Metal mining technical guidance: revised guidance for sample sorting, chapter 2

Laboratory Subsampling Methods

• When to subsample
• Recommended subsampling approaches for Environmental Effects Monitoring studies

When to subsample

The processing of quantitative samples of benthic invertebrates is very laborious, largely because the amount of time required to extract large numbers of invertebrates from a large quantity of inorganic (sand and gravel) and organic (algae, leaves, bryophytes) material (Rosillon 1987, Ciborowski 1991). The two main factors which influence the sorting time of a given sample are the number of organisms and the amount of detrital material in the sample. In fact, these two factors explained close to 95% (84 and 10 % respectively) of the variation in sorting time in a series of samples examined by Ciborowski (1991). Samples with clumping Cladophora algae can result in a 1/3 increase in sorting time. A number of strategies have been developed over the years to reduce this processing time (Resh et al. 1985), however, these time saving methods must not be at the expense of data quality. Some processing time savers include elutriation, fractionation, flotation, and dyes and are discussed in the previous section (Section 2.0). These methods help to speed up the processing procedures due to large amount of organic or inorganic material. To reduce processing time due to large numbers of invertebrates in a given sample, a variety subsampling techniques have been developed. These techniques significantly reduce processing time and associated expense, while adequately estimating whole sample densities and taxonomic composition (Vinson and Hawkins 1996). The objective of any subsampling program should be to minimize effort (and hence cost) while yielding the maximum information with statistically reliable results (Wrona et al. 1982). Thus, the decision to use techniques such as fractionation are based on the amount and type of debris in the sample, while the decision to subsample is based on the number of organisms in the sample. These decision points are schematically represented in Figures 2 and 3.

Recommended subsampling approaches for Environmental Effects Monitoring studies

The subsampling protocols in the following sections are recommended as a general approach to dealing with typical benthic samples from streams or lakes (Figures 2 and 3). However, the appropriateness of these general approaches will need to be assessed on an individual study, based on the type of material and numbers of organisms. The detailed reporting of subsampling accuracy and precision for all methods is essential to the QA/QC of EEM benthic invertebrate programs.

• Sieve Fractionation and Subsampling
• Minimum Number of Organisms
• Acceptable Error for Subsampling Protocols

Sieve Fractionation and Subsampling

For all samples for which subsampling is being considered, but especially for those which have large amounts or pieces of organic matter, it is strongly recommended that the samples be divided in the laboratory into appropriate size fractions to expedite the sorting process (see Section 2.0). Sieving also produces a more homogeneous fraction, allowing for the efficient withdrawal of random subsamples (Anderson 1990) from the appropriate fraction (Taylor and Bailey 1997). The selection of sieve sizes for fractionation should include a consideration of separating the fine and coarse organic matter, but in addition, an attempt to capture the majority of the organisms in one fraction will enhance the subsampling reliability (Rossilon 1987, Meyer 1990). Thus, the selection of coarse sieve size can be related to the size of animals in the sample. Generally a sieve series of 1.00 mm and 500 µm (500 µm = recommended fine mesh and sieve size for EEM) is adequate but a larger coarse sieve of 2.00 or 4.00 mm may also be appropriate if there are many large invertebrates in the samples. The appropriate subsampling technique is then applied to the fine fraction as the majority of small invertebrates will reside on this sieve. However, for samples for which fractionation is not applicable (e.g., samples with clumping algae) appropriate subsampling methods (e.g., wet weight) are available which do not require fractionation (Section 2.3).

Minimum Number of Organisms

Inherent with subsampling is a potential reduction in the accuracy of endpoint estimates. The potential for error is greater with small subsamples or for less common taxa (Wrona et al. 1982, Meyer 1990). Many studies have reiterated the suggestion by Lund et al. (1958) that, based on the Poisson distribution, reasonable accuracy can be obtained when > 100 organisms are enumerated (Hickley 1975, Elliott 1977, Wrona et. al. 1982, Rossillon 1987, Klemm et al. 1990). Previous guidance indicated that subsampling should continue until a predefined variance level is obtained following the approach outlined by Wrona et al. (1982). For each study/ subsampler approach, an assessment would include a description of the trend in variance as number of animals sorted increased. Wrona et al. (1982) found that, for the Imhoff cone subsampler, as the total number of animals counted exceeded 50 and approached 100, improvement in the error term for a given unit of additional effort decreased. However, for counts of less common taxa this variance may be larger at a given number sorted (Wrona et al. 1982). For the EEM program a standardized minimum number of organisms to sort has not been explicitly stated. Thus, depending on the technique used or variance in the sample, minimum numbers could vary widely. With the national objective of comparing EEM studies across the country, and the recommendation of effect endpoints, including taxa richness, further standardization of subsampling protocols will be beneficial, including a minimum number of organisms.

Many studies have attempted to provide recommendations of a minimum number of organisms which provide reasonable subsampling accuracy and/or precision. Much of the literature deals with the fixed count methods as the number of organisms counted is obviously critical to this method. Minimum number initially recommended in the literature was set at 100 (Plafkin et al, 1989), while more recently, recommendations have ranged from 100-300 (Caton 1991, Hannaford and Resh 1995, Vinson and Hawkins 1996, Grownes et al. 1997, Larsen and Herlihy 1998 Somers et al. 1998). In these recent assessments the endpoint of concern is usually taxa richness, as this metric is related to sampling (or subsampling) effort and the higher number (300) is generally recommended (Barbour Gerritsen 1996, Vinson and Hawkins 1996, Larsen and Herlihy 1998, Somers et al. 1998). As taxa richness is one of the EEM benthic endpoints, , it is recommended that a conservative approach be taken to setting minimum numbers of organisms. Therefore, although the most important aspect of a subsampling program is the accuracy of the endpoint estimates, the recommendation that a minimum number of 300 organisms be removed from a sample in any subsampling programprovides additional standardization across all methods and studies for the EEM program.Note that if this minimum number is reached part way through a subsample sort, the subsample must be sorted in it's entirety so that the fraction sorted is quantitative. A minimum proportion of the sample to sort (e.g., 25%) could also be adopted, however, in very large samples this could still represent a significant workload (Ciborowski 1991) and is not recommended as it is the accuracy of the endpoint estimate that is the ultimate measure of any subsampling method.

Acceptable Error for Subsampling Protocols

Regardless of the subsampling technique used, documentation of the accuracy of the estimate is essential to ensure that the data is comparable within and between studies. In fact, the main criteria to evaluate a subsampling technique is an evaluation of it's ability to accurately estimate the numbers and types of organisms in a sample. From the review of Cycle 2 Interpretative Reports, subsampling accuracy was generally not reported. Of those that reported some type of error, the majority reported the precision obtained when comparing two subsamples. For example;

1. a count in subsample A = 289
2. a count in subsample B = 316
3. the reported precision between theses two subsamples would be 8.5% (1-(289/316)) x 100

If all the subsamples from this particular sample had similar precision, then the accuracy will also be close to 9%. However, without sorting the remainder of the sample, accuracy cannot be determined. The studies which did report the accuracy of the subsamples, did so as suggested in the guidance documents. For 10% of all samples several subsamples were sorted and then the remainder of the sample was sorted entirely. Subsampling accuracy can then be calculated by comparing the estimates from the subsamples to the actual count. For example:

1. a count in subsample A = 289, representing 15% of the sample by volume, for an estimate of the total in the sample of 1927
2. a count in subsample B = 316, representing 15% of the sample by volume, for an estimate of the total in the sample of 2106
3. the count in the remainder of the sample = 1359, for a actual total of 1964
4. the reported precision would be the same as in the first example, 8.5 %
5. the reported accuracy would be -1.9% and +7.2% for sample A and B respectively

Both precision and accuracy of the subsampling methods is information which is essential to ensure that the subsampling program is accurately estimating numbers of organisms in the sample. The overriding objective of subsampling is to reduce the substantial effort involved in processing benthic samples but not at the cost of data quality. For acceptable subsampling error criteria, the majority of Cycle 2 studies applied the 20% precision rule suggested by Elliott (1977). That is, if the precision between two subsamples was < 20% the error was acceptable (see example on precision above). Although this has been applied to the precision, it can equally be applied to the accuracy of the estimates. Although accuracy and precision obtainable depend on many factors, including the inherent variability of field samples (Norris et al. 1996), the standardization of EEM techniques should minimize this variability. Many researchers have suggested that is desirable to achieve estimates that are within 20% of the true count (Hickley 1975, Elliott 1977, Downing 1979, Wrona et al. 1982, Resh and McElvry 1993) and in fact, many of the subsampling techniques reviewed and recommended have demonstrated that at least this level of accuracy is attainable (Hickley 1975, Wrona et al. 1982, Rosillon 1987, Meyer 1990). Therefore, the criteria for an acceptable subsampling protocol is that the estimates of each group of samples should be within 20% of the true counts. The factors which should be considered when determining similar groups of samples include: 1) subsampling technique and 2) type of sample (i.e., type and amount of organic matter). As with sorting efficiency, the effects of subsampling on the accuracy of abundance estimates should be examined for a minimum of 10% the samples (or sample groups) in each EEM study and be appropriately reported (Table 1). If an acceptable error level is not demonstrated for a particular technique or a particular set of samples (e.g., with clumping debris, disallowing random mixing) then the technique should be modified to achieve this level of precision and accuracy or the sample should be sorted in it's entirety.

A note is required here regarding a reasonable level of effort to demonstrate accuracy. The critical point is that there is documentation available regarding the accuracy of the estimates, which was lacking in many Cycle 2 reports. The suggested guidance would essentially add one estimate of subsample accuracy to a standard EEM study (assuming similar sample type, 10% of 5 exposure + 5 reference samples = 1). There is no intention to have the documentation of subsampling accuracy become more onerous than sorting all samples. Furthermore, if the technique is one that is well established with a variety of sample types, published in the primary literature, then this exercise is simply a confirmation that the technique is being applied appropriately. In fact, a particular contractor may be able to demonstrate a technique's applicability for a set of samples with similar characteristics spanning more than one EEM study (e.g., samples processed within one season with the same operator(s)).

Table 1: Example of recommended reporting for subsampling error.
Example is for a volume based method using the Imhoff cone (Wrona et al., 1982) where up to 10 subsamples are sorted and the remainder of the sample was sorted. Accuracy of each subsample, the minimum, maximum and mean subsampling accuracy as well as the range in precision between individual subsamples are all reported.