Department of Health Seal

TGM for the Implementation of the Hawai'i State Contingency Plan
Section 8.2
DATA QUALITY CONTROL AND DOCUMENTATION

8.2 DATA QUALITY CONTROL AND DOCUMENTATION

Table 8-1 summarizes the data quality levels and objectives of “field screening.” and “field analysis” data. Field analysis is used to discuss data quality control for quantitative field methods that can approach and even exceed laboratory quality data, though in many situations the cost of field analysis level data (with laboratory confirmation analyses) could exceed the cost of DU-MIS sampling with laboratory analysis.

Table 8-1 Data Quality Levels for Field Screening and Field Analysis Methods
Data Quality Level Purpose of Testing (Examples) Example Methods
Screening – Qualitative or Semi-Quantitative
  • Approximating known contaminate magnitude
  • Assist in identification of DUs
  • Initial estimates of excavation limits
  • Health and safety evaluations
Portable PID, VOC headspace analysis, in situ XRF, colorimetric analyses, and in situ sensor or probe equipment.
Field Analysis – Quantitative (with QC similar to a fixed laboratory analysis)
  • Site characterization
  • Excavation delineation
Ex situ XRF, and many immunoassay, colorimetric, and turbidimetric kits, with laboratory confirmation data

Field screening results can be presented in terms of presence or absence or in terms of relative concentration, for example above or below a pre-specified limit. The Photo Ionization Detector (PID), used to identify the presence and relative abundance of volatile organic chemicals (VOCs) in soil, is one of the most common field screening tools (see Subsections 8.4.2 and 8.4.5). Some PIDs can be used to measure both total VOCs in vapors emitted from soil as well as concentrations of a limited number of individual compounds (e.g., benzene). Another example of a field screening tool is a colorimetric field test kit for petroleum that is pre-set to a target concentration of TPH (e.g., 500 mg/kg). In some cases, screening may identify the total concentration of a family of contaminants but not concentrations for targeted, individual compounds needed for final decision making. This is true for some PAH field test kits. Due to potential matrix effects from moisture and large particles, the in situ use of a field XRF is considered “screening” level data, even though it reports specific concentrations of metals in soil.

Quality control procedures for the field screening methods typically consist of:

  1. Familiarity with the instrument operation and operations manual;
  2. Instrument calibration consistent with operations manual instructions; and
  3. Written documentation of calibration(s), any instrument maintenance, and data collected in the field.

Field analysis results are, in contrast, intended to match quantitative data that would be obtained from a laboratory. The objective of field analysis is to estimate the mean concentration of a targeted contaminant in a specific area and volume of soil (i.e. DU) and/or for an individual sample. Examples include immunoassay kits to test for PCBs or the use of an XRF to test processed, ex situ soil samples for metals. In the case of the immunoassay kits the mass of soil tested is similar to the correlative method used by a laboratory (e.g., ten grams). A single test may be adequate to represent the sample in terms of sampling theory, assuming that the sample is adequately processed and subsampled (see representative sampling information in TGM Section 4). Averaging multiple tests of a single sample will be required with use of a field XRF, however, since the mass tested during a single reading is relatively small (e.g. approximately one gram).

Data for discrete samples, while accurate for the specific mass of soil tested, are difficult to extrapolate to bulk sample submitted for testing or identify localized areas of contamination due to random, small-scale heterogeneity (see TGM Section 4 and XRF case study in Subsection 8.4.1). The collection of grid point samples of sufficient mass (e.g., 300+g) from multiple points within a small area (e.g. 1-2m2), rather than from single locations can help improve data representativenes and reduce this uncertainty. Note that such samples are sometimes referred to as "composites" by field workers, although in a strict sense of sampling theory (and in some regulatory applications) this term specifically infers the mixing of soil from otherwise separate DUs and its use as described above is discouraged (refer to Section 4.4.11). While potentially useful to help identify large-scale patterns of contamination and plan more detailed investigations or for initial identification and removal of areas of contaminated soil, HDOH considers this type of data to be adequate for screening purposes only. The designation of DUs and collection of MI sample data as described in Section 3 and Section 4 is recommended for final decision making.

Use of field analysis methods requires more attention and quality control in the field than screening methods, including:

  1. Calibration of instrument;
  2. Preparation of field standards using soil from the site or same soil type from near the site;
  3. Preparation of comparibility curves (e.g., XRF vs lab extraction data; updated as the project proceeds);
  4. Documentation of representative sampling/analysis methods
  5. Field replicates;
  6. Blank data;
  7. Documentation of data printouts or read-outs; and
  8. Field-laboratory data correlation.

Most field analysis method documentation includes a section that describes method precision and accuracy, as well as method limitations. These aspects of the method should be consistent with the goals established in the QAPP. The method of laboratory confirmatory analysis should also meet the project data quality objectives.

Traditional laboratory data are currently relied upon for final decision making purposes. Correlation analyses can be performed for field and laboratory confirmation data to utilize in data interpretation and decisions (if correlations are good). This might include, for example, correlation of the concentration of arsenic reported through use of ex situ field XRF analyses to USEPA Method 6010B (ICP) laboratory analyses (e.g., refer to discussion of Field XRF below). A minimum of ten to twenty samples should be submitted for laboratory analysis in order to carry out a comparability analysis of field screening data versus laboratory method data (see Subsection 8.2). Selected samples should span the range of metal concentrations estimated for the field samples and be processed and subsampled at the laboratory using MI procedures (see Section 4.2.6). Additional samples should be collected as needed to generate an acceptable correlation. Prediction lines that reflect 95% UCL and LCL confidence interval (or alternative values) should be added to the regression curve in order to assess the precision of an estimated value with respect to the target action level. For example, prediction lines might allow the ICP-equivalent concentration for a field XRF data point to be estimated within a range of +/- 25 mg/kg with a 95% confidence level. If the range of potential ICP concentrations predicted for an XRF data point spans both above and below the action level then a conclusion regarding the presence or absence of a potential environmental hazard (e.g., direct exposure risk) cannot be made for the specified degree of confidence. Certified laboratories are preferred or laboratories with equivalent documentation of QA/QC protocols. The proportion of field samples selected for laboratory confirmation will depend in part on project-specific data quality objectives.

Samples submitted to the laboratory for development of field versus laboratory data correlation analysis must be the same as used in the field or representative splits of the same material (see Section 4.2.2 on laboratory sub-sampling for guidance). Submittal of the entire sample that was analyzed in the field is preferable for cases where non-destructive field analysis methods are used (e.g., field XRF). Using a rigorous method to prepare any representative splits of samples is important in order to minimize subsampling error. This can greatly affect the correlation of field data to the laboratory data. Error due to variability in contaminant concentrations at the mass of soil analyzed may be unacceptable unless the same minimum (and representative) mass of soil is analyzed for both tests.

Because the minimum subsampling and analysis mass necessary to reduce fundamental subsampling/analysis error to a reasonable level is related to the maximum particle size in the sample, it is also important that the maximum particle sizes in the sample are known (to select the appropriate minimum subsampling/analysis mass) or that the samples are dried and sieved to a target maximum particle size before analysis in the field and lab, so a target subsampling/analysis mass can be selected. Typically, samples are air-dried and sieved to the < 2 mm particle size, before analysis in the field (e.g. ex situ XRF analysis) and laboratory, resulting in a target minimum subsampling/analysis mass of 10 grams. Due to the small mass typically measured by XRF analysis (~ 1 gram), averaging of multiple field XRF analyses (e.g. 10-20 analyses) across the same bulk sample will be necessary for comparisons to laboratory analyses that use a minimum of a 10 gram representative subsample mass for analysis (< 2 mm particle size samples). This same consideration for minimum subsampling/analysis masses for the maximum particle sizes in the sample may also be a significant issue for other types of field screening or field analysis methods, depending on the typical mass of sample the particular method/instrumentation utilizes, and the data quality objectives for the site investigation (see Section 4 for more information on representative subsampling issues).

Details of the sample processing, subsampling mass, and analytical mass used for the field and laboratory samples should be included in the reporting of field and laboratory sample correlation analyses, and uncertainties discussed. Replicate subsamples should also be used to assess the precision of data and subsampling error. Samples used in the correlation analyses should be selected from the lower, middle, and upper range of concentrations measured in the field analyses, as well as samples with contaminant concentrations at or near the site action levels, if any are found.

A comparison of laboratory and field analysis data can be carried out through either a simple correlation analysis or through a regression analysis, depending on the project objectives (e.g., refer to Yates et al, 2003). A correlation coefficient (r) is used to evaluate the strength of the relationship between field and laboratory data, within the limitations described above for error associated with subsampling of the parent sample. The correlation should be positive. The association between the two data sets strengthens as the coefficient approaches 1 (see Yates et al, 2003; Taylor, 1990). Coefficients less than 0.35 are generally considered to represent low or weak correlations. Coefficients values between 0.36 and 0.67 are considered to reflect modest or moderate correlations. Values between 0.68 and 1.0 are considered to be strong, with r coefficients at or above 0.90 considered to be very high.

The data plots should be reviewed to further interpret the nature of the correlation. A strong correlation but consistent variability in field versus laboratory data (e.g., field data consistently lower or higher) can be interpreted to indicate a true difference in the methods. A strong correlation with random variability (e.g., field data randomly higher or lower than laboratory data) could reflect heterogeneity in the soil and difficulty in obtaining representative splits of a sample for field versus laboratory analysis. The lack of a strong correlation between field and laboratory data does not necessarily indicate that the field screening data are “wrong.” This could instead simply reflect the presence of significant short-scale heterogeneity within the samples and error associated with processing and subsampling and/or variability and error in one or both test methods. Determining the exact source or sources of error will require additional studies and may or may not be cost and time beneficial for the subject site investigation.

Laboratory data are preferred for comparison to action levels or for use in risk assessments. If a strong, linear correlation exists between data sets (i.e., >0.68) then a regression analysis can be carried out to quantify this relationship and predict the laboratory-equivalent concentration of a chemical in soil based on field screening data (Yates et al 2003; see also Cutler 2009, 2011). This can significantly improve interpretation of field screening data and confidence in initial decision making. Combined with the use of DU and MIS investigation approaches and associated field and laboratory QA/QC this level of effort could be used to complete a site characterization investigation and/or site cleanup action based on field screening/analysis approaches alone, and avoid the delay and added cost associated laboratory data. The HEER Office may, however, recommend DU-MIS with laboratory analyses for final confirmation samples (after cleanup actions) in these cases. The use of field screening/analysis data in conjunction with or in lieu of laboratory confirmation data should be discussed with the HEER Office project manager on a site-specific basis and documented in the site investigation or removal/remediation report.