Incorporating Statistics into a Waste/Feedstream Analysis Plan

 

Gerald J. Drake

Melissa L. Douglas

 

Compliance Strategies & Solutions, Inc.

1301 Regents Park Drive, Suite 203

Houston, Texas 77058

 

ABSTRACT

 

Every hazardous waste combustion facility is required to develop and comply with a Waste Analysis Plan (WAP) and, in the future, a Feedstream Analysis Plan (FAP).  The written WAP/FAP is required to contain the procedures the facility uses to treat, store, and/or dispose of hazardous waste in compliance with regulatory and permit requirements.  Hazardous waste combustion facilities are limited to specified feed rates for such constituents as ash, metals, and chlorine.  Many facilities analyze each batch of material to be burned in the combustion device in order to demonstrate compliance with the constituent feed rates.  Compliance is typically demonstrated by specifying a maximum feed rate of feed streams to the combustion device that does not exceed any of the allowable constituent feed rate limits.

 

This paper presents a statistical approach to demonstrate compliance with the constituent feed rate limits that does not require the sampling and analysis of each batch of material to be burned.  The statistical approach presented uses Upper Tolerance Limit (UTL) statistics and includes discussions on the normal distribution of data, outlier analysis, maximum feed rate determinations, frequency of sampling and analysis, and special situations associated with the statistical approach.

 

Owing to the high costs associated with the sampling and analysis of each batch of material to be burned, this statistical method provides a cost-effective alternative to demonstrating compliance with regulatory and permit limits.

 

INTRODUCTION

 

Every hazardous waste combustion facility is required to develop and comply with a Waste Analysis Plan (WAP) and, in the future, a Feedstream Analysis Plan (FAP).1,2  The written WAP/FAP is required to include the parameters to be analyzed, the sampling and analysis methods to be used, the frequency that analyses will be repeated, and how the data will be used to document compliance with applicable feed rate limits.  Typical feed rate limits that hazardous waste combustion facilities must comply with include ash (except for cement kilns), metals, and chlorine.

 

Demonstrating compliance with a feed rate limit is typically performed by continuously monitoring and recording the feed rate of the appropriate feed streams and knowing the concentration of the regulated parameters in each feed stream.  This information is used to demonstrate that the feed rate of each regulated constituent is in compliance with the allowable constituent feed rate limits.

 

One method for demonstrating compliance with constituent feed rate limits is to analyze each batch of material prior to being fed to the combustion device.  The measured concentration of each regulated parameter is then used to calculate the maximum allowable feed rate for each feed stream to the combustion device that will maintain compliance with the constituent feed rate limitation.  This method can be used at off-site facilities, at facilities that use on-site laboratories, and at facilities with highly variable feed streams.

 

Another method for demonstrating compliance with feed rate limits is to use a statistical approach for determining the concentration of each constituent in the feed streams.  The statistically derived value is then used to calculate the maximum allowable feed rate for each feed stream to the combustion device that will maintain compliance with the constituent feed rate limitation.  This method can be used at on-site facilities, at facilities that use off-site laboratories, and at facilities with consistent feed streams.

 

This paper presents a statistical approach that does not require the sampling and analysis of each batch of material prior to being fed to the combustion device, and can be used to demonstrate compliance with constituent feed rate limits.  This statistical approach also establishes a sampling and analysis frequency based on the consistency of the feed stream.  The statistical approach presented is based on the EPA’s Waste Analysis Guidance for Facilities That Burn Hazardous Waste (hereafter, EPA Guidance) and uses Upper Tolerance Limit (UTL) statistics.3

 

STATISTICS METHODOLOGY

 

This section describes the methodology associated with the use of statistics in a WAP/FAP.  This section includes discussions on UTLs, the outlier analysis, the normality analysis, calculating the UTL, maximum feed rate analysis, frequency of sampling and analysis, special situations, and cost savings related to the use of statistics.

 

General UTL Information

 

Section 3.3.1 of the EPA Guidance presents the UTL based approach to using statistics for characterizing constituent concentrations in feed streams.  The UTL is generally defined as an estimate of the probability that a fixed percentage of a sample population will not exceed a certain value.  The EPA Guidance indicates that the minimum UTL value that should be used is the one-sided upper 95 percent tolerance bound (the confidence level) that exceeds at least 95 percent of the sample population (the coverage).  In other words, it can be stated with 95 percent confidence that 95 percent of all samples will not exceed the UTL.  It should be noted that the 95 percent confidence level and 95 percent sample population are the minimum recommended in the EPA Guidance.  Higher levels can be used such as a 99 percent confidence level for 97.5 percent of the population.

 

When using this statistical method, the calculated UTL value is considered to be the “known” concentration for each measured constituent in the feed stream.  The UTL value for each constituent is then used to demonstrate compliance with the constituent feed rate limits.  Because the UTL methodology is a statistical approach for demonstrating compliance, it must be noted that there is always a finite possibility that a constituent concentration in any given sample will exceed the UTL.  This occurs because 100 percent confidence cannot be achieved on 100 percent of the population.  The EPA Guidance acknowledges that this is an accepted fact when using statistical characterizations and facilities can expect this to occur periodically.  A discussion on values that exceed the UTL is presented later in this paper.

 

In order to use the UTL methodology, several pieces of information must be known.  The number of values, the arithmetic average, and the standard deviation must be known for the applicable data set.  Also, the data set must be normally distributed in order to use the UTL methodology.  Each of these is easily determined by developing a computer spreadsheet program to perform the required calculations.

 

The general WAP/FAP procedure presented for using the UTL statistical methodology is outlined as follows:


·         Evaluate the laboratory data (outlier analysis);

·         Evaluate the distribution of the data set (normality analysis);

·         Calculate the UTL;

·         Evaluate compliance with the feed rate limits (maximum feed rate analysis); and

·         Determine the frequency of sampling and analysis.

 

Each of these procedures is presented in the following sections.

 

Outlier Analysis

 

Each data point received from the laboratory should be evaluated as an outlier to determine if there are any errors in the sampling, analytical, or reporting methods.  Data that is attributable to poor analytical quality, sampling errors, and analytical errors should not be included in the data set and should not be used to document compliance with feed rate limits.  A well-written and implemented WAP/FAP should minimize the receipt of poor data.  However, out-of-calibration analytical instruments, non-representative samples, and the use of wrong or unapproved analytical methods have been known to occur, and data obtained from such should not be used.  It should be noted that the rationale for removing poor data from the data set should be documented in the facility operating record and presented to regulatory agencies, if requested to provide such information.

 

Section 4.1 of the EPA Guidance identifies ASTM Method E178-80 as one method for performing an outlier analysis.  The current edition of this method is ASTM Method E178-94.4  Section 4 of ASTM Method E178-94 discusses the outlier methodology associated with a single sample, which is appropriate for WAP/FAP analyses.  For the purposes of the outlier analysis, only values on the high side of the data set are to be tested as outliers.  This is considered appropriate because the purpose of the WAP/FAP is to demonstrate compliance with feed rate limits.  Non-compliance with a feed rate limit could only occur with high ash, metal, or chlorine data points.

 

The outlier analysis involves the evaluation of each new data point against the entire data set to determine if it is a statistical outlier.  This is done by calculating a value for the new data point (referred to as Tn) and comparing this value to a critical value (Tc).  The critical Tc value is obtained from Table 1 of ASTM Method E178-94.  As with the UTL values, different confidence levels can be used for the critical Tc values.  In keeping with the EPA Guidance, a minimum confidence level of 95% (specified as the Upper 5% Significance Level in ASTM Method E178-94, Table 1) should be used.  If the calculated Tn value is greater than the critical Tc value, there is statistical evidence that the new data point is an outlier from the data set.  Equation 1 is used to calculate the Tn value for the new data point.

 

Equation 1.  Calculating Tn for the new data point.

 

 

where:

 

Tn         =          Calculated value for the new data point;

 

Xn        =          New data point value being tested;

 

        =          Arithmetic average of the data set (see Equation 2); and

 

Sd         =          Standard deviation of the data set (see Equation 3).

 

The arithmetic average of the data set is calculated by summing each value in the data set and dividing by the number of data points.  Equation 2 is used to calculate the arithmetic average of the data set.

 

Equation 2.  Calculating the arithmetic average of the data set.

 

 

where:

 

        =          Arithmetic average of the data set for parameter X;

 

Xi        =          The ith data value for parameter X; and

 

n          =          Number of valid data points in the data set.

 

The standard deviation of a data set represents the variability of the data points in the data set as compared to the arithmetic average of the data set.  Equation 3 is used to calculate the standard deviation of the data set.

 

Equation 3.  Calculating the standard deviation of the data set.

 

 

where:

 

Sd         =          Standard deviation of the data set;

 

Xi        =          The ith data value for parameter X;

 

        =          Arithmetic average of the data set for parameter X (see Equation 2); and

 

n          =          Number of valid data points in the data set.

 

If a new data point is determined to be an outlier by this methodology, an investigation should be initiated to determine if an error in sampling or analysis occurred.  If an error in sampling or analysis is clearly and definitively discovered, the associated data point should not be added to the data set.  If no error in sampling or analysis is discovered, the decision to eliminate the data point should be made more cautiously.  The procedures and criteria for eliminating any data point from the data set should be developed by each facility and, preferably, should be documented in the WAP/FAP.  In all cases, the results of the outlier analysis and any associated investigation should be documented in the facility operating record.

 

Normality Analysis

 

One of the criteria for using the UTL statistical methodology is that the data set must be normally distributed.  A normal distribution of data can be represented as a “bell-shaped” curve with the right and left sides of the mean value being mirror images of each other.  This type of distribution is also called a Gaussian distribution.  Therefore, in order to use the UTL, the raw data in the data set must be evaluated for normality.  If the raw data in the data set is not normally distributed, the data should be log-transformed and re-evaluated for normality.  If the data is determined to be log-normally distributed, the log-transformed data should be used to calculate the UTL values.

 

One method for evaluating the data set for normality is the Coefficient of Variation test presented in EPA’s Statistical Analysis of Ground-Water Monitoring Data at RCRA Facilities guidance document.5  The Coefficient of Variation test is a quick and simple method for checking for gross non-normality of the data set.  The Coefficient of Variation is calculated using Equation 4.

 

Equation 4.  Calculating the Coefficient of Variation.

 

 

where:

 

CV      =          Coefficient of Variation;

 

Sd         =          Standard deviation of the data set (see Equation 3); and

 

        =          Arithmetic average of the data set (see Equation 2).

 

The calculated Coefficient of Variation is compared to the numeric value of 1.00.  If the calculated Coefficient of Variation is less than 1.00, the data set is considered to be normally distributed.

 

UTL Calculation

 

The UTL for the data set can be calculated after a new data point has been added to the data set and the data set has been determined to be normally distributed.  If the data requires log-transformation in order to become normally distributed, the UTL must be calculated using the log-transformed data.  The UTL is calculated using Equation 5 as presented in Section 3.3.1 of the EPA Guidance.

 

Equation 5.  Calculating the UTL for the data set.

 

 

where:

 

UTL    =          Upper Tolerance Limit;

 

1 – a    =          Desired level of confidence (i.e., 0.95, 0.975, 0.99, and 0.995);

 

r          =          The coverage as a decimal fraction of samples predicted to fall below the UTL (i.e., 95%);

 

n          =          Number of valid data points in the data set;

 

        =          Arithmetic average of the data set (see Equation 2);

 

K         =          The factor for a one-sided tolerance limit at the appropriate confidence level, coverage, and number of data points; and

 

Sd         =          Standard deviation of the data set (see Equation 3).

 

As previously discussed, a minimum confidence level of 95 percent (1 - a = 0.95) and a minimum coverage of 95 percent (r = 0.95) must be used.  One source of values for K is the Tables for Normal Tolerance Limits, Sampling Plans, and Screening by Odeh and Owen.6  The calculated UTL value for the data set can be used as the “known” value for determining compliance with a feed rate limit.

 

Maximum Feed Rate Analysis

 

One method for demonstrating compliance with a constituent feed rate limit is to calculate the feed rate contributed by each feed stream and summing the total.  The total constituent feed rate is then directly compared to its feed rate limit to demonstrate compliance.

 

Another method for demonstrating compliance is to back-calculate the maximum allowable feed rate for each feed stream using the feed rate limit for each constituent and the concentration in each feed stream (the UTL).  The calculated maximum allowable feed rate for each feed stream is then compared to its permitted feed rate limit.  As long as the calculated value is greater than the permitted feed rate limit, the combustion device can operate up to the maximum feed rate limit without exceeding a constituent feed rate limit.  If the calculated value is less than the permitted limit, feed rates must be reduced to the calculated value in order to ensure compliance.  It must be noted that this calculation must be performed for all applicable constituents.  The lowest allowable feed rate for any of the constituents must be used to ensure compliant operations.

 

A variation of the maximum allowable feed rate analysis described above is the maximum allowable constituent concentration analysis.  The maximum allowable constituent concentration method back-calculates the allowable constituent concentration in each feed stream using the feed rate limit of each constituent and the feed stream feed rate.  Compliance is demonstrated when the maximum allowable constituent concentration and the feed stream feed rate are below the constituent UTL and the permitted feed stream feed rate limit, respectively.

 

The utility of back-calculating the feed stream feed rate is that it readily shows the constraining constituent and its affect on the feed rate to the unit.  This method also provides a comparison for all the constituents against the feed stream feed rate.  This method is also easily implemented using a computer spreadsheet program.

 

The UTL value and the permitted feed rate limit for each of the regulated parameters (i.e., ash, metals, and chlorine) are used to back-calculate the maximum allowable feed rate for each feed stream to the combustion device.  The maximum allowable feed rate for UTL values with units of µg/g is calculated using Equation 6.

 

Equation 6.  Calculating the maximum allowable feed rate with units of µg/g.

 

 

where:

 

Maximum Feed Rate  =          Maximum allowable feed rate (lb/hr);

 

Permit Limit                =          Constituent feed rate limit (g/hr);

 

106                               =          Conversion factor for µg’s in one gram;

 

UTL                            =          Upper Tolerance Limit (µg/g); and

 

453.6                           =          Conversion factor for grams in one pound.

 

The maximum allowable feed rate for UTL values with units of percent is calculated using Equation 7.

 

Equation 7.  Calculating the maximum allowable feed rate with units of percent.

 

 

where:

 

Maximum Feed Rate  =          Maximum allowable feed rate (lb/hr);

 

Permit Limit                =          Constituent feed rate limit (g/hr);

 

100                              =          Conversion factor for percent;

 

UTL                            =          Upper Tolerance Limit (%); and

 

453.6                           =          Conversion factor for grams in one pound.

 

For example, assume that a combustion device has a chlorine feed rate limit of 1000 g/hr, a feed rate limit of 2000 lb/hr for a hazardous waste feed stream, and a data set UTL for chlorine of 1000 µg/g.  Using Equation 6, the calculated maximum allowable feed rate value is 2205 lb/hr.  This value is greater than the hazardous waste feed rate limit of 2000 lb/hr.  Therefore, the combustion device is allowed to feed the hazardous waste up to the feed rate limit of 2000 lb/hr and maintain compliant operations.

 

As another example, assume that the same combustion device has an ash feed rate limit of 1000 g/hr and a data set UTL for ash of 0.12 percent.  Using Equation 7, the calculated maximum allowable feed rate value is 1837 lb/hr.  This value is less than the hazardous waste feed rate limit of 2000 lb/hr.  Therefore, the combustion device is required to reduce the hazardous waste feed rate to less than 1837 lb/hr to maintain compliant operations.

 

For boilers and industrial furnaces subject to the 40 CFR Part 266 requirements, the aggregate risk of the carcinogenic metals (arsenic, beryllium, cadmium, and chromium) is currently required to be assessed.7  For the Tier I feed rate screening limits, the aggregate risk is defined as the summation of the ratio of a metal's feed rate to its regulatory limit.  The summation for all carcinogenic metals must be less than 1.0.  The aggregate risk for the Tier I metals is calculated using Equation 8.

 

Equation 8.  Calculating the aggregate risk for Tier I metals.

 

 

where:

 

AFRi   =          The actual feed rate of carcinogenic metal "i" (g/hr);

 

FRSLi  =          The Tier I feed rate screening limit for metal "i" (g/hr); and

 

n          =          Number of carcinogenic metals.

 

The feed rate of each carcinogenic metal (AFRi) is calculated using the UTL for each metal and the feed rate of the feed stream.  To back-calculate the maximum allowable feed rate for a feed stream, Equation 8 is converted into Equation 9.

 

Equation 9.  Calculating the maximum allowable feed rate for the aggregate risk.

 

 

where:

 

Maximum Feed Rate  =          Maximum allowable feed rate (lb/hr);

 

106                               =          Conversion factor for µg’s in one gram;

 

453.6                           =          Conversion factor for grams in one pound;

 

UTL                            =          Upper Tolerance Limit (µg/g);

 

Tier I                           =          Tier I feed rate screening limit (g/hr);

 

As                                =          Carcinogenic metal arsenic;

 

Be                                =          Carcinogenic metal beryllium;

 

Cd                               =          Carcinogenic metal cadmium; and

 

Cr                                =          Carcinogenic metal chromium.

 

Equations similar to Equation 9 can also be developed to calculate the maximum allowable feed rate using the aggregate risk equations for Adjusted Tier I feed rate screening limits, Tier II emission rate screening limits, Tier III metal control limits, and any combination of these limits.

 

Frequency of Sampling and Analysis

 

Section 3.3.1 of the EPA Guidance presents a statistical procedure for specifying an appropriate sampling and analysis frequency for feed streams being fed to combustion devices.  The sampling and analysis frequency is calculated using Equation 10.

 

Equation 10.  Calculating the sampling and analysis frequency.

 

 

where:

 

acalc      =          One minus the confidence level used to calculate the UTL; and

 

g          =          Days per year that the waste stream is generated.

 

For UTL values calculated using confidence levels of 95%, 97.5%, 99%, and 99.5%, the values of acalc are calculated to be 0.05, 0.025, 0.01, and 0.005, respectively.  Assuming that the waste stream is generated 365 days per year, the sampling and analysis frequency for the different confidence levels can be calculated.  Table 1 presents the calculated sampling and analysis frequency for each specified confidence level.

 

Table 1.  Sampling and Analysis Frequency.

 

Confidence Level

acalc

Frequency (Number of Samples per Year)

95%

0.05

18.25

97.5%

0.025

9.125

99%

0.01

3.65

99.5%

0.005

1.825

 

As shown in Table 1, as the confidence level increases for calculating the UTL values, the frequency of sampling and analysis decreases.  It should be noted that different confidence levels could be used for the different analytical parameters.  For example, metal analyses may use a 99% confidence level and ash analyses may use a 95% confidence level.  In this case, the number of samples required to be analyzed on an annual basis will be different for the metals (4 samples per year) versus ash (19 samples per year) because of the different confidence levels.

 

Special Situations

 

When developing and applying the statistical methodology presented above, several special situations are noted.  These situations are presented in the following sections.

 

Exceedance of UTL Values

 

Section 3.3.2 of the EPA Guidance presents a discussion on the procedure to be used when an analytical value is obtained that exceeds the UTL for the data set.  This discussion indicates that the previous UTL should continue to be used for calculating the feed rate limits, immediate daily sampling of the feed stream must be conducted until all regulated constituents are below the UTLs for three consecutive days, and, if the UTL is exceeded two times, discontinue use of the statistical methodology until a new feed stream profile is obtained.  The EPA Guidance states that there “is always a finite possibility that concentrations of constituents in any given sample will exceed the UTL”.  This is the case because 100 percent confidence cannot be achieved on 100 percent of the population.  Therefore, it is expected that this situation will occur periodically.

 

The underlying assumption of the exceeded UTL procedure is that the feed stream has changed in some fundamental way such that it is no longer the same feed stream.  The establishment of a new feed stream profile is the EPA’s basis for redefining the data set to account for these changes.  In reality, small variations in the generation process could cause a UTL value to be exceeded.  If a small change results in an exceeded UTL for two samples, continued sampling is required to re-establish the feed stream profile.  If the small variation in the generation process is corrected, the data set would begin receiving the lower values previously seen while maintaining the higher values from the new feed stream profile.  If the small variation in the generating process occurs again, the same series of events could occur.

 

Another problem associated with exceeded UTLs pertains to its affect on the permit limits.  If a UTL is exceeded for a constituent that is not close to its feed rate limit, a new feed stream profile would need to be established only to follow the EPA Guidance.  For example, assume that a combustion device has a barium feed rate limit of 1000 g/hr, a UTL for barium of 10 µg/g, and a feed rate limit of 1000 lb/hr for a hazardous waste feed stream.  The calculated barium feed rate for this UTL is 4.54 g/hr.  If an analytical value of 11 µg/g (4.99 g/hr) is received, immediate daily sampling would be required pursuant to the EPA Guidance procedure.  If two results above the 10 µg/g UTL are received, the statistical methodology would have to be discontinued for all the parameters until a new feed stream profile is developed.  This would be required even though the actual feed rate of barium is less than 5.00 g/hr and the allowable barium feed rate limit is 1000 g/hr.  In this case, it does not appear that establishing a new feed stream profile is warranted.

 

Analytical Methods

 

The data set for a constituent should be comprised of data that is generated using one analytical method.  When switching analytical methods, it is always preferable to establish a new feed stream profile for the new method.  The integrity of the data set is compromised when the data is generated using different methods.  For example, suppose that the chlorine data in a data set is generated using a titration method that produces a total halogen value instead of an ion-specific chlorine value.  If the analytical method is changed to one that produces an ion-specific chlorine value, the data in the data set will become contaminated with the non-identical data.  In this case, a new feed stream profile should be established for chlorine.

 

Another consideration relating to all analytical measurements is the method detection limit.  It is always preferable to maintain the same method detection limit for the entire data set, especially if non-detect values are prevalent in the data set.  If the method detection limit suddenly increases, a new data point could be determined to be an outlier or could exceed the data set UTL.  In this case, the increased detection limit should be realized when investigating the outlier or the exceeded UTL.

 

Another result that can be attributable to sudden method detection limit changes is a drastic change in the UTL value.  This occurs because of the way the UTL is calculated.  As seen in Equation 5, the UTL is a direct function of the data set average and standard deviation.  Therefore, any drastic change in the data set causes immediate changes to the standard deviation of the data set and could eventually change the data set average.  A sudden increase or decrease in detection limits would be seen as an increasing UTL.  Because the UTL value is used in determining compliance with feed rate limits, reduced feed rates for feed streams to the combustion device could result.  Therefore, changing method detection limits, whether caused by increasing laboratory accuracy or by switching analytical laboratories, will cause changes in UTL values.

 

Adding Outliers to the Data Set

 

A statistical outlier may occur that cannot be explained by an error in sampling or analysis.  The facility is then faced with the decision of eliminating the data point or including it in the data set.  The outlier may have resulted from an undiscovered change in the generating process or from an undiscovered error in sampling and analysis.  In this case, the conservative approach is to include the data point in the data set.  This is considered conservative because the UTL value would increase due to the increasing average and standard deviation of the data set caused by the outlier.  The increased UTL value then causes a decrease in the feed rate of feed streams to the combustion device in order to maintain compliant operations.

 

Robustness of the Data Set

 

In general, the more data points included in a data set, the harder it becomes to change the UTL with a single data point.  This is referred to as the robustness of the data set.  The more robust the data set, the more stable the results.  Therefore, data sets with relatively few data points will tend to see larger swings in the UTL values when compared to data sets with numerous data points.

 

Maximum Data Set Size

 

There are several reasons that the data set may need to be limited in size.  One reason is the table of critical Tc values used for the outlier analysis.  The table included in ASTM Method E178-94 is limited to 147 data points.  The data set is limited to 147 data points for this reason and, therefore, would be truncated based on the number of data points.  However, an option is to perform the outlier analysis only for the most recent 147 data points and continue to determine the UTL based on the entire data set.  This is possible because the referenced table of K values includes factors to infinity.6

 

Another reason for limiting the size of the data set is based on time.  For example, when using the 99.5% UTL values, only two samples per year are required to be analyzed.  At that rate, it would take almost 50 years of sampling and analysis to develop a data set with 100 data points.  Due to small but incremental changes in the generating process, analytical equipment, analytical reagents, analysis methods, etc., it may be prudent at some point in time to eliminate some of the oldest data.  This is referred to as truncating the data set based on time.

 

If a maximum size is placed on the data set, the method for doing such must be included in the WAP/FAP.  A complete description of the method or methods to be employed and the rationale should be described.

 

Possible Non-Compliance

 

When using a statistical approach for demonstrating compliance with feed rate limits, there is a finite possibility that non-compliance with a feed rate limit can occur.  The statistical approach reviews the history of data associated with the feed stream, and provides an indication of continued compliance at a certain level of confidence.  At any time, a single data point could be received that, when used to calculate a constituent feed rate, causes an exceedance of a feed rate limit.  As stated in the EPA Guidance, “If such a circumstance occurs, use of a statistical sampling and analysis strategy is not a shield against enforcement action and the adequacy of the analysis may be considered in penalty calculations.”  Therefore, it is imperative that a facility thoroughly understands the benefits and liabilities of using a statistical approach for demonstrating compliance.  It is also imperative that clear, well-defined procedures are developed and included in the WAP/FAP for the statistical approach, and that the procedures be diligently implemented.  It is also advisable for the facility to develop a procedure that addresses non-compliance situations and the actions to be taken to return to compliant operations.

 

Cost Savings

 

Incorporating statistics into a WAP/FAP can result in considerable cost savings for a facility.  For a facility that is currently analyzing each batch of material prior to burning, a reduction in sampling and analysis would be realized.  Depending on the confidence level used for the UTL determination, a minimum of two and a maximum of 19 samples per year would be required.  The higher the confidence level, the lower the number of samples required.  Therefore, for the most consistent feed streams, a significant reduction in the number of samples and analyses would be realized.

 

CONCLUSIONS

 

Incorporating statistics into a WAP/FAP is a viable method for complying with the sampling and analysis requirements for feed streams to hazardous waste combustion devices.  The EPA has provided guidance on using the UTL method of statistics as part of the waste analysis process.  This guidance is used to produce a process of waste analysis that includes an outlier analysis, normality analysis, UTL calculation, maximum feed rate analysis, and a sampling and analysis frequency.  Several insights are provided for developing, implementing, and interpreting the results of the data set.  The methodology presented can be incorporated into a WAP and FAP, and is easily implemented using a computer spreadsheet program.  Significant cost savings could be realized by using the UTL method of statistics due to the reduced number of samples requiring analysis.

 

REFERENCES

 

1.      40 CFR 264.13, 40 CFR 265.13, 40 CFR 266.102(a)(2)(ii), and 40 CFR 266.103(a)(4)(ii).

 

2.      40 CFR 63.1209(c).

 

3.      Waste Analysis Guidance for Facilities That Burn Hazardous Waste, Draft; United States Environmental Protection Agency; EPA 530-R-94-019; October 1994.

 

4.      ASTM Method E 178-94; Standard Practice for Dealing with Outlying Observations; American Society for Testing and Materials; September 1994.

 

5.      Statistical Analysis of Ground-Water Monitoring Data at RCRA Facilities, Interim Final Guidance; United States Environmental Protection Agency; EPA 530/SW-89-026; February 1989.

 

6.      Odeh, R.E.; Owen, D.B.; Tables for Normal Tolerance Limits, Sampling Plans, and Screening; Marcel Dekker, Inc.; pp 18-33.

 

7.      40 CFR 266.106.

 

KEY WORDS

 

Combustion Devices

 

Feedstream Analysis Plan

 

Hazardous Waste

 

Statistics

 

Upper Tolerance Limit

 

Waste Analysis Plan