, central basin: No significant trends were observed for any of the parameters analyzed. Mississippi River, north-central basin: No significant trends were observed for any of the parameters analyzed. Mississippi River, south basin: With a negative correlation (-0.54), the high group of lead displayed a declining trend over time. A moderate decrease (-0.46) was found in the percent of observations for chromium in the 5-50 ppb level. Mississippi River, south-central basin: No significant trends were observed for any of the parameters analyzed. Ohio River basin: Several trends were noticeable in the Ohio River basin. The proportion of observations in the low category for arsenic has been increasing, with a corresponding decrease in the high category. This implies that arsenic levels are decreasing. Cadmium concentration levels were also decreasing, with a moderate negative trend (0.50) in the high group. The low and high groups of lead showed an increasing trend, while the middle group showed a decreasing trend. This implies that in recent years, lead levels have been either in the 0-5 ppb range or in the >50 ppb range, which may be due to changes in the procedure for measuring lead concentrations. Rock River basin: No significant trends were observed for any of the parameters analyzed. Wabash River basin: Both the low and high groups for chloride showed negative trends at different levels of significance. Phosphorus concentration levels exhibited a moderate increasing trend (0.50). Illinois River basin: A negative trend was observed for COD. A negative trend for the low group and a moderate positive trend in the middle group indicate that chloride concentration levels have been increasing over time.
Since a trend was observed for nitrate nitrogen in a number of basins, the average concentrations of this parameter for four years, 1975, 1980, 1985, and 1990, are shown in figures 8 and 9. These figures provide some insight into the spatial variation in nitrate nitro-gen concentrations for the four time periods.
ANALYSIS OF LAKE WATER QUALITY IN ILLINOIS
A summary of the lake water quality data used for the trend analysis is shown in table 3. The results of the Kendall correlation analyses of selected water quality parameters for lakes averaged over the entire state of Illinois are given in Table 4.
No trends were observed for any of the parameters analyzed for lakes in Illinois. Data on DO, phenolics, chlordane, dieldrin, and DDT were very sparse, with many missing values, and thus could not be analyzed. Data values for pesticides such as chlordane, dieldrin, and DDT were less than or equal to 0.01 micrograms per liter for the few observations that were available. Annual water quality concentration levels for selected parameters in lakes are plotted as a function of time in figures 10-12.
ANALYSIS OF LAKE WATER QUALITY BY RIVER BASINS
The results of the Kendall correlation analyses for selected lake water quality parameters for 13 Illinois basins are given in tabular form in Ramamurthy (1993) and are summarized below.
For most of the basins, no significant trends were observed for any of the parameters analyzed. A moderate decreasing trend (-0.47) was observed for phosphorus in the Fox River basin. In the Kaskaskia River basin, a highly significant positive trend (0.50) was observed for nitrate nitrogen, which suggests that concentration levels are increasing over time. A highly significant positive trend (0.50) was found for ammonia nitrogen in the Mississippi River, north-central basin. A moderately significant positive trend (0.49) was observed for nitrate nitrogen in the Rock River basin. Similar trends for both ammonia nitrogen and nitrate nitrogen were also seen in the Wabash River basin.
SUMMARY AND CONCLUSIONS
Significant negative (or decreasing) concentration trends were observed for cadmium, mercury, chloride, and COD. Phosphorus and nitrate nitrogen concentrations exhibited equally significant positive (or increasing) trends. No significant trends were observed for arsenic, chromium, phenolics, fecal coliform, DO, pH, and TDS.
Significant temporal trends were observed in the Ohio, Des Plaines, Kaskaskia, Wabash, Illinois, Big Muddy, Mississippi south, Fox, and Kankakee River basins. The Des Plaines River basin, in contrast to the statewide trend, showed increasing cadmium concentration levels over time. An increasing trend in chromium concentrations was observed for the Kankakee River basin. A moderate increase in chloride concentrations was found in the Illinois River basin. Lead concentrations in the Ohio River basin increased for part of the observation period, but decreased in others. The Ohio River basin also revealed decreasing concentrations of nitrate nitrogen, which is contrary to the overall trend observed for the state.
No trends were observed for any of the water quality parameters analyzed for lakes in Illinois. Since data on some parameters were very limited with many missing values, they could not be analyzed.
Limitations of This Study
Water quality variables are generally known to exhibit seasonal trends. Therefore, water quality data series should be tested for inherent seasonal effects. If, for example, a periodicity of one year exists in water quality data, one should compare only observations within a particular month for each year. The Kendall tau test should be designed to reflect this seasonality (Smith et al., 1982). The correlation test used here has not been adjusted to allow for seasonality in water quality data.
Temporal trends have been analyzed with water quality concentration levels. Pollutant loadings that are influenced by both concentration and flow levels have not been analyzed. For parameters such as COD, phosphorus, and DO, concentration levels can be strongly influenced by flow levels. The concentration values should be adjusted for flow variations. Such a procedure requires several iterations to estimate the adjustment equations, and it has not been performed in this analysis.
As mentioned earlier, data recorded for some water quality variables are not precise. Consequently, the observations for such variables had to be grouped. The Kendall statistics therefore indicate movement in concentration levels from one group to another, rather than the decrease/increase that can occur within a group. Because percentage values are used in such cases, the data for each year are assumed to come from identical distributions, irrespective of the sample size for the year.
SUGGESTIONS FOR FUTURE RESEARCH
As outlined above, new and improved methodologies need to be introduced to extend the analysis of water quality undertaken in this study. Future work on this topic should comprise the following specific tasks:
1. Develop a seasonal Kendall estimator to test for seasonality in water quality data and estimate seasonal temporal trends. The number of seasons in each year and the duration of each season are em-pirical issues and should be determined using appropriate analytical procedures.
2. Develop pollutant loadings for different parameters using data on concentration levels and average streamflow values. The concentration values of certain parameters such as COD, phosphorus, and DO are strongly influenced by flow conditions. These concentration values should be adjusted for flow variations using iterative procedures suggested in the literature. The development of pollutant loadings and the use of flow-adjusted concentration values provide a more realistic measure of ambient water quality conditions.
3. There is no acceptable procedure to convert water quality measurement values for parameters such as metals to unique concentration values. Consequently, the observations for such parameters had to be grouped. Appropriate statistical procedures should be used to determine the expected concentration values by assuming an underlying distribution for such measurements. This procedure will yield a consistent set of concentration values that can be used to determine temporal and spatial trends.
4. Develop appropriate statistical methods to test and estimate the impact of surface water quality trends on selected water quality concentration levels and pollutant loads.
5. Integrate the results of water quality and water quantity trend analyses to identify environmental indicators based on chemical surface water quality. These can be combined with information developed on biological trends to identify biochemical environmental indicators.
Keith, L.H., W. Crummett, J. Deegan, R.A. Libby, J.K. Taylor, and G. Wentler. 1983. Principals of Environmental Analysis. Analytical Chemistry 55:2210- 2218.
Kendall, M.G. 1970. Rank Correlation Methods. 4th ed. Griffin, London.
Ramamurthy, G.S. 1993. Analysis of Ambient Water Quality Trends in Streams and Lakes. Unpublished report, Illinois State Water Survey, Champaign, IL. Smith, R.A., R.M. Hirsch, and J.R. Slack. 1982. A Study of Trends in Total Phosphorus Measurements at NASQAN Stations. U.S. Geological Survey Water-Supply Paper 2190.
Zipper, C.E., G.I. Holtzman, S. Rheem, and G.K. Evanylo. 1992. Surface Water Quality Trends in Southwestern Virginia, 1970-1989: 1. Seasonal Kendall Analysis. Virginia Water Resources Center VPI-VWRRC- BULL 173.
Figure 1. Streamwater quality stations in Illinois Ambient Water Quality Monitoring Network
Figure 2. Lake water quality stations in the Illinois Ambient Water Quality Monitoring Network
Figure 3. Major drainage basins in Illinois
Figure 4. Annual statewide average streamwater concentration levels for arsenic, cadmium, chromium, and lead
Figure 5. Annual statewide average streamwater concentration levels for mercury, chloride, phenolics, and total dissolved solids
Figure 6. Annual statewide average streamwater concentration levels for chemical oxygen demand, pH, dissolved oxygen, and fecal coliform
Figure 7. Annual statewide average streamwater concentration levels for phosphorus, ammonia nitrogen, and nitrate nitrogen
Figure 8. Nitrate nitrogen concentration levels in major river basins in Illinois, 1975 and 1980
Figure 9. Nitrate nitrogen concentration levels in major river basins in Illinois, 1985 and 1990
Figure 10. Annual statewide average lake water concentration levels for arsenic, cadmium, chromium, and lead
Figure 11. Annual statewide average lake water concentration levels for chemical oxygen demand (COD), pH, chloride, and fecal coliform
Figure 12. Annual statewide average lake water concentration levels for ammonia nitrogen, phosphorus, and nitrate nitrogen
Continue to Statewide Ground Water Quality