SITE CHARACTERIZATION USING STATISTICAL APPROACHES

ERDC TN-DOER-C15

July 2000

Bulk sediment data:

Volume of available bulk sediment or dredged material.

Grain size distribution (GSD) of the bulk material (prior to separation).

Concentrations of contaminants of concern (COC) in the bulk sediments.

BU specifications, including acceptable GSD and COC levels.

Concentrations of COC in material fractions, if separation is determined to be necessary to

meet BU specifications.

Use of existing information to obtain screening level estimates of MRP is described in Olin-Estes

and Palermo (2000a). This technical note addresses the case in which existing information is

inadequate for definitive determination of BU feasibility and MRP. Because data are rarely

available for in-CDF materials, and because physical and chemical data for in-channel materials are

not generally obtained specifically for determination of BU potential or physical separation

feasibility, most projects of any size will ultimately require an extensive sampling effort.

General sampling considerations (sampling methods and equipment, sample volume requirements,

analyte selection, depth of sampling, sample replication and compositing, and physical testing) are

the same for both the statistical and prescriptive site characterization approaches. These are

discussed fully in Olin-Estes and Palermo (2000a). This technical note address specifically the

statistical basis and procedures for developing a site sampling plan and interpreting and extrapolat-

ing data.

SITE CHARACTERIZATION USING STATISTICAL APPROACHES: In designing a sam-

pling plan, in addition to using available information about the site, it is often helpful to look at the

tools available for interpretation of the resulting data. A number of statistically based approaches

provide tools for determining the number of samples required to determine a measured parameter

with a specified degree of confidence, unbiased approaches for structuring a sampling plan, and

methods for interpreting and extrapolating data (Winkels and Stein 1997; Keillor 1995; Keillor

1993; Lubin, Williams, and Lin 1995; Isaaks and Srivastava 1989). Given the constraints of time

and budget, the number of samples required based on statistical considerations will often be much

larger than is physically or economically feasible to obtain or analyze, unless the variability of the

material is quite low. However, a sampling plan certainly should not be implemented without

considering a statistical design, even though modifications to that design may ultimately be required.

The resulting data will then lend itself to statistical analysis and available methods for extending

the data to unsampled areas. Appendix I presents a glossary of statistical terms used in the following

discussion.

Statistical Analysis. In general, the larger a data set is, the more it tends toward a normal

distribution. This is important because when it can be established that data are normally distributed,

there are a number of statistical tools to help interpret the significance of differences between

samples and to predict the likelihood of values falling outside a specified range. Among these are

the t-test, the Paired Difference Test, and Analysis of Variance (ANOVA). However, most

environmental data are not normally distributed. Due to cost constraints, the data sets are too small,

or may contain many zero values due to the heterogeneity of deposits and the difficulty of obtaining

representative samples. Because environmental data do not always meet the requirements and