• About this site|
• Site index|
• Contact us|
• Jobs

A User Guide for the Macroinvertebrate Community Index

Acknowledgements

Part 1: Background to Biotic Indices

1 Introduction

2 What are biotic indices?

3 New Zealand's MCI-type biotic indices

Part 2: Guidelines for using the MCI, QMCI and SQMCI

References

Appendix 1: Calculating the MCI: Excel Macros and User-defined Functions

Appendix 2: Use of MCI, QMCI and SQMCI Throughout New Zealand

3 New Zealand's MCI-type biotic indices

3.1 Origin and development of the MCI

A preliminary version of the MCI (the IHQI, or Invertebrate Habitat Quality Index) was included in the Taranaki ringplain freshwater biological report (Taranaki Catchment Commission 1984), but it was the Water and Soil Miscellaneous Publication prepared under secondment in 1984 to the Water Quality Centre (Hamilton) (Stark 1985) that proposed New Zealand's Macroinvertebrate Community Index (MCI) and its quantitative variant (QMCI) for assessing organic enrichment in stony riffles. [Riffle: a shallow part of a stream or river with broken water flow.] The concept was derived from the United Kingdom's BMWP Score System (BMWP 1978), although genera are mainly used for scoring in New Zealand indices in contrast to families for the BMWP Score System. The MCI is analogous to the ASPT (Average Score Per Taxon) variant of the BMWP Score System (Armitage et al 1983).

Subsequent research funded by the Public Good Science Fund through the Foundation for Research, Science and Technology (FRST) focused on characterising the performance and precision of the MCI and QMCI (Stark 1993b).

Stark (1993b) used macroinvertebrate data from both the North and South Islands to investigate the influences of sampling method, water depth, current velocity, and substratum on the MCI and QMCI. When calculated from macroinvertebrate samples collected by hand-net or Surber sampler from stony riffles, the MCI and QMCI are independent of depth, velocity, and substratum; a major advantage when assessing water pollution or enrichment. The statistical precision of MCI and QMCI values obtained in these ways was defined, along with two methods for detecting statistically significant differences between index values (Stark 1993b).

A more cost-effective variant of the QMCI called the Semi-Quantitative Macroinvertebrate Community Index, or SQMCI was developed in 1998 (Stark 1998). The SQMCI uses a five-point scale of coded abundances (i.e. Rare, Common, Abundant, Very Abundant, Very Very Abundant). This index produces values very similar to the QMCI, but at less than 40% of the cost, due to reduced numbers of replicate samples being required to achieve the desired precision, and savings in macroinvertebrate sample processing time. Stark (1998) also re-evaluated the statistical precision of the MCI and QMCI from hand-net and Surber samples, based on a larger sample database than was previously available. Similar information was provided for the SQMCI.

Recently, Stark and Maxted (2004, [Note that the MCI-sb described by Stark and Maxted (2004) is a preliminary version that isnotthe same as the final version (Stark and Maxted 2007). We have simplified the tolerance value derivation process and have derived tolerance values to the nearest 0.1 (rather than integers) to improve the performance of the MCI-sb and to reduce the possibility of confusion between the HB (integer) and SB (nearest 0.1) tolerance values.] 2007) developed new biotic indices for assessing the health of soft-bottomed streams. These indices are analogous to the MCI, SQMCI and QMCI, and are denoted by the addition of "-sb" to the respective index names (i.e. MCI-sb, SQMCI-sb and QMCI-sb). New Zealand appears to be the only country with qualitative, semi-quantitative and quantitative versions of the same biotic index, and different versions for hard- and soft-bottomed streams (Stark 1985, 1993b, 1998; Stark and Maxted 2004, 2007).

3.2 Assigning tolerance values to taxa

Most biotic indices require tolerance values to be assigned to macroinvertebrate taxa. These tolerance values are related in some way to stream condition or an environmental gradient; for example, from unmodified native forest (the reference condition), through to highly intensive urban or rural land use. Well-performing biotic indices have been developed using a variety of methods for deriving tolerance values, including:

• professional judgement (Chutter 1972; Hilsenhoff 1977; Chessman 1995)
• numerical proportioning applied to taxon occurrences and/or abundances along pollution gradients, or among site groups differing in their pollution status (Stark 1985; Chessman et al 1997; Chessman 2003; Stark and Maxted 2004, 2007)
• associating taxon occurrences or abundances with water quality data (Lawrence and Harris 1979)
• canonical correspondence analysis (Suren et al 1998; Davy-Bowker et al 2005).

3.2.1 Deriving tolerance values for new biotic indices

For the MCI, tolerance values were determined initially by a weighting procedure based on the relative percentage occurrence of taxa at three site groups differing in their enrichment status (i.e. clean and un-enriched, slight to moderate pollution, moderate to gross pollution) (Stark 1985). Tolerance values for less common taxa, for which this procedure was unreliable (Stark 1985), or those added subsequently (Stark 1993b, 1998) have been assigned by professional judgement.

Stark and Maxted (2004, 2007) used an iterative rank correlation procedure developed by Chessman (2003) (hereafter referred to as the "Chessman process") to derive tolerance values for the MCI-sb using data primarily from the Auckland region, with data from other regions (Northland, Waikato, Bay of Plenty, Hawke's Bay, Taranaki, and Otago) to provide tolerance values for taxa that were not recorded in the Auckland data set. It is worthwhile pausing here to explain this useful procedure for objectively deriving tolerance values.

A prerequisite for using the Chessman process is a macroinvertebrate data set that covers the full range of disturbance, from the best to the worst sites in the region. The resulting tolerance values will be derived in response to the dominant gradient within the data set. Often, in natural systems the gradient is confounded by a variety of variables. In other words, it is due to a complex of interacting environmental factors, which may include enrichment, sedimentation effects (bed sediments tend to become less coarse progressively downstream), altitude, water temperature and other water quality variables, changes to riparian vegetation and condition, and the effects of stream order (velocity, depth). Most biotic indices developed from real-world data sets respond to this complex of factors.

Our implementation of the Chessman process proceeded as follows. First, the sites or samples need to be ordered from best to worst in terms of the environmental gradient of interest. We used MCI values calculated by a user-defined function (or a macro) on an Excel spreadsheet. Spearman rank correlations (r_s) were calculated between the MCI values and the abundances of all taxa across all samples using STATISTICA 7.1. [Note that Excel cannot normally be used to calculate Spearman rank correlations, not only because it does not have a function to do so, but also because the work-around (involving linear correlation of ranks) using Excel's RANK function provides incorrect results because it does not handle tied ranks correctly. A solution to this problem is presented here:http://udel.edu/~mcdonald/statspearman.html. However, given the number of rank correlations required when using the Chessman process, a spreadsheet-based approach would be tedious in the extreme.] Because it is mathematically impossible for rare taxa to achieve large positive or negative correlations (Chessman 2003), each r_s was expressed as a proportion of the maximum possible r_s for a taxon recorded from the same proportion of samples. The taxon with the highest adjusted positive r_s was assigned a tolerance value of 10, and the taxon with the lowest adjusted negative r_s was assigned a tolerance value of 0.1. The remaining taxa were assigned tolerance values (rounded to the nearest 0.1) between these extremes in proportion to their adjusted r_s values. The resulting tolerance values were pasted back into the Excel spreadsheet and new biotic index values were calculated for each sample. This procedure was repeated until the tolerance values stabilised (i.e. no tolerance values changed from one iteration to the next), and these became the tolerance values that were adopted for the new biotic index.

The Chessman process entails an apparent circularity, since all samples in the data set are ordered from best to worst using the MCI. Ideally, this would be done independently of the biological data, but if there were an easy way of doing this there would be no need for biotic indices. Chessman (2003) used SIGNAL to determine the initial site order, noting that SIGNAL was a proven indicator of stream health. We used the MCI, because in New Zealand the MCI has shown high correlations with indicators of organic enrichment (e.g. Quinn and Hickey 1990), and it performs adequately in soft-bottomed streams (Maxted et al 2003). The final set of tolerance values derived by this process is not overly sensitive to the starting condition if there is a strong environmental gradient in the data set. If there is more than one strong gradient in the data, say enrichment and altitude, the algorithm can result in an index of altitude when an index of enrichment was desired (Bruce Chessman, personal communication), but the starting point remains unimportant. In our experience with Auckland Regional Council's State of the Environment (SoE) data, and data from soft-bottomed streams from other regions, running the Chessman process on sample data, site-averaged data and various subsets of the data all produced tolerance values that were similar. This suggested that these data embodied a strong environmental gradient, and gave confidence that the process was likely to produce a useful result. Subsequent testing of the new indices (by rank correlations with environmental variables) confirmed that the Chessman process does produce biotic indices that perform well.

3.2.2 Deriving new tolerance values for existing biotic indices

Once a biotic index has been developed, it is inevitable that new taxa (i.e. not previously scored) will be encountered. How should tolerance values for these new taxa be derived? There are several options here.

1. Adopt tolerance values from another biotic index. For example, MCI tolerance values are likely to be a reasonable substitute if a particular taxon has not yet been assigned a tolerance value for the MCI-sb. It would be better to substitute tolerance values in this way than to exclude unscored taxa from the index calculations. When developing the MCI, Stark (1985) used family scores from the BMWP Score System as a guide for assigning tolerance values when no better information was available.

2. Professional judgement. Most of the additional tolerance values for the MCI (i.e. those added to the list provided by Stark [1985]) were assigned by the professional judgement of one or more experienced freshwater macroinvertebrate ecologists (Stark 1993b, 1998; Stark et al 2001; Winterbourn et al 2006). These tolerance values are not necessarily unreliable or incorrect, but this process is subjective rather than objective, and has been criticised for that reason (Hickey and Clements 1998; Joy and Death 2003).

3. The Chessman process was designed to assign tolerance values objectively when developing new biotic indices (Chessman 2003), but can be used to derive additional tolerance values for previously unscored taxa. To be practical, any procedure that sets tolerance values needs to be quick and conservative (i.e. cause little or no change to existing tolerance values), because it would be very undesirable if the index was re-invented each time a new tolerance value was required.

Of these options, the last is the most objective, but there remains the issue of how best to carry it out. The initial development of the MCI-sb was based on 2000−2004 data (117 taxa x 179 samples) from soft-bottomed streams in Auckland. Auckland Regional Council's 2005 SoE monitoring data set (45 samples) included seven new taxa that did not have MCI-sb tolerance values. We added the seven samples containing the new taxa to create a 124 taxa x 224 sample data matrix and re-ran the Chessman process. We then adopted the seven new tolerance values from this analysis, while retaining existing tolerance values (Stark and Maxted 2004). The fact that 86% of existing tolerance values were unchanged and 99% changed by less than ±1 justified this approach.

For the final version of the MCI-sb, however, Stark and Maxted (2007) adopted a different approach. The entire Auckland soft-bottomed data set (2000−2005, 224 samples) was used to derive tolerance values for 124 taxa using the Chessman process. These tolerance values were calculated to the nearest 0.1 rather than to the nearest integer (e.g. the MCI-sb tolerance value for the mayfly Acanthophlebia is 9.6, cf. 7 for the MCI), because this improved the performance of the resulting indices (i.e. it gave higher correlations with environmental variables). Tolerance values for an additional 35 taxa were derived by running the Chessman process on all of the soft-bottomed data available to us - a total of 1,159 samples from Northland, Auckland, Waikato, Bay of Plenty, Hawke's Bay, Taranaki, and Otago. These data contained 35 new taxa. Comparison of the 124 existing tolerance values (derived from the Auckland analyses) with those produced by this analysis showed that 21% were the same, with 57% within ±1, 82% within ±2, and over 93% within ±3 of the Auckland-derived tolerance values. This agreement is good enough, in our view, for us to retain the existing tolerance values and adopt the 35 new ones. [Chessman (2003) derived tolerance values (grades) from 24 regional data sets and expressed scores as means with a standard error (SE) provided as a measure of confidence in the averaged national SIGNAL2 grade. Most SEs were less than one unit; the higher SEs (up to 3.2) were usually for rarer taxa. We could not adopt a similar approach because there were insufficient regional data sets available. However, the variability in scores that we encountered based on analyses of various data sets (and combinations of samples) corresponded to SEs between 0 and 2.5, with SEs for over 77% of taxa ≤ 1. Thus, we believe that the approach we used provided tolerance values that should be fairly reliable.]

3.3 Calculating the MCI, QMCI and SQMCI

The MCI is calculated from presence-absence data as follows.

where S = the total number of taxa in the sample, and a_i is the tolerance value for the ith taxon (see Table 1).

The QMCI is calculated from count data as follows.

where S = the total number of taxa in the sample, n_i is the abundance for the ith scoring taxon, a_i is the tolerance value for the ith taxon (see Table 1) and N is the total of the coded abundances for the entire sample.

The SQMCI is calculated in a similar way to the QMCI, except that coded abundances (assigned to the R, C, A, VA and VVA [R = Rare; C = Common; A = Abundant; VA = Very Abundant; VVA = Very Very Abundant.] abundance classes) are substituted for actual counts:

where S = the total number of taxa in the sample, n_iis the coded abundance for the ith scoring taxon (i.e. R = 1, C = 5, A = 20, VA = 100, VVA = 500), a_i is the tolerance value for the ith taxon (see Table 1), and N is the total of the coded abundances for the entire sample.

Versions of the MCI developed specifically for soft-bottomed (SB) streams are calculated in exactly the same way, except that a different set of taxon tolerance values is used (see column SB in Table 1). Most taxa commonly encountered in soft-bottomed streams have been assigned tolerance values. If a taxon that has not been scored is encountered, the hard-bottomed tolerance value can be used. Alternatively, if data containing the unscored taxa are available, the first author of this report (John Stark) could derive new tolerance values using the Chessman process.

QMCI and SQMCI values range from 0 to 10 and are directly comparable with each other (Stark 1998). MCI values range from 0 to 200 (Stark 1985). Only when no taxa are present are these indices zero. In practice it is rare to find MCI values greater than 150 (or SQMCI and QMCI >7.5) and only extremely enriched stony riffle sites score less than 50 (QMCI and SQMCI <2.5). The soft-bottomed versions are analogous (Stark and Maxted 2004, 2007). The different scales for the indices were chosen deliberately to avoid inappropriate comparisons.

Table 1: Tolerance values for MCI-based biotic indices in hard-bottomed (HB) (Stark et al 2001) and soft-bottomed (SB) (Stark and Maxted 2007) streams

3.4 Interpreting the MCI

The interpretation of index values when applied to stony (MCI, SQMCI, QMCI) or soft-bottomed (MCI-sb, SQMCI-sb, QMCI-sb) streams throughout New Zealand is given in Table 2. The quality thresholds are the same for hard-bottomed and soft-bottomed streams, making the new indices easy to implement. These thresholds do not work well when the hard-bottomed indices are applied to soft-bottomed streams, however. For example, soft-bottomed reference sites (which should be high quality) had MCI scores <119, indicating possible mild pollution (Maxted et al 2003; Stark and Maxted 2004, 2007). This provided the motivation for developing a separate set of tolerance values for taxa found in soft-bottomed streams.

Although Stark (1998) provided interpretive descriptions based on enrichment or pollution, we now prefer to use the quality classification used by Stark and Maxted (2004, 2007) (see Table 2). This recognises that the MCI (and its variants) respond to an interacting complex of environmental variables including (but not limited to) enrichment.

Table 2: Interpretation of MCI-type biotic indices

The index values corresponding to divisions between the four quality classes were selected initially by Stark (1993b) based on professional judgement. However, Stark and Maxted (2004, 2007) used an objective procedure based on the statistical distribution of biotic index values at references sites, together with an estimation of the lowest practical index value, to determine divisions between quality classes. A similar procedure had been used previously in the United States by Maxted et al (2000). In brief, the "excellent" quality class was set at the 25th percentile of the reference site biotic index distribution. This means that 75% of all reference samples have higher index values than this threshold and are assigned to the "excellent" quality class. The midpoint between "excellent" and the "lowest practicable" index value was set as the threshold between "fair" and "poor". The range between the "excellent" and "fair/poor" thresholds was then bisected to set the threshold between the "good" and "fair" classes. When this procedure was applied to the MCI-sb scores for Auckland soft-bottomed streams, it resulted in the same thresholds that Stark (1998) had provided for the MCI. [The "worst site" (an unnamed tributary of Wairau Creek, off Goldfield Road, North Shore City) had an MCI-sb of 40, and the 25th percentile of the reference site biotic distribution was 126.1 (which was rounded to 120).]

We believe that you should be flexible when interpreting the divisions between quality classes (see Table 2), and that it is best to regard the boundaries between them as fuzzy. This concept is not new: see Figures 1 to 3 in Stark (1985), where it was suggested that the divisions between three site groups (which were, in effect, pollution classes) should be 120 ±5 units, 100 ±5 MCI units, and so on.

The same suggestion was made by Wright-Stow and Winterbourn (2003) following their examination of the correspondence between the MCI and QMCI using fixed-count data from 230 stream and river sites in Canterbury. The two indices ranked sites similarly (r_s = 0.86), but the MCI placed most sites in the "good" and "fair" pollution classes, whereas most sites were assigned to the "excellent" and "poor" classes by the QMCI. Wright-Stow and Winterbourn concluded that either the MCI was a more conservative index, or that the boundaries between pollution classes were not equivalent. The latter reason was considered more likely, and given the difficulties inherent in defining classes based on continuous distributions and the fact that there is no way of knowing which index gives the "right" answer, Wright-Stow and Winterbourn suggested a return to fuzzy boundaries between classes (MCI: Excellent 125−200, Good 105−115, Fair 85−95, Poor <75; QMCI: Excellent 6.2−10, Good 5.2−5.7, Fair 4.2−4.7, Poor <3.7).

Alternatively, when comparing large numbers of sites, as in SoE monitoring, Wright-Stow and Winterbourn (2003) suggested that the percentile within which the site of interest falls could be stated. A site with an MCI of 130, for example, could be described as being within the top 10% of sites in the region.

Fuzzy boundaries are desirable because there is always error when estimating biotic indices. Stark (1998) has shown that the MCI from a single hand-net sample has a precision of approximately ±10%. For example, an MCI of 117 taken at face value would assign a site to the "good" quality class, but given the ±10% error inherent in the index estimate, the true MCI could have been anywhere from 105 to 129. The balance of probability would still place that site in the "good" class, but it could possibly be classified as "excellent". We quantified fuzzy boundaries for soft-bottomed streams and found the error twice as high for the QMCI-sb (±12−22%) compared to the MCI-sb (±4−9%) (Stark and Maxted 2004, 2007). This error in QMCI-sb estimates is a major reason why the MCI-sb is recommended for assessing soft-bottomed streams.

In such cases, how should you decide which quality class to assign the site to? Consider, for example, SoE reporting that is based on coloured dots on maps - green dots denote "excellent" stream condition, yellow "good", orange "fair", and red "poor". If a site has an MCI of 126 in year one, 119 in year two and 124 in year three - values that are unlikely to be significantly different − the site would be regarded as "excellent" in years one and three but only "good" in year two if these values were interpreted strictly in terms of the guidelines in Table 2. If there was no reason why there should have been a decrease in stream health in year two, then we believe that the site could remain classified as "excellent". Alternatively, it could be described as "good-excellent" with a symbol that was 50% green and 50% yellow.

Thus, for borderline biotic index values (i.e. threshold ±5 MCI units or 1 SQMCI or QMCI unit), we suggest that the ecologist should be able to choose the more appropriate pollution class to assign the site to, based on other information (such as knowledge of water quality, catchment land use, or the existence of point or diffuse sources of enrichment). A borderline site alternating between two quality classes from year to year is undesirable when annual SoE reports are prepared because it is more likely to reflect sampling error (combined with the quality class threshold effect) than indicate any real change in stream condition.

This is not an issue with more sophisticated analyses of biotic indices (such as time series analyses) because the assignment of sites to pollution classes based on single estimates of index values is not required.

3.5 Strengths and weaknesses of the MCI and variants

The MCI, QMCI, and SQMCI were developed to assess organic enrichment in stony streams by sampling in stony riffles, where the greatest variety of the most sensitive macroinvertebrates may be expected (Stark 1985, 1998). The MCI-sb, SQMCI-sb, and QMCI-sb have been developed for assessing the condition of soft-bottomed streams (Stark and Maxted 2004, 2007). These indices are designed to be used with samples collected according to the national protocols (Stark et al 2001).

The MCI and MCI-sb respond to any perturbation that alters the list of taxa (i.e. taxonomic composition) present at a site. The QMCI and SQMCI, and their soft-bottomed variants, respond to changes in taxonomic and numerical composition or relative abundances.

Because the MCI reflects changes in community taxonomic composition, not numerical composition, it is less sensitive to subtle changes in community composition than the QMCI or SQMCI. Sometimes this is an advantage (e.g. for SoE monitoring, see Section 0, Part 2), but for compliance monitoring, where subtle changes in community composition need to be assessed, the QMCI (or SQMCI) would be more appropriate. Overall, it is important to use a version of the MCI that meets the aims of the investigation.

High MCI values can be derived from taxonomically poor communities (e.g. situations where a few individual mayflies and stoneflies are found). High MCI values when taxa richness is low (say <5 taxa per sample) may be an indication of impairment and should be interpreted with caution.

Note that the MCI-type biotic indices have been developed to assess nutrient enrichment/sedimentation in stony- and soft-bottomed streams: they have not been evaluated for other habitats (e.g. lakes, ponds, wetlands, large non-wadeable rivers, hot springs), or for other types of disturbance (e.g. toxic discharges, flow variation). It is possible that the MCI (or similar indices) might work in these other situations, but they should be used and interpreted with extreme caution. For example, Maxted et al (2003) found that the MCI and SQMCI performed acceptably in soft-bottomed streams, but that the interpretation differed (e.g. MCI >100 indicated a soft-bottomed stream in reference condition, cf. >120 in a stony stream). However, the MCI-sb, which was developed specifically for soft-bottomed streams, performs much better with an interpretation that is consistent with the stony-stream MCI, an expanded range (permitting greater discrimination in stream health both between and within land-use classes), and higher correlations with environmental factors and land uses that are known to affect stream communities (Stark and Maxted 2007).

The main criticism of indices (including biotic indices) is the inevitable loss of information compared with the raw data from which index values are derived. Such criticisms generally are made by biologists who can make sense of raw data, but who may not always appreciate the needs of water managers who cannot. We maintain that it is better to convey 40% of the information so that all of it is understood, than all of the information in a form in which only 10% is understood.

Unfortunately, the attractiveness of biotic indices to water managers can lead to their misuse and misinterpretation. For example, if the objective is to assess between-site differences in water quality, and one site is a stony riffle while the other is silty or sandy, then the difference in MCI will not be entirely due to the quality of the water. Invariably, biotic indices respond to a complex of factors (primarily water quality, substrate, and disturbance), so interpretation can be difficult and should be made only by those with suitable training and experience.

Biotic indices should not be the sole means of analysing or depicting biological data if a comprehensive assessment is required. In addition, we recommend looking at EPT richness (either the number or percentage of taxa richness comprising mayflies, stoneflies and caddisflies [EPT = Ephemeroptera, Plecoptera and Trichoptera.] ) and macroinvertebrate densities (if available), and discussing or tabulating the dominant (say top five) taxa at monitoring sites. Total taxa richness is not, in our view, a particularly useful indicator of stream health. Multivariate analyses (clustering and/or ordination) may also be useful to "let the data tell their own story".

The major problem with the use of biotic indices is establishing that they actually measure features of the environment that are of interest, and that they reflect environmental change in some ecologically meaningful way (hopefully linearly) (Norris and Norris 1995). One measure of the performance of a biotic index is to determine whether interpretations based on indices are consistent with those produced by other methods. For example, Stark (1985) found that the MCI produced interpretations consistent with those based on the more traditional quantitative and descriptive analyses used prior to its introduction, by workers such as Hirsch (1958), Winterbourn et al (1971), Winterbourn and Stark (1978), and Marshall and Winterbourn (1979). The MCI (and variants), however, do also have a track record of proven performance in the scientific literature (see Appendix 2).

The tolerance values for most biotic indices usually assume a particular type of pollution (frequently organic enrichment), a particular habitat type (e.g. stony riffles), and a particular geographic area. Be careful when using indices to assess different kinds of pollution (e.g. sedimentation, inorganic chemicals, metal toxicity), or in different habitats (e.g. weed beds, swamps, lakes and estuaries). Applying biotic indices to different regions will require the derivation of tolerance values for the taxa encountered, but as yet there is no convincing evidence to suggest that different indices would be required for different parts of New Zealand (although it is almost certain that better-performing indices for use in different eco-regions could be derived from suitable regional data sets).

A strong correlation of biotic indices with chemical pollution measures may seem desirable. Indeed, some workers have suggested that "subjective tolerance estimates" should be replaced by "quantitative tolerance determinations" by extensively examining the correlations between species presence and water quality (Herricks and Cairns 1982). Others have questioned the validity of this approach, however. Washington (1984) noted that biotic indices have been developed primarily to assess organic pollution, so they should have high correlations with biological oxygen demand (BOD), dissolved oxygen (DO) and total organic carbon (TOC), but not necessarily with other chemical parameters. In fact, there is no a priori reason why a biotic index should correlate only, or primarily, with chemical data, because chemical changes are not mirrored uniformly by biological organisms or communities (Washington 1984).

Benthic macroinvertebrate communities respond to changes in water quality and bed sediments (Katoh 1992). When using biotic indices to assess water quality, it is essential to minimise or eliminate between-site variation in other factors (particularly substrata) otherwise it will be difficult to determine the causes of any changes in biotic indices. Artificial substrates can be used for this purpose (De Pauw et al 1986). Often, however, it is the overall quality of the habitat (i.e. both the substrate and water quality) that is of interest, and biotic indices are suitable for this purpose.

3.6 Alternative or complementary approaches

3.6.1 Predictive models

Although this report is about the use of the MCI, there are alternative or complementary ways of undertaking biological assessments. Indeed Stark (1985) concluded by noting:

Finally, I must stress that a biotic index (such as the MCI) must not become the be-all-and-end-all of biological monitoring programmes. A biotic index can be a useful management tool but if progress is to be made, especially in the understanding of habitat requirements and tolerances of macroinvertebrate species, then it is essential that detailed quantitative and taxonomic studies continue to be undertaken whenever possible.

The MCI has stood the test of time and has been the most often used measure of stream health in New Zealand since it was developed in the mid-1980s (see Appendix 2). However, it has not been without its critics. The Ministry for the Environment (1997), for example, in a discussion document outlining the proposed Environmental Performance Indicators Programme, noted:

In New Zealand scientists have developed the Macroinvertebrate Community Index (MCI), but this was developed explicitly to assess nutrient enrichment for Taranaki streams. At the time it was developed the MCI was considered "state of the art", but techniques overseas have now moved well beyond the MCI.

This statement is somewhat misleading, because although the MCI was developed initially using a Taranaki ringplain macroinvertebrate data set, it was tested on data from Manawatu, Canterbury and Southland, leading Stark (1985) to conclude that it showed potential for application throughout New Zealand - an assertion that was validated subsequently by Quinn and Hickey (1990) (see Appendix 2). Furthermore, biotic indices are far from obsolete, are widely used around the world, and are still being developed, not only for freshwater (e.g. Artemiadou and Lazaridou 2005; Davy-Bowker et al 2005; Jiang 2006), but also for marine ecosystems (e.g. Borja and Muxika 2005).

AUSRIVAS was trialled in the Waikato region of New Zealand (Coysh and Norris 1999), but it has not been adopted nationwide. Joy and Death (2003) are strong advocates of undertaking biological assessments using predictive models derived from the British RIVPACS (Clarke et al 2003) or Australian AUSRIVAS (Davies 1997). In our view, predictive models and biotic indices are complementary, and both may be part of stream health assessment programmes. In addition, multivariate data analyses using canonical correspondence analysis or non-metric multi-dimensional scaling can be used to analyse raw macroinvertebrate data. These methods can provide further insight into the data and the summary measures (e.g. MCI) derived from them.

3.6.2 Multi-metric indices

A multi-metric index comprises several metrics that incorporate biological components that are sensitive to a broad range of human activities (e.g. sedimentation, organic enrichment, toxic chemicals, or flow alteration). The QMCI or MCI is included as one of seven metrics in NIWA's adaptation of the Index of Biotic Integrity (IBI) of the US EPA Rapid Bioassessment Protocol III (Plafkin et al 1989) (e.g. Quinn et al 1997a).