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

A User Guide for the Macroinvertebrate Community Index

Acknowledgements

Part 1: Background to Biotic Indices

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

1 Applying MCI indices in different freshwater environments

2 Taxonomy

3 Monitoring and reporting

4 Design of monitoring programmes

5 Detecting trends

6 Reporting the MCI

References

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

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

5 Detecting trends

One question that crops up after SoE monitoring has been undertaken for several years is whether or not in-stream conditions have deteriorated, improved or stayed the same. To answer this question, a class of statistics called time-series or trends analysis is used.

Different techniques for time-series analyses vary in their data requirements and complexity. Linear regression-based parametric methods should only be used when the trend is expected to be linear (often it is not), unless data transformations are used. There are also problems with parametric methods such as linear regression if there is heteroscedasticity in the data (i.e. variance differs with time). Non-parametric techniques for trend analysis are much better able to handle non-normal data with censored, tied and missing values, so they have been favoured for analysing trends, particularly in water quality data.

A popular non-parametric trend test for water quality data is the seasonal Kendall trend test - a technique described by McBride (2005) and implemented by Bill Vant of Environment Waikato (see Vant and Smith 2004). However, this technique requires monthly data collected for at least three − and preferably five − years or more (i.e. 36 to 60 data points). This is seldom the case for biotic data, where sampling may be seasonal at best, and is more often undertaken only once or twice a year. Although the seasonal Kendall test could be adapted for detecting trends in biotic data collected much less frequently, it is doubtful whether it would be worth the effort compared with less complicated methods that are likely to give a similar result.

Stark and Fowles (2006) examined several simple statistical approaches for detecting significant trends in stream macroinvertebrate biological indices. These included non-parametric tests based on the Mann−Kendall or Spearman rank correlations (Collier and Kelly 2006) and a parametric approach using linear regressions.

When trend testing is undertaken at a number of sites, it is probable that some significant results will be obtained by chance. The FDR analysis is undertaken when multiple comparisons are made in order to eliminate significant positive or negative trends that may have arisen by chance. The positive trend in MCI shown on Figure 2 had a probability of 0.011, which would normally be considered significant, but since it was one of a batch of 60 trends tests, the FDR cut-off value of P was 0.0065, so this significant result was deemed non-significant (i.e. it could have occurred by chance).

When a trend has been identified there are two vital questions that should be asked (Rutherford 1985):

• What caused the trend?
• Will it continue into the future?

Unfortunately, trend analysis alone cannot provide definite answers to these closely related questions. To find out why a significant trend has occurred you will need additional information, which could include data on stream flows, weather patterns, catastrophic erosion events in the catchment, physicochemical water quality, or changes to land or water management practices that may have resulted from water management initiatives, or industry or farming activities. It is worth emphasising that the final decision on whether or not any trend should be considered ecologically significant is reliant on the best practice judgement of an experienced freshwater ecologist. We caution water managers about interpreting statistically significant trends in stream biological health, particularly if these trends are only marginally significant, if they cannot be explained and/or may be unrelated to initiatives aimed at improving stream condition.

If the cause(s) of the trend can be determined, then one may be able to predict the future with some confidence. Conversely, if one cannot determine why a trend has occurred, then extrapolation into the future may be most unwise because there remains a chance that the trend is simply an artefact of natural variability or sampling error.

Figure 2: Scatterplot of MCI versus time for the Huatoki Stream at Hadley Drive (Site HTK000350) in Taranaki with a LOWESS (tension = 0.4) fitted line

View Scatterplot of MCI versus time for the Huatoki Stream at Hadley Drive at full size including text description.

Experience with trend analysis of macroinvertebrate data is limited (because there are few long-term data sets), so the minimum number of sampling times required before meaningful trends can be detected is uncertain. Collier and Kelly (2006) suggested that a time series of five occasions is the minimum, but Stark and Fowles (2006) consider that trend testing is best undertaken with a time series of 10 or more. Scarsbrook et al (2000a, b) examined data from annual sampling of macroinvertebrates from 66 of the National River Water Quality Network sites from 1989 to 1996 (inclusive). These data were sufficient to detect trends in various macroinvertebrate community measures, many of which coincided with general trends in water quality over the same period (Smith et al 1996), suggesting that at least some of the measured indices (such as MCI and %EPT) were appropriate biological indicators of trends in water quality at a national scale. However, no causal links were established, and it is possible that some of the trends were artefacts of the short time series (i.e. only eight data points each, one year apart).