Barton Springs, the major discharge point for the Barton
Springs segment of the Edwards Aquifer, is primary habitat
for the endangered Barton Springs salamander (Eurycea
sosorum), supplies a portion of Austin's drinking water,
provides winter and drought baseflow to the Colorado River
downstream, and is an important recreational resource.
Significant time trends in Barton Springs have not been
previously identified. However, the passage of time,
continuing development, and increased data collection efforts
combined with recognition of the importance of variable
recharge/discharge conditions has led to additional data
analysis. The current analysis of long term water quality
records (1975-1999) from Barton Springs indicates
statistically significant changes in water quality
potentially related to watershed urbanization. Increasing
specific conductance, sulfate, turbidity, and total organic
carbon trends were noted to be significant. A decreasing
trend in dissolved oxygen concentrations was also significant
in Barton Springs. Significant trends were not noted in other
parameters that are commonly considered pollutants, such as
nutrients and total suspended solids. However, when older,
less verifiable data is included in the analysis, a long term
increase in nitrate nitrogen is statistically significant.
Constraints associated with using this type of data analysis
for future predictions are discussed.
1 by Martha Turner, P.E., Project Coordinator, Environmental Resources Management Division Update of Barton Springs Water Quality Data Analysis Austin, Texas ABSTRACT Barton Springs, the major discharge point for the Barton Springs segment of the Edwards Aquifer, is primary habitat for the endangered Barton Springs salamander (Eurycea sosorum), supplies a portion of Austins drinking water, provides winter and drought baseflow to the Colorado River downstream, and is an important recreational resource. Significant time trends in Barton Springs have not been previously identified. However, the passage of time, continuing development, and increased data collection efforts combined with recognition of the importance of variable recharge/discharge conditions has led to additional data analysis. The current analysis of long term water quality records (1975-1999) from Barton Springs indicates statistically significant changes in water quality potentially related to watershed urbanization. Increasing specific conductance, sulfate, turbidity, and total organic carbon trends were noted to be significant. A decreasing trend in dissolved oxygen concentrations was also significant in Barton Springs. Significant trends were not noted in other parameters that are commonly considered pollutants, such as nutrients and total suspended solids. However, when older, less verifiable data is included in the analysis, a long term increase in nitrate nitrogen is statistically significant. Constraints associated with using this type of data analysis for future predictions are discussed. INTRODUCTION COA staff, and others (Barrett, 1996; COA, 1997; TNRCC, 1995;) have examined Barton Springs data at various times in the past for evidence that water quality is changing due to increasing urbanization in contributing watersheds and have not found any statistically significant time trends. With the continuing development in the contributing and recharge zone and additional data collection at Barton Springs an update of the trend analysis was determined to be appropriate at this time. This analysis is also timely given the current effort to complete a Recovery Plan for the Barton Springs Salamander. The focus of the current analysis was also expanded from investigating common non-point source pollution parameters (solids, nutrients, etc.) to all water chemistry parameters available for a significant period of record. Additional data treatment including grouping by hydrological condition and removing questionable data influenced by short term abnormal events was performed to isolate long term trends.
2 THE SPRING DATA: Treatment of Barton Springs water quality data requires an examination of the data source, storm conditions, recharge conditions, and impacts induced from short term activities such as spills or pool maintenance. Base Data Sources (COA and USGS) Barton Springs data was extracted from the COA field sampling database. The COA Springs Project data, the USGS Barton Springs data, and the Austin-Travis County Health Department fecal coliform samples were selected. These are the largest and most comprehensive data sets available for Barton Springs, in both period of record and the number of parameters. These data are used in the analyses reported in this report. In addition, time series data collected every 15 minutes for several years with a Datasonde is presented, for one parameter, dissolved Oxygen to aid in interpretation of long term data. Other available data sets not used in the analyses (TNRCC data, short term COA sampling projects, citizen monitoring, etc) are more sporadic in sampling frequency, do not meet quality control standards, and/or tend to be limited to just a few constituents. Exclusion of Abnormal Conditions Data reflecting two abnormal conditions, Barton Springs pool maintenance drawdown and a major sewer line failure, were removed before additional analysis was begun. Data from these events are not representative and could obscure trends in parameter levels caused by watershed impacts rather than localized short-term influences. This was conducted as a prudent evaluation of outlier conditions that were evaluated, documented, and removed before further analysis. The rationale for excluding drawdown data is that it does not represent the normal discharge quality from Barton Springs. These data were originally obtained as part of the COA Springs Project to gauge the short-term effects of drawdown for maintenance cleaning on spring discharge quality. Routine COA and USGS sampling is not conducted during pool drawdown. When Barton Springs pool is lowered, specific conductance and turbidity usually increase, and dissolved oxygen decreases (COA, 1997). Including these data would bias the analysis towards short duration impacts rather than the capture of long-term trends. Therefore, drawdown sampling data from September 17 and 18, 1998 events were not used. In addition, dates where water quality is suspected to have been affected by documented sewer line breaks were removed from the data set (McReynolds, 1986; USGS 1986). These sewer lines were repaired in April and November of 1982; however, neither the duration of the discharge nor the length of time necessary to flush the portion of the aquifer that was impacted can be determined. Plots were examined for high levels of ammonia (NH 3 ), total kjeldahl nitrogen (TKN), total suspended solids (TSS), and fecal coliform bacteria during baseflow to isolate the affected data. Concentrations greater than 0.06 mg/L NH 3 0.5 mg/L TKN, 10 mg/L TSS and 100 colonies/100ml fecal coliform during baseflow were considered indicative of the sewer line break impacts based on examination of the entire data record. During 1981 and 1982 almost all
3 baseflow concentrations were above these limits. Thus all data from 1981 and 1982 was removed from the data set. Hydrologic Condition Data Separation The remaining data were then separated into three flow categories: baseflow without recharge, baseflow with recharge and storm flow. This separation was performed because the factors affecting water quality in the springs differ under these three major flow conditions. During recharge, the water quality at the springs partially reflects the current water quality in the creeks, whether it is baseflow or storm flow (COA, 1997). Under baseflow without recharge, the spring discharge would primarily reflect the long-term changes in aquifer water quality. S torm and baseflow: Because neither rainfall in the large contributing area, flow in the recharging creeks, or Barton Springs flowrate itself is a conclusive measure of the influence of stormflow conditions on the springs, discharge quality was used as a conservative separation indicator. If Fecal Coliform counts were greater than 100 mg/L or TSS were greater than 10 mg/L then the data was labeled as indicative of storm conditions. If both Fecal Coliform counts and TSS concentrations were missing the data from the sample was not used. In addition, a concentration of intensive spring sampling had been done during three discrete periods when storm flow conditions were expected from the sampling design (personal communication: David Johns, COA). Data from these dates, 11/20/92-11/21/92, 5/30/96-6/1/96, 9/11/98-9/14/98, was examined in detail, and if the pattern of increasing and decreasing fecal counts and TSS concentrations had a single peak, the data were split into storm and baseflow; otherwise, the data from these periods were labeled storm flow. This partitioning of the data into flow categories is not precise because no definitive indicator of when the discharge from the springs should be considered storm flow exists. Rainfall histories throughout the contributing and recharge zone are not consistently available over the period of record. Flow paths vary with flow rate and have not been determined for the entire Barton Springs Recharge Zone through ongoing dye studies. Also, known travel times from the various recharge points range from 10 hours to 8 days adding to the uncertainty. Traditional baseflow separation from gaging data was also not applicable in this case due to the flow and aquifer level variable recharge pathways and the desire to determine storm influenced water quality conditions rather than purely hydrologic storm flow conditions. In most cases, it was determined to be more appropriate to place the transition period of mixing storm and baseflow into the storm flow influenced category. In all cases, data placed in the wrong flow category will likely increase the variability in that category and obscure any actual trends occurring under either flow condition. Recharge and non-recharge conditions. Daily flow in Barton Creek at Loop360 was examined to determine if recharge was occurring. If the flow was greater than zero (0) then it was assumed that the data could be categorized as recharge. At times recharge may have been occurring from other watersheds when no flow was recorded at Barton Creek at Loop 360. However, as with the storm segregation, the absence of distributed rainfall records, uncertainties in transport through the aquifer, and data gaps in flow
4 records make determination of recharge conditions problematic. From examination of the available flow records, separation based on the single continuous gage will be accurate in the vast majority of cases. Again, data placed in the wrong category will likely increase variability and obscure actual trends occurring in either recharge or non-recharge conditions. Additional analysis of flow data for other recharging creeks is ongoing. DATA ANALYSIS METHODS Trend Analysis Multiple Regression Multiple linear regression analysis was used to determine if parameter levels were changing over time, the direction of change, and the level of significance of the change. Regression analysis was performed for each parameter for three flow classes: baseflow with recharge, baseflow without recharge and stormflow. A relationship between spring discharge levels and parameter levels for some parameters has been previously demonstrated at Barton Springs (Senger and Kreitler, 1984; COA, 1997). Therefore, spring discharge was entered first in the regression model followed by time. Using this method, the variation due to time was distinguished from that due to spring discharge Since turbidity had a large number of values at the detection limit, Cox regression was also used to confirm the results for this parameter. Cox regression is a semiparametric method which is recommended for use with censored data (Allison, 1995). Significant relationships are identified as direct or inverse with time. For direct relationships the variables increase together. For inverse relationships as one variable increases, the other decreases. Relationships in which both the model and the date coefficient are significant at the 0.05 level are discussed. In addition, several relationships in which the model is significant at the 0.05 level but the date coefficient is only significant at the 0.10 level, were also identified, because more accurate storm and baseflow separation or the acquisition of more data may determine that a significant trend exists at the 0.05 level. The results were then examined by plotting the data and looking for the trends determined analytically. The trends are not always visible and plots must be viewed with caution since statistically significant differences in Barton Springs discharge are present in different time periods and many of the parameters vary significantly with spring discharge under some flow conditions. In addition, the frequency pattern for all discharges does not match the frequency pattern for flow when samples were collected. Figure 1 illustrates changing flow patterns over time and the relationship between spring discharge and one parameter, dissolved oxygen, under baseflow conditions. While, DO concentrations vary with flow, the greater incidence of lower values through time at similar discharge levels results in a significant inverse relationship for DO which is not evident from the scatter plot. Magnitude of Change In order to illustrate the magnitude of the changes in the constituent concentrations over time, three different methods were utilized. Normalized period medians were compared, the regression equation was used to predict the concentrations in 1980 and 2000, and the normalized period means with outliers removed were compared. The use of multiple methods in examination enhances the confidence in the results and was recommended by the Barton Springs Salamander Recovery Team. For all methods, the effect of changes in Barton Springs discharge was considered. When regression was used to predict the concentrations, the discharge was set
5 to a round value of 50 cfs near the long term average for the discharge. For the means and medians, if the constituent had a significant relationship with spring discharge, the slope of the regression equation was used to normalize the concentrations to those that would have been expected at an average discharge of 50 cfs. The equation used to normalize the data was: Normalized concentration = original concentration + (regression coefficient for discharge)*(50 cfs spring discharge). Median concentrations were calculated for each five year period from 1975 to 1999 using the original data if the relationship with spring discharge was not significant, and the normalized data if it was significant. Medians were used rather than means to reduce the impact of outliers and values that are in the wrong category due to uncertainties in the data separation process. Five year periods were selected as a common planning increment consistent with TNRCC evaluations of surface water bodies for evidence of water quality impairment in the Clean Water Act Section 303(d) assessments and NPDES permit actions. Mean concentrations were calculated for each five year period from 1975 to 1999 using the original data if the relationship with spring discharge was not significant, and the normalized data if it was significant. Outliers were identified and removed prior to estimating the concentration means. Influence diagnostics 1 were used to identify individual values that overly affected parameter estimates from the regression. The earliest sampling dates are in either the 1975-1979 or the1980-1984 period depending on parameter. The increase or decrease in a parameter was determined by difference in the period medians from the earliest and most recent five-year period. The percent change in the parameter concentrations was determined from the size of the change in the concentrations divided by the median concentration level during the earliest period. Figure 1 Discharge (---) Compared With DO (.) Concentration Levels
6 1 If the studentized r esidual was larger than two in absolute value, or if the DFFITS statistic was greater than the size adjusted cutoff of 2 2/n with n = number of observations, the data point was removed (SAS 1989). RESULTS Parameters with significant changes over time included conductivity, dissolved oxygen, organic carbon, sulfate and turbidity. Regression r-squares, model and coefficient probabilities, and coefficient estimates and standard errors are shown in Table 1. Numbers denoted with an asterisk (*) indicate regression coefficients for date (time trend) which were significant at the 0.10 level but not at the 0.05 alpha level. The multiple linear regression model using discharge followed by date was significant at the 0.05 level in all cases. The model r-square is not high in most cases indicating that many factors, such as antecedent weather conditions, which affect the water quality of the spring discharge are not included in the model. These factors can not be adequately characterized over the entire period from 1975 to 1999 and thus cannot be included in the model. Hence, the regression model should not be used to predict future water quality concentrations. In addition it should be noted that the model is linear. Water quality changes in response to environmental stresses may be linear over a certain range of stress levels and then change abruptly once a threshold is reached. However a significant time coefficient in parameters of consequence to drinking water or aquatic life uses would demonstrate a trend for the worse in Barton Springs water quality.
Table 1 Regression R-squares, Model and Coefficient Probabilities, and Coefficient Estimates Regression Coefficients Model Discharge Date Parameter Flow Condition Pr > F RSquare Pr > |t| Coefficient Estimat e Std Error Pr > |t| Coefficient Estimat e Std Error Conductivity Baseflow without Recharge <0.0001 0.34 <0.0001 -1.19 0.14 0.0663 0.0037 0.002 Baseflow with Recharge 0.0002 0.18 0.0304 -0.45 0.21 <0.0001 0.0106 0.0024 Storm Flow <0.0001 0.29 <0.0001 -0.98 0.14 0.0257 0.0051 0.0023 Dissolved Oxygen Baseflow without Recharge <0.0001 0.59 <0.0001 0.03 0.004 0.0016 0.00015 0.00004 Organic Carbon Storm Flow 0.0404 0.1 0.7538 0.01 0.03 0.0116 0.0009 0.0003 Sulfate Baseflow with Recharge 0.0062 0.36 0.0163 -0.15 0.06 0.0016 0.0023 0.0006 Turbidity Storm Flow <0.0001 0.19 0.0001 -0.12 0.03 0.064 0.001 0.0005
8 Significant relationships to spring discharge are also listed in Table 1. In general, dissolved oxygen increases with increasing discharge, whereas conductivity, sulfate, and turbidity decrease with increasing discharge. Organic carbon is not significantly related to spring discharge under any flow condition. The size, percent and direction of the change in these five parameters with significant time trends are summarized in Table 2 Predictions of future conditions should not be made by extrapolating the rate of change during the past 20 years. Future rates of change will depend on the rates of change in environmental stress and possible threshold conditions The paragraphs below describe the significant changes identified for each individual parameters by this analysis Conductivity Conductivity has increased during all flow conditions over the past 20 to 25 years. The largest change is observed during baseflow with recharge and is estimated to be less than a 15% change. Storm flow changes are estimated to be less than 7%, and during baseflow without recharge, the change is less than 5%. The median concentration estimate during baseflow without recharge increased from 655 to 677 uS/cm. For comparison, these concentration both lie between the mean baseflow concentrations of 566 for much smaller rural springs and 867 uS/cm for much smaller newer urban springs, respectively, in the Jollyville plateau (COA ,1999). However, the increase noted in Barton Springs may be an indicator of future change in Barton Springs to more of an urban signature. Scatter plots of these data are provided in Figures 2 through 5. Dissolved oxygen (DO) Scatterplots of DO data are provided in Figures 6 through 9. DO has decreased over time during baseflow, when recharge was not occurring. During non-recharge, at low spring discharge levels, the measured DO sometimes drops below 4 mg/L. DO is significantly directly related to spring discharge levels, but DO is decreasing both at high discharge levels and at low ones. The median dissolved oxygen concentration has decreased approximately 1.1 mg/L over the last 25 years, from 6.8 to 5.7 mg/L. This is a decrease of 16%. Sampling has been much more frequent recently, leading to a higher probability of observing extreme events. Therefore it is possible that the change is a sampling artifact. However, DO concentrations in Barton Springs, tracked with a Datasonde (data at 6-hour intervals over month long periods) have been below 4 mg/L 11% of the time during an approximately four year period of record as indicated in Figure 9. The plots of the Datasonde data, which was not included in the regression or magnitude of change calculations, compared with the discrete DO data show that low DO levels may predominate during periods without much recharge. The Datasonde data has yet to be scrutinized carefully for drift or calibration problems, but it does indicate the potential for the occurrence of low DO in the springs Naturally a long term change in DO of greater than one mg/L is significant in any isolated aquatic habitat.
Table 2 The Magnitude and Percent Change in Constituent Levels Over 20 to 25 Years Normalized Period Medians Predicted from Regression at 50 cfs Normalized Period Means with outliers removed Parameter Flow Conditio n 1975-1979 or 19801984^ Median 1995-1999 Median Change over approx. 20 years Percent Change prediction 1-1-1980 prediction 1-1-2000 Change over 20 years Percent Change 1975-1979 or 1980-1984^ Mean 1995-1999 Mean Change over approx. 20 years Percent Change Conductivity (uS/cm) Baseflow without Recharg e 655 677 22 3% 642 668 27 4% 651 658 7 1.1% Baseflow with Recharg e 590^ 646 56 9% 574 651 78 14% 569^ 645 76 13.0% Storm Flow 624 642 18 3% 601 638 37 6% 624 640 16 2.6% Dissolved Oxygen (mg/L) Baseflow without Recharg e 6.8 5.7 -1.1 -16% 6.5 5.45 -1.1 -16% 6.4 5.6 -0.8 -12.5% Organic Carbon (mg/L) Storm Flow 1.5 3.4 1.9 127% -0.68 5.8 6.5 799% 1.5 4.2 2.7 180.0% Sulfate (mg/L) Baseflow with Recharg e 28.3^ 38.8 10.5 37% 25.1 41.7 16.6 66% 28.3 ^ 37.6 9.3 33.0% Turbidity (NTU) Storm Flow 5.3 7 1.7* 32%* 3.7 11.2 7.5* 203%* 5.3 7.3 2* 37.7%* significant at the 0.1 level but not at the 0.05 level ^ acually 1980, 1983 and 1984 since 1981 and 1982 were removed from the analysis due to a sewer line break
10 Figure 2 Conductivity During Baseflow Without Recharge Figure 3 Conductivity During Baseflow With Recharge
11 Figure 4 Conductivity During Storm Flow Figure 5 Normalized Conductivity During Storm Flow
12 Figure 6 Dissolved Oxygen During Baseflow Without Recharge Figure 7 Normalized Dissolved Oxygen During Baseflow Without Recharge
13 Figure 8 Datasonde Dissolved Oxygen in Barton Springs Figure 9 Datasonde Dissolved Oxygen Frequency
14 Organic Carbon Organic Carbon has increased during stormflow only. The size of the increase in median concentration over the last 25 years is 1.9 mg/L, from 1.5 to 3.4 mg/L. This is an increase of 127%. Perhaps increased deposition of degradable organic carbon in the aquifer during storm flow, may lead to decreases in DO during baseflow when there is no recharge occurring. Scatterplots of these data are provided in Figures 10 and 11. Sulfate Sulfate has increased during baseflow when recharge is occurring. Median sulfate concentrations have increased approximately 10.5 mg/L, from 28.3 to 38.8 mg/L. This is an increase of 37% over a 20-year period. Sulfate levels have been found to be fairly consistent indicators of urbanization in much smaller springs in the Jollyville Plateau region. Mean concentrations in rural springs ranged from 12 to 26 mg/L, whereas mean concentrations in newer urban springs ranged from 43 to 59 mg/L (as read from a graph) (COA, 1999). The current median concentrations in Barton Springs lie between these two groups. Again, this increase may be an early indicator of the effects of watershed urbanization that are not reflected in more commonly considered pollutants. Scatterplots of these data are provided in Figures 12 and 13. Turbidity Turbidity has increased significantly over time during storm flow. Turbidity is significantly inversely proportional to spring discharge. Sampling has been much more frequent recently leading to a higher probability of observing extreme events. However the frequency of high turbidities is such that the observed increase is unlikely to be a sampling artifact. The average increase in storm water turbidity is 1.7 NTU, from 5.3 to 7 NTU. This is an increase of about 32% over the past 20 years. Scatterplots of these data are provided in Figures 14 and 15. It should be noted that the influence of recent data on storm condition results may be significant due to an effort to obtain representation of turbidity over the storm flow hydrograph. This can be compared to previous sampling strategies whereby only single grab samples were obtained for storm events. Replacement of storm event data with median values causes the regression to be nonsignificant at the 0.05 level; however, the regression is still significant when these events are replaced with the maximum single grab taken over the storm event. While the changes in turbidity during baseflow are not significant due to the variability of the data and the large number of very low concentrations, there is some indication that change is occurring. Table 3 shows the percent of the turbidity measurements that fell within various ranges for three periods of time. Prior to 1990, under baseflow conditions, 82% of the turbidity levels during recharge were less than 2 NTU and all storm flow turbidities were less than 12 NTU. In the past five years 74% of the baseflow turbidities levels during recharge conditions were between 2 and 12 NTU, and 34% of storm flow turbidities were between 12 and 50 NTU. Although short term turbidity increases are expected during storm conditions as a watershed is
15 Table 3 Percent of Turbidity Concentrations in Selected Ranges for Three Time Periods Baseflow without Recharge Baseflow with recharge Storm flow Period 0-2 NTU >2 NTU 0-2 NTU >2 NTU 0-12 NTU >12 NTU 1975-1989 100% 82% 18% 100% 1990-1994 97% 3% 85% 15% 95% 5% 1995-1999 75% 23% 28% 72% 67% 33% Figure 10 Organic Carbon During Storm Flow
16 Figure 11 Organic Carbon During Storm Flow Without Outliers Figure 12 Sulfate During Baseflow With Recharge
17 Figure 13 Normalized Sulfate During Baseflow With Recharge Figure 14 Turbidity During Storm Flow
18 Figure 15 Normalized Turbidity During Storm Flow urbanized, baseflow increases in turbidity may also be an early indicator of such watershed changes. Also, the inclusion of data removed due to lack of corresponding coliform and TSS data impacts the turbidity regressions. Including these data as baseflow resulted in a significant increasing trend for non-storm, recharge conditions. ADDITIONAL ANALYSES ON SPRING DATA Several additional analyses were done as checks on the validity of our results on parameters and flow conditions that were shown to have significant changes over time in the results presented above. The multiple regressions were rerun on two subsets of the data: USGS data and all data prior to 1995. In addition all flow conditions were lumped together and two different regression models were investigated. In one model, the two independent variables were discharge followed by date. This is the model used in all the analyses discussed previously. In the other model, date was the only independent variable. The regression coefficients for date from these analyses are presented in Table 4. Trend Analysis on USGS Data Most of the data prior to 1995 was gathered by the USGS, whereas in recent years most samples have been collected by other agencies. Time trends identified by analyses on all the data may be due to method or lab differences. To investigate this possibility the analyses were rerun on just the USGS data. Significant results provide an important confirmation of the original analyses. If the date regression slope is no longer significant then additional investigation is needed. Method or lab differences should be considered. However the loss of significance may be due simply to the decrease in the number of data points. If this is the case we would expect the slope of the time trend to be similar to that found on the entire data set. Time trends for dissolved oxygen,
19 organic carbon, and sulfate were confirmed by regression on the USGS data, as were time trends for conductivity during baseflow. Time trends for conductivity during storm flow and turbidity were not significant when only the USGS data was considered. The regression coefficients for date on the USGS data have the same sign and are approximately half the size of the coefficients for the entire data set. This may imply that the change over time is not as large as indicated by the entire data set, or that with more USGS data the trend will be confirmed, or that the trend does not exist. These parameters under these flow conditions could warrant more investigation. Trend Analysis on 1975-1994 Data No significant time trends at the 0.05 level were found when the data from the last five years was eliminated. This result would explain why previous analyses did not observe such trends. However for most parameters and flow conditions, the slopes were similar in magnitude and had the same sign. This would imply that the trends were there but that the number of data points was insufficient to confirm the significance of the trend. The parameters and flow conditions where this was not true were conductivity during storm flow and turbidity. These are also the time trends not confirmed by the analysis on the USGS data. Trend Analysis on Un-separated Data Since the split of the data in to the three flow categories is imprecise, the multiple regression was run on all flow categories lumped together with discharge and date as the independent variables. In addition regression with date for the independent variable was done. Significant time trends were identified for dissolved oxygen and conductivity with both regressions. When the data is lumped, no trends are observed for organic carbon, sulfate, or turbidity. Table 4 Regression Coefficients for Date Flow Condition All conditions lumped Baseflow without Recharge Baseflow with Recharge Storm Flow Model Independent Variables date Discharge and date Discharge and date discharge and date Discharge and date Discharge and date discharge and date discharge and date Discharge and date Discharge and date discharge and date Data All All All USGS 1975-1994 All USGS 1975-1994 All USGS 1975-1994 Conductivity 0.006 0.006 0.004* 0.006 0.001 0.011 0.014 0.011* 0.005 0.004 -0.003 Dissolved Oxygen -0.00012 -0.00012 -0.00015 -0.00015 -0.00014 Organic Carbon 0.0009 0.0004* (not shaded) 0.0002 Sulfate 0.002 0.002 0.002 Turbidity Not shaded 0.001* 0.0006 0.0002 Cells are blank when none of the regressions for that parameter and flow condition were significant Shaded cells indicate significant regressions at the 0.05 level indicates significance at the 0.10 level
20 SUMMARY OF BARTON SPRINGS DATA ANALYSIS The analysis of long term water quality records from Barton Springs using two primary data sources now indicates statistically significant changes in water quality which could be related to watershed urbanization. Increasing conductivity, sulfate, turbidity, and total organic carbon trends were noted to be significant. A decreasing trend in dissolved oxygen concentration was also found to be significant. Significant trends were not noted in other parameters that are commonly considered pollutants, such as nutrients and total suspended solids. Significance and presence of trends is variable depending on flow conditions (i.e. baseflow vs. stormflow, recharge vs. non-recharge). REFERENCES Allison, Paul D., Survival Analysis Using the SAS System: A Practical Guide Cary, NC: SAS Institute Inc., 1995. 292 pp. Barrett M.E., and Charbeneau, R.J. 1996. A Parsimonious Model for Simulation of Flow and Transport in a Karst Aquifer : The University of Texas at Austin, Center for Research in Water Resources. Technical Report 269. 149 p. COA, 1997. The Barton Creek Report COA, DUD, ERM. Water Quality Report Series COAERM/1997. COA, 1999. Jollyville Plateau Water Quality and Salamander Assessment COA, WPD, ERM. Water Quality Report Series COA-ERM/1999-01. Johns, David A. Personal Communication with Martha Turner concerning dates of recharge condition and storm influenced sampling of COA. City of Austin. Watershed Protection Department. September 3, 1999. McReynolds, M., and Slade, R. 1986? Barton Springs: A Case Study in Aquifer Contamination and Development of a Groundwater Management Program. City of Austin. Water/WasteWater Department. Santos and Assoc., Loomis and Assoc., LOK, Tetra Tech and Glenrose Engineering, 1995. Barton Springs Zone Retrofit Master Plan Study. Santos and Assoc., Loomis and Assoc., Austin Texas. SAS Institute Inc., SAS/STAT Users Guide, Version 6, Fourth Edition, Volume 2, Cary, NC: SAS Institute Inc., 1989. 846 pp. Senger, Rainer K., and Kreitler, Charles W., 1984 Hydrogeology of the Edwards Aquifer, Austin Area, Central Texas Bureau of Economic Geology, Report of Investigations No. 141
21 TNRCC 1995, Letter from Mark Jordan, Director Water Quality Policy and Regulation Division to Jana Grote, USFWS concerning proposed listing for the Barton Springs Salamander and absence of material decline in water quality of Barton Springs over 20 year period of nine studies between 1976 and 1994. May 17, 1995. USGS 1986. Hydrology and Water Quality of the Edwards Aquifer Associated with Barton Springs in the Austin Area, Texas. USGS, Water-Resources Investigations Report 86-4036.
A 1 Appendix A Trends in Creek Concentrations in the Contributing and Recharge Zones Preliminary investigation of USGS surface water quality data at sites in the contributing and recharge zones indicates that dissolved oxygen, turbidity, and organic carbon are decreasing and sulfate and conductivity are increasing during both storm and baseflow. The direction of the trends over time in the creeks matches those in Barton Springs for DO, sulfate and conductivity. However the trends for turbidity and organic carbon in the creeks are in the opposite direction from the trends in Barton Springs. Trends which are significant when the data from all the sites is combined may not be significant when the data from each site is considered separately. In some cases even the direction of the trend is different at a particular site. For example, both sulfate and conductivity have decreased, from abnormally high values, in Barton Creek at Lost Creek Blvd during baseflow, whereas the trend in the combined data is increasing. These data are still under investigation and the results on the combined data may also be influenced by inconsistent frequency and timing of samples between sites. Plots of the data are included as follows in this Appendix for information and review.
B 1 Appendix B Additional Analyses on Nitrate Concentrations in Barton Springs. EARLY NITRATE DATA (1937 1971) Analysis of COA, ATCHD and USGS data from 1975-1999 showed no trends in nitrate concentrations. The data that was included in these analyses was selected because it was both comprehensive and collected by agencies with QA/QC procedures, leading to a greater degree of confidence in the analysis results. However additional nitrate data is available from earlier periods. Nitrate concentrations were measured 10 times between 1937 and 1973 (see attached Table 7. Water Quality Analyses for Barton Springs prior to the beginning of the USGS sampling program in 1978 from Slade, 19??). These 10 samples were collected by several different agencies. We do not have QA/QC information for these samples; thus the quality of this data is unknown. It has been suggested that these concentrations may be high estimates of nitrate since the holding times may have been longer is currently allowed and the measured concentration may be total nitrogen rather than nitrate. Discharge levels were recorded for these samples and storm and recharge conditions were estimated from daily rainfall at Austin airports (see Table 1). Table 1. 1937-1973 Nitrate Data with Flow Condition Estimates DATE RAIN Storm Recharge DISCHARGE (cfs) Quali fier Dissolved NO3 mg/L as N August 23, 1937 No No No 32 < 1.13 September 7, 1937 Maybe 4 days after 2" Yes Yes 31 < 1.13 September 9, 1937 No 6 days after 2" No No 31 < 1.13 October 27, 1939 Yes 2 days after 1.2 Yes Yes 16 < 1.13 November 9, 1939 No No No 12 < 1.13 October 1, 1941 Yes? 1.9" on date Yes Yes 55 0.99 June 10, 1948 No No No 19 1.02 January 18, 1955 Yes .21" on date, .85" on previous day Yes Yes 21 1.02 April 22, 1971 No No No 30 1.47 February 6, 1973 No No No 69 1.24 Three separate analyses were done on the expanded data set with these early nitrate concentration included: 1) Regression on the entire data set, 2) Analysis of variance on period means under low Barton Springs discharge and baseflow conditions, and 3) Estimation of the probability that the nitrate concentration distribution has not changed over time.
B 2 Regression on the expanded Data Set: 1937-1999 When the early data is included in the regression analyses, significant trends for baseflow without recharge and storm flow are found (Table 2). There is no early data in the baseflow with recharge category. Table 2 Regression R-squares, Model and Coefficient Probabilities, and Coefficient Estimates. Regression Coefficients Model Discharge Date Paramet er Flow Condition Pr > F RSquar e Pr > |t| Estimate (Std. Error) Pr > |t| Estimate (Std. Error) Nitrate Baseflow without Recharge 0.000 7 0.11 0.073 7 -0.00085 (0.00047) 0.0002 0.00001 (0.000003 ) Baseflow with Recharge No 1937-1973 data Storm Flow 0.011 0 0.10 0.008 5 -0.0029 (0.001) 0.0785 0.00001 (0.000007 ) Baseflow Concentrations At Low Discharge Levels Examination of nitrate concentrations under low discharge levels in 1999 showed a slight increase over levels in 1996 under similar flow conditions. To better examine the hypothesis that nitrate levels are increasing under low discharge conditions, nitrate data from 2000 was added to the 1937-1999 data set. Nitrate concentrations during baseflow with discharge levels less that 40 cfs are plotted in Figure 1. Figure 1.
B 3 The recent data was divided into five-year periods, and the early data into longer periods (see Figure 1). Analysis of variance confirmed that the nitrate concentrations in the different periods are not equal (P < 0.0001). To determine which periods were significantly different, five contrasts were investigated: Table 3. Period Differences as determined from ANOVA contrasts Contrast Pr > F Before 1950 vs. after 1950 P < 0.0001 Before 1998 vs. after 1998 P < 0.0001 1993-1997 vs. 1998-2000 P < 0.0001 1978-1982 and 1998-2000 vs. 1983-1997 P < 0.0001 1978-1982 vs. 1998-2000 P = 0.4418 The nitrate levels before 1950 were significantly lower than the lumped data after 1950. Also the nitrate levels after the start of 1998 were significantly higher than the lumped concentration levels before 1998. Indeed, the nitrate levels after the start of 1998 were significantly higher than during the previous five year period 1993-1997 when flow conditions, sampling frequencies, and lab and analysis methods were quite consistent. However, concentrations in the five-year period from 1978 through 1982 were not significantly different from those in 1999 and 2000. It should be noted that a major sewer line failure occurred and was fixed in 1982. Data from 1981 and 1982 was removed from the analysis. However the start data for the sewer line failure is not known. Adequate data exists in 1981 and 1982 to demonstrate that water quality was affected by sewage but data from 1980 is sparse and the water quality signature is not so clear. The nitrate concentrations may have been affected by the sewer line break but they may not have been. The single low discharge storm flow sample taken during the 1978 through 1982 period had a higher nitrate level than the baseflow samples. This may indicate sewer line problems since nitrate concentrations typically decrease during storm flow. But we do not know, and thus we can not tell if nitrate concentrations at low discharge have increased during the last 25 years or not. Probability that the Nitrate Concentration Distribution has not changed over Time All nitrate concentrations prior to 1960 were less than 1.13 mg/L. However similarly low nitrate concentrations can be observed today. Was the distribution of nitrate concentrations prior to 1960 really the same as todays distribution? It is possible but the probability is not high. Table 4 shows the proportion of nitrate concentrations above and below 1.13 mg/L as N for both time periods.
B 4 Table 4. Number of Nitrate samples above and below 1.13 mg/L for two time periods Time Period Nitrate Level Baseflow without recharge Baseflow with recharge Storm Flow < 1.13 mg/L 4 No data 4 Before 1960 > 1.13 mg/L 0 No data 0 < 1.13 mg/L 7 27 22 After 1960 > 1.13 mg/L 116 41 67 In the pre-1960 period there were only four samples for each of two flow conditions: baseflow without recharge and storm flow. The probability that all four samples would be below 1.13 mg/L if the distribution were similar to the post-1960 distribution is (7/123) 4 = 0.0000105 or approximately 1 in 100,000 for baseflow without recharge. For storm flow the probability is (22/89) 4 = 0.0037 or approximately 1 in 300 for storm flow. These probabilities, that the nitrate distribution has not changed over time, are rather small and thus it appears safe to say that nitrate concentration have increased over time.