As of January 2004 au.riversinfo.org is archived under Precision Info

RIVERSINFO AUSTRALIA ARCHIVE

This archive provides non stylised versions of selected reference documents (formerly) resident on au.riversinfo.org.

Contents

Foreword & Publication Details

Abstract & Acknowledgements

1. Introduction

• 1.1 Project Objectives

2. Methods

• 2.1 Study Area and Site Selection
• 2.2 The Data
• 2.3 The Model
• 2.3.1 Classification Analysis
• 2.3.2 Discriminant Function Analysis
• 2.3.3 Rare taxa
• 2.3.4 Validating and defining the population of reference sites
• 2.3.5 O/E taxa bands representing biological health
• 2.3.6 Test site assessment evaluation

3. Results and Discussion

• 3.1 The Data
• 3.1.1 Model construction using mixed-method data
• 3.2 The Model
• 3.2.1 Test site evaluation

4. Conclusions

• 4.1 The Data
• 4.2 The Models

5. References

Appendix 1

• Reference and test sites used to create the AUSRIVAS Urban Family and Lowest taxon predictive models

Appendix 2

• The taxa used in the construction of the urban models.

Appendix 3

• List of sites where paired sample processing data were available

List of Figures & Tables

• List of Figures
• List of Tables

More About Ausrivas

 AUSRIVAS web site

AUSRIVAS (Australian River Assessment System) is a prediction system used to assess the biological health of Australian rivers. AUSRIVAS was developed under the National River Health Program (NRHP) by the Federal Government in 1994, in response to growing concern in Australia for maintaining ecological values. The NRHP involves the major environmental agency in each state and territory and is centrally administered by Environment Australia (EA) and the Land and Water Resources Research and Development Corporation (LWRRDC). The AUSRIVAS predictive software was developed at the Co-operative Research Centre for Freshwater Ecology (CRCFE). (Added by B Hall for the Electronic Version).

• Foreword
• Publication Information

Foreword

This report describes the outcomes of a research project conducted under the Urban Research and Development sub-program of the National River Health Program (NRHP).

The NRHP is an on-going national program established in 1993, managed by the Land and Water Resources Research and Development Corporation (LWRRDC) and Environment Australia. Its mission is to improve the management of Australia's rivers and floodplains for their long-term health and ecological sustainability, through the following goals:

1. To monitor and assess the health of Australia's rivers.
2. To enhance the management of river flows, water allocation and water use to ensure the sustainability of river and floodplain ecosystems.
3. To encourage active management to improve the health of Australia's rivers, based on a sound understanding of the ecological and hydrological processes.
4. To evaluate the effectiveness of river management actions at a national scale.

Urban streams and estuaries (i.e. those affected by runoff and discharges from urban areas) are an important subset of Australia's waterways. Most are degraded biologically, physically and chemically and therefore require appropriate methods to be developed for health assessment and management. The Urban R&D Sub-program, managed by the Water Services Association of Australia, comprises 8 research projects which were developed to meet research priorities for urban streams and estuaries within the goals of the NRHP and to complement existing NRHP projects on non-urban rivers. Thus, research focuses on development of standardised methods for assessing the ecological health of urban streams and estuaries which can be linked with data on water and sediment quality. The urban R&D projects commenced in 1996.

The definition of health in urban waterways used is "the ability to support and maintain a balanced, integrative, adaptive community of organisms having a species composition, diversity and functional organisation as comparable as practicable to that of natural habitats of the region".

The eight projects of the Urban Sub-Program are:
 

Decision support system for management of urban streams	Dr John Anderson Southern Cross University, Lismore
RIVPACS (River InVertebrate Prediction and Classification System) for urban streams	Dr Peter Breen CRC for Freshwater Ecology, Monash University, Melbourne
DIPACS (Diatom Prediction and Classification System) for urban streams	Dr Jacob John Curtin University, Perth
Sediment chemistry- macroinvertebrate fauna relationships in urban streams	Dr Nick O'Connor Water EcoScience, Melbourne
Classification of estuaries	Dr Peter Saenger Southern Cross University,Lismore
Literature review of ecological health assessment in estuaries	Mr David Deeley Murdoch University, Perth
Estuarine health assessment using benthic macrofauna	Dr Gary Poore Museum of Victoria, Melbourne
Estuarine eutrophication models	Dr John Parslow CSIRO Marine Laboratories, Hobart

Publication Details

Published by:
Land and Water Resources Research and Development Corporation
GPO Box 2182 Canberra ACT 2601
Telephone: (02) 6257 3379
Facsimile: (02) 6257 3420
Email: public@Iwwrrdc.gov.au
WebSite: www.lwrrdc.gov.au
© LWRRDC

Published Electronically on au.riversinfo.org by the Environmental Information Association (Incorporated) with the permission of LWRRDC and Environment Australia. Environment Australia assisted by providing copies of the manuscript for electronic publication. The Natural Heritage Trust provided project funds which were used to assist in publishing this material. In the case of variation between this document and the hard copy original the original takes precedence. (Bryan Hall).

Disclaimer:
The information contained in this publication has been published by LWRRDC to assist public knowledge and discussion and to help improve the sustainable management of land, water and vegetation. Where technical information has been prepared by or contributed by authors external to the Corporation, readers should contact the author(s), and conduct their own enquiries, before making use of that information.

Publication data:
Urban AUSRIVAS: An evaluation of the use of AUSRIVAS models for urban stream assessment. CRC for Freshwater Ecology P. Breen^1,2, C. Walsh^1,2, S. Nichols³, R. Norris³, L. Metzeling⁴, &. J. Gooderham^{2 1}Melbourne Water Corporation, ²Water Studies Centre, Monash University, ³University of Canberra, ⁴Environment Protection Authority, Victoria
Report No 5, LWRRDC Occasional Paper 12/99.

ISSN 1320-0992
ISBN 0 642 26766 9

Managing Agencies

• The Urban Sub Program was managed by Water Services Association of Australia under a contract with Land and Water Resources Research and Development Corporation.
• Environment Australia funded the research projects of the Urban Sub Program of the National River Health Program.
• Hard copy designed by: Nuttshell Graphics, North Melbourne
• Hard copy printed by: Aristoc Offset, Mt Waverley
• Electronically Published by: Environmental Information Association (Incorporated)

Abstract

The feasibility of developing and using area specific urban AUSRIVAS models was evaluated using existing stream macroinvertebrate datasets for the Melbourne metropolitan and hinterland area. To obtain an adequate number of reference sites for model development a number of datasets had to be combined. Melbourne hinterland sites with predominantly rural landuse were adopted as reference sites for urban areas. Datasets from the CRC for Freshwater Ecology (lab-sorted samples) were combined with the Victorian EPA National River Health Program dataset and the Melbourne Water Corporation Streamwatch dataset (both field-sorted samples). It was determined that lab-sorted and field-sorted datasets had some quantitative differences in the taxa recorded for the same sites. However, despite these systematic differences, multivariate analyses of community composition indicated that the lab-sorted and field-sorted datasets contained similar information about the patterns in macroinvertebrate community composition.

Successful urban models were constructed using both family and lowest taxonomic level data for hinterland sites. O/E scores were negatively correlated to catchment imperviousness and biochemical oxygen demand. These variables are considered to be indicators of urban density and efficiency of pollutant delivery to streams. The experimental models suffered from a lack of appropriate reference sites and a poor physiographic coverage of reference sites. This is likely to be a problem for many large cities. Results from the urban family model and the EPA Victorian family model were not significantly different.

A potential alternative approach is presented which uses existing regional models to assess urban environments. This approach uses a model with minimally impacted reference sites and develops a relationship that accounts for the impact of urbanisation on the model output. Catchment imperviousness measures the degree of catchment landuse change and is generally un-related to the type or quality of drainage infrastructure. Consequently it can be used as a general measure of urbanisation and as a covariate against which other ecological indicators can be set. Deviation from this relationship can be used to identify sites that are more (or less) disturbed given a particular level of imperviousness.

Acknowledgements

The Water Services Association of Australia is acknowledged for project funding. The CRC for Freshwater Ecology, Melbourne Water Corporation and the Victorian Environment Protection Authority are thanked for access to their data. John Dean is thanked for making available data from the Yarra River. We also thank Dr Richard Marchant and Dr Phillip Suter for reviewing the report.

1. Introduction

The National River Health Program (NRHP) was formed in 1993 to assess the health of Australian rivers at the national scale. Under this program the Cooperative Research Centre for Freshwater Ecology developed the Australian River Assessment System (AUSRIVAS) to predict the health of rivers at the state and regional scale (Coysh et al. 2000). The current AUSRIVAS models were not specifically developed for use in urban areas. It was hypothesized that a model based on reference sites in the immediate region of Melbourne may provide greater sensitivity to urban disturbances than larger-scale regional models. This hypothesis was tested by comparison of the results of the model developed here with an existing Victorian model (Marchant et al., 1997). A major objective of the Urban R&D sub-program is to foster the development of stream health assessment techniques suitable for use at a smaller scale in urban areas.

AUSRIVAS is broadly based on the strategy of Wright et al. (1984). The procedure is summarised as follows. A series of minimally impaired sites are chosen as reference sites. Chessman (1995) describes a procedure for reference site selection. Reference sites are then classified into groups based on their macroinvertebrate community composition by cluster analysis. Environmental predictor variables that best discriminate among the groups of reference sites are then identified by such techniques as multiple discriminant function analysis. Environmental variables that are commonly influenced by human activity are usually excluded from the analysis. To use the model any number of test sites are then selected and, the limited set of environmental characteristics selected in the last step above are measured and the biota sampled. The environmental variables are used to match the test sites with groups of reference sites and test site probabilities of group membership are calculated. Once group membership of a test site has been determined from the environmental data, the taxa (at whatever taxonomic level has been used) that should occur at an individual site are predicted. Minimally impaired reference sites are selected to create the predictive models, thus the AUSRIVAS models predict the expected macroinvertebrate community structure at a test site independent of impact from anthropogenic factors such as changes in water quality or flow patterns. The difference between the observed community structure (O) at a test site and that expected (E) by the model can be used as an indicator of impact. Test sites equivalent to reference condition can be expected to have O/E taxa ratios close to one, whereas impacted sites can be expected to have O/E taxa ratios much less than one.

1.1 Project Objectives

The primary aim of the project was to develop a modeling approach for urban stream health assessment. Specific objectives included:

• The development and testing of an AUSRIVAS (Australian River Assessment System) model for urban streams using existing macroinvertebrate data
• Preparation of models consistent with those being prepared for the NRHP

An important first step in the process was the task of constructing an adequate database to be able to undertake the modeling. A critical issue associated with this task was ensuring the compatibility of macroinvertebrate data collected using different processing techniques, eg. preserved laboratory sample sorting (lab-sorted) versus live sample sorting in the field (field-sorted).

2. Methods

• 2.1 Study Area and Site Selection
• 2.2 The Data
• 2.3 The Model
• 2.3.1 Classification Analysis
• 2.3.2 Discriminant Function Analysis
• 2.3.3 Rare taxa
• 2.3.4 Validating and defining the population of reference sites
• 2.3.5 O/E taxa bands representing biological health
• 2.3.6 Test site assessment evaluation

2.1 Study Area and Site Selection

The study area was the Melbourne region of southeastern Australia. Reference sites were selected from the rural environment surrounding Melbourne. Test sites were within the Melbourne metropolitan area. Reference sites were in the Melbourne hinterland and were selected to represent typical to good quality rural landuse sites free from obvious sewage or stormwater discharges, dams, mining or forestry activities.

AUSRIVAS models operate best when the number of reference sites in the model is high (preferably above 30) and the reference site spatial distribution is similar to the distribution of the test sites. In order to compile a dataset with an adequate number of sites (within the study area, with similar habitat and sampling time) it was necessary to combine data from a number of organisations. Data for this project was pooled from a CRCFE study of urban stream ecology in the Melbourne region (Walsh et al., in press), Victorian EPA monitoring as part of the National River Health Program (L. Metzeling, unpublished data), and from Melbourne Water Corporation Streamwatch data collected by Water Ecoscience (Smith, Vertessy & Hardwick, 1997). The pooled dataset contained 31 reference sites and 31 test sites (Appendix 1).

The rationale behind using reference sites from surrounding rural areas lies in the proposition that rural streams, while subject to some anthropogenic disturbances, represent an achievable target for restoration of urban streams. The use of non-pristine sites as reference sites has been demonstrated in Britain (Wright, 1995). Such an approach assumes that the patterns of biotic community composition across the non-pristine reference sites remain primarily non-anthropogenic. The validity of this assumption is discussed in section 4.2.

It is further hypothesized that a model based on reference sites in the immediate region of Melbourne may provide greater sensitivity to urban disturbances that larger-scale regional models. This hypothesis will be tested by comparison of the results of the model developed here with an existing Victorian regional model (Marchant et al., 1997).

2.2 The Data

Because data for this study were derived from three institutions, the voucher collections of the EPA, Water EcoScience and the CRC FE at the Water Studies Centre were compared to eliminate cases of synonymy, or differences in identification, prior to analysis. The lowest level of taxonomic resolution used here was the lowest taxonomic level routinely identified by all three institutions (Appendix 2).

All samples were taken from 10m transects of riffle, using hand nets with 250 µm mesh (Rapid Bioassessment methods; Chessman, 1995). Samples taken by the CRC FE were preserved in the field and processed in the laboratory. Each sample was sub-sampled (Marchant, 1989) and sorted to remove 10%, or 300 organisms from each sample (Walsh, 1997). Samples taken by the EPA and the Streamwatch Program were sorted unpreserved in the field for 30 minutes.

Riffles were the only habitat sampled at enough reference sites for model construction. Samples were collected over three seasons: spring 1994 and 1995, and autumn 1995.

A primary requirement of combining data from the three institutions was to ensure compatibility of the two sample processing methods: field-sort and lab-sort. Therefore a subset of data was compiled consisting of sites that were sampled by each method within a period of two months (Appendix 3). The two samples (field-sort and lab-sort) from each site in each season is termed a _5method-pair_6. Two method-pairs exist for several sites, as samples were collected over two seasons. Lab-sort data for seven Yarra River samples and three Gardiners Creek samples were derived from the complete sorting of residues preserved after field-sorting. This data was reduced to the same level of processing used for all other samples using a computer program that simulates the Marchant subsampler (Walsh, 1997). Twenty-five pairs of riffle samples were used in the comparison.

Numbers of taxa recovered by the two methods were compared by a paired t test using un-transformed data. To evaluate any systematic taxonomic differences between the methods, the fidelity (Boesch, 1997) of each taxon to each of the methods was calculated with the following equation:

Where A is the number of samples in which the taxon occurs for one method; B is the total number of samples using that method; C is the total number of occurrences for both methods; and D is the total number of samples.

Patterns of community composition in the method-pair data were assessed using semi-strong hybrid multi-dimensional scaling (MDS) ordinations (Belbin, 1994) based on Bray-Curtis similarities for presence-absence data.

Three questions were explored in this assessment:

1. Did the two methods result in the same patterns of community composition across the 25 samples?
2. Were the differences in community composition within method-pairs small compared to the differences between samples?
3. Can the differences between method-pairs be minimized by data manipulations such as a) the removal of taxa more commonly collected by one processing method, or b) the removal of rare taxa?

The removal of rare taxa (arbitrarily defined here as taxa occurring in less than 5 samples or 10% of the dataset) is a common manipulation that may be employed in multivariate analyses to simplify data sets and reduce background noise (Marchant et al,. 1997).

The above analyses were conducted for both lowest-taxonomic level and family level resolution. These preliminary analyses were used to decide the most appropriate data manipulation for the construction of models.

2.3 The Model

2.3.1 Classification Analysis

The first step in the model building process was to classify reference sites that had similar invertebrate compositions. This was done for two data sets, one based on lowest taxonomic level and the second at family taxonomic level using presence/absence data and PATN multivariate analysis package (Belbin 1994). The Bray-Curtis association measure was used on the recommendation of Faith et al. (1987) as a robust measure of association for cluster analysis and ordination. Groups were selected based on the dissimilarity level of the clusters, the tightness of the reference site clusters and how well agency staff thought the groups represented a _5type_6 of reference site. The agglomerative clustering technique, flexible Unweighted Pair-Group arithMetic Averaging (UPGMA) recommended by Belbin and McDonald (1993) was used to form the reference site groupings. The classifications were viewed as dendrograms allowing the fusion level, which divides sites into groups, to be selected.

2.3.2 Discriminant Function Analysis

The AUSRIVAS models use habitat features (predictor variables) from a site to predict which taxa should occur at that site in the absence of environmental stress. Habitat variables that may be affected by human activity should not be used as predictor variables. Variables such as turbidity, dissolved oxygen and phosphorus concentrations are often affected by human impacts and might provide spurious predictions if used to predict the membership of test sites to the reference site groups. In contrast, habitat features such as altitude, distance from source, latitude and longitude often make good predictor variables because they are rarely affected by human impacts.

The reference site groups from the UPGMA classification were entered as a variable into the reference habitat data set and a Stepwise Multiple Discriminant Function Analysis (MDFA) performed to select the predictor variables used in an AUSRIVAS model. The Stepwise MDFA was performed in the SAS statistical package version 6.12 (SAS Institute 1995). This procedure selected a subset of habitat variables that best discriminated among the groups of sites formed from the faunal classifications. The stepwise procedure included habitat variables one at a time, selecting at each step the variables that gave the best group discrimination. At each step of the analysis the significance of variables already included was checked and variables that were no longer significant were removed. The significance levels for variables to enter and be retained by the Stepwise MDFA both were set at 0.05.

The subsets of habitat variables from the Stepwise MDFA were then tested in a MDFA to predict the probabilities of group membership for a reference site. Biased discriminations were avoided by using the cross-validation option that predicts group membership of each site separately. A subset of habitat variables that produced the lowest error in predicting the group membership of reference sites was obtained from this procedure. However, the actual value of the misclassification error is not critical because AUSRIVAS uses all the probabilities of a site belonging to each group for site predictions, rather than the allocation to a single group as performed by the cross-validation procedure. Thus, sites with an affinity for two or more groups can be misclassified but still provide adequate predictions for a model.

The subsets of habitat variables obtained from the stepwise MDFA were used as predictor variables for the AUSRIVAS model under construction. The predictor variables and the reference site invertebrate-classification form the foundation of AUSRIVAS, allowing predictions of the taxa expected at new "test" sites. The list of predictor variables used in model construction is shown in Table 1.

2.3.3 Rare taxa

Rare (infrequently occurring) taxa are commonly removed from the invertebrate data sets. Rare taxa can have a high variance because of their sparse distribution and are often removed from multivariate analyses to reduce noise (Gauch 1982) and usually their exclusion has little effect on results (Smith et al. 1988). However, in this study the influence of data transformation was specifically evaluated. It was determined that rare taxa would be included (see 3.1.1).

2.3.4 Validating and defining the population of reference sites

The initial _5working_6 models were run in SAS because this was the primary tool used in developing the predictive rationale. The full reference invertebrate and habitat datasets minus sites deleted at the classification stage, were used to create the first version of the model. The same reference data set was then entered into the model as if they were test sites to validate reference site condition. Reference sites with observed to expected taxa ratios (O/E ratios) below 0.75 were investigated more closely and deleted from the model if found to be unsuitable as reference sites. The 0.75 cut off was based on the value judgement that a site missing 25% of the taxa expected to occur there was unlikely to represent reference conditions. The models were then reconstructed with the failed reference sites removed. The reduced reference site data sets were then run through the revised model. The O/E ratio output at this stage was considered to represent the distribution of ecological health for the population of reference sites.
 

Table 1. Description of predictor variables used in model construction
Variable	Definition	Units
ALTITUDE	Altitude	m
AMGEAST	Australian map grid easting
AMGNORTH	Australian map grid northing
DEPTH	Mean water depth of riffle	cm
DISCHARGE	Mean annual discharge	ML
GRANITE	% catchment area with Granitic surface geology	%
RAINFALL	Mean annual rainfall	mm
SEDALLOC	% catchment area with "course" (gravel, sand) sedimentary alluvial surface geology	%
SEDMARM	% catchment area with Sedimentary marine (mudstone) surface geology	%
SEDMARS	% catchment area with Sedimentary marine (sandstone) surface geology	%
SLOPE	Slope of site taken from map contours (1:50000)	%

2.3.5 O/E taxa bands representing biological health

To simplify interpretation AUSRIVAS presents the O/E taxa ratios as bands that represent different levels of biological condition. The widths of the bands are based on the distribution of the O/E taxa values for the reference sites of each particular model. Test sites that fall between the 10^th and 90^th percentiles from the population of reference sites used to create the model are considered equivalent to reference condition (band A). Impaired test sites will fall into a band equivalent to the severity of impact that the site is experiencing. The next two bands (B and C), which represent increasing levels of impairment, are the same width as band A. Band D will vary in width depending on the range of the reference O/E taxa values. Test sites with biological condition richer than reference condition (i.e. > 90^th percentile) are placed in band X.

2.3.6 Test site assessment evaluation

The objective evaluation of model test site assessments is a particular problem. To evaluate model output requires an independent site impact assessment. This study has developed an interim approach. In this study region, the work of Walsh et al. (in press) has identified a series of environmental variables as being correlated with the community structure of stream macroinvertebrate communities. The two major variables indicative of urban impact on macroinvertebrates were catchment imperviousness and the water quality variable BOD. Relationships between each of these variables and O/E were assessed. To assess their combined influence, these variables were combined into a composite variable (BODIMP).

BODIMP = Catchment imperviousness (proportion of catchment area) * BOD (mg.L^-1)

The range of BODIMP is 0 (where BOD » 1 and impervious area = 0, ie. a typical undisturbed state) and 9.5 (where BOD = 10 and impervious area = 0.95, ie. polluted water with central business district level of urban development).

The range for AUSRIVAS index is 0-1, where 0 = no expected taxa observed and 1 = a match between the number of taxa expected (E) and the number of observed taxa (O). However, AUSRIVAS may produce an O/E taxa value greater than 1 when more than the expected number of taxa are observed in the sample.

This variable was calculated for a test dataset (Appendix 1) consisting of 16 lab-sorted and 15 field-sorted samples.

3. Results and Discussion

• 3.1 The Data
• 3.1.1 Model construction using mixed-method data
• 3.2 The Model
• 3.2.1 Test site evaluation

3.1 The Data

A quantitative comparison of sample processing methods indicated there were some systematic differences between lab-sorted and field-sorted samples. Using lowest taxonomic resolution, lab-sorted samples contained more taxa (mean 21.9) than field-sorted samples (mean 18.3: t = -3.823, P < 0.05). The numbers of families collected by the two methods were not different (lab-sorted mean 12.1, field-sorted mean 11.7: t = -1.113, P > 0.05).
 

Table 2. Fidelity analysis for twenty-five pairs of samples, each pair taken in the same season from the same site. One of each sample pair was field-sorted, and one was lab-sorted. Taxa that occurred more than twice as frequently in samples processed by one method than the other are listed. The analysis was conducted separately for two taxonomic levels
	No. occurrences in				No. occurrences in
Lowest-taxon level	field-sort	lab-sort	Fidelity	Family level	field-sort	lab-sort	Fidelity
More common in lab-sort
Pisidium spp.	0	8	2.00	Sphaeriidae	0	8	2.00
Empididae SRV sp.1	0	6	2.00	Empididae	3	14	1.65
Austrolimnius sp.L25E	1	6	1.71	Ceratopogonidae	1	4	1.60
Nanocladius spp.	1	6	1.71	Nematoda	2	6	1.50
Cladotanytarsus spp.	2	10	1.67	Ancylidae	6	17	1.48
Austrolimnius group A spp.	1	5	1.67	Ecnomidae	3	7	1.40
Empididae SRV sp.6	1	5	1.67
Tanytarsus spp.	1	4	1.60
Oecetis spp.	1	4	1.60
Notriolus quadriplagiatus	2	7	1.56
Ecnomus continentalis	2	7	1.56
Nematoda	2	6	1.50
Cheumatopsyche sp.4	2	6	1.50
Ferrissia sp.	6	17	1.48
Botryocladius grapeth	6	15	1.43
Parakiefferiella spp.	3	7	1.40
Oligochaeta spp.	24	25	1.02	Oligochaeta	24	25	1.02
More common in field-sort
Parastacidae spp.	6	1	1.71	Parastacidae	6	1	1.71
Baetidae genus 1 spp.	5	1	1.67	Aeshnidae	8	2	1.60
Corbiculina spp.	6	2	1.50	Corbiculidae	6	2	1.50
Paratya australiensis	6	2	1.50	Atyidae	6	2	1.50
Simsonia longipes	5	2	1.43
Illiesoperla spp.	5	2	1.43

Lab-sorted samples more commonly included a number of small or immobile taxa than did field-sorted samples (Table 2). A smaller number of taxa were more commonly found in the field-sorted samples (Table 2), all of which were large or highly mobile taxa.

Despite these systematic differences, both processing methods produced similar patterns of community composition, (Fig. 1). Samples fell into four groups:

1. Yarra river sites (Y1 in spring 1994, and Y5, Y6 and Y7 in spring 1994 and autumn 1995);
2. Eastern hinterland small stream sites (D1, O2 and W5 in spring 1994 and autumn 1995);
3. Western hinterland small stream sites (E5 and J3 in spring 1994 and autumn 1995);
4. Urban small stream sites, all within the Melbourne metropolitan area (G3, G4, G5 and U5 in spring 1994, and M5 and D5 in spring 1994 and autumn 1995).

The major difference between patterns for the two methods was the greater dissimilarity within the western hinterland communities of in field-sorted samples than in lab-sorted samples (Fig 1).

Figure 1. MDS ordinations

(original size image)

Figure 1. MDS ordinations of 25 sites used in the comparison of processing methods, based on lowest taxonomic level resolution, using all taxa. (a) Field sorted samples, (b) lab-sorted samples. Four groups of sites (eastern hinterland small streams, western hinterland small streams, Yarra River, and metropolitan sites) are grouped by kernel density contour curves, P=0.9.

The four groups of samples remained evident when method-pairs were included in the ordination (i.e. two sets of data were included for each sample: one field-sorted, one lab-sorted. Figs. 2a, 3a). For lowest-taxonomic level resolution, most differences between method-pairs were small compared to differences between sample groups (Fig 2a). For family-level resolution, the same was true for the hinterland small stream groups, but systematic differences between method pairs were evident, particularly in the urban group, but also in the Yarra group (Fig. 3a).

Biased taxa were arbitrarily defined as those that were ³ 3 times as common in one processing method than the other (Fidelity ³ 1.5, Table 2). In addition, Oligochaeta were present in all lab-sorted samples, but absent from one field-sorted sample. In light of the ubiquity of oligochaetes, and the increased chance of their omission from field-sorted samples, these were also included as biased taxa. Removal of biased taxa prior to calculating ordinations tended to reduce method-pair differences within the urban group, but the effect on method-pair differences in the other groups was small (Fig. 2b, 3b). However, the removal of biased taxa tended to reduce differences between hinterland groups, particularly at the family-level.

Metropolitan sites support depauperate macroinvertebrate communities that showed little pattern that was explicable by environmental variation (Walsh et al., in press). In the absence of strong patterns associated with environmental variation, it was not surprising that method-derived bias had a strong influence on patterns within urban sites. The lesser effect of biased taxa for hinterland sites at the lowest taxonomic level of resolution suggests that patterns associated with environmental variation among these sites were strong compared to patterns associated with method bias. The relatively greater effect of biased taxa at the family level may be due to the fewer number of taxonomic groups in the analysis and that some of the biased families contributed to group separation.

At the lowest-taxonomic level, the removal of rare taxa reduced some method-pair differences but also increased others. The removal of rare taxa had little effect on the differences between sample groups (Fig. 2c). The removal of rare families from the family-level ordination both increased and decreased method-pair differences, but more importantly, obscured differences between sample groups (Fig. 3c).

Removal of both rare and biased taxa did not improve ordination patterns at either taxonomic level (Figs. 2d, 3d).

3.1.1 Model construction using mixed-method data

Because the inclusion of all taxa, either families or species, resulted in the best group separation in community composition analyses, the models were based on all taxa. Removal of rare and biased taxa tended to obscure community patterns, particularly at the species level.

Figure 2. MDS ordinations

(original size image)

Figure 2. MDS ordinations of 25 sites used in the comparison of processing methods based on lowest taxonomic level resolution. Lines join method-pairs: i.e. samples taken from the same site in the same season but processed by different methods (field-sort and lab-sort). Large symbols indicate field-pick samples. Grouping conventions as in Figure 1. (a) Using all taxa, (b) biased taxa (with fidelity to one method of ³ 1.50, and Oligochaeta) removed, (c) rare taxa (occurring in <10% of samples) removed, (d) biased and rare taxa removed.

Figure 3. MDS ordinations

(original size image)

Figure 3. MDS ordinations of 25 sites used in the comparison of processing methods based on family-level resolution. Lines join method-pairs: i.e. samples taken from the same site in the same season but processed by different methods (field-sort and lab-sort). Large symbols indicate field-pick samples. Grouping conventions as in Figure 1. (a) Using all families, (b) biased families (with fidelity to one method of ³ 1.50, and Oligochaeta) removed, (c) rare families (occurring in <10% of samples) removed, (d) biased and rare families removed.

3.2 The Model

The reference site groupings (Table 3) produced in the classification analysis of the biological data reflect some clearly recognisable physiographic and geomorphologic characteristics of the sites (Table 4). The descriptions of the reference site groups reflect the predictor variables chosen in the models (Table 5). This may be expected and acts to increase user confidence in the model, however it is not necessarily a characteristic of a successful model.
 

Table 3. Reference site group membership
Family Model			Lowest Taxon Model
Site Code	Site	Group	Site Code	Site	Group
BI1	1	1	BI1	1	1
BI3	2	1	BI3	2	2
BI4	3	1	BI4	3	2
BI5	4	1	BI5	4	2
CA1	5	1	CA1	5	3
CT1	6	1	CT1	6	3
DA1	7	2	DA1	7	1
DP2	8	3	DP2	8	4
DP3	9	3	DP3	9	4
DP4	10	3	DP4	10	4
DP5	11	3	DP5	11	4
JA1	12	3	JA1	12	4
JA2	13	3	JA2	13	4
JA3	14	3	JA3	14	4
JA5	15	3	JA5	15	4
JA6	16	3	JA6	16	4
LA2	17	3	LA2	17	3
LI1	18	1	LI1	18	3
MK1	19	3	MK1	19	4
OL1	20	2	OL1	20	1
OL2	21	2	OL2	21	1
OL3	22	1	OL3	22	3
PL1	23	1	PL1	23	2
SH1	24	2	SH1	24	Excluded
WC2	25	3	WC2	25	3
WO1	26	2	WO1	26	1
WO2	27	2	WO2	27	1
WO3	28	2	WO3	28	1
WO4	29	2	WO4	29	1
WO5	30	2	WO5	30	1
WO6	31	1	WO6	31	Excluded

 

 

Table 4. General physiographic description of model groups
Family Model		Lowest Taxon Model
Group 1	Low slope streams on granitic geology with homogeneous sandy substratum and moderate rainfall	Group 1	Eastern ranges high slope streams on granitic geology with heterogeneous gravel substratum and moderate rainfall
Group 2	Eastern ranges high slope streams on granitic geology with heterogeneous gravel substratum and moderate rainfall	Group 2	Low slope streams on granitic geology with homogeneous sandy substratum and moderate rainfall
Group 3	Western plains, low rainfall, low slope streams on basalt geology with gravel/cobble substratum	Group 3	Moderate slope deep water streams on mixed, but predominantly granitic geology, with heterogeneous gravel substratum and moderate rainfall
		Group 4	Western plains, low rainfall, low slope streams on basalt geology with gravel/cobble substratum

The list of predictor variables and the misclassification errors for both the Family and Species models are provided in Table 5.
 

Table 5. Predictor variables used in particular models
Model	Error	Predictor Variables
Family	0.20	AMGeast	AMGnorth	Granite	Rainfall	Sedmarm	Sedmars	Slope
Species	0.28	Altitude	AMGeast	Basalt	Depth	Discharge	Sedallc

When setting the O/E bands only O/E scores between the 10^th and 90^th percentile from the population of reference sites used to create the model are considered as equivalent to reference condition (Band A). Therefore, assuming the validation reference sites are representative of the reference site population then up to 20% may be expected to fail or be misclassified. The bands used for the urban models are shown in Table 6.
 

Table 6. Bands representing different levels of biological condition for the urban models
	O/E Bands
Model	X	A	B	C	D
Family	>1.15	0.85-1.15	0.54-0.84	0.23-0.53	0.00-0.022
Species	>1.22	0.78-1.22	0.33-0.77	0.00-0.32	No band D

Family model group membership displays some inconsistencies when explaining the groups in terms of the predictor variables, for example PL1 in group 1, SH1 in group 2, and LA2 and WC2 in group 3. PL1 has high slope and the lowest rainfall of the group and at least half the catchment is on marine sedimentary sandstones. SH1 is predominately on marine sedimentary sandstones. At least half the catchment of LA2 is on coarse sedimentary alluvium. This latter geological group is very poorly represented in the reference sites and LA2 is the only site with a high proportion of this geology in its catchment. WC2 is predominately on marine sedimentary sandstones. This suggests that for at least some sites group membership may be determined simply by geographic proximity. SH1, LA2 and WC2 are also all field-sorted sites. The introduction of a fourth group in the species model resolves many of these physical inconsistencies. However, while the inclusion of an extra group in the species model (Group 3) provides more consistent physiographic groups it also further reduces the number of sites in the groups. Both Groups 2 and 3 in the species model are numerically under represented in the reference dataset and models. Also while the introduction of Group 3 may be physiographically accurate, five out of the six sites in the group are field-sorted sites. Even though the lab-sort and fields-sort data have been shown to contain similar community patterns (Figure 1) this kind of field-sorted grouping raises some concerns about how systematic differences in the data may be influencing the models.

3.2.1 Test site evaluation

Walsh et al. (in press) found that median BOD (14 monthly samples)(hypothesised to be an indicator of the efficiency of pollutant delivery by stormwater drainage systems) and catchment imperviousness explained patterns in macroinvertebrate community composition between metropolitan and hinterland sites in Melbourne. These variables and the combined variable BODIMP are used to objectively evaluate urban model ratings of test sites. The relationship between O/E ratios and BOD and catchment imperviousness were non-linear (Figure 4). Three groups of sites account for a large proportion of the scatter in these relationships (Figure 4b).

• Group 1 (DA3, DB4, GA3, GA4, GA6) have low O/E ratios, low BOD, and high catchment imperviousness.
• Group 2 (EC2, KO2, MK3, MK4, MK5, MK6) have low O/E ratios, low catchment imperviousness, and high BOD.

This emphasizes the multifaceted nature of urban impact. By combining BOD and catchment imperviousness into a composite urban impact variable (BODIMP) the influence of these out lying groups is reduced and the relationship is more linear. The Pearson correlation coefficients for the O/E _0 BODIMP relationship for the lowest taxon and family models are _0 0.555 and _00.562 respectively.

A third group of sites is also evident.

• Group 3 (MU3-7) have medium O/E ratios, low BOD, and high catchment imperviousness.

These sites have higher O/E ratios than expected given the degree of urban development in their catchment. A potentially significant feature of these sites is that they all have physically stable rocky benthic habitat and relatively intact native riparian habitat and suggests the impact of urbanisation does not necessarily follow a uniform pattern.

The mean rank difference between BODIMP and urban model assessments was high (>6), suggesting some substantial differences between the assessments (Table 7) but the overall ordering of the sites was good given our understanding of the sites (Figure 4). The sites in good condition were clearly identified (CT2, DP6, JA4, MA2) by both models and the BODIMP variable. Agreement on the assessment of intermediate sites was good. Most discrepancy appears to occur among the most impacted sites. For instance, sites in both Dandenong (DA2, DA3, DA5) and Gardiners (GA3, GA4, GA6) Creeks had variable outputs from the lowest taxon and family models relative to BODIMP. There is some suggestion that benthic stream communities in highly urbanised catchments like Dandenong and Gardiners Creeks are in a constant state of recovery from a whole range of different impacts (including some not incorporated in the BODIMP variable). For example MK3 and MK4 have medium BOD and low catchment imperviousness and appear to be more severely impacted than the simple BODIMP variable would suggest. Contradictions such as this provide useful opportunities for identifying specific urban impacts beyond those expected from a particular level of development. However while the models and the independent assessment variable (BODIMP) showed some considerable variation in detail, the overall broad evaluations of the relative impact or condition of the test sites was similar
 

Table 7. Test site assessment by urban models and an objective urban impact variable (BODIMP)
Site	Urban _0 fam. O/E	Urban _0 fam.Band	Urban _0 fam.Rank	Urban _0 spp O/E	Urban - spp Band	Urban - spp Rank	BODIMP	Inverse BODIMP Rank	Rank differences
									fam. & spp.	BODIMP & fam.	BODIMP & spp.
BC1*	0.42	C	15	0.38	B	20	1.1192	1	5	14	19
CT2*	0.86	A	28	0.92	A	31	0.0294	27	3	1	4
DA2	0.17	D	3	0.19	C	8	0.8517	5	5	2	3
DA3	0.25	C	6	0.12	C	2	0.7427	7	4	1	5
DA5	0.42	C	16	0.12	C	3	0.7118	8	13	8	5
DB3*	0.47	C	20	0.36	B	19	0.3382	15	1	5	4
DB4*	0.17	D	4	0.12	C	1	0.2998	16	3	12	15
DP6*	1.12	A	31	0.84	A	30	0.0032	31	1	0	1
EC2*	0.22	D	5	0.21	C	9	0.1833	19	4	14	10
GA3	0.33	C	11	0.17	C	5	0.85	6	6	5	1
GA4	0.17	D	1	0.24	C	12	0.95	2	11	1	10
GA6*	0.17	D	2	0.14	C	4	0.8603	4	2	2	0
JA4*	0.9	A	29	0.81	A	29	0.0093	29	0	0	0
KO2*	0.26	C	7	0.18	C	6	0.1314	21	1	14	15
MA2*	0.9	A	30	0.69	B	28	0.0072	30	2	0	2
MK2	0.6	B	25	0.42	B	25	0.0161	28	0	3	3
MK3	0.3	C	10	0.21	C	10	0.061	24	0	14	14
MK4	0.26	C	8	0.18	C	7	0.0768	22	1	14	15
MK5*	0.34	C	12	0.21	C	11	0.2697	17	1	5	6
MK6	0.3	C	9	0.27	C	13	0.3469	14	4	5	1
MU3	0.38	C	13	0.42	B	24	0.5928	11	11	2	13
MU4*	0.55	B	23	0.27	C	14	0.5859	12	9	11	2
MU5	0.5	C	21	0.32	C	16	0.8626	3	5	18	13
MU6	0.55	B	22	0.38	B	21	0.7022	9	1	13	12
MU7	0.45	C	19	0.44	B	26	0.5611	13	7	6	13
OL5	0.42	C	17	0.4	B	23	0.1399	20	6	3	3
PL6	0.56	B	24	0.38	B	22	0.0364	26	2	2	4
PL7	0.43	C	18	0.34	B	18	0.0666	23	0	5	5
SL1*	0.39	C	14	0.3	C	15	0.623	10	1	4	5
TY1*	0.69	B	27	0.33	C	17	0.216	18	10	9	1
WA1*	0.66	B	26	0.62	B	27	0.0597	25	1	1	2
								Mean	3.87	6.26	6.65
(* Indicate field-sort site)								SD	3.69	5.43	5.60

 

Figure 4. Scatter plots

(original size image)

Figure 4. Scatter plots showing the relationship between urban model O/E ratios and BOD, Catchment Imperviousness and BODIMP. Enclosed groups of sites are discussed in text

 

Table 8. Summary of Wilcoxon test results comparing urban model outputs against an objective urban impact variable
	BODIMP (1) vs Family O/E (2)	BODIMP (1) vs Species O/E (2)	Family O/E (1) vs Species O/E (2)
Count of differences
1 >2	19	25	27
2>1	12	6	4
Z	-0.852	-3.567	-4.184
Probability	0.394	<0.005	<0.005

Results of Wilcoxon paired sample test comparisons between the model assessments and site condition evaluation using the (inverse) BODIMP variable are shown in Table 8. The results of the Wilcoxon test indicate there was a significant difference between the means of the BODIMP variable and species O/E ratio, and the mean O/E ratios of the family and species models. Species model O/E ratios were generally lower than the family O/E ratios as indicated by the count differences in Table 8.

The method pair dataset (Appendix 3) was used to compare the Urban models to the Victorian family model (Table 9). There was no significant difference between the O/E ratios of the family models (Wilcoxon test p = 0.099) and the relative O/E ratios were highly correlated (Pearson correlation coefficient = 0.943). The O/E ratios from the urban family model and the Victorian family model were both reasonably well correlated with BODIMP with Pearson correlation coefficients of _00.708 and _00.574 respectively. Marchant et al. (1997) found similar correlation between O/E ratios for sites in the Yarra catchment and a range of long-term water quality variables. Part of the rationale for using hinterland sites, with rural landuse, as reference in the urban models was the expectation that metropolitan sites would be severely downgraded and poorly resolved in the regional models. From the data available for this study it appears this is not the case.

The similar performance of the Urban and EPA Victorian models suggests broader regional AUSRIVAS models adequately assess urban impacts in metropolitan areas. The use of the broader regional models would avoid the considerable problem of selecting reference sites for urban streams.

A basic assumption of the reference site approach is that they should have minimal impact from human activity. This allows patterns in community structure to be well explained by relatively simple (physical) variables un-related to anthropogenic influences. There are two major difficulties with using genuine reference sites for urban streams. Firstly in most large cities whole physiographic regions can be developed leaving no local reference sites for certain terrain types. Secondly where genuine reference sites are available they set an unreasonably high target. Urban development by its very nature is going to change the physical characteristics of a catchment. This means that no matter what remedial management actions were undertaken, urban sites could never approach genuine reference site condition.
 

Table 9. Comparison of Urban models and the EPA Victorian family model
Site	Data Type	Urban _0 fam O/E	Urban - fam Band	Urban _0 spp O/E	Urban - spp Band	VIC _0 fam O/E	VIC - fam Band
DA1	Lab.	1.03	A	1.04	A	0.72	B
	Field	0.65	B	0.63	B	0.77	B
DA5	Lab.	0.41	C	-	-	0.25	C
	Field	0.25	C	-	-	0.20	C
DP5	Lab.	0.95	A	1.05	A	0.80	B
	Field	1.08	A	0.84	A	1.07	A
GA3	Lab.	0.33	C	0.17	C	0.22	C
	Field	0.33	C	0.18	C	0.30	C
GA4	Lab.	0.17	D	0.24	C	0.12	C
	Field	0.22	D	0.15	C	0.12	C
	Field	0.26	C	0.19	C	0.18	C
GA7	Lab.	0.34	C	-	-	0.26	C
	Field	0.22	D	-	-	0.35	C
JA3	Lab.	0.86	A	0.87	A	0.88	A
	Field	0.86	A	0.81	A	0.96	A
MK5	Lab.	0.30	C	0.27	C	0.15	C
	Field	0.17	D	0.12	C	0.15	C
OL2	Lab.	0.80	B	1.11	A	0.66	B
	Field	0.79	B	1.01	A	0.66	B
WO5	Lab.	0.89	A	1.10	A	0.90	A
	Field	0.97	A	1.05	A	1.03	A
Y5	Lab.	0.82	B	-	-	0.82	B
	Field	0.56	B	-	-	0.65	B
Y6	Lab.	0.7	B	-	-	0.61	B
	Field	0.7	B	-	-	0.74	B
Y7	Lab.	0.63	B	-	-	0.65	B
	Field	0.63	B	-	-	0.73	B

Given this understanding our initial response to this problem was to choose reference sites that were perceived as a reasonable target for the urban environment. In general these were hinterland sites of the Melbourne metropolitan area that were predominately rural landuse but inevitably also contained some satellite urban development. In a practical sense this approach worked; satisfactory models were constructed. However, the reference sites were clearly influenced by urbanisation which is an impact the models are intended to evaluate. Walsh et al. (in press) demonstrated that catchment imperviousness, a measure of urban density, was a dominant correlate explaining patterns of macroinvertebrate community composition among hinterland sites to the east of Melbourne. Thus the patterns of community composition among many of the sites used as reference in this study could be well explained by urbanisation, the same impact the models were being constructed to detect.

An alternative is to use a model with minimally impacted reference sites and to develop a relationship that accounts for urbanisation. Catchment imperviousness measures the degree of catchment landuse change and is generally unrelated to the type or quality of drainage infrastructure. Consequently it can be used as a general measure of urbanisation that can be used as a covariate against which other ecological indicators can be set (Walsh et al. .in press). O/E ratios from the EPA Victorian family model were negatively correlated to catchment imperviousness of both reference and test sites from this study (Figure 5). However, a number of sites had both low O/E ratios and low catchment imperviousness, suggesting impacts in addition to that explained by imperviousness alone.

Among hinterland sites, where stormwater drainage is largely through open earthen drains or small localized pipes, a decline in O/E is evident up to a maximum imperviousness of 0.12. Three eastern hinterland sites (Cockatoo 1, Woori Yallock 3 and Olinda 2) have low O/E scores, which may be a result of impacts not directly associated with catchment imperviousness.

O/E is low for almost all metropolitan sites (where stormwater is intensively drained through pipe networks). The western metropolitan site with the relatively high O/E score is Taylors Ck. Most urban areas of this catchment drain into a series of lakes which act as stormwater treatment ponds. The eastern metropolitan sites with highly impervious catchments and relatively high O/E scores are Ruffey Ck and sites on Mullum Mullum Ck. Ruffey Ck has a stormwater flood retarding basin and stormwater treatment wetland in its upper catchment. Mullum Mullum Ck is characterized by moderately well preserved riparian zones and good rocky instream habitat that is resistent to the increased flows experienced in urban streams.

The observed spread of O/E ratios around the negative relationship with catchment imperviousness creates an opportunity to band O/E ratios weighted by catchment imperviousness. This has been done in a subjective way in Figure 5 where a series of bands have been hand fitted. The slope of the bands allows for the handicap of catchment imperviousness. The spread of O/E ratios around the general catchment imperviousness relationship represents variation in stream community condition not explained by catchment imperviousness. Factors explaining this spread include water quality and the nature of the drainage infrastructure (Walsh in press, Walsh et al. in press) and water quality treatment ponds and wetlands (see Ruffey Creek in Figure 5). The band interval can be used as a measure of the degree of variation between the observed condition (O/E ratio) and that expected given a particular level of catchment imperviousness. The best existing condition in Melbourne streams is represented by band A, and bands B and C represent progressively poorer condition. The band A+ is suggested as a hypothetical target for catchments where stormwater best management practice has been adopted. However, there has been virtually no assessment of the efficacy of catchment scale stormwater best management practice on the health of receiving waters.

A systematic means for determining band limits needs to be formalized. One approach would be to develop a dataset with an even number of sites across the catchment imperviousness range and a spread of stream community conditions (O/E ratios) within any narrow range of catchment imperviousness. The number of sites would need to be high enough so that within a narrow range of catchment imperviousness O/E percentiles could be calculated. Regression lines could be developed for various percentiles as a method of establishing bands. The regression relationships would not need to linear. This approach remains a topic for future research.

The use of regional AUSRIVAS model O/E ratio versus catchment imperviousness relationships is a potential alternative to specific urban models. This relationship provides an informative use of the Victorian AUSRIVAS model for managers of catchments containing any urban land use. It provides targets for better stormwater practices based on best available information. However, no catchments have yet been studied that have been developed using contemporary best management practices. Scores within the A+ band may be targets for such catchments where Water Sensitive Urban Design principles have been applied.
 

Figure 5. Relationship between O/E ratios

Figure 5. Relationship between O/E ratios from the EPA Victorian family model and catchment imperviousness for a selection of reference and test sites. Preliminary banding is shown: A+ Better than expected/Potential target for stormwater best management practices A Best existing condition B Poor condition C Very poor condition

4. Conclusions

• 4.1 The Data
• 4.2 The Models

4.1 The Data

Laboratory and live-pick field sorted sample processing result in quantitative and systematic differences in the taxonomic composition of datasets. Lab-sorted samples typically contained more smaller and immobile taxa, whereas field-sorted samples contained more large and highly mobile taxa. However, separate analysis of community of composition showed lab-sorted and field-sorted samples contained similar community patterns and resulted in similar site groupings in MDS ordinations. Ordination analysis of the combined method-pair dataset indicates that differences between community groups were greater than differences between the method pairs. Data manipulations to remove rare taxa, method biased taxa or both, did little to increase the coupling between method pairs, and generally resulted in obscuring the differences between groups.

Although particular differences were evident in the community data analyses, the broad patterns in datasets were similar. It was considered the datasets could be combined and used for evaluating the potential for urban AUSRIVAS modeling. However the differences would suggest that data sets of this type should not be routinely combined and that the differences between the datasets may have caused some of the variation in O/E scores for samples from the same site collected by different methods.

4.2 The Models

The reference site groupings produced in the classification analyses for model construction reflect some clearly recognisable physiographic and geomorphologic characteristics of the study area. However, it is clear some regions are under-represented in the reference site dataset. For example Groups 2 and 3 in the lowest taxon model (Table 4). In spite of the small size of the reference site dataset misclassification errors for the models were satisfactory.

Both the family and lowest taxon urban models produced O/E scores that were negatively correlated to BOD and catchment imperviousness. While there were some ranking difference between the assessments the overall evaluation of sites were similar and generally reflected impact from identified disturbance factors in the study area (Walsh et al. in press). O/E scores from the urban family model and the Victorian family model (Marchant et al. 1997) were not significantly different in rank-order, and their ranges were similar. Thus the urban models developed here were not found to be any more sensitive to urban impacts than a larger-scale model.

Arising out of this study and the work of Walsh et al. (in press) is a potential alternative to the development of area specific urban models. This involves the development of a general relationship between community composition and catchment imperviousness. The assumption is that even with the best management practices used in urban development, once catchment imperviousness increases above a certain level some community impact will be inevitable. Catchment imperviousness is used as a general measure of urban development. Community composition could be assessed using various approaches but in this case O/E ratios from the EPA Victorian model were used. Test sites would be assessed using established regional AUSRIVAS models and the resulting O/E ratios evaluated against an O/E _0 catchment imperviousness relationship. In this sense catchment imperviousness is used as a "handicapping" variable for the O/E ratios. A major attraction of this approach is that it avoids the problem of selection and availability of reference sites to construct urban specific models for assessment of urban streams.

The development of urban AUSRIVAS models is clearly possible and the models can be useful in discriminating between the relative condition of sites within metropolitan areas. However the present study suggests that area specific models (eg. urban), as developed here, need to be built on a larger more representative group of reference sites. The number of reference sites (31) in the present study appears to be too small to adequately cover the geographic range of the Melbourne metropolitan area. This study was restricted to using existing data but with a more strategic approach to sampling and reference site selection improved models may be possible. However, the problem of patterns of community composition among hinterland sites being strongly correlated with the degree of catchment urbanisation is likely to be common to large, sprawling cities such as Melbourne. Such relationships compromise the appropriateness of using hinterland sites as reference sites for models designed to assess urban effects.

5. References

• Belbin, L. (1994) PATN, Technical Reference. Commonwealth Scientific and Industrial Research Organisation, Division of Wildlife and Ecology, Canberra, Australia.
   
  
• Belbin, L. and McDonald, C. (1993) Comparing three classification strategies for use in ecology. Journal of Vegetation Science, 4: 341-348.
   
  
• Boesch, D.F. (1977) Application of numerical classification in ecologial investigations of water pollution. United States Environmental Protection Agency. Washington D.C.
   
  
• Chessman, B.C. (1995) Rapid assessment of rivers using macroinvertebrates: A procedure based on habitat-specific sampling, family-level identification and a biotic index. Australian Journal of Ecology, 20, 122-129.
   
  
• Coysh, J.L., Nichols, S.J., Simpson, J.C., Norris, R.H., Barmuta, L.A., Chessman, B.C. and Blackman, P. (2000). AUStralian RIVer Assessment System (AUSRIVAS) National River Health Program Predictive Model Manual. Cooperative Research Centre for Freshwater Ecology, Building 15, University of Canberra, ACT, 2601 http://ausrivas.canberra.edu.au/ausrivas
   
  
• Faith, D.P., Minchin, P.R. and Belbin, L. (1987). Compositional dissimilarity as a robust measure of ecological distance. Vegetatio, 69: 57-68.
   
  
• Gauch, H.G. (1982) Multivariate Analysis in Community Ecology. Cambridge University Press, Cambridge.
   
  
• Marchant, R. (1989) A subsampler for samples of benthic invertebrates. Bulletin of the Australian Society for Limnology, 12, 49-52.
   
  
• Marchant, R., Hirst, A., Norris, R.H., Butcher, R., Metzeling, L. & Tiller, D. (1997) Classification and prediction of macroinvertebrate communities from running waters in Victoria, Australia. Journal of the North American Benthological Society, 16, 664-681.
   
  
• S.A.S Institute. (1995) SAS/STAT Guide for Personal Computers, Version 6.11, SAS Institute, Carey, North Carolina.
   
  
• Simpson, J., Norris, R., Barmuta, L. & Blackman, P. (1997) Australian River Assessment System: National River Health Program Predictive Model Manual. http://ausrivas.canberra.edu.au/ausrivas
   
  
• Smith, R.W., Bernstein, B.B. and Cimberg, R.L. (1988) Community-environmental relationships in the benthos: applications of multivariate statistical techniques. In: Marine Organisms as Indicators, eds D.F. Soule and G.S. Kleppel. Springer-verlag, New York.
   
  
• Smith, T., Vertessy, D. & Hardwick, R. (1997) StreamWatch Biological Monitoring Program. Annual Report 1996. Melbourne Water Corporation. Melbourne.
   
  
• Walsh, C.J. (1997). "A multivariate method for determining optimal subsample size in the analysis of macroinvertebrate samples." Marine and Freshwater Research 48: 241-248
   
  
• Walsh, C.J. (in press) Urban impacts on the ecology of receiving waters: a framework for assessment, conservation and restoration. Hydrobiologia.
   
  
• Walsh, C.J., Sharpe, A.K., Breen, P.F. & Sonneman, J.A. (in press) Effects of urbanization on streams of the Melbourne region, Victoria, Australia. II. Benthic macroinvertebrate communities. Freshwater Biology.
   
  
• Wright, J.F. (1995) Development and use of a system for predicting the macroinvertebrate fauna in flowing waters. Australian Journal of Ecology, 20, 181-198.
   
  
• Wright, J.F., Moss, D., Armitage, P.D. & Furse, M.T. (1984) A preliminary classification of running-water sites in Great Britain based on macro-invertebrate species and the prediction of community type using environmental data. Freshwater Biology, 14, 221-256.

Appendix 1 Reference and test sites used to create the AUSRIVAS Urban Family and Lowest taxon predictive models.

Appendix 2 The taxa used in the construction of the urban models.

Appendix 3. List of sites where paired sample processing data were available.

Figures & Tables

List of Figures

1. MDS ordinations of 25 sites used in the comparison of processing methods, based on lowest taxonomic level resolution, using all taxa. (a) Field-sorted samples, (b) Lab-sorted samples. Four groups of sites (eastern hinterland small streams, western hinterland small streams, Yarra River, and metropolitan sites) are grouped by kernel density contour curves, P=0.9.
2. MDS ordinations of 25 sites used in the comparison of processing methods based on lowest taxonomic level resolution. Lines join method-pairs: i.e. samples taken from the same site in the same season but processed by different methods (field-sort and lab-sort). Large symbols indicate field-sorted samples. Grouping conventions as in Figure 1. (a) Using all taxa, (b) biased taxa (with fidelity to one method of ³ 1.50, and Oligochaeta) removed, (c) rare taxa (occurring in <10% of samples) removed, (d) biased and rare taxa removed.
3. MDS ordinations of 25 sites used in the comparison of processing methods based on family-level resolution. Lines join method-pairs: i.e. samples taken from the same site in the same season but processed by different methods (field-sort and lab-sort). Large symbols indicate field-pick samples. Grouping conventions as in Figure 1. (a) Using all families, (b) biased families (with fidelity to one method of ³ 1.50, and Oligochaeta) removed, (c) rare families (occurring in <10% of samples) removed, (d) biased and rare families removed.
4. Scatter plots showing the relationship between urban model O/E ratios and BOD, Catchment Imperviousness and BODIMP. Enclosed groups of sites are discussed in text.
5. Relationship between O/E ratios from the EPA Victorian family model and catchment imperviousness for a selection of reference and test sites.

List of Tables

1. Description of predictor variables used in model construction
2. Fidelity analysis for twenty-five pairs of samples, each pair taken in the same season from the same site. One of each sample pair was field-sorted, and one was lab-sorted.
3. Reference site group membership
4. General physiographic description of model groups
5. Predictor variables used in particular models
6. Bands representing different levels of biological condition for the urban models
7. Test site assessment by urban models and an objective urban impact variable (BODIMP)
8. Summary of Wilcoxon test results comparing urban model outputs against an objective urban impact variable
9. Comparison of urban models and the EPA Victorian family model

Riversinfo Australia (au.riversinfo.org): Publication Information:

• Webserver publication date (version): Saturday, April 20th 2019.
• The publication of material on au.riversinfo.org was supported by the Commonwealth Government via the National Landcare Program and the National River Health Program.
• End users are advised to expect variations in formatting and layout depending on both the format of the original material, as supplied by the authors, and the end users particular software and hardware configuration(s).
• In the case of variation between this document and a hard copy original the original takes precedence.
• Except when otherwise stated, in writing, information provided by au.riversifno.org is provided "as is" without warranty of any kind to anyone. The entire risk as to use of the information rests with the user. Should the information prove defective, the user of the information assumes the cost of all necessary repair or correction. In no event, unless agreed to in writing, will anyone associated with au.riversinfo.org be liable for any damages arising out of the use or inability to use the information, even if au.riversinfo.org has been advised of the possibility of such damages.
• As of January 2004 au.riversinfo.org is archived under Precision Info.

1. Authors and the public are reminded that the purpose of au.riversinfo.org is to compile and reference Internet information about rivers in an easily accessible and comprehensive manner to assist all those seeking to preserve and publicize Australia's Rivers.
2. au.riversinfo.org has no interest in acquiring or assuming exclusive copyright over any authors material published on au.riversinfo.org. The copyright of material appearing on au.riversinfo.org rests with existing copyright holders who give approval for publication on au.riversinfo.org.
3. The copyright status of articles is generally stated within the article and those seeking to reproduce the article further should, if required, seek permission from the copyright holder prior to any further publication or circulation.
4. Where no explicit copyright statement has been made copyright will be assumed to be held by the author of the article.
5. au.riversinfo.org principally converts and marks up existing material from various formats. The au.riversinfo.org content delivery scripts, mark up code and presentation are copyright.
   • Consistent with the aims stated in point 1 and provided you do not attempt to substantially reproduce large sections of the au.riversinfo.org Internet site you may, provided you have the permission of the copyright holders, copy and distribute verbatim copies of the files appearing on au.riversinfo.org provided that you conspicuously and appropriately publish on each copy an appropriate notice acknowledging the source of the material.
   • Consistent with the aims stated in point 1 and provided you do not attempt to substantially reproduce large sections of the au.riversinfo.org Internet site you may, provided you have the permission of the copyright holders, copy and distribute modified versions of the files appearing on au.riversinfo.org provided that you conspicuously and appropriately publish on each copy an appropriate notice acknowledging the source of the material and stating the date and nature of any changes you may have made to the material.
6. If you systematically copy and or re-distribute information (or modified information) originating from au.riversinfo.org then you must:
   • Accompany the distribution with a written offer to transfer the information to any third party, wishing to use it in a manner consistent with the aims stated in point 1, a complete machine-readable and interchangeable copy of the corresponding material.
   • Any charges to the third party arising as a result of the transfer must not exceed the cost of your physically performing the transfer.
   • The offer must be valid for at least three years from the date of your last systematic distribution of the material.
   • Acceptable machine readable and interchangeable file formats include .html and or .txt and or .rtf formats. PDF format is NOT considered an acceptable machine interchangeable file format.
   • You must cause any work that you distribute or publish, that in whole or in part contains information derived from au.riversinfo.org, to be available under the principles of these same distribution terms
   • Third parties receiving material deriving from au.riversinfo.org should be advised that they will be bound by the principles of these distribution terms

Current at December 2002

 au.riversinfo.org has been developed to facilitate public access to scientific and technical information on Australia's Rivers. Much of this material was paid for using public funds and is therefore owned by the Australian people. Environment Australia have been instrumental in making available all the publications from:

• The National River Health Program
• The Ausrivas Assessment Scheme

 The au.riversinfo.org Internet site is in-part supported by the Natural Heritage Trust. The Natural Heritage Trust focuses on five key environmental themes - land, vegetation, rivers, coasts and marine, and biodiversity. The programs of the Natural Heritage Trust play a major role in developing sustainable agriculture and natural resource management, as well as protecting our unique biodiversity through improved management and delivery of resources.

 The au.riversinfo.org Internet site is in-part supported by Sydney Airport who have co-funded a Natural Heritage Trust supported project intended to develop Internet based environmental information integration and management systems. This will assist in developing an overall perspective on the demands and uses for environmental resources