Intercalibration of RIVPACS for the Irish Ecoregion
April 2005


RIVPACS (River Invertebrate Prediction and Classification system) is a software package devised by scientists at the Centre for Ecology and Hydrology (CEH, Dorset) formerly the Institute of Freshwater Ecology. The system produces a site-specific prediction of the macroinvertebrate fauna to be expected in the absence of environmental stress based on a combination of geographical, physical and chemical features for that site.

RIVPACS III was developed in 1995 and included a 614 reference site database for Great Britain and 70 reference sites for Northern Ireland. This was further expanded to include an additional 40 Northern Ireland sites resulting in the final Northern Ireland RIVPACS classification consisting of 110 sites (Wright et al., 1995).

Recent research at University College Dublin (UCD) assessed a wide array of reference sites with detailed macroinvertebrate inventories which have the potential to develop a predictive system conceptually similar to RIVPACS. The main objective of this study was to produce a collated macroinvertebrate species database for the entire island and investigate whether the development of a single prototype system which uses RIVPACS protocols could be used for WFD reporting for the Irish Ecoregion.

Macroinvertebrate family and presence/absence data for three seasons spring, summer and autumn for the 110 Northern Ireland sites were supplied by J. Davy-Bowker (CEH). The environmental dataset was supplied by Imelda O’Neill (EHS) and Chris Burns (EHS). The macroinvertebrate data for the 97 Republic of Ireland sites were supplied by Catherine Bradley and Mary Kelly-Quinn (UCD). The environmental data were also supplied by Catherine Bradley and Martin McGarrigle (EPA).

A total of 519 macroinvertebrate taxa representing 106 families were recorded in the combined dataset. Representative taxa from all of the major aquatic macroinvertebrate groups were present in each of the datasets. Over 300 of the taxa identified to species/genus level occurred in less than 5% (10 sites) of the total sites indicating the presence of a complex noisy dataset.

Differences were evident at the species/genus level between the two datasets. In particular, the three stoneflies Protonemura meyeri (Pictet), Leuctra hippopus (Kempny) and Siphlonoperla torrentium (Pictet) which occurred in over 90% of the ROI sites had much lower occurrences in the Northern Ireland dataset. Perla bipunctata Pictet and the trichopteran Odontocerum albicorne (Scopoli) occurred in less than 5% of the Northern Ireland sites but occurred in over 50% of the ROI sites.

This research project highlighted several issues that need to be addressed before the development of a predictive model for the entire island could be realised. The absence of normal distributions within several of the environmental variables within the presently available dataset confounded their potential use in any development of a predictive model at this time.

TWINSPAN was used to generate a site classification using the macroinvertebrate dataset. Two separate classification procedures were carried out using the sum-option data consisting of the standardised species data plus the family log category abundance data and the max-option procedure consisting of the standardised species data plus the maximum family categorical data for the combined spring and autumn seasons.

Classification of the sum-option dataset resulted in the generation of fourteen endgroups after four TWINSPAN divisions ranging in size from one to 55 sites per group. High heterogeneity within each of the endgroups however was apparent. The max-option classification which emphasized the maximum seasonal family log category observed, generated 16 groupings ranging in size from 2 to 34 sites per group after four division levels. Heterogeneity within the endgroups again was generally high. The high number of endgroups produced in order to reach an acceptable level of heterogeneity in both analyses is not preferable when trying to
develop a predictive system. The small size of some of these endgroups is also undesirable.

Multiple Discriminant Analysis was then performed to find combinations of environmental variables that would best predict the defined site groupings. Several problems occurred when the assumption of homogeneity of covariance matrices between groups was not met. Five of the environmental variables could not be normalised rendering them inadequate for analysis. Variables that were highly correlated were also removed.

The remaining environmental variables were not very successful in predicting the membership of the sites to the groups within the classifications. The issue of normality needs to be addressed for any future development of a predictive system. The addition of more sites in future analyses would be beneficial.

Copies of this report are available from the Foundation, in electronic format on CDRom at 20.00 + VAT or hard copy at 35.00, less 20% to FWR members.

N.B. The report is available for download from the SNIFFER Website