9.1 Case Study 1. PCB-Contaminated Landfill

Site Name: Green Island Landfill and Reburial Pit, Kure Atoll, Hawaii

Contact Name: Roger Brewer, HDOH

Site Location: Kure Atoll is the northernmost island in the Hawaiian Island chain, located approximately 1400 miles northwest of the island Oahu and 56 miles northwest of Midway atoll. The atoll consists of a lagoon encircled by a reef and a single vegetated island, Green Island. Green Island is just under 1.5 miles long and about 0.35 miles in width and has a maximum elevation of 15 feet.

Background: A U.S. Coast Guard (USCG) station was located on the atoll from the 1960s through the 1990s. A ½-acre area located on the southwest corner of the island was used to dispose of old electrical components and scrap metal. Discreet confirmation soil samples identified concentrations of PCBs as high as 170 mg/kg within the formal landfill footprint. Soil, sediment, and biota samples collected in the surrounding area indicated that PCB contamination was primarily restricted to the landfill site. Debris and approximately 700 yd3 of PCB-contaminated soil were removed from the site in 1993.

A follow-up study of the former landfill area was carried out in 2008. As part of the site investigation, USCG took the opportunity to evaluate the potential advantage and limitations of incremental soil sampling approaches over traditional discrete sampling approaches. The investigation focused on the use of DU and ISM investigation strategies published by HDOH (2008b).

Statistical Evaluation: A statistical evaluation of discrete vs. incremental sample data was conducted by Anita Singh, a contractor to USEPA with Lockheed Martin in Las Vegas and member of the ITRC ISM Team. One objective of the review was to compare estimates of the mean concentration of PCBs in the DU soil based on a specific number of discrete samples vs. one to three incremental samples drawn from the same data set. Another objective of the Kure atoll data set was to determine the equivalent number of discrete samples to a triplicate set of 30–50 point incremental samples. This effort will help to evaluate the cost-effectiveness of collecting incremental samples over discrete samples.

Lessons Learned—ISM Data Collection:

  • Isolation of areas of suspected higher contamination is important at a site-wide scale but not at a DU scale.
  • Identify and investigate suspected spill areas separately via historical knowledge and/or preliminary sampling.
  • Subdivide remaining area into risk-based DUs based on human or ecological health concerns.
  • Incorporate an adequate number of increment points within a DU to capture contaminant distribution and heterogeneity.
  • A range of 30–50+ increment points is required to adequately characterize a DU—anything less is probably just sampling the mode.
  • Use replicate samples to verify that contaminant heterogeneity has been adequately characterized.
  • Tight grids of discrete samples can be useful for an initial screening of sites and DU designation, as well as subdivision of “hot” DUs for smaller areas for isolation and characterization of concentrated contamination.

Lessons Learned—ISM Simulation:

Some combination of both discrete and ISM sampling data was ideal for estimating the PCB mean.

  • Include at least 30–50 increments per ISM field replicate sample for initial DU characterization.
  • Always collect and use replicate sample data (e.g., triplicates) from one or more DUs at a site to evaluate the representativeness of incremental sample data.
  • Determining the appropriate number of ISM increments and replicates is critical to ensuring that the ISM sample is representative of the conditions in the field and to assess precision. Between 60 and 90 increment replicate samples are needed to ensure the incorporation of isolated hot spots.
  • ISM helped identify the primary spill area, but ISM for the entire core would have yielded the same answer.
  • It is not possible to test data representativeness of field data with a small set of discrete samples (e.g., <30 samples), as lognormal outliers would likely be missed.