v>EPA
United States
Environmental Protection
Agency
Environmental Monitoring
Systems Laboratory
P.O. Box 93478
Las Vegas NV 89193-3478
May 1990
TECHNOLOGY SUPPORT PROJECT
Assessment
of Errors in
Soil Sampling
This fact sheet is based on "A Rationale for the Assessment
of Errors in the Sampling of Soils" by J. Jeffrey van Ee, Louis
J. Blume, and Thomas H. Starks, 1990.
INTRODUCTION
The sampling and analysis of soils for inorganic contaminants
is a complex procedure from experimental design to the final
evaluation of all generated data. Sources of error abound but
they can be successfully mitigated by careful planning or
isolated by intelligent error assessment. Error (or variability)
can be either bias or random. Biased error is indicative of a
systematic problem that can exist in any sector of soils
analysis, from sampling to data analysis. The first step in
analysis of variability (or error) is to establish a plan that will
identify errors, trace them to the step in which they occurred,
and account for variabilities to allow direct corrective action to
mitigate them. In anticipation of errors, it is essential to ask two
questions:
1 How many and what type samples are required to assess
the quality of data in a field sampling effort?
2. How can the information from these samples be used to
identify and control sources of error and uncertainty in the
measurement process?
Error assessment should be understood by the field scientist
and the analyst. To aid scientists in the estimation and
evaluation of variability, the Environmental Monitoring Systems
Laboratory-Las Vegas (EMSL-LV) has developed a computer
program called ASSESS. By applying statistical formulae to
data entered, ASSESS can trace errors to their sources and
help scientists plan future studies that avoid the pitfalls of the
past.
-------
RANDOM ERRORS
BIAS ERROR
PREVENTION
Random errors can result in variations from
tjie true vikje*that are either positive or
negative but di3 not follow a pattern of
variability. During the measurement process,
random errors may be caused by:
1) sampling variations
2) handling discrepancies
3) transportation vagaries
4) preparation dissimilarities
5) subsampling problems
6) analytical discrepancies
7) data handling eccentricities
The greatest source of error is usually the
sampling step. In the Comprehensive
Environmental Response, Compensation,
and Liability Act of 1980 (Superfund)
(CERCLA) and the Resource Conservation
and Recovery Act (RCRA), site
investigations, analytical, and data handling
variability are checked by the CLP protocol.
When more than one laboratory is involved,
handling, transportation, subsampling, and
preparation can be checked at the Level IV
CLP step, too. But how can the analyst know
that the sample in the jar is representative of
the surrounding samples at the site? How can
the field analyst know that the more (or less)
concentrated soil didn't stick to the auger or
split-spoon?
It is strongly recommended that the traditional
approaches used in mitigating the error in the
last six steps be applied to sampling itself, i.e.,
use of duplicates, splits, spikes, evaluation
samples, and calibration standards. A certain
amount of random error is inherent in samples
themselves, in fact, the total variance equals
the measurement variances plus the
population variances
(ot2 = om2 + ap2)
We can address the variance in measurement;
the population variance, however, is a true
picture of the complexity of the soil.
Some sources of error are systematic, that is,
in a given situation conditions exist that
consistently give positive or consistently give
negative results. This skewing of data can be
initiated early in a sampling regime, e.g., by a
sampling device that alters the composition of
the soil matrix. It can occur in the middle of
the sampling regime, e.g., by the preferential
handling of a sampler who isn't trained in the
intricacies of sample handling and
preparation. Or it can be introduced in the
later, analytical stages, where it is easier to
trace because of interlaboratory comparisons
and frequent calibration checks. The
pervasive quality of an early bias error is its
resistance to detection and the fact that other
variabilities are added throughout the
process until, finally, the reported data may
be significantly non-representative of the true
value. Bias errors can be traced to:
• faulty sampling design
• skewed sampling procedure
• systematic operator error
• contamination
• degradation
• interaction with containers
• displacement of phase (or chemical
equilibria)
• inaccurate instrument calibration
To avoid both random and bias errors (or at
least to be able to pinpoint their occurrence
and estimate their extent), it is wise to plan a
study well, anticipating possible sources of
error. The inclusion of quality assurance
samples used for quality assessment and
quality control can help isolate variability and
identify its effect.
An effective technique is to concentrate
duplicate sampling early in the study and
send the samples off for rapid CLP analysis.
Dependent on the results, it may not be
necessary to include as many quality
assessment samples after these samples
demonstrate reliability. This allowance for
early detection of sources of error can help
the field scientist customize the remainder of
the study to meet the specific needs of the
project.
-------
QUALITY ASSESSMENT
SAMPLES
A Remedial Project Manager (RPM) must
ask: how many samples are needed to
adequately characterize the soil at this site?
The key word is "adequately." By
determining the data quality objectives
(DQOs) in advance, the RPM can assure
adequate sampling at a site. Too little
sampling, as well as too much, is a waste of
time and money. The extent of QA/QC effort
is dependent on the risk to human health, the
nearness of action levels to detection limits,
and the size, variability, and distribution of
contamination. Ultimately, the number of
quality assessment samples is determined by
the DQOs for the site. The following table
explains various types of quality assessment
samples and their uses.
QUALITY ASSESSMENT SAMPLES
• ALLOW STATEMENTS TO BE MADE CONCERNING THE QUALITY OF THE MEASUREMENT
SYSTEM
• ALLOW FOR CONTROL OF DATA QUALITY TO MEET ORIGINAL DQOs
• SHOULD BE DOUBLE-BLIND:
Types of Samples Description
Field Evaluation (FES) Samples of known concentration are introduced in the field as ear'y
as possible to check for measurement bias and to estimate preciso •
Low Level Field Low concentration FES samples check for contamination in
Evaluation (LLFES) sampling, transport, analysis, detection limit
External Laboratory Similar to FES but without exposure in the field, ELES can measure
Evaluation (ELES) laboratory bias and, if used in duplicate, precision
Low Level External Similar to LLFES but without field exposure, LLELES can
Laboratory (LLELES) determine the method detection limit, and presence of laboratory
contamination.
Field Matrix Spikes (FMS) Routine samples spiked with the analytes of interest in the field
check recovery and reproducibility over batches.
Field Duplicates (FD) Second samples taken near routine samples check for variability at
all steps except batch.
Preparation Splits (PS) Subsample splits are made after homogenization and are used to
estimate error occurring in the subsampling and analytical steps of
the process.
• SHOULD BE SINGLE-BLIND:
Field Rinsate Blanks (FRB) Samples obtained by rinsing the decontaminated sampling
equipment with deionized water to check for contamination
Preparation Rinsate (PRB) Samples obtained by rinsing the Blanks sample preparation
apparatus with deionized water to check for contamination
Trip Blanks (TB) Used for Volatile Organic Compounds (VOCs), containers filled with
American Society for Testing and Materials Type II water are kept
with routine samples through the sampling, shipment, and analysis
phases
MAY BE NON-BLIND: AS IN THE INORGANIC CLP PROTOCOL
-------
SOME STATISTICAL
CONCERNS
SAMPLE
COLLECTION
CONSIDERATIONS
ASSESS
Confidence in quality assessment sample
data can be expressed as an interval or as an
upper limit. All confidence levels/limits are
based on the number of degrees of freedom
and the limits get lower (or the intervals get
smaller) as the number of degrees of freedom
gets higher. The number of degrees of
freedom for any particular statistic is
dependent upon the experimental design.
For example, if 15 samples are taken at a
site, and each is divided into 2.preparation
splits, and each split is extracted twice at a
CLP laboratory, and 2 injections of each
extraction are made into an Inductively
Coupled Plasma/Mass Spectrometer (ICP/
MS), the total number of degrees of freedom
associated with this experimental design
would be calculated as:
15 samples x 2 preparations splits = 30
X 2 CLP extractions =60
. X 2 injection replicates = 120
120 degrees of freedom for the whole
process. But, if only the population variability
in tee field samples (which includes the
sampling error) is being estimated, the
number of degrees of freedom is 15-1, or 14.
There are 15 independent samples but one
degree of freedom is lost with the estimation
of the mean. Therefore, there are 14 degrees
of freedom for the sampling variance
estimate. As another example, to estimate
the variability in the extraction step, one has
30;independent pairs of numbers, each pair
associated with one extraction. Thus, there
are 30 degrees of freedom associated with
the extraction error. In fact, this error
represents any error associated with the
injection step as well because, obviously, no
data are available until injections are made
and results are reported.
Consulting the Rationale Document or a
statistical manual can give an RPM a feel for
how many quality assessment samples must
be sprinkled among the routine site
characterization samples.
If Level IV CLP analysis is performed on the
soil, we can assume that very little error is in
the analytical stage. This focuses our
attention on sources of error in the sampling,
handling, and preparation steps. The two
major considerations in collection of
environmental samples are:
1. Will the collected data give the answers
necessary for a correct assessment of the
contamination or a solution to the problem?
2. Can sufficient sampling be done well and
within reasonable cost and time limits?
The EMSL-LV has developed an easy-to-use
program to apply the necessary statistics to
the generated data for an accurate
determination of precision and bias.
ASSESS is a public domain, Fortran program
that is available from EMSL-LV. It may be
applied to cases where no field evaluation
samples are available as well as cases
where they are. ASSESS is user-friendly and
its use will greatly aid both field scientists and
RPMs in .decision-making based on soil
' studies.
cf
° I echnology "^
\
\
support
reject
^
FOR FURTHER INFORMATION:
This fact sheet is based on "A Rationale for
the AssessjosoLof Efror&jn the Sampling of
Soils"b/0^JeffreyvariE^J.ouis J. Blume,
and ThortaTiTSTaTKsri99u~. ' '
This document is available in hypertext for
IBM PCs to enhance readability and aid
users at all levels in knowledge of this
material.
For additional copies, hypertext disks, or
more information, call, or write, EMSL-LV:
^
Technology Support Center
U.S. Environmental Protection Agency
P.O. Box93478
Las Vegas, NV 89193-3478
(702) 798-2270
FTS 545-2270
FAX 545-2637
------- |