Poorly designed composite sampling

Unfortunately, as commonly practiced, composite sampling seldom considers the spectrum of sampling errors or requests that laboratory subsampling be done in a way that addresses microscale heterogeneity. Also, techniques long used to "homogenize" soil samples such as cone-and-quartering have been shown by experiment to be ineffective and are no longer recommended (Gerlach and Nocerino 2003). In summary, composite sampling as conventionally implemented is characterized by unspecified sample collection and analysis procedures that do not adequately consider the following:

  • the number of soil increments to be collected
  • the intended "area of inference" for the composite samples
  • the size and boundaries of the DU
  • particle size selection or reduction measures
  • bulk sample mass requirements
  • field and laboratory subsampling techniques

As routinely applied, composite sampling is viewed primarily as a way to reduce analytical costs, without taking more important sampling goals into account. It is not surprising that over the years composite sampling has developed an unfavorable reputation. It is important to understand that ISM differs greatly from the practices common to poorly designed composite sampling applications. It is worth noting that routine applications of composite sampling also differ significantly from the composite sampling designs recommended in USEPA guidance. Yet, ISM transcends even most USEPA compositing guidance because ISM prominently calls out specific error-controlling steps. These were not yet well researched when most USEPA statistical sampling design documents were written.

However, the primary reason that ISM and composite sampling as typically practiced cannot be considered equivalent is that typical composite sampling rarely involves enough aliquots of soil to manage contaminant heterogeneity over an entire DU. Therefore, even when the goal of compositing is to determine an average over some area, it is less likely to estimate the mean concentration with the precision needed by data users. Empirical studies and sampling theory suggest that composite sampling (with inadequate consideration of the steps listed above) simply does not perform as well as ISM sampling. Indications are that low-increment number composite samples combined with insufficient mixing and processing procedures perform about as well as discrete samples. However, it is acceptable to use the composite sampling approach if it meets the user-defined goals for precision and accuracy. Composite sampling approaches should include a methodology for estimating the total precision and take measures to ensure that an unbiased mean is obtained.