Advanced Design Application Data Analysis for FieldPortable XRF

Advanced Design Application & Data Analysis for Field-Portable XRF A Series of Web-based Seminars Sponsored by Superfund’s Technology Innovation & Field Services Division Contact: Stephen Dyment, OSRTI/TIFSD, dyment. stephen@epa. gov 1

How To. . . u Ask questions » “? ” button on CLU-IN page u Control slides as presentation proceeds » manually advance slides u Review archived sessions » http: //www. cluin. org/live/archive. cfm u Contact instructors 2

Module 1: Introduction 3

Your Instructors……………. u Deana Crumbling - USEPA/OSRTI » Technology Innovation Field Services Division — crumbling. deana@epa. gov u Robert Johnson, Ph. D - Argonne National Lab » Environmental Assessment Division — rlj@anl. gov u Stephen Dyment - USEPA/OSRTI » Technology Innovation Field Services Division — dyment. stephen@epa. gov 4

Who Will Benefit from this Course? u Regulatory project managers and quality assurance reviewers who use XRF data u Consultants and regulatory staff responsible for » Designing and approving work plans that use XRF » Interpreting XRF data 5

Take Away Points… u Spatial heterogeneity is a primary source of data uncertainty u Traditional data strategies often not cost-effective for addressing this data uncertainty u More effective, efficient data designs involve » dynamic/adaptive field decision-making » real-time data generation and management tools u XRF is one of these tools u Use of appropriate sampling designs, QA/QC, and collaborative data allow higher certainty and defensible decisions with XRF 6

Web-Seminar Sessions and Schedule u Session 1 » Module 1 - Introduction and Module 2 - XRF Basics » Monday, August 4, 2008. 1 PM-3 PM EST. » Stephen Dyment, Robert Johnson u Session 2 » Module 3. 1 - Representativeness Part 1 » Thursday, August 7, 2008. 1 PM-3 PM EST. » Deana Crumbling u Session 3 » Module 3. 2 - Representativeness Part 2 » Monday, August 11, 2008. 1 PM-3 PM EST. » Deana Crumbling (continued) 7

Web-Seminar Sessions and Schedule u Session 4 » Module 4, Demonstration of Method Applicability and QC » Thursday, August 14, 2008. 1 PM-3 PM EST. » Stephen Dyment u Session 5 » Module 5, XRF and Appropriate Quality Control Strategies » Monday, August 18, 2008. 1 PM-3 PM EST. » Stephen Dyment u Session 6 » Module 6. 1 - Dynamic Work Strategies Part 1 » Thursday, August 21, 2008. 1 PM-3 PM EST. » Robert Johnson (continued) 8

Web-Seminar Sessions and Schedule u Session 7 » Module 6. 2 - Dynamic Work Strategies Part 2 » Monday, August 25, 2008. 1 PM-3 PM EST. » Robert Johnson u Session 8 » Q&A for Session 7, In Depth Q&A Review for All Seminars, and Resources » Thursday, August 28, 2008. 1 PM-3 PM EST. » Deana Crumbling, Robert Johnson, Stephen Dyment 9

Session Logistics u Each session will be 2 hours long u Questions should be submitted by email or chat u Some questions may be answered at the end of the current session u Most questions will be answered during the first 30 minutes of the subsequent session 10

Session Breakouts u Session 1 » Presentation » Answers to some questions u Session 2 through 7 » 30 minutes answering questions submitted for previous session » 1 hour and 30 minute presentation for current session » Answers to some questions for current session u Session 8 » 30 minutes answering questions submitted for Session 7 » Q&A review for Sessions 1 - 7 » Review of resources 11

Instrument and Software Disclaimer Referring to specific XRF instruments or software packages is for information purposes only and does constitute endorsement. u Manufacturers Niton and Innov-X u Excel (Microsoft Office) u Visual Sampling Plan (Pacific Northwest Lab: http: //dqo. pnl. gov/) u BAASS (Argonne National Lab: http: //www. ead. anl. gov/project/dsp_topicdetail. cfm? topicid =23) u Surfer/Grapher (Golden Software: www. goldensoftware. com ) u Arc. View 3. x or 9. x (ESRI: www. esri. com) u Freeware can be found at http: //www. frtr. gov/decisionsupport/ 12

Module 2: Basic XRF Concepts 2 -1

What Does An XRF Measure? sample u Elements emit characteristic x-rays in response u Characteristic x-rays detected u Spectrum produced (frequency and energy level of detect x-rays) u Concentration present estimated based on sample assumptions Source: http: //omega. physics. uoi. gr/xrf/english/images/PRINCIP. jpg u X-ray source irradiates 2 -2

Example XRF Spectra 2 -3

Bench-top XRF 2 -4

How is an XRF Typically Used? u Measurements on prepared samples u Measurements through bagged samples (limited preparation) u In situ measurements of exposed surfaces (continued) 2 -5

How is an XRF Typically Used? u Measurements on prepared samples u Measurements through bagged samples (limited preparation) u In situ measurements of exposed surfaces 2 -6

What Does an XRF Typically Report? u Measurement date u Measurement mode u “Live time” for measurement acquisition u Concentration estimates u Analytical errors associated with estimates u User defined fields 2 -7

Which Elements Can An XRF Measure? u Generally limited to elements with atomic number > 16 u Method 6200 lists 26 elements as potentially measurable u XRF not effective for lithium, beryllium, sodium, magnesium, aluminum, silicon, or phosphorus u In practice, interference effects among elements can make some elements “invisible” to the detector, or impossible to accurately quantify 2 -8

How Is An XRF Calibrated? u Fundamental Parameters Calibration – calibration based on known detector response properties, “standardless” calibration, what is commonly done u Empirical Calibration – calibration calculated using regression analysis and known standards, either sitespecific media with known concentrations or prepared, spike standards In both cases, the instrument will have a dynamic range over which a linear calibration is assumed to hold. 2 -9

Dynamic Range a Potential Issue u No analytical method is good over the entire range of concentrations potentially encountered with a single calibration u XRF typically under- reports concentrations when calibration range has been exceeded u Primarily an issue with risk assessments 2 -10

Standard Innov-X Factory Calibration List Antimony (Sb) Iron (Fe) Selenium (Se) Arsenic (As) Barium (Ba) Lead (Pb) Manganese (Mn) Silver (Ag) Strontium (Sr) Cadmium (Cd) Mercury (Hg) Tin (Sn) Chromium (Cr) Molybdenum (Mo) Titanium (Ti) Cobalt (Co) Nickel (Ni) Zinc (Zn) Copper (Cu) Rubidium (Ru) Zirconium (Zr) 2 -11

How Is XRF Performance Commonly Defined? u Bias – does the instrument systematically under or over- estimate element concentrations? u Precision – how much “scatter” solely attributable to analytics is present in repeated measurements of the sample? u Detection Limits – at what concentration can the instrument reliably identify the presence of an element? u Quantitation Limits – at what concentration can the instrument reliably measure an element? u Representativeness – how representative is the XRF result of information required to make a decision? u Comparability – how do XRF results compare with results obtained using a standard laboratory technique? 2 -12

Analytical Precision Driven By… u Measurement time – increasing measurement time reduces error u Element concentration present – increasing concentrations increase error u Concentrations of other elements present – as other element concentrations rise, general detection limits and errors rise as well 2 -13

Lead Example: Concentration Effect 2 -14

Lead Example: Concentration Effect 2 -15

XRF Detection Limit (DL) Calculations u SW-846 Method 6200 defines DL as 3 X the standard deviation (SD) attributable to the analytical variability (imprecision) at a low concentration u XRF “measures” by counting X-ray pulses u XRF instruments typically report DLs based on counting statistics using the 3 X SD definition u SDs and associated DLs can also be calculated manually from repeated measurements of a sample (if concentrations are detectable to begin with) 2 -16

The 3 Standard Deviation Concept Frequency of XRF Responses When Element Not Present 2 -17

DL <> Reliable Detection 2 -18

DL <> Reliable Detection 2 -19

DL <> Reliable Detection 2 -20

For Any Particular Instrument, Detection Limits Are Influenced By… u Measurement time (quadrupling time cuts detection limits in half) u Matrix effects u Presence of interfering or highly elevated contamination levels Consequently, the DL for any particular element will change, sometimes dramatically, from one sample to the next, depending on sample characteristics and operator choices 2 -21

Examples of DL… Analyte Antimony (Sb) Arsenic (As) Barium (Ba) Cadmium (Cd) Calcium (Ca) Chromium (Cr) Cobalt (Co) Copper (Cu) Iron (Fe) Lead (Pb) Manganese (Mn) Mercury (Hg) Molybdenum (Mo) Nickel (Ni) Innov-X 1 120 sec acquisition (soil standard – ppm) (alluvial deposits - ppm) (elevated soil - ppm) 61 55 232 6 7 29, 200 NA NA NA 34 30 598 NA NA NA 89 100 188, 000 54 121 766 21 17 661 2, 950 22, 300 33, 300 12 8 447, 000 56 314 1, 960 10 8 481 11 9 148 42 31 451 2 -22

To Report, or Not to Report: That is the Question! u Not all instruments/software allow the reporting of XRF results below detection limits u For those that do, manufacturer often recommends against doing it u Can be valuable information if careful about its use…particularly true if one is trying to calculate average values over a set of measurements 2 -23

XRF Data Comparability usually refers to comparing XRF results with standard laboratory data u Assumption is one has samples analyzed by both XRF and laboratory u Regression analysis is the ruler most commonly used to measure comparability u SW-846 Method 6200: “If the r 2 is 0. 9 or greater…the data could potentially meet definitive level data criteria. ” 2 -24

What is a Regression Line? 2 -25

Regression Terminology u Scatter Plot: graph showing paired sample results u Independent Variable: x-axis values u Dependent Variable: y-axis values u Residuals: difference between dependent variable result predicted by regression line and observed dependent variable u Adjusted R 2: a measure of goodness-of-fit of regression line u Homoscedasticity/Heteroscedasticity: Refers to the size of observed residuals, and whether this size is constant over the range of the independent variable (homoscedastic) or changes (heteroscedastic) 2 -26

Heteroscedasticity is a Fact of Life for Environmental Data Sets 2 -27

Appropriate Regression Analysis u Based on paired analytical results, ideally from same sub-sample u Paired results focus on concentration ranges pertinent to decision-making u Non-detects are removed from data set u Best regression results obtained when pairs are balanced at opposite ends of range of interest 2 -28

Evaluating Regression Performance u No evidence of inexplicable “outliers” u Balanced data sets u No signs of correlated residuals u High R 2 values (close to 1) u Constant residual variance (homoscedastic) 2 -29

Example: XRF and Lead u Full data set: » Wonderful R 2 » Unbalanced data » Correlated residuals » Apparently poor calibration u Trimmed data set: » Balanced data » Correlation gone from residuals » Excellent calibration » R 2 drops significantly 2 -30

Converting XRF Data for Risk Assessment Use u Purpose: making XRF data “comparable” to lab data for risk assessment purposes u To consider: » Need for “conversion” may be an indication of a bad regression » XRF calibrations not linear over the range of concentrations potentially encountered » Extra variability in XRF data not an issue (captured in UCL calculations when estimating EPC) » Contaminant concentration distributions are typically skewed… lots of XRF data may provide a better UCL/EPC estimate than a few lab results even if the regression is not great 2 -31

A Cautionary Example… u Four lab lead results: 20, 24, 86, and 189 ppm u Pro. UCL 95%UCL Calculations: » Normal: 172 ppm » Gamma: 434 ppm » Lognormal: 246 – 33, 835 ppm » Non-parametric: 144 – 472 ppm u Four samples are not enough to either understand the variability present, or the underlying contamination distribution 2 -32

Will the “Definitive” Data Please Stand Up? One of these scatter plots shows the results of arsenic from two different ICP labs, and the other compares XRF and ICP arsenic results. Which is which? 2 -33

Definitive Data, Please Stand Up! 2 -34

Take-Away Comparability Points u Standard laboratory data can be “noisy” and are not necessarily an error-free representation of reality u Regression R 2 values are a poor measure of comparability u Focus should be on decision comparability, not laboratory result comparability u Examine the lab duplicate paired results from traditional QC analysis - The split field vs. lab regression cannot be expected to be better than the lab’s duplicate vs. duplicate regression 2 -35

What Affects XRF Performance? u Measurement time – the longer the measurement, the better the precision u Contaminant concentrations – potentially outside calibration ranges, absolute error increases, enhanced interference effects u Sample preparation – the better the sample preparation, the more likely the XRF result will be representative (continued) 2 -36

What Affects XRF Performance? u Interference effects – the spectral lines of elements may overlap u Matrix effects – fine versus coarse grain materials may impact XRF performance, as well as the chemical characteristics of the matrix u Operator skills – watching for problems, consistent and correct preparation and presentation of samples 2 -37

What Are Common XRF Environmental Applications? u In situ and ex situ analysis of soil samples u Ex situ analysis of sediment samples u Swipe analysis for removable contamination on surfaces u Filter analysis for filterable contamination in air and liquids u Lead-in-paint applications 2 -38

Recent XRF Technology Advancements… u Miniaturization of electronics u Improvements in detectors u Improvements in battery life u Improved electronic x-ray tubes u Improved mathematical algorithms for interference corrections u Bluetooth, coupled GPS, connectivity with PDAs and tablet computers 2 -39

…Contribute to Steadily Improving Performance Antimony (Sb) 40 55 Innov-X 1 120 sec acquisition (soil standard – ppm) 61 Arsenic (As) 40 60 6 Barium (Ba) 20 60 NA Cadmium (Cd) 100 NA 34 Chromium (Cr) 150 200 89 Cobalt (Co) 60 330 54 Copper (Cu) 50 85 21 Iron (Fe) 60 NA 2, 950 Lead (Pb) 20 45 12 Manganese (Mn) 70 240 56 Mercury (Hg) 30 NA 10 Molybdenum (Mo) 10 25 11 Nickel (Ni) 50 100 42 Analyte DL in Quartz Sand by TN 900 (60 to Method 6200 100 sec) – ppm (600 sec – ppm) 2 -40

Q&A – If Time Allows 2 -41

Thank You After viewing the links to additional resources, please complete our online feedback form. Thank You Links to Additional Resources Feedback Form 2 -42