Uncertainties in Trace Analysis Presented by Dr George

Uncertainties in Trace Analysis Presented by: Dr. George Duncan, P. Geo. , C. Chem. , MCIC, MRSC, Q. P. Environmental Consultant Performing Phase 1 & 2 Environmental Site Assessments

What are the issues? n n n Phase 2 Environmental Site Assessments (ESA) include sampling and analysis of soil, sediment and ground water. The “Qualified Person” (QP) must interpret the analytical results to determine if the site meets the “generic” (Province-wide) criteria or sitespecific, “risk-assessed” criteria. Exceedence of a criterion---even by the slightest amount can result in very expensive clean-ups.

Defining “Uncertainty” n n n What kinds of uncertainty are we dealing with? What are the sources of these? What do they mean in PRACTICAL terms to the “Qualified Person” (QP) who must interpret the lab results to assess the environmental condition of the site?

What kinds of uncertainty? To the Analytical Chemist. . Uncertainty: A non-negative parameter associated with the result of a measurement that characterizes the dispersion of the values that could reasonably be attributed to the measurand. (International vocabulary of metrology – Basic and general concepts and associated terms; ISO/IEC Guide 99: 2007 (VIM 2007)).

What kinds of uncertainty? n Standard Uncertainty: Uncertainty components are evaluated by the appropriate method and each is expressed as a standard deviation and is referred to as a standard uncertainty.

What kinds of uncertainty? n n Combined Standard Uncertainty: Standard uncertainty components are combined to produce an overall value of uncertainty known as the combined standard uncertainty. It is an estimated standard deviation equal to the positive square root of the sum of variances of all uncertainty components. This uncertainty is what the labs report.

What kinds of uncertainty? n Expanded Uncertainty: Expanded uncertainty (U) is obtained by multiplying the combined standard uncertainty by a coverage factor “k” to provide an interval within which the value of the measurand is believed to lie, with a specified level of confidence (e. g. 95%).

The Sources of Uncertainty n n n These exist at every stage of the analysis: In the balance used to weigh the standards and samples; In the volumetric pipets used to deliver and dilute the solutions; In the instrument’s readings; Etc.

In Practical terms… n The uncertainties quoted by the laboratories to their clients are typically reasonably “tight” (narrow range) for most parameters: Arsenic @ 18 mg/Kg = +0. 5 mg/Kg (3%) Cadmium @ 2 mg/Kg = +0. 08 (4%) Benzo-a-pyrene @ 0. 3 mg/Kg =+0. 022(7%) Benzene @ 0. 32 mg/Kg = +0. 063 (20%) TCE @ 0. 55 mg/Kg = +0. 061 (11%) NOTE: trace organics are much higher than metals

In Practical terms… n n n Uncertainty of Measurement is NOT a satisfactory measure of the UNCERTAINTY OF THE REPORTED RESULT. Accredited laboratories generally are excellent at minimizing uncertainty in the measurements they make---they have to be to stay accredited! Uncertainty of measurement as described above is a measure of the unavoidable variability inherent the analytical method BUT NOT IN THE SAMPLE.

In Practical terms… n n QP’s generally accept the analytical results on the C/A as “correct” and representative of the site condition in the area sampled. Under the current “Pass/Fail” system of O. Reg 153/04 (December, 2009), any lab result that exceeds a limit by ANY amount must be considered as correct and dealt with either through remediation or risk assessment--both of which can be very expensive.

In Practical terms… n n Problems arise when the C/A shows results just above the allowable limit. One or two repeat analyses on the SAME sample (if enough sample is left) often show some exceedences becoming “Non-detects”, some falling below the limits and some rising even higher, well beyond the uncertainty in the analytical method:

In Practical terms…

Real Uncertainty…inhomogeneity n n 1. 2. To the lab chemist, uncertainty in measurement is a measurable quantity that can be reported but it represents only a small part of the REAL uncertainty in the analytical result. The QP must also consider the uncertainty in the reported result created by two important criteria: Sample inhomogeneity Field inhomogeneity

Real Uncertainty…Sample Inhomogeneity n n Sample Inhomogeneity relates to how (un)well the component being analyzed is distributed throughout the sample. Where the sample (e. g. metals), can be ground to -355 μm (0. 355 mm) this alleviates the problem but only a little. If the sample cannot be ground (trace organics), serious problems arise. Slight, or even major exceedence of a limit may not be “real” and repeat analyses can bounce wildly.

Real Uncertainty…surrogates n n Surrogates are used for organic tests. All samples are spiked with compounds (usually deuterated analogs) representative of the analytes being determined but not found in environmental samples. The surrogates are spiked into the sample prior to any sample preparation steps and carried through the entire analytical process. Surrogate Recovery = ([measured concentration] / [theoretical concentration]) x 100

Real Uncertainty…what is real? n n Question: If the acceptable surrogate recovery is 50 – 150%, what is the “acceptable” deviation from the TRUE concentration in the sample? For PAHs & VOCs, percent surrogate recoveries in soil and water should be between 50 – 140%; for hydrocarbons 60 – 140%, for mercury 70 130% (matrix spike).

Real Uncertainty…the consequences n n Most QPs are engineers and geoscientists (non-chemists) and are unaware of these issues so the reported lab results are accepted at face value and any exceedences, including the slight ones, are treated as “real”. A “Record of Site Condition” cannot be registered for any site with an exceedence of ANY allowable limit. No RSC? ---no building permit will be granted

Real Uncertainty…the consequences n O. Reg 153/04: Data quality objectives outline the overall level of uncertainty that a QP will accept in collecting field data in order to develop a CSM. Data quality objectives are set to determine precision, accuracy, reproducibility, representativeness, and completeness for field data (i. e. , relative percent difference (RPD), matrix spikes, matrix spike duplicates, data qualifiers, etc. ).

Real Uncertainty…the consequences n n n “Data quality objectives for sampling and analysis plan are to be set beforehand” The data quality objectives for any field data collected during the phase two ESA must ensure that the decision making is not affected and the overall objectives of the investigation are met. O. Reg. 153/04 gives no instructions on how to achieve this. For example:

Real Uncertainty…the consequences n n n Stockpiles: “Sampling locations must be chosen so as to ensure uniformly distributed and representative sampling collection throughout the stockpile”. Question #1: How many samples should be gathered and what size should they be (Weight? Volume? ) ? Question #2: How do you sample the following?

Real Uncertainty…the consequences

Real Uncertainty…the consequences Lab sample soils are screened through a 2 mm sieve before analyses is begun (except for organics such as F 2 -F 4 hydrocarbs and PAHs).

Real Uncertainty…the consequences

Real Uncertainty…the consequences n n n Question #1(b): ………. what size (Weight? Volume? ) should they be? Reg 153/04 Answer: “Any sample(s) collected analyzed must be representative samples collected at locations and frequencies, following a sampling plan determined by the QP The lab answer: Fill the 100 m. L soil jar to the top and we’ll use that for metals, F 2 -F 4, and PAHs, etc. For VOCs put 2 – 3 g in a VOC vial!

Real Uncertainty…reducing it Reducing Uncertainty in the lab The labs are already doing all they can to produce accurate results. The major source of real uncertainty is sample inhomogeneity. Fine grinding of the sample helps but this is not allowed for organic COC analyses. The few grams of sample actually analyzed is often not representative of the 100 g of sample in the soil jar. n

Real Uncertainty…reducing it If one of the large ptles is lead paint and the rest are native soil from the site, would the analysis result be close to the “true value” if the true value is 1000 ppm?

Real Uncertainty…reducing it Reducing Uncertainty at the site This is currently an impossible task because each tiny sample submitted for analysis is forced to represent many tons of soil. Soils are notoriously heterogeneous and many contaminants occur as discrete particles. The Reg pays only lip-service to collecting REPRESENTATIVE samples but truly representative samples will be far too large (>10 100 Kg) to be analyzed at the lab. n

Real Uncertainty…reducing it

Real Uncertainty…reducing it n n For COCs in coarse soils at low ppm levels (e. g. As, Be, Cd. . ), unless you grind ALL the sample very fine (e. g. <-200 mesh or 74 μm) and perform repeat analyses, the C/A can be way off the mark. It explains why results at near-limit values can bounce wildly, making cleanup decisions very difficult. Consider the following site investigation of a sandblasting site used to remove lead paint.

The problem is statistics n n n The site soils are coarse sand of average particle size 1 mm. Let’s assume the ACTUAL Pb concn = 1000 mg/Kg (O. Reg limit = 120 mg/Kg) In 1 cc of soil (~1. 4 g) there are 1000 ptles max. Actually it’s only ½ this due to the airspace between particles.

The problem is statistics n A 1. 4 g soil sample (1 cc) CANNOT PRODUCE THE CORRECT ANSWER since it will contain 0, 1, 2 or more 1 mm ptles of lead paint! Each paint ptle weighs 8 x more than a sand ptle so if the sample has no paint particles, the C/A indicates 0 mg/Kg but if the sample has 1 ptle of paint, it will produce a lab result of ~8000 mg/Kg (a 1 mm 3 paint ptle weighs ~11 mg)!

The problem is statistics n n The only way to get the correct result is to take a much larger sample! Remember: In the 1. 4 g of soil the grain size is 1 mm so there are only ~500 ptles in the sample. The true Pb concn. is 1000 mg/Kg so you need ~8 x 1. 4 g sample (and be lucky enough) to catch one ptle of lead paint and get the correct result!

The problem is statistics n n You can grind the 11. 2 g sub-sample as fine as you like and get very reproducible analysis results but they will all be equally wrong if: Your sub-sample has only 5 or 6 ptles of paint or 12 or 14 ptles of paint! Remember: O. Reg. 153/04 screens the whole soil sample through a 2 mm screen and the 5 10 g sub-sample is taken from this. What happens if the Pb ptles are 1. 9 mm?

The problem is statistics n n The limit for lead in soil is 120 mg/Kg The cardinal rule in sub-sampling is: once you lose sample representativeness, you can NEVER regain it---not matter how fine you grind. If the <2 mm sub-sample taken for grinding is non- representative of the whole lab-sample, grinding is useless! We haven’t yet discussed field sampling!!

The problem is statistics n n n Consider the highly toxic metals such as arsenic, beryllium, cadmium, etc. The allowable limits for these are in the low ppm range (1 – 20 mg/Kg). 1 ptle of arsenic in 1 cc = ~37 mg/Kg (assuming no density difference). The limit is 18 mg/Kg for Table 2 soils. How much sample are you going to have to collect and send to the lab to be representative? REMEMBER: ANY EXCEEDENCE = SITE CLEANUP

The problem is statistics n n NOTE: none of the above discusses the much more serious issue of how to get a 3 g sample of methanol-preserved soil to represent 50 m 3 (3 truckloads or ~ 65 Tons) from a soil pile or a borehole you just drilled. YOU CAN’T! Statistically you are not even in the ballpark. The mining exploration industry would never base any decision to build a mine on such paltry numbers and sizes of samples.

Major changes to O. Reg 153/04 are needed n n n Much larger sample sizes are needed to begin to be representative but labs cannot handle these. Leach-testing of large samples may provide a more reliable and a more relevant estimate of a site’s condition. For MEANINGFUL lab analyses to be done, the sample size MUST be established at the SITE, not at the lab!

Major changes to O. Reg 153/04 are needed n n The polling industry (Gallup, etc. ) first calculates the # and location of people to be polled (“sample size and distribution”) before commenting on the accuracy of the poll (“correct within 3% points, 19 times out of twenty). The ESA industry has this backwards by limiting the size of the sample no matter the size and distribution of the population!

O. Reg 153/04 is currently under Review n n n Hopefully, we will see the changes necessary to correct these issues. Until then we will continue to struggle with C/As showing dubious exceedences of regulatory limits resulting in many $ millions spent on needless investigations and cleanups. REMEMBER: If the C/A shows an exceedence of any limit---clean it up or risk assess it!