Groundwater Statistical Analysis Plan

Transcription

1 Groundwater Statistical Analysis Plan Milton R. Young Station Coal Combustion Residuals Waste Disposal Facility Center, ND Prepared for Minnkota Power Cooperative, Inc. October 2017

2 Groundwater Statistical Analysis Plan Milton R. Young Station Coal Combustion Residuals Waste Disposal Facility Center, ND Prepared for Minnkota Power Cooperative, Inc. October West Century Avenue Bismarck, ND Phone:

3 Groundwater Statistical Analysis Plan Milton R. Young Station Coal Combustion Residuals Waste Disposal Facility Center, ND October 2017 Contents 1.0 Introduction Data Evaluation Methods Preliminary Data Evaluation Data Summary Statistics Data Distribution Outlier Testing Time Series Plots Statistical Tests Analysis of Variance (ANOVA) Tolerance Intervals Prediction Intervals Control Charts Verification Retesting Data Considerations Interwell and Intrawell Site-Specific Considerations Non-Detects Duplicates Seasonal Trends SSI Evaluation Reporting References... 9 P:\Mpls\34 ND\33\ Minnkota Ash Management\WorkFiles\CCR Groundwater 2\Statistical Certification\Statistical Certification Report.docx i

4

5 1.0 Introduction This Groundwater Statistical Analysis Plan (Plan) represents the certification of statistical method required by the Coal Combustion Residuals (CCR) Rule, 40 CFR (f)(6), Groundwater Sampling and Analysis Requirements: The owner or operator of the CCR unit must obtain a certification from a qualified professional engineer stating that the selected statistical method is appropriate for evaluating the groundwater monitoring data for the CCR management area. The certification must include a narrative description of the statistical method selected to evaluate the groundwater monitoring data. The Plan provides a description of the statistical methods for evaluating the groundwater monitoring data for the active CCR Unit at the Milton R. Young Station (Site) located near Center, North Dakota, in accordance with the CCR Rule. For additional details of the groundwater monitoring system (GWMS) see the GWMS Certification Report (Barr, 2017). Baseline groundwater quality monitoring consisted of the collection of eight samples from each monitoring well, as required by the Rule, prior to October 17, A summary of the data collected each year following baseline data collection will be prepared in an annual groundwater monitoring report by January 31 each year. This Plan outlines a means to consistently evaluate the data and to establish criteria that will identify statistically significant increases (SSIs) that may indicate a potential release from the facility. The Plan also defines a protocol for verification of SSIs. Statistical evaluations of groundwater samples will be conducted in accordance with CCR Rule Section (f) and other technical documents, including the 2009 United States Environmental Protection Agency (US EPA) document Statistical Analysis of Groundwater Monitoring Data at RCRA Facilities: Unified Guidance (EPA 530/R ; US EPA 2009). Background groundwater quality concentrations will be utilized to determine if there is a SSI in downgradient wells for Appendix III parameters (CCR Rule Appendix III to Part 257 Constituents for Detection Monitoring), and if necessary Appendix IV parameters (CCR Rule Appendix IV to Part 257 Constituents for Assessment Monitoring). Based on our review of the site conditions, intrawell control charts are the preferred statistical method and will be used to evaluate for SSIs for each parameter, provided that the data set satisfies the statistical assumptions and requirements described in this Plan, in the CCR Rule, and in guidance documents. 1

6 2.0 Data Evaluation Methods 2.1 Preliminary Data Evaluation Several preliminary data evaluation techniques are described below Data Summary Statistics The summary statistics may include number of observations, number of values below the detection limit (non-detects), minimums, maximums, means, medians, measures of variability, and quantiles. Graphical displays, such as box plots and quantile-quantile plots, may be utilized to visually evaluate the data, including the presence and degree of the outlier(s). In box plots, the box represents 25 th and 75 th percentiles, the middle line shows the median (50 th percentile), the whiskers represent the entire range Data Distribution It is necessary to define the type of distribution (normal, log-normal, non-parametric, etc.) for each parameter at each well prior to choosing statistical analysis testing methods. Testing methods may be more or less appropriate for different types of distributions. A Goodness-of-Fit test will be performed at a significance level of Data distribution determination will be updated each time statistical analyses are conducted Outlier Testing Outlier testing can be conducted on data sets that are normally distributed or that are normally distributed once outliers are removed. Outliers will be evaluated via the Dixon or Rosner test methods, both of which are outlined in the US EPA Unified Guidance (US EPA 2009). Outliers will not be evaluated statistically for non-normally distributed data sets Time Series Plots Time series plots are useful graphical representations of the water quality over time; they can assist in visual evaluations of trends over time as well as outliers. 2.2 Statistical Tests Following the collection and analysis of the groundwater samples, Appendix III and/or Appendix IV parameters will be analyzed using statistical methods to determine if there are statistically significant increases (SSIs) in each parameter concentration for each downgradient well relative to baseline (historical data collected at the individual well). The proposed statistical methods will be in accordance with the guidance documents listed in Section 4.0. The CCR Rule allows for several types of statistical evaluations to be conducted to determine if an SSI has occurred for individual parameters in each downgradient monitoring well. This includes the following tests: 2

7 1. Parametric analysis of variance (ANOVA) followed by multiple comparison procedures ( (f)(1)); 2. Analysis of variance based on ranks followed by multiple comparison procedures ( (f)(2)); 3. Tolerance or prediction interval ( (f)(3)); 4. Control chart ( (f)(4)); and 5. Another statistical test method that meets the performance standards outlined in (g). The performance standards of the statistical method must, as per the CCR rule ( (g)): 1. Be appropriate for the distribution of the data; 2. Minimize the Type I (false positive) error: if an individual well comparison method is used, the Type I error level must not exceed 0.01 and if a multiple comparison method is used, the Type I error level must not exceed 0.05; 3. Control charts, tolerance intervals, and prediction intervals, if used, must be at least as effective as any other approach; 4. Account for non-detect data, utilizing the lowest concentration reasonably achieved within the specified limits of precision; and 5. Control and/or correct season and temporal correlations Analysis of Variance (ANOVA) ANOVA is a test used to determine if there is a statistically significant difference between the mean concentrations of specified wells or groups of wells. ANOVA is included in the CCR Rule as one of the acceptable methods for evaluating SSIs, however the Unified Guidance states that it does not recommend this method for formally making regulatory decisions about compliance wells or regulated units ; rather, the Guidance recommends utilization of prediction limits, control charts, and confidence intervals (discussed in sections below; US EPA 2009, pg. 17-1). However, ANOVA has been identified as a useful tool to identify spatial variation Tolerance Intervals A tolerance interval is defined as the concentration range designed to contain a pre-specified proportion of the underlying population from which the statistical sample is drawn (e.g., 95 percent of all possible population measurements) (US EPA 2009, pg ). Two values are reported with the interval, the coverage (proportion of the population covered by the interval) and the degree of confidence or confidence level. However, as with the ANOVA test method, the Unified Guidance does not recommend utilization of tolerance limits in making regulatory decisions Prediction Intervals Prediction intervals (or limits) are utilized to predict the upper limit of possible future values based on a background or baseline data set and comparing that limit to compliance point measurements or statistics prediction limits explicitly account for the degree of variation in the background population and the size of the sample of measurements used to construct the limit (ITRC 2013, pg. 87). Both parametric and non-parametric prediction limits can be determined. For parametric prediction limits, the mean and 3

8 standard deviation of the background or baseline dataset are used, whereas non-parametric prediction limits are based on order statistics (i.e., highest detected concentration; ITRC 2013, pg. 87). It should be noted that non-parametric prediction limits require a larger sample size than parametric prediction limits in order to obtain a desired confidence level. Additionally, for intrawell comparisons, ASTM recommends at least 13 baseline samples prior to reporting the non-parametric limit or to utilize a Poisson Prediction Limit (ASTM 2017). The Unified Guidance, however, states that the intrawell approach can be utilized despite its inadequate power until the intrawell background dataset is sufficiently large via periodic updates (US EPA 2009, pg. 6-34). Finally, inclusion of outliers should be avoided, either through removal from the dataset or, for non-parametric limits, through use of a different order statistic, such as the second-highest detected concentration (US EPA 2009, pg ) Control Charts Control charts are constructed similar to prediction limits, where the limit is estimated from background or baseline and compared to compliance measurements. They are designed for datasets with known distributions and a high detection frequency (>85% detection; US EPA 2009, pg ). Although not specified in the CCR Rule, the Unified Guidance recommends the utilization of the Shewhart-CUSUM control chart (US EPA 2009, pg. 6-46). The Shewhart-CUSUM control chart conducts two tests against the control limit, the Shewhart control line, which identifies sudden spikes in concentrations or changes in trend (such as a new release), and the cumulative sum control chart (CUSUM), which identifies more gradual trend changes over time (plume migration; ITRC 2013, pg. 136). The Shewhart-CUSUM control chart utilizes three parameters: h (the value the CUSUM is compared against), c (related to the displacement that should be quickly detected), and SCL (the value the upper Shewhart line is compared against). Historically, the h and SCL values were slightly different, however based on more recent research, both ASTM and the Unified Guidance allow for setting these parameters as follows: h = SCL = 4.5 and c = 1. These values are utilized in order to quickly identify a parameter concentration changes exceeding two standard deviations, which correlates to a 95% confidence for normally distributed datasets (ASTM 2017, p. 12). The ASTM Standard Guide (2017) states that the combined Shewhart-CUSUM control chart is a good method for conducting intrawell comparisons. In particular, it notes that the method is sensitive to both gradual and rapid releases and is also useful as a method of detecting trends in data (ASTM 2017, pg. 11). However, the ASTM Standard Guide recommends that the control chart method only be used on data sets with a detection frequency greater than 25%. Additionally, the data should be screened for outliers and historical trends prior to computing a limit with the control chart. For datasets containing less than 25% detections, either additional data is required prior to computing a limit (at least 13 background samples, the number required to provide a 99% confidence non-parametric limit with one resample), or the non-parametric prediction limit should be used, as either the maximum or second maximum quantified value (if there is at least one detection) or the detection limit (if there are no detections; ASTM 2017). If data do not fit a normal distribution and cannot be transformed to normality, a non-parametric prediction limit will be used as an alternative to a control chart. 4

9 Control charts are the preferred statistical method for this facility and will be used provided that the data set satisfies the statistical assumptions and requirements described above, in the CCR Rule, and in guidance documents (e.g., US EPA 2009) Verification Retesting Based on the CCR Rule timeline and guidance, the 1-of-2 retesting strategy will be utilized. That is, if a potential SSI is observed, a verification resample will be collected and compared to the limit. If the potential SSI is verified, then an SSI will be declared. However, if the potential SSI is not verified, the initial sample will be flagged and not included in subsequent analysis and the verification resample will be utilized in its place (US EPA 2009, pg ). 2.3 Data Considerations Interwell and Intrawell SSI evaluations for individual parameters can be conducted using one of two methods: interwell analysis (comparison of pooled upgradient to individual downgradient wells) or intrawell analysis (comparison of baseline/background data to compliance data within a single well). Selection of the appropriate method depends on site-specific conditions. Geologic environments are naturally heterogeneous in that differences in lithology exist at a variety of scales in the subsurface. Therefore, assumptions that the geology below a site will be homogenous and isotropic are not generally an accurate description of site geology. This observation is made more apparent because the components of leachate from CCR units all occur naturally in the subsurface and can be assumed to be present naturally at varying concentrations for all Appendix III and Appendix IV parameters. In some cases, the natural differences aggregate into a relatively consistent geochemical signature. That signature can change if there is a change in chemical equilibrium (e.g., either a natural process or a release from a CCR unit). The Unified Guidance (US EPA 2009) outlines several conditions to consider when choosing the appropriate methodology, based on both the site conditions as well as potential spatial variability for individual parameters. More specifically, the Unified Guidance states: For upgradient-to-downgradient well comparisons, a crucial detection monitoring assumption is that downgradient well changes in groundwater quality are only caused by on-site waste management activity. Up- and down-gradient well measurements are also assumed to be comparable and equal on average unless some waste-related change occurs. If other factors trigger significant increases in downgradient well locations, it may be very difficult to pinpoint the monitored unit as the source or cause of the contaminated groundwater (US EPA 2009, pg. 6-29). Therefore, use of interwell analytical methods is most appropriate when the following conditions are true: All wells are screened in same aquifer at similar hydrostratigraphic position (US EPA 2009, pg. 6-29); 5

10 Definable, consistent groundwater flow path from upgradient to downgradient (US EPA 2009, pg. 6-29); Significant groundwater flow velocity such that upgradient and downgradient wells are drawn from the same statistical population (US EPA 2009, pg. 6-30); Lack of significant spatial variability, demonstrated by lack of differences in mean and variance (US EPA 2009, pg. 5-6 and 6-31); The first three conditions relate to the characteristics of the monitoring network as a whole, whereas the fourth condition, regarding spatial variability, can change parameter to parameter. Several steps can be taken to determine if these conditions can be met or if intrawell analysis should be considered. There are some disadvantages to utilizing the intrawell method, namely, that it may be difficult to differentiate between historical contamination and more recent contamination. Therefore, the Unified Guidance recommends that intrawell testing be utilized only when it is clear that spatial variability is not the result of recent contamination attributable to the regulated unit (US EPA 2009, pg. 6-32) Site-Specific Considerations Assessment of the monitoring network was conducted through a review of the site conceptual model, including review of cross sections, boring logs, groundwater elevation data, hydrographs, and hydraulic conductivity test results, as described in the GWMS document (Barr, 2017). The potential spatial variability for individual parameters was evaluated using box plots and ANOVA (US EPA 2009). A discussion of ANOVA is included in Section Some of the above-mentioned conditions for using interwell analytical methods are not true for the site, and therefore intrawell testing was judged to be more appropriate based on current data. For example, groundwater velocities are estimated to be about 0.02 feet/yr (Barr, 2017), which is not a significant rate. As such, the travel time from the downgradient edge of the active CCR unit to the downgradient wells, about 500 feet away, may be over 20,000 years; therefore, we concluded that the downgradient wells have not been contaminated by the regulated unit and are suitable for intrawell testing. Furthermore, both chemical and physical spatial variability has been noted in the water table aquifer (WTA). Statistical analysis of water quality data from well upgradient of the active CCR unit showed significant spatial variability in levels of Appendix III parameters. ANOVA tests (parametric or nonparametric, depending on data normality) showed significant variation among the means of all seven Appendix III parameters measured in upgradient wells. ANOVA tests (Levene s Equality of Variance) also showed significant differences in variance among upgradient wells for calcium, sulfate, and total dissolved solids. Because these tests compared wells upgradient of the CCR units, the spatial variability cannot be the result of recent contamination from the regulated unit. In addition, hydraulic conductivities measured for the screened zones of the six CCR unit groundwater monitoring wells vary by nearly two orders of magnitude (Barr, 2017, Table 3). These results suggest significant spatial variability, both chemical and physical, are present in the WTA; accordingly, intrawell testing is appropriate. 6

11 2.3.3 Non-Detects Parameters with concentrations below the detection limit (non-detect data) will be reported at the detection limit and will be appropriately accounted for during the statistical analysis. No substitution method will be utilized in the handling of non-detect data Duplicates Duplicate samples are typically collected as a quality control check on field and laboratory sampling methods, not for data analysis purposes. If Barr conducts a QA/QC review of the data and the results of a duplicate sample do not sufficiently match that of the original sample, a qualifier is typically applied and, based on the type of qualifier, the original data may be excluded from further analysis. Duplicate sample results will not be utilized in data evaluations beyond QA/QC procedures Seasonal Trends As required by the CCR rule, seasonal trends will be evaluated and adjusted, if detected. 2.4 SSI Evaluation For the purpose of this Site, Minnkota or its contractors will be conducting some combination of control charts and non-parametric prediction limits to determine if an SSI has occurred. Determination of a SSI in downgradient monitoring wells will be based on whether or not the control chart or prediction limit (whichever approach is applicable at each well location) for a given parameter has been exceeded at a downgradient well. Following the receipt of the final analytical report from the laboratory for all the monitoring locations, it will undergo a QA/QC evaluation by Barr. Once the validity of the data has been verified, time series plots, summary statistics, a distribution analysis, and a trend test will be conducted. If a trend is detected, additional evaluations may be warranted to determine if there is an off-site/non-ccr unit contamination source If an exceedance of a limit occurs in one or more downgradient wells, the suspected well(s) will be resampled. As previously mentioned, the 1-of-2 retesting strategy will be utilized. As recommended by ASTM, a statistically significant exceedance is not declared and should not be reported until the results of the verification resample are known (ASTM 2017, pg. 9). If the verification resample for a given parameter at a given well exceeds the control limit, then an SSI will be declared for that well-parameter pair. If not, detection monitoring will continue. 7

12 3.0 Reporting An annual groundwater monitoring report will be uploaded to the facility s operating record by January 31, 2018, and on an annual basis for subsequent years. The report will summarize the results of the year s groundwater monitoring, describe any modifications to the monitoring network, and propose changes to the monitoring network, parameters, or frequency of monitoring activities based on the evaluation of the annual data. The annual reports will also note the occurrence of any SSIs and proposed actions to resolve the SSI. Revisions to this Plan, including statistical methods, will be conducted on an annual basis and will be updated in the facility s operating record along with the rationale for the proposed change(s) as applicable. 8

13 4.0 References ASTM, D Standard Guide for Developing Appropriate Statistical Approaches for Groundwater Detection Monitoring Programs at Waste Disposal Facilities, ASTM International, West Conshohocken, PA, Barr, Groundwater Monitoring System Certification Report. ITRC, ITRC Guidance Document: Groundwater Statistics and Monitoring Compliance. US EPA, 2009, Statistical Analysis of Groundwater Monitoring Data at RCRA Facilities - Unified Guidance. EPA 530/R US EPA, 2015a. Hazardous and Solid Waste Management Systems; Management of Coal Combustion Residuals From Electric Utility, CFR Parts 257 and 261, Federal Register, Vol. 80, No. 74, April 17, US EPA, 2015b. ProUCL Version Technical Guide: Statistical Software for Environmental Applications for Data Sets with and without Nondetect Observations. EPA/600/R-07/041. October