Frontispiece: Kangeroo Swamp, site number 48. Air oblique photograph by Neville Rosengren, March 1981. NUMERICAL ANALYSES OF MACROPHYTE VEGETATION

IN VICTORIAN

IN RELATION TO E~VIRONMENTAL FACTORS

by

Michele Mary Barson B.A.(Hons),M.Sc.

1984

A thesis submitted for the degree of

Doctor of Philosophy

at the

University of CONTENTS

Page

Declaration i

Acknowledgements ii

Abstract iv

List of Figures vii List of Tables x

Chapter 1 Introduction

1. The value of wetlands 1

2. Types of classifications 5

3. Aims of the study 8

4. Same definitions 9

5. Description of the study area 9

Chapter 2 Classification of wetlands vegetation

l. Data collection 13

Selection of wetland sites 13 Sampling considerations 14

Sample size 16

Choice of attributes 17

2. Classification of floristic data 19

Choice of unit to be classified 19 Scale of measuranent 20

Choice of strategy 20

3. Validation of groups produced by numerical

classifications 21

4. Methods of analysis

5. Results 23

Stopping rules 24

Floristic groups 25

6. Discussion 32 Criteria for the assessment of a

classification 32

Adequacy of a class representation 33

Evidence for relevance of the model

and adequacy of fit 34

7. Conclusions 38

Chapter 3 The effects of data reduction on classification

1. Introduction 40

2. Application of data reduction techniques 43 3. Results 47 A. Data set after reduction by EIDENT 47

B. Results of classification of aquatic

and semi-aquatic species 53

4. Discussion 56

5. Conc:lusions 60

c.'1apter 4 ordination of wetland sites

1. Introduction 61

2. Choice of strategy 64

Data transfonnations and sbnilarity

measures 65

3. Results 68

Principal Coordinates Analyses 68

Detrended Correspondence Analysis 71

4. Discussion 72

5. Conclusions 77

6. Summary of results of floristic analyses 78 Chapter 5 Physical and chemical characteristics of

aquatic macrophyte habitats in

1. Introduction 79

2. Methods 81

A. Sample collection and analyses 81

B. Numerical analyses of data 83

3. Results

A. Laboratory analyses 84

B. Numerical analyses 87

4. Discussion 93

5. Conclusions 98

Chapter 6 Vegetation-environment relationships

1. Introduction 100

2. Choice of strategy 101

A. canonical Correlation Analysis 101

B. Analysis of variance 103

C. Discrlininant analysis 104 3. Results 108

A. canonical Correlation Analysis 108

B. Analysis of variance 110

c. Discrbninant Analyses III

4. Discussion 113 A. canonical Correlation Analysis 113

B. Analysis of Variance 113

c. Discrbninant Analyses 114

5. conclusions 117

Chapter 7 Conclusions 119

1. Summary of results 119 A. Floristic analyses 119 B. water chemistry analyses 120 c. vegetation-environment relationships 121

2. Evaluation of method.s am results of the study 122 A. Variability of wetlands vegetation 123

B. Assessment of ~~e classification of wetlands vegetation 124 c. Vegetation-environment relationships 129 D. Relationships with other Australian wetlands 132 3. Recommendations for conservation 134

Bibliography 137

Appendix I 155

Species recorded fram the Victorian ~~tlands sampled DECLARATION

I hereby dec~e that this thesis is my own work~ except where specifically stated to the contrary~ and that it is not substantially the same as any other thesis which has already been submitted to any other university.

MICHELE BARSON ii

ACKNOWLEJX;EMEt-.."TS

It is a pleasure to acknowledge the assistance of the following people: Dr. P. Y. Ladiges, my supervisor, for her patience and encouragement and Dr. M.B. Dale of the CSIRO Division of Computing Research for the guidance provided through the maze of nunerical methods. Dr. D. Ratcliff, CSIRO Division of Mathematics and Statistics, who carried out the analyses for Chapter Six.

Andrew Corrick of the Victorian Fisheries and Wildlife Division and Lex Thompson of the Department of Forestry, University of

Melbourne provided information about many of the wetlands wnich were sampled.

Ian Clarke, Mark Ellaway , Ros Gleadow, Steve Gloury, Laur ie Koster, Sigrid Kraemers, Phil Ladd, Vuong Nuygen, Neville Rosengren and Dick Williams braved snakes, uncertain substrates and turbid waters to provide assistance and company in the field.

Equipment and advice regarding L,e analysis of water samples was provided by Dr. M. McCormick of the E.P.A. Laboratory, Latrobe University and Dr. J.D. Smith of the Department of Marine Chemistry, University of Melbourne. I am grateful to Lyn McKinley for assistance in the preparation of samples.

Irene Folie, Colin Summerbell and John Myers of the Botany School are thanked for their patience and advice. Thanks are also due to the Chairmen of the Departments which supported this work, Dr. D.M. Calder and professor T.C. Chambers (Botany School) ard Dr. J.R.V. Prescott iii and Dr. T.M. perry (Geography Department) •

Special thanks are due to Lois Davey for the inter library loan service, Rob Bartlett who drew the diagrams and to Shirley Fricke,

Ruth Terrell-Phillips, Jenny Gilbert and Robyn Cotter for typing.

Jenny Ziviani and David wadley provided friendship and hospitality in Brisbane. Dr. Eric Bird, Professor Eddy van der Maarel and Mr. LeO Devin are thanked for their constant encouragement.

This work was supported by a Commonwealth pOstgraduate Research

Award. iv

ABSTRACT

This study was undertaken to investigate the variability of vegetated victorian wetlands and to establish the relationships between this variation and major environmental factors.

Criteria for the selection of the 55 wetlands sampled included the presence of aquatic macrophyte or helophyte vegetation, the presence of at least an intermittent water body and, comparative lack of disturbance of the si te. Si tes were also chosen to reflect the considerable lithological and climatic variation found across lowland victoria.

At each site, species presence/absence data, water depth and water transparency were recorded within 1m square quadrats positioned at Sm intervals along transects located to ~est sample the vegetation. At each transect, water samples were collected for the analysis of major ions, and substrate samples were taken for the estimation of texture and measurenent of pH and percentage salts. The maximtm depth of t..'1e basin when flooded and its water regime were estimated and the geology and rainfall of the catchment were recorded.

An information statistic strategy was used to classify the large, relatively sparse floristic set of data. The classification recognised five distinctive, relatively homogenous and ecologically interpretable groups of wetlands, which were characterised as having saline, very saline, turbid, acidic or calcareous waters, and a further three freshwater groups which were closely related to one another. v

The application of data reduction techniques suggested that the infonnation statistic model was unable to adequately define some of the freshwater groups of sites primarily because of the highly heterogenous nature of the data set. TWO more data sets were produced by deleting species with low "eident" values (Dale and Williams 1978) and by deleting species regarded as terrestrial. However, classification of these reduced data sets did not provide markedly better results.

The relationships between the groups (and their members) generated by classification was examined through indirect ordination of the floristic data. Inspection of the results indicated that six of the eight groups identified by classification of the floristic data could be recognised. However, two groups of sal ine si tes could not be separated, largely because they were both species-poor. Six sites were identified as the probable cause of overlap of some of the freshwater groups.

Laboratory determination of the major ionic constituents of the waters of the 55 wetlands indicated that the orders of anion dominance were Cl>HC03+C03>S04 (freshwater sites) or Cl>S04>HC03+C03 (saline and coastal freshwater si tes) and those for cations were Na>Mg>Ca>K or

Na>Ca>Mg>K (freshwater) and Na>Mg>Ca)K (saline sites). The dominance of chloride and sodiun ions in the waters sampled suggested that salinity was a major factor affecting the distribution of aquatic macrophytes in Victoria.

Numerical classification of the wetlands on the basis of their vi

water chemistry was undertaken to provide a comparison with the eight group floristic classification. However, two of the intuitively recognised groups, the turbid and calcareous waters, were not identified by classification of the water chemistry data, and me:nbership of the two independently generated sets of groups was not identical.

The nature of the hypothesized joint pattern between the floristic and the water chemistry data was further investigated by canonical correlation analysis, analysis of variance and discrlininant analyses. These analyses ccnfirmed that overall, the variance observed in the vegetation of the wetlands sampled was significantly influenced by water chemistry. However, the level of vegetational variation identified as a result of classification of the floristic data (eight groups) did not correspond well with the measured differences in the water chemistry variables. Vegetation differences which could be attributed to water chemistry differences were those between the saline groups, the turbid water group, the acidic water group, the freshwater complex of three groups, and possibly the calcareous group.

The salinity variable largely separated the saline groups from t.."1e rest, whilst pH separated the acidic water and calcareous water groups from each other and the freshwater complex. The turbid water group was separated by a combination of pH and salinity variables.

An evaluation of the methods used is presented and other factors which may be linportant in explaining the heterogeneity of the saline and freshwater floristic complexes are discussed. The wetland vegetation sampled in Victoria is compared with t.."1e plant communities of Tasmanian wetlands. vii

LIST OF FIGJRES

No. Shortened titles Facing p3.ge

1.1 Physiography of Victoria 9

1.2 Average annual rainfall map of Victoria 10

2.1 Distribution of wetland study sites 14

2.2 Dendrogram showing major attribute

contributions at each dichotomy 26

2.3 Dendrogram showing ecological affinities of the

floristic groups 30

2.4 The effect of stand richness on Infonnation Level 33

2.5 OCOA of simulated data set 36

3.1 Dendrogram showing major attribute contributions at

each dichotomy in the "eide.'1t" classification 48

3.2 Dendrogram showing ecological affinities of t."1e

floristic groups proouced by classification of the

"eident" data set 49

3.3 Dendrogram showing ecological affinities of the

floristic groups proouced by classification of the

aquatic-semi-aquatic species 54

3.4 Dendrogram showing major attribute contributions at

each dichotomy in the aquatic-semi-aquatic species

classification 55

4.la Scatterplot generated by OCOA ordination of the

entire data set 70 viii

4.lb OVerlay showing group boundaries after reallocation

of misclassified sites 70

4.2a Scatterplot generated by ECOA ordination of the

lIe ident" data set 71 4.2b OVerlay showing group boundaries after reallocation of misclassified sites 71 4.3a Scatterplot generated by ECOA ordination of the

aquatic-semi-aquatic data set 72 4.3b OVerlay showing group boundaries after reallocation of misclassified sites 72 4.4a Scatterplot produced by Detreooed Correspoooence

Analysis of the entire data set 13 4.4b OVerlay showing areas of ordination space occupied by

members of the floristic groups after reallocation of

misclassified sites 73

5.1 Deoorogram produced by classification of the water

chemistry data 87 5.2 Deoorogram showing contribution of water chemistry

attributes at each dichotany 88 S.3a ECOA ordination of water chemistry data 92

S.3b OVerlay showing ordination space shared by five water

Chemistry groups. 92 ix

LIST OF TABLES

No. Shortened titles facing page

2.1 Location of study sites 14 2.2 Membership of groups generated by NIASM floristic classification 25 2.3 Diagnostic species of floristic groups 31

3.1 "Eident" values for wetlam species 44 3.2 Life-foDD classification systems for wetland species 46 3.3 List of species and their life-foDD categories 47 3.4 Me:nbership of groups generated by NIASM classification of species with high "eident" values 50

3.5 Diagnostic species of 1I eident" groups 52

3.6 Me:nbership of groups generated by NIASM classification of aquatic am semi-aquatic species 53

3.7 Diagnostic species of aquatic-semi-aquatic groups 56 3.8 Membership of floristic groups after reallocation 59 3.9 Sites allocated to different groups by classification of entire, "eident" am aquatic data sets 60

4.1 Species contributions to PCOA vectors 68

5.1 M:thods used in the analysis of water samples 82 5.2 Water chemistry of 55 Victorian wetlam sites 84 5.3 Membership of groups generated by classification of the water chemi stry data 89 x

5.4 Ranges of values for water chemistry parameters 90

6.1 Canonical correlations for the three floristic data

sets wi th water che:nistry data 109

6.2 Canonical correlation analysis results 110

6.3 Back correlation of attributes wi th the first

canonical vectors III

6.4 Back correlation of attributes with the secom

canonical vectors 112

6.5 Results of the analysis of variance for the water che:nistry variables (S groups) 113

6.6 S1.m11ary of the results of tr.e direct am the stepwise

discriminant analyses 114

6.7 Jacknifed classification of wetlands 115

6.S Discriminant function coefficients 116

6.9 Results of the nine variable stepwise discriminant

analysis. 117 1.

CHAPTER ONE

Introouction

1. The value of wetlands

Wetlands are a valuable national resource. They provide habitats for our distinctive, diverse and often endemic aquatic biota (Williams

1980), and they have considerable significance as areas for scientific research ( 1980, Kusler 1979). Wet:ands also support fauna, such as the insect eating Ibis, which are of direct benefit to man (Brown

1973). Wetlands, like other natural ecosystems, contain a rich store of genetic material for future study and possible use (Frankel and

Bennett 1970), and in an undisturbed state can serve as baselines for envirormental monitoring (Jenkins and Bedford 1973). As natural basins, they have a role in flood retardation and as filters of surface runoff pollutants which would otherwise enter groundwaters

(Department of Conservation and Envirorment 1980), and they have a role in global eycling of carbon, nitrogen and sulphur (Odun 1979).

Wetlands also have a nunber of econanic values, sane yet to be realised. The potential for aquaculture in inland waters has been little explored and presently non-ccnmercial species may be of econanic importance in the future. Wetlands are also often among t.'e

last open spaces in rural and urban areas suitable for teaching

purposes, whilst their use for recreational pursuits has been '~ll docunented• 2.

Until recently, lack of appreciation of the importance of wetlands has resulted in alteration and often large scale destruction of these habitats. In Victoria, the extent of wetlands has been significantly reduced since settlement. In 1967 it was estimated that approximately

500,000 acres (202350 ha) of swampland had been drained (Fisheries and

Wildlife Department in Brown 1973). Much of this drainage has been carried out for agricultural purposes; large swamps which have been

"reclaimed" incltrle the Carrun Swamp south east of ~lbourne, the t-be

Swamp in and the Condah, Eumeralla, t-byne, ~rri, ~punga

Buckleys and Heifer Swamps in south western victoria, and the Orbost

Fast, Bete Belong South and Newneralla Swamps in Fast Gippsland (East

1935). The conversion of the extensive Kooweerup and Cardinia Swamps on the northern shores of Westernport to farmland following the construction of an extensive system of drains has been well docunented

(Key 1967). Many smaller wetlands have been and are still being drained privately; during the course of this sttrly several wetlands identified fdr sampling were drained and converted to crop land.

The construction of impoundments has also destroyed a number of wetland habitats in Victoria; wetlands or which have been converted to water storages include waranga Dam, Lake Fyans, Batyo

Catyo and Lake Whitton (Anon. 1983). While such water storages increase ~~e areas of open water, their usually steep banks and large fluctuations in water level frequently make them unsuitable as habitats for many wetland species.

Wetlands near urban areas are frequently utilised as dunps for unwanted soil and danestic refuse; a number of Muehlenbeckia daninated wetlands, once typical of the basalt plains west of ~lbourne have 3. been obliterated in this manner. wetlands have also been drained and filled to provide land for housing and industrial sites.

wetland ecosystems may also be significantly altered by land use practices, by the introduction of exotic species of flora and fauna, by interruption of their hydrological regime and through their utilisation for wastewater disposal. Changes in catcrment landuse can result in an increase in ~utrient inputs to wetlands through surface runoff, erosion and the discharge of animal wastes (Cullen, Rosich and

Bek 1978). Increased. levels of phosphorus and nitrogen may also result fran point source 1ischarges of sewage, industrial wastewater and stODD drainage. Increases in these nutrients stimulate algal and macrophyte growth which may result in detrunental changes to aquatic ecosystems, including the decline or elimination of certain species and a consequent loss of diversity (Lake 1980). Excessive growth of several species of nuisance aquatic macrophytes in urban lakes at Albert Park and near has been correlated with the increase in levels of nutrients in urban runoff (Cullen and Rosich

1980) •

The introduction of exotic species of both flora and fauna may adversely affect wetland ecosystems. Numerous species of freshwater fish have established viable populations in Australian inland waters, their effect on the bdigenous aquatic biota has gone largely unrecorded although it is likely to include a reduction in native fish and invertebrate populations, habitat alteration and the introduction of diseases (Lake 1980). A large nunber of aquatic plants have been introduced, including eleven species which cause serious problems

(Mitchell 1977). Alien species often replace native species, and may 4. disturb the structure and functioning of an ecosystem, but seldan provide suitable habitats for native fauna (Mitchell 1980). victorian wetlands appear to be largely free of serious aquatic weed infestations, although a number of native and introduced species have been recorded as nuisance weeds in irrigation channels, and rice paddies (Graham 1973) •

Other impacts on wetland systems inclooe interruption of the h}ldrologic regime. r-t:xlification of water regimes following the implementation of drainage schemes has affected wetland habitats in the Lake COoper area (Corrick and COwling 1978) and in the district (Brown 1973), and river impoundments on the Murray and Goulburn have resulted in the drying of mallY billabongs (Shiel 1980). Sane wetlands in the Kerang area have also been altered by their use as disposal sites for saline irrigation waters.

At present a number of stooies are examining the feasibility of utilising Australian wetlands as filters to remove suspended materials, nutrients, bacteria and viruses fran wastewaters. However, changes in the biota of same experimentally treated wetlands overseas indicate the possibility of undesirable disturbances, and caution must be exercised in the widespread use of natural wetlands for the treatment of waste if these valuable resources are not to be endangered (Sloey, Spangler and Fetter 1978).

The report of the Committee of Inquiry into the National Estate (1974) noted that "further inroads into rare ecosystems still in their natural condition, particularly wetlands, should not take place until their conservation potential has been properly assessed". , 5. as a signatory to the Convention on Wetlands of International Dnportance especially as Waterfowl Habitat (Australia, Department of Foreign Affairs 1976), also has an international obligation to pramote the conservation of v.;etlands. While there are several general descriptions of Australian v.;etland vegetation (Beadle 1981, Briggs 1981), and a number of detailed local studies have been carried out

(Eardley 1943, Dodson and Wilson 1975, Congdon and M:Camb 1976, Kershaw 1978, Ladd 1979, paijrnans 1581), is the only state for which an extensive survey of v.;etland plant communities has been

UIrlertaken (Kirkpatrick and Harwood 1983). An extensive survey of the larger Victorian v.;etlands as bird habitats is currently being undertaken by the Victorian Fisheries and Wildlife Department (A. Corrick pers. camm.) and same surveys of the fauna of Victorian lakes have been published. Hov.;ever, the vegetation of Victorian v.;etlands

(excepting salt marshes) has been little studied.

2. Types of v.;etland classifications

Realisation of the value of v.;etlands has resulted in widespread interest in the documentation of those remaining, and has stimulated the development of a number of inventory and classification schemes

which can be used to describe v.;etlarrls, document thei r variabili ty arrl provide data for planning and conservation purposes and as a baseline for moni toring •

Various classification systems have been developed for v.;etlands, the criteria proposed for classification being dependent on the nature of the interests of the user, the scale of the study and the variability present in the area surveyed. Many early classifications 6. were concernerl with wetlands as wildlife (especially wildfowl) habitats, ego Martin, Hotchkiss, Uhler and Bourne (1953), Shaw and Frerline (1956), Goodrick (1970), Golet and Larsen (1974), Corrick and Cowling (1975), Millar (1976) and Coward in , carter, Golet and LaRoe (1979), or for land classification purposes (welch 1978, Boissoneau and Pala 1979, Zol tai 1979). Major cdteria utilised in these classifications incltrle physiography, lithology of the substrate, depth and duration of flooding, salinity of the water, the lifefonn of the daninant vegetation and scmetimes percentage cover of the water

surface by vegetation - attributes considererl to be important in the description of wildlife habitats.

other classifications have been developed by 1imnologists, prerlaninantly for application to lakes which often include deep water habitats. Thienemann (1925) utilised the trophic state of the lake, Hutchinson (1957) the origin of the lake basin, Brundin (1958) the bottan fauna, Hansen (1961) the nature of the lake sediment, Williams (1964) salinity, Maemets and Raitvir (1977) trophic status and thennal stratification, Topping and Scudder (1977) physico-chemical

characteristics, (Winter 1977) hydrology, Legendre, Long and Beauvais

(1980), physical and geomorphological characteristics, Heywood, Dartnall and Pr iddle (1980) nutrient status, stmller transparency, seasonality of vegetation and vegetation type, while Rai and Hill

(1980) used microbiological and physico-chemical criteria to classify lakes.

Additionally, there is a long history of classification of aquatic habitats on the basis of their vegetation. previous approaches to the classification of wetlands on this basis incltrle a large number of 7.

studies in the Scandinavian phytosociological tradition which have produced regional classifications for lakes largely based on macrophyte canposition, although sane have also included envirormental characteristics. Sane of the more important classifications of

Swedish lakes for particular regions include Samuelsson (1925),

Almquist (1929) and Lurrlh (1951). Maristo (1941) developed a detailed system of classification for Finnish lakes while regional classifications were dev~loped by Cedercreutz (1934, 1937, 1947 in

Jensen 1979) and Jaatinen (1951 in Jensen 1979). A classification of

Danish lakes was produced by MathieSen (1969) utilising Samuelsson's

(1925) typology.

There is also an extensive literature devoted to the syntaxonamy and ecology of aquatic carmunities, inclooing den Hartog and Segal

(1964), Segal (1966), MalIne (1975), zutshi (1975), Pietsch (1977,

1978), Makirinta (1978a), Verhoeven (1980), O'Connell (1981), Snoeijs and van der Ster (1981}, Wiegleb (1981a) and Wheeler (1981a, 1981b).

However, the major interest of these workers has been to classify vegetation itself, rather than produce a scheme to classify wetlarrls.

Sheldon (1973) discussed the application of numerical methods to the classification of inland waters. M:>re recently, a variety of numerical methods which have been widely used in the analysis of terrestrial vegetation have been applied to aquatic vegetation, both to stooy carmunity relationships within sites (eg. Howard-Williams and walker 1974; Adam, Birks, Huntley and Prentice 1975, Makirinta 1978a) and to canpare carmunities between sites. Classification and ordination techniques have particularly been applied to northern

hemisphere peatlands (Vitt and Slack 1975, Pakarinen 1976 1 Pakarinen 8.

and Ruvhijarvi 1978, Daniels 1978) to proo.uce typolClg'ies and to examine major envirormental gradients wi thin peatlands. Similar methoo.s have been applied to herbaceous WE!tland vegetation in

Saskatchewan, Canada (Walker and Coupland 19.70), to lake vegetation in southern Sweden (Jensen 1979), to the WE!tlands of the Camargue region,

France (Britton and POdlejski 1981) and to standing waters in north

WE!st Africa (.M::>rgan arrl Boy 1982) •

The distinctive contribution of nunerical methoo.s to the classification of vegetation is to allow the data themselves to indicate the most efficient criteria for classification (Greig-Smith

1980) . This approach results in a classification which might be te:oned intrinsic, whereas many previous approaches to the classification of WE!tlarrls have utilised extrinsic criteria (eg. water depth, water pe:onanency) to construct a framework wi thin which sites can be placed. Further, nunerical methods facilitate the ordering and manipulation of large heterClg'enous or complex data sets.

3. Aims of the study

This study docunents the variability of vegetated Victorian

WE!tlands and attempts to establish the connections between this variation arrl envirormental influences. A primary aim of the study is to produce a classification of some of the remaining Victorian

WE!tlands which could provide a means of assessing wetland resources for conservation purposes, and order the large data set to facilitate examination of the relationships between species distributions and envirormental factors. This work also provides an opportunity to assess the perfonnance of a variety of numerical methods applied to a MURRAY ~ I- " .. -' ••• 't ,. .. .. • " .... It .. : ...... VICTORIA V-4 I" • - • • • 'l.. (0::. Lf:: y PHYSIOGRAPHIC DIVISIONS \...... " ...... •I'~/-:~~ \ , , I ;. " '" ~ " ...... , ~\ ! !.. " .. ..' ,. .. ) ~ ! · ...... - 'v1-t, lIL__ i'- 35'S I .. a ..,...... "" , " .' ' .', • '" • " ~.-39'S , ", .., '\~ ~. ... "" ...... " ... ,....,.,... , Location map '0 ...... ,," .."" ...... ) '\ I. , '.w' • ....,.1 I \. \ ' ••••• Up! ", " (V', I " I ...... " ...... : :9~Y'." (\. '"\., (,/...J."', j \ .. ""." J .I' •• \. _ .. ') V' J' ,,' •• ,f ,',I)•• r ''i) ,,'I{ ,,\ ~~ N I •• • ...... ;("'..:\l .. "·i~r ....·:~ 84./1 ~ ,..,,--, ,_, """-\ I.· ./ <.:. \ 'i' \.::: (" II (yf ~rv : NORTHERN PLAINS """ \ 07..-;' , ~ '\ ,i -~ , \:J t.} 'Q' I) {/' 'p ~' ' \ >-..~~ ~ " , I ,__ ;_-;_;~,:,.j ./;)... "?':}Aj~S~ ," ,"<:{ ~':::~"~'''' ,) I :', .• ;.~ • ~i) WIMMERA J / " } ~/ t'J).-:/=:z. ~",~" ~'\ ". " , '-.,,!~:t \.,_J"- ,.) ///, / ), ~Y'//,/ /" _<.~ S~',', '~"'-." I ,r" / / //'7' / /, ,/ \ /\-\\.-:>.t-\0~ V '" ,'" \ - _ .,. t/> .. / ,/ // /CENTRAL/'// 11A\:'\~ ',~ ;./,(//. 'S>\ '~" 'j("'~::< <~~.~ I ~ -~ 7/ -'/- ,/ / ~ I , I' "V.,,-, 'X", ~ ; / r //-"'}' )j"RAM~NS // ///,//~ IQ ,~,'" ''-,' ~'"-¥ 3~,~ p.t-\O"'~' '" ,,"-, I " I / / / ~(pa' / • q, .'" ~ ~ HIGH\. , ~ ," -. ~ i ""',",,"''-.,, I tWEST~~V/ / //(!/t--.::.( -/..'HIGLJ~'/ qLANDS /~('~,~~~EAST'E~~~/g,~1.'~ R~~l 1/-\') ~'''''' , "'-. '---,.' '",'-."~~, )- ~ I \V:><~'d- (/'l~---V;A,>~,>~" ,,,",O~(.f'Z~~/L~-?;-.!-~-:::)-:,;:;,)..y.,/' ~! '" .'" v1"'\,/,:>.!... 1', '~\' ",", ,,~VC"'\ t-\V ; c '7 VOLCANIC PLAIN~--"----' \ ''', ' r,,\'-.,,' /' ",,51..:" '" /vV"'''TERN--OISTRICr-PLAIN ) ~~~I ,,,,,-,~/ '\ G' ~ <5'?-\,,"'-f"'~ • I .! ~v I C..--:; Ep..S / '1(,v--..\,... .. rr::-:-:=--" _~\j'\\"l~. l3CUTHC... "'--, / ~\ ~ -I'~,,=- -'" _"::,0 ~ Q ~IP..E'SlAND I / ~eLA.INS ,:;,,;;3.,,__ ''JEST --;J:I~41\\ 'JHI~HLrN0.o/' '------1'0" ~"f3 G;PPSLANO ~\ ?5}~ 0 100 200km. ~\h PLAINS YvV'~"" ; . WILSONS . PROMONTORY

Fig, 1.1 Physiography of Victoria (Source: After Hills 1975) 9. large but particularly sparse data set containing a few relatively common species and a large number of uncommon species.

4. Same definitions

wetlands have been defined as "lands transidonal between

terrestrial and aquatic systers where the water table is usually at or near the surface or the land is covered by shallow water" (Cov.rardin et ale 1979). In these environments at least periodic inundation with water is the major determinant of the nature of the plant and animal communities and of soil development. Wetland habitats incluae tidal flats, wet vegetated basins variously described as marshes, bogs and swamps, river and lake margins, seasonally flooded agricultural land, basins and flats and inland saline flats and lakes.

For the purposes of this study, tidal wetlands, same of which have

been classified for Victoria (Bridgewater 1975) were excluded, as were wetlands associated wi th flowing waters, seasonally flooded agricultural land and flats, wetlands associated with impoundments and alpine wetlands (largely daminated by Sphagnum). Deepwater habitats

(> 2m) were not sampled, although aquatic macrophytes (particularly

Chara and Nitella spp.) were occasionally observed at depths exceeding

2 m.

5. Description of the study area

Victoria is situated in south eastern Australia be~Neen latitudes

35 0 8 and 390 S (Fig. 1.1). The major topographic determinant of victoria's climate is the southern extrenity of Australia's Great VICTORIA AVERAGE ANNUAL RAINFALL

300 N ~ I T I

Ga~bo I. 1000

o 100 200km. ----======> Isohyets are 10 millimetres

Fig. 1.2 Average annual rainfall in Victoria (Source: Anon. 1983) 10.

Dividing Range (the Eastern and western Highlands in Fig. 1.1) which runs east-west across the State, rising to nearly 2000 m in the eastern half. This range acts as a barrier to moisture laden south east and south west winds and together wi th its prox imi ty to the coast, results in southern Victoria receiving more rain than the north

(Anon 1983). The Southern CCean has a moderating effect on VictoriaIS winter climate. There is no permanent snow and snow is rare below

600 m.

Annual average rainfall varies fram 250 rnm in the north west of

Victoria to over 1400 rnm in the ranges north east of ~lbourne (Fig.

1.2) • However, both rainfall and runoff are highly variable and unreliable. With the exception of , Victoria receives more rain in winter than in summer.

Evaporation exceeds the annual average rainfall in inland areas, especially in the north and nort.1t west, by about 1000 rrtn. In the highland areas and in the W:stern District the discrepancy is much less marked, and in the Central District and lowlands of East Gippsland annual evaporation only exceeds annual average rainfall by 200-400 rnm. In winter, rainfall is greater than evaporation in most of Victoria except the north and north west.

Fig. 1.1 shows the major physiographic divisions of Victoria. The Murray Basin Plains include the areas known as the Mallee, wirnmera and Northern Plains. The surficial sediments of the Mallee r83ion are predaninantly fine sands which form dunes and sand plains. Low flow volume and high rates of evaporation and infiltration result in most streams in this area failing to reach the ; instead they 11.

terminate in shallow, brackish or saline lakes. The Wlinmera consists of low alluvial fans and alluvial plains and abandoned river channels.

The Northern Plains are the coalescing alluvial plains of the Murray,

OVens, Broken, Goulburn, carnpaspe and Leddon Rivers; the Murray River in particular has extensive areas of associated riparian wetlands and features such as billabongs (ox bow lakes) are common.

Tile Central Highlands incltrles the mountainous Eastern Highlands where elevations commonly exceed 1200 m, and the lower (generally <600 m) Western Highlands. Lakes are uncammon in these highland regions which have not been subject to glaciation. However, Sphagnum dominated bogs are common at higher elevations.

The volcanic plains of the Western District is one of the largest areas of volcanic plain in the I.

Pliocence to Holocene basalt flows and sane basaltic ash, and are almost horizontal. Lakes associated with volcanic features are very ccmnon in this region. North of Colac is a s:nall area of mostly sal ine lakes associated wi th the endorheic drainage system of the basalt plains (Williams 1967). The coastal plains of the Western

District are the flat or undulating, uplifted surface of Tertiary sedimentary rocks, incltrling limestones ccmnonly associated with

Quaternary dune limestones and sands. South west of Portland, coastal dune lakes are carmon, and further inland wetlands are ccrnmon in dune swales and in karst topography.

The Gippsland plains are the upper surface of a ':'ertiary and

Quaternary basin in which thick sequences of marine and freshwater sediments have accunulated. The plains are generally covered with 12.

sand, sandy clays and gravels. Block faulting in the Moe and westernport areas resulted in the fonnation of extensive swamps which have largely been drainerl. Extensive wetlands are also associated with the coastal .

The Southern Uplands, steep, mountainous regions with a characteristic rounderl topography I contain relatively few lakes or wetlarrls I arrl no wetlarrls were samplerl here.

In the following chapter the victorian wetlands samplerl are

classified on the basis of their vegetation. Chapter Three examines

the effect of data rerluction on this wetlands classification, and in

Clapter Four, the structure of the vegetation data is further analysed

using ordination techniques. The physical and chemical

characteristics of aquatic macrophyte habitats in Victoria are

describerl in Chapter Five; in Clapter Six the postulated vegetation ­

environnent relationships are analysed. Clapter Seven examines the

efficacy of the wetlands classification producerl and suggests

guidelines for the selection of wetland sites for conservation

purposes in Victoria. 13.

CHAPI'ER '!WO

Classification of wetlands vegetation

1. Data collection

Selection of wetland sites

Prior to data collection, consideration was given to what

Mueller-Danbois and Ellenberg (1974) have described as the essential steps in sampling vegetation, 1. recognition of the entities to be sampled, 2. selection of samples within the recognised sites,

3. decision as to what size and shape the sample should be,

4. the choice of attributes to be recorded.

The sixty wetland sites sampled (Fig. 2.1) were chosen to reflect the lithological and clbnatic variation found across lowland victoria. Sampling was restricted to lowland sites since, with the exception of Sphagnum bogs, there are few natural wetlands wi thin the Great Dividing Range, and they are difficult of access. Tidal wetlands and seasonally flooded pastures were also exclooed. .M:l.jor lithological types represented on the lowlands and coastal plains inclooed Quaternary coastal sands, Quaternary inland aeolian sands, Quaternary alluvial deposits, Recent volcanics and Tertiary sediments. Within each of these provinces, wetlands were selected for sampling by field reconnaissance. The m:mber of sites chosen 'Hi thin each area was proportional to the total nunber of wetlands in the area. o 100 200km. t. I N

MALLEE ·1 Wetland Study Sites (for site names see Table 2-1).

"J..­ ?'..<'l'..... ~ ·25 ..-F' ~ ·;,>9

·24 WIMMERA .18 ·30 -, 27· 32' .21 -, 26' 37' ·33 ", -4 ", ,2] 22, .19 ", '5 o Ballarat EAST GIPPSLAND '2 WESTERN DISTRICT .49 .12 .44 .38.11 BASALT .17 PLAINS ',8 ·53 '~1 ::.d· •10 14.3 .34 ·55 '54 13 __.___ .16 D,scovlir V '35 ( ...... ____ '7 BaV "'~ '6 .41 ~ ·52 Port Call1pbell

Fig- 2.1. Distribution of wetland study sites in Victoria. Table 2.1 Location of sites sampled for Victorian Wetlands Study. Latitudes and Longitudes are given. Sites marked * had no official or local names, and have been referred to by locality names.

La t. (S) Long. (E) La t. (S) Long. (E) I. Palparra Settlement Swamp'< 37°58' 141 °01' 29. Loch Garry 36°13' 143°18' 2. Break Number 2 Swamp* 37 38 141 13 30. St. Mary's Lake 36 46 141 45 3. Tremaines Swamp 38 06 141 25 31. Middle Lake 35 40 140 53 4. Brown Reedy Swamp* 37 20 141 20 32. The Stony 37 07 141 40 5. Kanawinka Swamp'" 37 28 141 10 33. Lake Kanagulk 37 05 141 51 6. Floating Island 38 21 143 23 34. Heywood Go 1f Club Swamp" 38 10 141 40 7. Cobrico Swamp 38 18 143 00 35. Lake Bridgewater 38 20 141 24 8. Deep Lake 37 56 143 10 36. Cotters Lake 33 56 146 15 9. Lake Elusive 37 45 149 27 37. Bitter Lake 37 05 141 46 10. Lake Bong Bong 38 08 141 11 38. Swan Lake (E.Vic.) 37 45 149 02 11. Dock Inlet 37 47 148 50 39. Rockbank Swamp* 37 44 144 41 12. Ewing Marsh 37 48 148 21 40. Army Reserve* 37 39 144 40 13. Swan Lake (W.Vic.) 38 13 141 19 41. Bleak House * 38 18 143 44 14. Little Lake Malseed* 38 13 141 17 42. Lake Tutchewcp 35 31 143 45 15. Ge IIi onda 1e S~lamp* 38 38 146 33 43. Lake Barracouta 37 33 149 52 16. The Long Swamp 38 15 146 56 44. Lake Beadle 37 46 148 24 17. Dereel Lagoon 37 49 143 45 45. Brisbane Range Swamp* 38 51 144 16 18. Two Tree Swamp 36 30 144 54 46. Fernbank Swamp 37 51 147 21 19. Cockpit Lagoon 37 27 143 37 47. MacKenz i e Road Swamp'" 38 26 144 45 20. Lake Lookout 38 41 143 40 48. Kangaroo Swamp 38 22 144 53 21. The Swannee 37 07 141 43 49. Banyule Billabong 37 47 145 03 22. Bittern Lagoon 37 24 143 30 50. Little River* 38 55 144 30 23. Victoria Lagoon 37 23 142 10 51. Round Lake 38 08 143 11 24. Meander* 36 23 146 22 52. Bay of Islands Swamp. 38 34 142 50 25. 35 45 141 58 53. Lake Struan 38 01 143 25 26. The Duck Holes 37 07 141 27 54. Lake Winchel sea 38 08 143 57 27. Champion Lake 37 58 141 21 55. Lake Terangpom 38 08 143 19 28. Lake Wandella 35 45 143 52 56. Black Swamp 36 44 144 12 14.

Criteria for the selection of wetlands incluCled. the presence of aquatic macrophyte or helophyte vegetation, the presence of at least an intermittent body of open water, and canparative lack of disturbance. These criteria were not always strictly adhered. to, sane sites with artificially maintained. water regimes being incluCled. if they supported an interesting flora.

wetlands sampled but rejected prior to final analyses included several highly saline sites which supported no aquatic macrophytes, and Black Swamp (site 56, Fig. 2.1) which was dry when sampled.. Preliminary analyses indicated that the latter was an outlier likely to disturb subsequent ordinations (see Whittaker and Gauch 1978, van der Maarel 1982) •

sampling considerations

The anticipated. vegetation heterogeneity and differences in size of the wetlands made the choice of an appropriate sampling strategy important. Many previous wetland studies (e.g. Walker and Coupland 1968, 1970, and Walker and wehrhahn 1971, Musil, Grunow and Borman 1973, Howard-Williams and Walker 1974, Kershaw 1978, Schwintzer 1978) have been based on subjectively delineated, representative stands within which samples were either randanly or systematically located. The lack of rigour and dependence on subjective notions of vegetation entailed in the selection of representative stands has frequently been criticised (Smartt 1978). M::>re objective methods using randan or systematic sampling procedures are attractive in that they make no prior assunptions about the vegetation, but t..'iey are frequently 15.

uneconomic in teons of time and effort, especially for large scale vegetation surveys.

Several recently proposed sampling models erect sampling patterns independently of the vegetation to be studied, and allow more intensive sampling of areas with the greatest vegetational heterogeneity. snartt's model (1978) uses infotmation on the total range of vegetational variation within the study area, which is obtained from an initial framework of points and t."len used to locate the remaining samples. Jensen (1977) derived an objective method for

D.le sampling of macrophytic lake vegetation using morphological parameters. This method enables calculation of the number of profiles needed. for a representative description of vegetation in lakes of different area and lakeshore development.

However, both these methods require prior knowleclge of the sites to be sampled. The distances travelled. to field sites sampled. for this study precluded. the separate reconnaissance trips requirecl to collect the infotmation necessary for the use of snartt I s sampling system. Although the morphometric parameters used by Jensen can be obtained. from maps and aerial photographs, the difficulty of identifying the landward limits of wetlands, and the computation required made the use of this system impractical.

To make best use of the available time, transects were located to sample the vegetational variation wi thin each wetland. Each transect was positioned. at right angles to the shoreline, and samples were recorded. at five metre intervals. Use of a constant intersample distance precluded. the subjective assessment inherent in 16.

representative sampling. Data 'Were collected along three transects at each site. In 'Wetlands where the vegetation was particularly homogenous two transects 'Were used. Sampling was carried out by wading along the transect line, usually to a depth of one and a hal f metres. At some sites diving was required, and at several deeper lakes sampling was done fram a boat.

Sample size

Vegetational heterogeneity, both between and wi thin sites, made it difficult to establish a standard quadrat size using a minimal area technique. Sane VoUrkers (Spence 1964, Zutshi 1975) have overcame this problem by varying quadrat size according to the nature of the vegetation being samploo. However, in the present study, trials indicated that one metre square quadrats were a practical size for assessing aquatic vegetation which may consist of several submerged layers. This size is consistent wi'b.'1 that adopted by some other

VoUrkers (e.g. Kershaw 1978, Jensen and van der Maarel 1980) for aquatic macrophytes.

The sampling program resulted in the campilation of floristic data at three levels: (1) quadrat data, (2) transect data from the amalgamation of quadrat data, and (3) wetland data from the further amalgamation of transect data. The three-tiered nature of the data alloYlE!d a nunber of different approaches to the numerical classification of 'Wetland vegetation. 17.

Choice of attributes

Descriptors

M::>st ecolCXJical stud.ies characterise vegetation by recording species presence or abundance. Other methoos of vegetation description which have been enployed for classificatory purposes includ.e higher taxonomic ranks and structural characterisations. Van der Maarel (1972), Dale and Clifford (1976) and del M::>ral and Denton

(1977) have suggested that the use of higher taxonomic ranks in the numerical analysis of vegetation may provide a sufficient level of interpretability, whilst reducing computing time and the noise which accompanies large, heterogenous data sets. Structural characterisations based on life-fonn categories have been proposed for aquatic macrophytes by Raunkiaer (1907), du Rietz (1930), den Hartog and Segal (1964), Dansereau, Buell and Dagon (1966), and Makirinta

(1978) ; Lambert (1971) has suggested that such descriptors may be usefully classified. Since it is possible to generate taxonomic and life-fonn categories from species lists, the decision was merle to record species data.

Species nomenclature folloW'S Aston (1973) for aquatic plants, and

Willis (1970, 1972) for terrestrial plants. Angiosperms and pteridophytes were identified to species level, but the absence of reproouctive structures usually preclud.ed identification of mosses, liverWJrts and Cl1araceae below generic level. Voucher speci..."'Ilens have been deposited in the Herbariun, Botany School, Un i.versity of

Melbourne. 18.

Measurement scales

Floristic data may be qualitative (presence/absence), or quantitative (measures of abundance). Quantitative data clearly provide more information about the vegetation under study, but are more tnne-consuming to collect.

Difficulties 'Ifolere experienced early in the sampling program in the visual estimation of cover values for submerged species, especially where several species grew together in more than one metre of water, or where a view of the botton vegetation was obscured by suspended material.

Studies which have examined the effect of data type on numer ical classification suggest that, for the survey of large or heterogenous regions, qualitative data will suffice (Williams, Lance, Webb and

Tracey 1973). Smartt et. al. (1974) and Frenkel and Harrison (1974) conclt:.ded that for the data sets examined, presence/absence information provided ecologically interpretable groupings. However,

Jensen (1977) has pointed out that "the use of presence/absence data may prevent finer classification on the basis of quantitative differences in species performance".

Observations have indicated that the abundance of certain aquatic species is dependent on the time of sampling (A. Corrick, pers. ccmn.) • Since sampling necessarily extended throughout Spring,

Summer and Autumn of 1979-1981, it was thought that percentage cover values would not be meaningful in this context. A decision was made

to record presence/absence data and to note t.l)e species with the 19.

highest :t;:ercentage cover (the daninant) •

Water and substrate samples were collected along each transect, and general environmental characteristics recorded. These data are discussed in subsequent chapters. Of the 60 wetlands sampled, 55 provided data recorded as 1277 quadrats along 158 transects for floristic analysis. The sites sampled are listed in Table 2.1.

2. Classification of floristic data

Choice of unit to be classified

Approaches to the nunerical classification of macrophyte communities have been discussed by Jensen and van der Maarel (1980).

Previous investigators have classified quadrats (Musil, Grunow and

Bornman 1973, Daniels 1978, Kershaw 1978), or data fram each stand sampled (Walker and Coupland 1970, Jensen 1979) • Data collected fram

Victorian wetlands could be analysed at quadrat, transect or lake level, but the size of the data set (1277 quadrats x 140 s:t;:ecies) imposed sane constraints on methodology. The canputations required for a matrix of these dimensions made tl1.e use of available agglanerative programs too ex:t;:ensive, and preliminary analyses carried out using a subset of the quadrat data indicated that the high level of vegetational heterogeneity made interpretation of the reSUlting classification difficult. Since t."e aims of the study ...ere to establish a classification of wetlands on the basis of their aquatic vegetation, and to relate any resulting patterns to causal environmental factors, it was decided to amalgamate the quadrat data for each wetland to classify wetlands first. 20.

Scale of measurement

The species presence/absence data from 1277 quadrats were canpressed by the program GPRES (from the C.S. I.R.O TAXON package,

Ross 1982) to provide floristic infonnation for the 55 wetlands.

GPRES calculated the probabili ty of occurrence of each species, and the resulting values .Nere scaled by a factor of 100, producing percentage frequency of occurrence for each species in each site.

This procedure effectively nonnalised the data, reducin:; the effect of the variable number of quadrats recorded for each site.

Choice of strategy

Clifford and Williams (1976) have emphasised the irnp::>rtance of choosing the most appropriate measure of dissimilarity for a given data set. Since it was planned to carry out the initial analyses following a priori grouping of the quadrats, which would generate frequency data, the infonnation statistic suggested by Dale (1971) for use wi th this type of quantitative data was employed.

Infonnation statistic strategies are known to be intensely clustering, and disadvantages associated with the use of infonnation gain as a dissimilarity measure include double zero rnatchin:; and group size dependency (Clifford and Stephenson 1975). A number of authors have evinced concern for the extent to which certain similarity coefficients regard patterns of joint absences of species as equally imp::>rtant as joint presences. calculation of the proportion of zero entries in the 140 x 55 data matrix showed that joint absences would 21.

make a large contribution to dissimilarity, and this was considered undesirable. However, the particular infonnation statistic chosen was as:ym:netric, and "while affected by presence/absence effects, uses only presence in the calculation of heterogeneityll (Dale and Anderson

1972) •

Group size dependency arises as a large group becomes increasingly difficult to join and separates frcm the hierarchy at a level determined partly by its relative affinities with other groups, and partly by the nunber of sites it contains (Clifford and Stephenson

1975) • 'Itlis frequently results in a "ragbagll group forming, whose members share little in cammon and may require reallocation.

3. Validation of groups produced by numerical classifications

Classification techniques attempt to discover natural groupings within a data set. However, same data may exhibit little tendency to cluster, so that classification results in the imposition of

inappropriate structure.

criteria for determining the appropriate nunber of groups in a classification may be employed at the level of individual cluster (see

Bailey and Dubes 1982), or at the global level (e.g. Sandland and

young 1979), or a canbination of these approaches such as the

Similarity Ratio proposed by Popna, Mucina, van TOngeren and van der

Maarel (1983) may be used. However, many of the published

stopping rules require that the data meet special conditions and thus

are not widely applicable. 22.

Fatkowsky and Lance (l978) have describErl a criterion for determining the number of groups in a classification Which allows a global canparison of cluster validity. The contributions of each attribute to a classification are calculatErl, the contributions of all attributes averagErl, and a scaling factor applied to allow for varying numbers of groups. In the TAXON package, the diagnostic program

CRAMER provides the Cramer value (a measure of attribute contribution) fran Which scalErl Cramer values are calculated by the progra~ RATLAN at each level of the hierarchy. The maximal scalErl Cramer value indicates the number of groups to be acceptErl.

The Fatkowsky-Lance criterion can be applied to qualitative, multistate am nuneric data, am unlike many other procErlures does not attach special corrlitions to the nature of the data. It provides a measure of global separability of clusters, canparing cluster isolation (separation between clusters) and canpaction (cohesiveness amongst data items) simultaneously at the global level. However, as the authors indicate, the value of the test can only be assessed by empirical investigation.

Hill's modification (Hill 1980) of the Fatkowsky-Lance criterion,

-which restricts the number of groups canparErl to two, allows canparison of sibling groups, thus providing a measure of cluster isolation. However, this test applied to the floristic classification produced a very large number of groups, so the results were not taken into account when the nunber of groups was determined.

The use of information statistic as a clustering strategy provides an additional approach to the determination of appropriate levels of 23.

fusion. 'IWO information measures are associated with each group generated, the group heterogeneity (information level) and the increase in heterogeneity associated with fusion (information change) •

At the local disjunction the degree of isolation between sibling groups may be inferred from the information change, whilst information level provides a measure of group compactness or homogeneity.

In deciding which groups to accept, the Ratkowsky-Lance criterion, information gain and the information level of each group was examined.

However, the overriding criterion for the acceptance of a group was ecological interpretability, especially in view of the a~knowledged problem of group size depemency.

4. ~thods of analysis

The availability of computer programs and the nature of the data collected ultimately dictated the analytical approach. The analyses were carried out in association with Dr. M.B. Dale of the C.S.I.R.O.

Division of Computing Research, St Luc ia, , using the TAXON

Package of programs.

Programs employed in the classification of floristic data were

GPRES, which, as previously mentioned, compressed floristic records from vegetated quadrats to wetlarrl floristic frequency data, NIASM am the associated NIAP polythetic agglomerative information statistic classificatory programs, and RATLr...N, which pr:::lvided the

Ratkowsky-Lance global comparison of cluster validity. TABOUT produced two-way tables for each group generated, NEAREST listed the nearest seven neighbours of each irrlividual, anc1 the programs GCOM aoo 24.

CRAMER were used to provide diagnostic infonnation. Documentation for tr..ese programs is given in the TAXON Users Manual alition P3 (Ross

1982) •

5. Results

stopping Rules

The program NIASM was run wi th the number of groups requested set

to ten. A slight excess of the probable nt.lllber of groups was generated in vrder to ascertain the level at which the maximal scale:3.

Cramer value occurre:3.. Application of the Ratkowsky-Lance criterion by the program RATLAN suggested that truncation of the classification at the seven group level was optimal. A nine group solution was

suggested by the infonnation change associated with the fusion. At

this level the groups appeare:3. sufficiently isolated. The degree of canpactness of the groups as indicated by their relative levels of

heterogeneity was very variable at both the seven and nine group

levels. An examination of the group memberships suggested that the

seven group level would provide an ecolcx;,ically interpretable

classification. However, a decision was made to split the final group

of saline sites (the last group to fuse at the eight group level) ,

since it appeare:3. to consist of two canpact and isolated subgroups

which were floristically distinctive. The levels of heterogeneity of

the resulting groups were canparatively low, reflecting the floristic

poverty of these sites. Table 2.2 Membership of groups generated by NIASM floristic classification

Group Site Group Site No. No.

102 9 Lake Elusive 84 8 Deep Lake 12 Ewing Marsh 20 Lake Lookout 22 Bittern Lagoon 25 Lake Albacutya 32 The Stony 28 Lake Wandella 41 Bleak House Lagoon 38 Swan Lake (E.Vic.) 46 Fernbank Swamp 53 Lake Struan 54 Lake Winchel sea 97 10 Lake Bong Bong 14 Little Lake Malseed 96 21 The Swannee 35 Lake Bridgewater 27 Champion Lake 36 Cotters Lake 30 St. Mary1s Lake 43 Lake Barracouta 37 Bitter Lake 52 Bay of Islands Swamp 40 Army Reserve Rockbank 42 94 16 The Long Swamp 51 Round Lake 34 Heywood Golf Club Swamp 55 Lake Terangpom 47 MacKenzie Road Swamp 48 Kangaroo Swamp 99 18 Two Tree Swamp 24 King River Meander 100 2 Break Number 2 Swamp 29 Loch Garry 3 Tremaines Swamp 31 Middle Lake 4 Brown Reedy Swamp 33 Lake Kanagulk 5 Kanawinka Swamp 39 Rockbank Swamp 11 Dock Inlet 50 Li ttle River 13 Swan Lake (W.Vic.) 17 Dereel Lagoon 101 1 Palparra Settlement Swamp 19 Cockpit Lagoon 6 Floating Island 23 Victoria Lagoon 7 Cobrico Swamp 26 The Duck Holes 15 Gell iondale Swamp 44 Lake Beadle 49 Banyule Billabong 45 Brisbane Ranges 25.

Floristic groups

Membership of the eight groups generated is shown in Table 2.2, and Fig. 2.2 shows their floristic relationships. The convention of reading the hierarchy downwards has been adopted. The first group to fODD, 84, is floristically comparatively homogenous, and consists of species-poor saline sites dominated by Lepilaena bilocularis. Swan

Lake East (38) and Winchelsea Lake (54) join the group last. Species characteristic of grou~ 84, Lepilaena bilocularis and Chara sp., were recorded for both these sites as well as species associated with other groups.

Triglochin procera, Myriophyllum propinquum and Phragmites communis, typically fresh to brackish water species, were found in the upper reaches of Swan Lake (38). Although the lake is not tidal, saline water penetrates via a channel which flows into Sltienham Inlet.

Lake salinities as high as 12 0/00 have been recorded, but the influence of the saline water quickly dnninishes towards the head of the lake (M. Williams 1973). Swan Lake contains the easterrmost occurrence of Lepilaena bilocularis (most records are fram western

Victoria), the most sou~~erly occurrence in the world of Najas marina CAston 1979), and Cladophora echinus, an algal species known fram

Europe, Japan and Mainland Asia (Ducker, Brown and calder 1977) which has a very lnnited distribution in southern Australia. It is thought that water birds may have been responsible for the introduction of these species •

The occurrence in Lake Winchelsea of Ruppia maritima, a species otherwise confined to members of group 96, indicates that this lake 65362 --­

Fig. 2.2

20 34 Dendrogram produrrd by NIASM rla.silieation of entire floristic

d .. ta set liHing 5pecie~ in order of their m<>gnltude "f contribution to each dichotomy Only 'p<"ci". whose contribution to group

heterogeneity was >IOi_ and "5~, to the sibl jng group are shown.

I. A(I"ootis (1",,",,1_<1 40. Myriophyllum mueHeri

~. A:JoUa [iZi('!tI~oider, 41. Myriophyllum propinquwn 5. Baumea artieulata ~3. Naaturtium o[[icimte 6. Bmmrt!'o .iunee" ~~. Ni tetta Bpp. 7 . 8a""""a rW)ig;rzosa ~8. PiI1'O.gmi teo """m"l'Ii8 34737­ 101, 10. Carex apl'rBo6o 5 I • Ral'lwwu lUB inul'Ula tUG 12. Cham oH'. 52. Ral'lul'lculuB rivutaris 51 39 16. Crall8ula helmsii 53. lIestia tetraphyllu8 20. F:ragro8ti P a"si_l'alafliew; 5~. lIiaoia sp.

2 I. Gtyeeria ilia' t,·oII:(l 117. oS,,; '7'Wl [Lui tails

1()!I 2~. lIydrocotyl MrfpU" 61. .'ipMgl1"'" sp. 4 30. Demnl1 miner 63. 7'rigl.o<,hin pT'IJ"ara 24986-- - -I 6 - 31. Lpmna tl'ioulea 65. Typha riomil'l!!,'lls to ::r0 1: 32. r,Bpidoer"mno tongi[oZia 66. Typha arip.ntatp 5 33. [,ppi tal"11a bllceu/oris 67. Urtica inciaa 32 1.05_ 3't . [""'rllf'lCM cy 1{'f1drOCal~[Ja 68. 1Itl'ioular,:a aw,tlYlti:; 15103 -- - -I 1104­ , N"elllenbeoha Olmninglu

Information Only the fi rst forty species to make contributions Were listed, other level ~91 198 971 94 species further down the d"ta list may have been important. 26. has some floristic affinities with these more saline sites. Eragrostis australasicus, a species characteristic of turbid waters

(group 99) , was also recorded here.

Group 84's sibling, group 96, consists of sites with generally higher salinities, which are characterised by the daninance of Lepilaena cylindrocarpa or Ruppia maritima, both of which ccmnonly occur with MYriophyllum muelleri. This saline group is also species poor.

The freshwater groups (i.e. those where salinity <3.0 0/00) which incltrle 94, 97, 99, 100, 101, and 102 are more difficult to characterise floristically, since sane camnon species are widely distributed, and unique assignation of species or species groups did not always occur within the classification.

1 The Cramer values, which are calculated as (B/T) /2, where B is

the between group variance and T the total variance, indicate the relative im.portance of attributes over the canplete set of classificatory groups. These values show that, although a number of widespread species, such as MYriophyllum propinquum, Triglochin

procera, Eleocharis sphacelata, Scirpus flu'::' tans and Lepidosperma longifolia all occur in sites belonging to groups 100, 101 and 102, the level of attribute contribution of same of these species varies significantly between groups. In addition, 101 and 102 contain species which are either uniquely assigned to them, or which occur only at low mean frequencies «4%) within the other freshwater groups. 27.

All the sites of the largest freshwater group, 100, contain

Triglochin procera and Eleocharis sphacelata, or Baumea articulata, which has otherwise only been recorded fran site 1. Baumea

rubiginosa is also carmon. Myriophyllum propinquum is characteristic

of most sites wi thin this group, exceptions being Cockpi t Lagoon (19) which has Myriophyllum verrucosum, and Lake Beadle (45). Tinms

(1973) noted sparse beds of Myriophyllum propinquum here, but they were not .recorded within the transects, possibly because sampling was carried out while the lake level was abnonnally high.

GrOUp 101 was characterised by the presence and usually abundance

of Azolla filiculoides. Other species which contributed to group

similarity include Crassula helmsii, Lemna minor and Lemna trisulca.

The occurrence of these species was largely confined to this group and

the closely related group of turbid sites (99). Triglochin procera,

cannon to four of the five wetlams in group 101, was fouOO in most of

the freshwater groups. Sites 6, 7 and 15 were dominated by stams of

Phragmites australis or Typha spp. Banyule Billabong (49) shared few

species with the remainder of the group. The occurrence of the

comparatively rare Riccia sp. has placed site 49 in this group.

Palparra Settlement Swamp (1), a wetland which shared many of the

species carmon to members of group 102, and whose envirormental

characteristics were similar to those of members of group 97, has

joined 101 because its most frequently recorded species 'Here those

which made the greatest contribution to the similarity of this group.

Group 102 is characterised by comparatively high frequencies of

occurrence of Myriophyllum propinquum, a widespread species of fresh

water. Lake Elusive (9) and Bleak House Lagoon (41) only shared t:.vo 28.

other species each with the ranainder of the group. The fusion of these outliers occurred well up in the hierarchy, they had no strong floristic affinities wi th any of the other wetlanCis sampled. Lake

Elusive is a deep (21 m) coastal dune lake, no macrophytes grow in its highly coloured hunic waters, and vegetation is confined to the shoreline near several creek mouths. Bleak House Lagoon, most of which was only sparsely vegetated, supported Lepilaena bilocularis arrl

Myriophyllum muelleri which are carmon in more saline waters, as well as some species typical of freshwater corrlitions.

The majority of the sites within group 102 (12, 22, 3~, 46) had a dense cover of floating anCi suhnerged species, including Villarsia reniformis, Potamogeton tricarinatus, Nitella sp. Myriophyllum propinquum and Scirpus fluitans. These plants occur in other freshwater groups, but were particularly abundant in the shallow waters of this group of sites. The widespreoo. emergent, Eleocharis sphacelata was also fourrl in all these wetlarrls.

The wetlanCis of group 97 are all located wi thin dune systems less than one kilometre from the coast. The sites are characterised by the presence of frequently broad zones of Baumea rubiginosa and often

Baumea juncea. The Discovery Bay sites, Lake Bong Bong (10), Little

Malseed (14), and Lake Bridgewater (35), which are comparatively deep p::rmanent lakes with calcareous substrates and very clear waters, supported same dense beds of Chara and Ni tella, and Myriophyllum elatinoides was cannon. Also in Western Victoria, the Bay of Islarrls site (52), a depression amongst cliff-top dunes on the Port campbell llinestone coast, is floristically similar to ~,e Discovery Bay lakes, but due to its shallowness lacks the dense beds of C1arophytes. 29.

cotter's Lake (36), in the Yanakie dunes on Wilson's Promontory, the final site to join this group shows some floristic affinities with these ~stern victorian sites, but lacks subnerged aquatic species because de'fX)si tion of aeol ian sand has raised t.'e level of the lake floor (Leech 1973) and there are now no areas of open water.

Lake Barracouta (43) in Far East Gippsland is surrounded partly by siliceous sands and is less diverse floristically than the other group members with which it shares Baumea spp. and Triglochin procera.

The 'lirtual absence of macrophyte species, (several small be:ls of potamogeton australiense were noted) may be due to strong wave action and consequent turbidity.

Group 94 has no single characterising species. The waters of all the sites are acidic (pH 4.07-6.2) and have low conductivities.

Heywood Golf Club SWamp (34) and MacKenzie Road. Swamp (47) are characterised by the presence of Restio tetraphyllus and Sphagnum sp. growing on peaty substrates Mlich appear to b.ave been burnt periodically. Similar Restio tetraphyllus-Sphagnum daminated swamps have been re'fX)rted from French Island (S. Krae:ners pers. comn.) t."le

Otway Ranges (N. Rosengren pers. comn.) and East Gippsland (J.

Grindrod pers. comn.), and fram and

(Crocker and Eardley 1939). They appear to be uncommon but may have been under-represented in the sampling. Kangaroo Swamp (48) and The

Long Swamp (16) I although geographically close to site 47, only share several species each with 47 and 34. However, they share nine and

seven species respectively with group 100 (although not Triglochin

procera, characteristic of all members of 100). Since the next group

to join 100 is 94, the effects of group 5ize dependency may have fresh waters pH neutral aCId

sItes onI coastal plains or tertIary Inliers south of Great D,VIding Range saline or turbid I on basalt plains or north of Great DIviding Range iexcept 101)

1071

'I

permanent water I

annual I fiillng I i: shallow fresh water basIns n I \ I ~brown i salinIty water I 02-2 1%0 1 i clear salinity water >30/00 1105 I ~UCtIVlty calcareous I . <1000 ~mhos coastal ~ conductivity I >1000 ~mhosfresh brown water 102 I I organic substrate fresh brown turbid hOl water sandy 103 substrate

very aCidIC sal.ne saline

89 98 97 94 100841 \a6 99 90 95

Fig. 2.3. Dendrogram produced by NIASM classification of floristic data showing ecological affinities of the groups. 30.

resulted in the fusion of 16 and 48 wi th 47 and 34 rather than fusion wi th group 100.

r£he members of group 99 also share no single species in cammon.

The highly turbid Middle Lake (31), Lake Kanagulk (33), Rockbank (39) and Little River (59), have been linked with the billabongs King River Meander (24) and Loch Garry (29) by the shared occurrence of floating and floating-leaved species such as Azolla filiculoides, Lerona minor,

Ludwigia palustris and Ludwigia peploides. Two Tree SWamp (18), a shallow basin which dries out in swmer, is the last site to join group 99. It shows floristic affinities with members of group 102

(shallow southern Victorian wetlands), but the large attribute contribution fram Eragrostis australasicus places it in group 99.

The dendrogram

Fig. 2.3 indicates the ecological affinities of the eight groups generated by the NIASM classification of the floristic data. The first split was essentially into freshwater wetlands on sandy substrates and more saline or turbid sites. All the freshwater

wetlands in groups 89, 94, 97, 98, and 100 are situated on coastal plains or Tertiary inliers south of the Great Dividing Range, and are generally neutral to acid in pH, with comuctivities below 1000 ps. Of the other major division, groups 84, 96, and 99 occur on the basalt plains of western Victoria or north of the Dividing Range, and have saline or turbid waters which are neutral to alkaline with

comuctivities exceeding 1000 fS. Group 101 contains freshwater sites

(salinities 0.2-1.00/00) which show some floristic affinities with the

turbid group of sites, 99. Table 2.3 Groups generated by NIASM classification of the floristic data showing diagnostic species (shared by all members

of a group) at the 8 group level.

Group number Group type Shared species

84 Sa 1i ne Lepilaena biloau~is

96 Very sal ine ;'c

97 Calcareous coastal Baumea rubiginosa

94 Acidic water *

99 Turbid -l:

100 Fresh brown water Trigloahin proa~ra

sandy substrate

101 Fresh brown water AzoUa fiUattloides

organic substrate

102 Shallow fresh water basins Myriophyllum propirnr~um

* No species common to all group members 31.

The secOrxl dichotamy wi thin the freshwater group separated shallow, completely vegetated basins fram sites with more permanent water regimes. The next division separated very clear, calcareous waters fram sites with brown water. The latter further divided into sites with acidic waters and those with a pH closer to neutral. The shallow basins were further differentiated, but this distinction was not maintained.

Wi thin group 108, the sites wi th high salinities were separated fram the turbid sites and the anamalous group 101. The saline .:3i tes divided again on the basis of their salinity levels.

The site groupings are regarde:1 as ecologically meaningful. The position of group 101, fresh brown water sites which appear on the saline-turbid side of the dendrogram, is probably due to a group size dependency effect. Group 101 has been fused with 99 (turbid waters) although it shares more s;::ecies wit."1. its other potential sibling groups, 89, 94,97, 98 and 99, because t.1-tis fusion resulted in a minimal information gain; that is, it caused least disturbance.

Table 2.3 surmarises the eight flor istic groups which were recognise:1, their ecological typology and the species which were common to all members of a group_ The more saline wetlands (group 96) contain Lepilaena cylindrocarpa or Ruppia madtima , but share no singIe species in carmon_ The less sal ine sites (group 84) are characterised by the presence of Lepilaena bilocularis. The calcareous coastal group of sites (97) is characterised by the occurrence of Baumea rubiginosa, whilst all ~he shallow, freshwater 32. basins of group 102 share Myriophyllum propinguum. The fresh, brown water sites with sandy substrates all support Triglochin procera, whilst those with organic substrates all share Azolla filiculoides.

Neither the turbid sites (group 99) nor the acidic water sites have species cammon to all their manbers.

6. Discussion

criteria for the assessment of a classification

It would be desirable to have a measure which would indicate an optimal classification for a given data set. Unfortunately, there is yet no computationally feasible algorithm which guarantees an optimal

solution (Jardine and Sibson 1971) • criteria which have been used to assess the adequacy of a class representation include group compactness and isolation (groups should be clear cut and

comparatively hcmogenous), and the ecological acceptability of the proposed classification (Williams, Lambert and Lance 1966, Frenkel and Harrison 1974). In order to be ecologically acceptable, a classification needs to be interpretable in the light of ecological

experience. The groups generated should probably not be too different

fram those proposed intuitively; they should be at similar levels in

the hierarchy, indicating in vegetation ter:ms that each group is responding to the same scale of envirormental factors. It has also

been suggested (Silvestri and Hill 1964) that a good classification

should be objective, so that conclusions are unambiguous and independent of interpretation, stable, and therefore not unduly affected by the addition of more data, predictive, remaining largely unaltered when 1:.."1e same entities are measure6. using a new set of I

102·

-l-­ 101· ~--1------t---- 99. ' c. I ;,e CD '0 97· -8 ~ '" .~"

Q'" .8 E ;, c: -; ~ 84.

Information Level

Fig. 2.4. The effect of stand richness (total number of species recorded for each group) on information level of the eight groups generated by the NIASM classification of the floristic data. 33.

attributes, and robust (Orloci 1978), small changes in the data resulting in only minor changes in the classification.

The results of a numerical classification should also be examined for evidence of the relevance of the mathematical mcd.el used and the adequacy of its fit, since the nature of the mcd.el used will influence

the interpretation of the groups generated.

Mequacy of a class representation

Classification was chosen as an initial approach to the elucidation of patterns within the data set since inspection of the

raw data suggested that natural groupings existed.

The infOl::mation level associated wi th each of the eight final groups ranged fram 1755 (measured in natural infoonation units) for

saline group 84 to 9879 (shallow freshwater group 102), indicating

that the degree of compactness of the groups was variable. Fig. 2.4

suggests that this is a function of stand richness, as the more

compact groups were floristically poorer.

The change b infoonation associated with the fusion of final

groups at the next level in the hierarchy (Fig. 2.2) suggests that the

groups ~re comparatively isolated, although the distribution of

certain widespread species across the groups reduced their floristic

distinctiveness.

GrO'l.."Ps 94 and 99 did not meet one of the criteria proposed by

Williams and Dale (1965) to distinguish bet'Neen classification and 34.

partitioning of a data set. Neither group containerl a species which was sharerl by all manbers of the group.

The proposed v.letlands classification is interpretable in ecological terms, and the groups recognised largely accord with intuitive groupings. Most of the manbers of each of the v.letland types share similar environmental characteristics, such as salinity, turbidity, pH, conductivity, water depth and water regime. HOv.lever, the freshwater groups 100, 101 and 102 which share a number of species are also more difficult to distinguish on environmental grollI'Xls. The success of the classification will be further examinerl by car.onical correlation of the environmental variables measured with the floristic classification.

EVidence for relevance of the model and adequacy of fit

The classification generated is ecologically acceptable although sane misclassifications indicate that it is not optimal. Lake

Elusive (9), The Long Swamp (16), Bleak House Lagoon (41) am Kangaroo

Swamp (48) appear to be more closely related to manbers of groups outside their own, and may require reallocation. The classification also produce:3 some groups (94 and 99) lNhich v.lere difficult to characterise floristically, and some distinctive but llI'Xler-represente:3 types v.lere fuse:3 with dissimilar sites. These problans may be attributed to the nature of the information statistic model used: it

~as agglomerative, asymmetric, polythetic, and allowed an interaction between species presence and abundance values. 35.

Agglomerative programs are frequently prone to minor misclassifications (Williams 1976a) because fusion be:;Jins at the inter-individual level where the fOssibility of error is greatest.

Lake Elusive may have been placerl in group 102 instead of the more appropriate 101 as a result of its initial fusion with Ewing Marsh

(12) •

Choice of an asyrrmetr ic version of information statistic was deliberate, since the probability of occurrence of most of the 140 species recorded in anyone wetland was low, and it was thought undesirable to incllile double zero matches in the calculation of heterogeneity. l-bst of the imfOverished sites shared sufficient species to form compact am isolated groups (84 and 96). However, the assignnent of Lake Barracouta (on quartzose sands) to the calcareous group 97 may have resulted fram the effects of double zero suppression. Same of the few species recorderl for Lake Barracouta were possessed by the site it fused wi th. The absence in Lake

Barracouta of any submerged macrophyte species cammon to other members of group 97 was not taken into account.

polythetic assignment (baserl on more than one species) of sites to groups has led to difficulties in characterising some of the

freshwater groups because they do not have a single species or group of species in cammon. Assignment to groups is further complicated by

presence/abundance interactions.

The presence/abundance interaction of the NIASM algorithm, which

calculated heterogeneity using relative amounts of species (the

profOrtion contributed by a species to the ve:;Jetation of a stand) was 9 H10 H20 Percentage freQuen / H30 / ey of d . 8 H9 / 100 80 60 Otnmant S / H19 H40 40 Pee; / 29 / o F\' F\11 F2\1 F31 20 es /H H39 A50 -t.. F41 10 S \ F51

~~ G38 LJ­ %.\-::--2-E3", E'\ E2\ E33 G21 JMBjA58

'\<).~ "" ~ " \ / C<3 G37 \S' 3 E4 1:14 ~ 5 C6 C16 / M7 ..o~9 ~ E2~ '" /"" "9 4 ~c~ \34 / '" "I :15""<'C25 C36 / /

C54 C46 --C45-,,~ /'/' ~C55 i" 1[)

Fig. 2.5. Principal Coordinates Analysis ordination of 60 quadrat simulated data set. letters reFer to the nrnlln tn whl,-h .,,,,,-h nll",rlr!Ot """'.:: !l11n'-!'ltt>rI hv th., NIA<::M .... l"'c:c:ifi .... !'ltinh 1: 36.

investigated using a simulated data set. The data generated consisted of 60 quadrats. The first species listed in each quadrat was naninated the daninant species and assigned a constant frequency of occurrence within each set of ten quadrats, 100, 80, 60, 40, 20, and 10 percent frequency respectively. Up to nine additional species with frequencies of ten percent were added to successive quadrats in each set.

The quadrats were classified using NIASM and ordinated by

Principal Coordinates Analysis (see Chapter Four). Inspection of the change in infonnation associated with fusion of groups in the hierarchy produced by classification suggested that a six group solution was optimal. EXamination of the group membership indicated that allocation of a quadrat to a group was heavily dependent on the number of additional species present in the quadrat. It is expected that their relative abundances would also affect allocation.

Fig. 2.5 shows the two dimensional spatial contiguity of the quadrats. Quadrats one and ten have been allocated to groups F and

H, both have 100 percent frequency of occurrence of the daninant

species, but contain none and nine additional species respectively.

The most dissimilar sites had relative frequencies of the daninant

species of 20 percent or less I and had more than five additional

species present (quadrats 47-59, 57-60). Here, the naninated daninant made no larger contribution to the classification than the

additional species.

The problem of partitioning infonnation into qualitative and

quantitative canponents has been discussed by w.T.Williams (1973). 37.

The program NIAP contains a general infonnation statistic which allows partitioning of the measure into its qualitative and quantitative components, so ~~t the relative infonnation content of presence/absence and "abund.ance given presence" data may be detennined. The model is ~etrical, and therefore includes double zero matches. To examine the effects of the presence/abund.ance interaction on the assignment of sites to groups, the floristic data used for the NIASM classification was re-analysed using NIAPi the qualitative am quantitative components were analysed separately, then together.

The classifications produced by NIAP were truncated at the eight group level to simplify comparison with the eight groL."PS frcm the

NIASM classification. The hierarchies produced by the NIAP

(presence) and NIAP (presence and abund.ance) classifications were identical prior to the fusion fOITIIing group 102, where the order of the groups fusing was reversed. The group membership was identical.

The results of the NIAP (abund.ance) classification, were nonsensical.

Saline sites were grouped wi th freshwater sites and the groups generated shared few species.

The classifications produced by NIAP (presence) and (presence and abund.ance) recognised calcareous, saline, turbid and shallow freshwater groups similar to those generated by NIASM, although more sites were considered to have been misclassified. The other freshwater groups produced were more difficult to characterise floristically and interpret ecologically, and the species-poor sites and acid sites were included in a "ragbag" group. However, all the sites which had been considered intuitively to belong to the shallow 38.

freshwater type ¥Jere grouped together, although they formal t".vo separate clusters at the eight group level.

Williams (1981) noted that the algorithms used to classify frequency data are sensitive to sane unknown aspect of the configuration of the data, and that canparisons using different algorithns have shown "wildly discrepant results". Both the NIAP am

NIASM programs recognised saline, calcareous and turbid groups, but

NIASM, within the limitations describal above, generated groups I.tbich were more coherent am amenable to ecological interpretation, am thus it was judgal the more relevant model.

Methods I.tbich examine the goodness of fit of a classification require sane estimate of how ¥Jell the hierarchy generated fits the original similarity matrix. The matrix calculated by NL~SM was not made available for inspection, so no measure of fit of the classification could be made without recalculation.

7. Conclusions

The classification producal by NIASM recognised eight groups of ¥Jetland sites I.tbich can be characterised as very saline (96), saline (84), turbid (99), calcareous (97), acidic (94) and freshwater (100, 101, 102). Apart fran minor misclassification, the first five groups were distinctive, relatively homogenous and ecologically acceptable, although there is same evidence that certain sites are intermaliate, am thus difficult to assign uniquely to a group. 39.

The three freshwater groups 100, 101, 102 are closely related to each other floristically, and may not be sufficiently isolated to warrant separate recognition. The failure of infonnation statistic (which has an intensely clustering strategy expected to accentuate the differences between potential groups) to ooequately define groups wi thin the freshwater canplex was thought to be due to the high heterogeneity of the data set. It was therefore decided to examine the effects of data reduction on the classification, and to assess the relative distinctiveness of classificatory boundaries using ordination techniques. 4~.

CHAPI'ER THREE

The effect of data reduction on classification

1. Introduction

The use of data reduction techniques prior to nlJIlerical analyses has been justified where it has been desirable to reduce computation expanses, or necessary to reduce a data matrix to a size where it can be handled by a particular classificatory strategy or computer, or

necessary to reduce the effect of large nlJIlbers of zero entries in a matrix. Data reduction methods may also be used to define a meaningful subset of attributes which would serve to reproduce the

whole pattern, thus minimising data collection (Webb, Tracey, Williams

and Lance 1967). A further application is to reduce noise which may

be obscuring underlying patterns in large, heterogenous data sets.

Stephenson, Williams and Lance (1970) have pointed out that

site-species associations may be masked by the presence of nlJIlerous

"irrlifferent species" which contribute little to classification.

Data reduction has most ccmnonly resulted in the ranoval of

individual species fram a data set, and various criteria have been

used to select t.'1e species for omission. Species which only occur

once within a data set, or are confined to a single site, contribute

little to a classification and are frequently excluded. Often all

rare species are excluded. Arbitrary ranoval of species with a low

frequency of occurrence has been criticised by Dale and Williams

(1978) who noted that if many of the rare species are jointly confined 41.

to a small number of sites, the aberrant nature of such sites may be concealed. In large scale surveys many species of low frequency occur, and arbitrary methods of species exclusion based on frequency of occurrence or abundance do not provide a useful solution.

other methods order species by calculation of same measure, and those species which fail to obtain a given threshold value are exclooed. Maasures of species imp::>rtance (e.g. the prop::>rtion of the sun of squares or other quantity specifically accounted for by a species in a sample), which may be derived fram a variety of analytical methods (Orloci 1978), have been used as a means of ranking species. Stephenson and Cook (1980) eliminated those species with the highest average dissimilarities, since they were the least likely to characterise the sample groups. "Eident" values, measures of the extent to which species discriminate between site groups, also provide a measure of species imp::>rtance (Dale and Williams 1978).

Extrinsic criteria may also be used to excltxie species fram

subsequent nunerical analyses. This approach has most commonly been applied in order to divide data into subsets. Webb et ale (1967) generated data subsets based on groups of species which occupied the

same synusia within rainforest. The observation that different layers

(life-forms) of macrophyte vegetation are distributed independently of

each other (Almquist 1929, Thunmark 1931, Lillieroth 1950), has led

some Scandinavian workers to separate groups of macrophyte species on

the basis of shared life-forms, each group providing a subset for

classification. Jensen (1979) classified Scanian lakes (Southern

Sweden) using data sets subdivided on the basis of life-form

categories. 42.

Prior to ordination, Seddon (1972) separated emergent species such as Phragrnites conmunis from suhnersed aquatics, suggesting that the emergents were more dependent on the substrate in which they were rooted, while suhnersed species would be more directly dependent on the IIaqueous mediurn" as a source of nutrition.

These approaches have maintained the use of species as descriptors. Lambert (1971) has summarised the advantages of the use of species as descriptors in ecological studies. Species are "widely accepted entities with characteristic properties whicL automatically

ll impart a range of other information , and they are externally defined by current taxonomic practice.

An alternative approach to data reduction involves the use of more general categories of descriptor. Species are united on the basis of sane shared taxonomic or morphological feature, and the subsequent lli.,i t is used as the descriptor. This method of data description can reduce data collection costs, and computation time, and provide a means by which regional surveys can be related to each other. Van der

Ma.arel (1972), Dale and Clifford (1976), and del Moral and Denton

(1977) have shown that hig~r taxonanic categories can provide suitable data for numerical analyses.

Species may also be grouped according to their life-form or growth

form classes on the basis of their similarities in structure and

function. A plant life-form is usually understooo to be a growth fom

which has an obvious relationship with envirormental factors

(Mueller-Danbois and Ellenberg 1974). Ma.ny different life-fom 43.

classifications have been develoflErl, including several for aquatic macrophytes (Raunkiaer 1907, Du Rietz 1921, Iverson 1936, den Hartog and Segal 1964, Dansereau, Buell and Dagon 1966). Although life-form criteria have been used to determine subsets of species data for numerical treatment, the categories themselves appear not to have been used as descriptors for numerical classification.

Physiogncmic or structural features have also been used in the numerical classification of vegetation. Webb, Tracey, Williams and

Lance (1970) classified eastern Australian rainforests using a number of structu=al features and same life-form categories as descriptors, concluding that the method enabled a rapid assessment of site potential and was as successful in recovering environmental data as the species classification.

other potentially useful categories may be defined in relation to individual organs rather than whole plants. Systems proposed incllrle eco-organs, based on the vegetative features of organ status

(Attanapola in Lambert 1971), and the eco-morphological criteria of

Makirinta (1978b).

2. Application of data reduction techniques

TWO methods of data reduction were chosen in the present study in order to select species which were expected to make the best contributions to classification. "Eident" values (Dale and Williams

1978) provide a measure of the extent to which a species discriminates between sites or groups of sites. They are calculated from a site-by-species matrix of freqJency data. For each cell within the Table 3.1 Eide.nt values for 140 wetiand species 1 (sced in descending order of magnitude

Order Spec j at Ei dent Order Spe.c i e$ Ei dent no. value nO. value 63 1428.09 Tr>ig~a"hin praaera 71 84 82.43 :iamdus repo". 33 1119.24 l.epila.lina biZoauZaria 72 119 82.31 Sr;e l l.aPia caespi toaa 41 1017.42 Myriopht(lZ,um propiru.r""um 73 137 81.87 :4elaZe:uaa. squ.tllTosa. 44 935.93 :Iitslla ~ 51 79.76 Ranuncutus irp.JJ1tiat"'dS 12 891.96 Cha.r::r. ~ 56 72.73 Sagi ttar"~a gRlmirlea 6 4 810.12 AzoHa fiU",,"oide8 ~ 36 71 .62 LobeZia alata: 7 775.96 Baumea r"...JJigir.oacz 77 59 69.12 Scirp'Us va Z. idua 18 734.7J Eteochar~8 sphace lata ~ 113 67.80 Pseudoraphis Spi'l"l2SCenS 9 50 724.64 PotaJ1foglitort. tp"~cClrinatu8 ~ 99 67.34 Polygor:um prostrotum 10 72 639 •• 1 Villarsia rel1iformia ~ n 65.6) Unknown 11 133 635 ••7 R'uppia mari tima 31 28 61. 22 Juncus paltidus 12 30 618.20 [,61'tma minor ~ 132 59.43 Soirpus ,,f1uviatilis 13 39 511 .64 Myriophy i i:um e la tinaides 83 79 59·02 Juru::ua mat"i ti,'tfua 14 57 503.54 Sairpu$ .l:'luitans ~ 29 56.94 Junous proaerus­ 15 68 495.39 :Jtricu lari::. aU$t;ru i is 85 76 56.89 ·]unaus hoio:J:.::hoen:u.s 16 40 462.65 ,\fYl"icphy l lUJ1l mue lZeri 86 140 56.25 R,:munauZWJ tr'ichoFhylZa 17 20 459.16 cpa,grosr;is ;;u8tP:lla~iou.s 87 23 47.44 GrotioLa pedunauLClta 18 34 446.61 Leri l.i.lena cy l indrocarpa • 127 47.22 Sium Lati/'oU~orr 19 26 403.55 J:A.nau.s articuLatus M 85 .7.14 CI-adiiAm proce~ 20 17 375.30 Eleocharis acuta % 74 46.84 Putamoge"ton. .)C!nreat'... e 21 3 I 371.12 [,enma crist.. [,.ca 91 114 45.61 lH.,,1T<.ilus rqpens 22 5 364.33 Baumea aP"t'f-cuZ~tta 92 37 45·22 pmt1:oiries 23 32 333·51 : epiacsperma iOtifji.toZ ~u: n 125 41. 7Z vahn. ia fi Lum 24 66 310.37 ti:/Pha. ori~ntal ~iJ ~ 128 41.33 ~'~"PQCh.oer1:;;; i·~,l"' ;-,z;:;a 15 310.18 Agro3-;i,~ 1<31'1'1U,,:<1 H 103 41.00 ::,er;iJo.1peJ:>r:a ·l",'esii 26 35 281.84 ;:'i'la.eupsis r([,L.a.ntha % 73 39.98 ;·:clff'{,;, .l1.4Jtrr.zl::Grta 27 16 256.86 CroasuZa i:elm:;ii. ~ 134 39·52 ~.;unC'U8 ;;auai ;'Zr)1't;l 28 78 227.56 Nueh !.enbeckia -::-o<.nningr.ami £ ~ 98 37. 48 3o,;r~obol".A.s :JiI'gim:;:Ju8 29 6 226.86 Eaumea .,/'.AnceJ. 122 37·33 ",~,"!£.rptta c:m6r-£aanus 30 14 224.78 .:~"",)tQnu.7. J.LH3:::rat\;v'i,;:u 100" 126 35.42 ~'cit--puS' r!.cJ():J~t;J 31 .5 221 _28 :::'ypy>,a, iu"l!'in'Jer!si~j 101 33·13 _'.:.ll :'tric).:t.: .;tJgro,alis 32 21 208.20 ;!~CgY'1;:1 ;U$tr>a:"c.-G 1~ 136 32.32 ~:6:a.l.::;".aa ~.]~ ':a 33 61 203.82 3phl;Jn:J1! 103 115 32.16 ?I":l;iJ :?:Jti';",ro:.< i.-:;.t:J 34 32 : 98. ,,4 .Ju!'oat:...; ,,;':ga.n.cea I" 60 29·90 ,:.;neC~~Q "r~ '1 • 'T"", ...~ 35 48 195.36 Phragr'irea _'o!mT'Ar.is I~ 81 29.90 :::!,~'.l;)biWrt (narrow l 36 64 188.01 ;'p"~'JlGcr:in ztY'i.:;;t.:;: 106 130 29.14 :"J tr.:<:ri.::; ~~Q;" L :c:,I"'~;;" 37 13 167.31 ·:'hori::andY'Q ;urrDr:.r1/1 107 100 29. I 1 :.:,Y(i·ptW J;;:';zt;'_'.2f·PUB 38 ,.'0 165.72 ,?-::.!rruncn(Z~$ -~'-:':'~Z'{1',:,3 I~ 33 27.69 1~;rost"i3 39 55 162.55 F'w7'ex b':c'eJ1-,'1 I~ 91 'Z,7 .. j 2 , .;." ,:ntt.; "Ii.de:::

41 Ii 156.23 c,J.rttelZa :'cr,i:'..::';i:<.1 111 .:.7 26.26 'JL<>1.C""".J "~,---"l""J

42 5~ 154.96 .'Ii~"ic. ; 12 III 21;.60 F't'('u;;it.l '":or.(J-).-,)::'

43 67 154.21 '!r~£,Ja. ~~~-:;:,:u I; 3 123 24_60 ,,;'-;ii~s "4 70 15 L 7t 'iat7..ilJr:ersa 114 124 24.60 -CC''nc1 45 24 147.40 H.yJ'J.'oco ty;' 115 138 23.95 46 25 lLt5.97 H'f1U:r,,):.Jar;~Z ~t3C'('G~~ 116 1S 13.47 .j/.<.nCU8 47 22 11.&5.28 Grar;:a:;;:. p.Jr'U~licl'L: I ! 7 45 23.12 fl;:r;pr.o:-d.2J }t'e~.a.;...: 48 129 f35.62 .Cfydrocotyl :Jlabr~ liS B3 20.71 :T'::::'.arsia '..Imt'p·:~otf2 "9 n i33.76 '/il.lc;:I's·£a 6:::a~:;.ar~ 119 121 19.76 50 118 130.64 JiA.nUUS {/I,gr?Yi3 120 90 t9·lj3 'IaZ;;ra.;:!,~.· 51 96 ~ 29.65 Ps€udoroph'iD 121 1,07 18. 19 ;:;ctien'ta Jr·~a:'{.~$

52 10 127.7\ ';G.I."f!.X ~;:preS3,'7. 122 109 113.19 E'Z0oah::Zl'i.a ,:'U8:;'::..:: 53 58 126.27 .)oirrr-A..i: ::numhl~~w 123 92 IS.91 .;::.ai<....uJ ?~$.r:(-'!);;,~·:..ts 54 53 126.19 F?,zst1.'.o -;etf':lpftbtZlu,J 124 97 ,4.42. K(Jrippa 55 38 123.93 :"u.d:..:-ig·~a peploi,;.eE , 25 30 13.90 ',fenr::ha i;d~ql:i:Ja 56 46 119. 1'3 QvC:£f-)iiu 126 ] 12 12. I) :.: W'asol'l,1..UII" "7inus 57 86 '16.92 Poll~gOy;.un s1;1'i:;~3u.w: 127 106 11,38 31V1(JY;ycOm!? ,)dcalf;i:;.'a 58 87 113.69 'J'Jode)Ji;;. :;U'1,CU3 128 94 10',<.5 ,lL:.srnG pl-ar;["J.'dc-,,~q".A,;;;,"t:-;:;2 59 ~ 3 110.64 ,'iast;urc'ium 129 120 9.B8 :'udwi:;ia ;-aiuc';z'i3 60 a9 108.10 ]cftoenu:; DY'8V:CU :,~;J 130 95 9.61 Poi•. pogol" tl'Icrl.areUensis 61 110 107.62 :;~ter'.£1nth~:r·a ienl';ieui-:::::.:r:a 131 131 9.41 J:::irrU3 .zrrtCil'ot-:".:1US 62 47 105.78 ?23Dai:.." 132 lQI 8.20 3'J.W1'ej ti! tJ~ag::m(l 63 15 102.2.6 r::;:::7A.~a :!~l'arl.O:Ji~·~:;;::':l 133 139 7.33 Ha['A'a''JiJ b>o.·,;n:i. 64 108 96.94 \1-zrsii"C! ar!fJ!

Q­ 66 ,j 58.33 ?,Jt&;;tOnur; rin:.ts 136 lOS 5.59 ..J.:;ter Ju.bul;.;:r:-~s 67 19 87 .Q4 E,::,i:()b::~,rr; red 137 5.50 .~a~~:tp·i.Jhe :;,'1r'!1tLc.::a 68 69 85.58 -·'r-;"~1;;,.i" 138 4. 7':) ~"'!'!!')i! :.[.'I'Ci'T'.... s 69 62 83.12 ; 39 116 !.." :"7 ;;l ~ 'wn ; ;~~tJ :',Ji;,;;;":{.!: :­ 70 II l3 _03 141; "9 J.00 • '::;;dtiAJ7T 44.

matrix, a theoretical expected value based jointly on the associated

row and cohron totals is calculated. The deviation of any site/species fran expectation is obtained by subtraction fran the actual data matrix. Large deviations indicate that there is some major heterogeneity associated with the species. The absolute values of these deviations are summed for each species, and provide a measure of the importance of the species in the carmunity structure. See Dale

and Anderson (1973) and Dale and Williams (1978) for a canplete description of the canputational procedure.

"Eident" values were calculated for each of the one hundred and

forty species recorded using the program EID fran the TAXON package

(ROSS 1982). The data had already been canpressed to frequencies by

the program GPRES (see Olapter 2). The list of "eident" values generated was examined (Table 3.1). Large changes in these values may

serve to indicate appropriate cut-off points. The largest change was

fran 256 to 227, hOYJever this leads to a reduced data set of only 28

species and would result in the ranoval of all species records for

some sites. Thus the "eident" value of 126.19 was selected for the

cut-off point Which gave 54 species with equal or higher values for

subsequent classification. At the 54 species level, all the species

Which were considered intuitively to contribute towards the

distinctiveness of wetland sites were inchrled, and most species

recorded for a single site were eliminated.

The distribution of terrestrial species, which ~~re recorded in

quadrats on the margins of wetlands, is unlikely to be influenced by

the same envirounental factors as aquatic macrophytes, and their

presence in the data set might be expected to obscure underlying 45. patterns. It was decided to delete terrestrial species from the data to see if the groups produced were more distinctive than those generated by classification of the complete data set.

Sculthorpe (1967) has discussed the difficulties inherent in the definiHon of aquatic vascular plants. A nunber of authors (Du Rietz 1921, Raunkiaer 1907, Iversen 1936, den Hartog and Segal 1964, Aston 1973) have differentiated aquatic species from essentially terrestrial plants by their dependence on the presence of permanent water. However, many such species are able to grow where seasonal fluctuations in water-table levels may result in gradual transition fram dry through waterlogged to submerged soils. The morphological responses of certain species to fluctuations in water level are well docunented. (Sculthorpe 1967). In the Victorian wetlands sampled, such polymorphisn was noted in Mimulus repens, Myriophyllum propinquum, Potamogeton tricarinatus, Scirpus fluitans and Triglochin procera.

In spite of the diversity of habit and considerable variation in morphology, a nunber of systans based on life-fom or growth-fom have been elaborated for the classification of aquatic species, including those of Du Rietz (1921), Raunkiaer (1907), Iversen (1936), den Hartog and Segal (1964) and Dansereau, Buell and Dagon (1966) (see Table 3.2) • Most of these classifications excluded plants with submerged basal parts but essentially aerial leaves (e.g. Phragmites, Typha) to which the term "helophyte" has been applied. (Raunkiaer 1907, den Hartog and Segal 1964), and semi-aquatic species, i.e. plants lNhich "can probably wi thstand near-permanent shallow waters, and only require periodic temporary inundation for survival" (Aston 1973). Table 3.2 Life-form classification systems for wetland species.

Du Rietz (1921) den Hartog and Segal (1964) Nymphaeids Isoetids Elodeids Vall isnerids Isoetids Elodeids Lemnids Myriophyll ids Bat rach i ids Dansereau (1966) Nymphaeids Rooted (submerged) Ceratophyll ids Rooted (floating leaves) Hydrocharids Rooted (emergent) Stratiotids Non-rooted (submerged) Lemnids Non-rooted (emergent) Ricciell ids Bryoid Crust (floating) Raunkier (1934) Hydrophytes 1. rooted 2. free swimming with roots 3. free swimming without roots

Iversen (1936) Amphiphyten - plants which normally possess aerial and aquatic leaves but which can develop water forms. Limnophyten - "lake plants" include Schwimmblattgewachse (plants with floating leaves) Grundsprossgewachse (plants with submerged short shoots) Wassersprossgewachse (plants with submerged long shoots)

Life form categories distinguished for Victorian wetland species 1. Aquatic rooted (floating leaves) 2. Aquatic rooted (submerged) 3. Aquatic non-rooted (submerged) 4. Aquatic crust (floating) 5. Semi-aquatic 6. Terrestrial shrubs 7. Terrestrial graminoids 8. Terrestrial small-leaved herbs 46.

These definitions suggest that helophytes and semi-aquatic species may differ in their requirements for inundation.

Since a nunber of species recorded in Victorian wetlands were clearly se:ni-aquatic or helophytic, a wider definition of aquatic

species as plants which grow in water, in soil covered with water or

in soil which may be saturated (modified fran VEaver and Clements

1938) was used.

For the purposes of nunerical classification, plants in the

semi-aquatic and helophytic cate']ories were grouped and a modified

version of Dansereau's classification (1966) generated (Table 3.2) to

provide an extrinsic criterion for the selection of species to be

anitted fran the data set. The species were scored according to t"1e

life-form categories 1-8 shown in Table 3.2. The species in

categories 6, 7 and 8 were masked prior to classification, leaving the

species listed in Table 3.3.

The frequency data for the 54 species with high "eidentll values

and the 64 species categorised as aquatic or semi-aquatic were

classified using the same sequence of programs outlined in Chapter Two

( NIASM, RATLAN, OCOM, CRAMER, NEAREST, TABOUT) and used in the

initial analyses of the entire data set. Table 3.3 List of species and their life-form categories used in the aquatic-semi-aquatic classification.* See Table 3.2 and text for explanation of the categories.

Life-form Species Species Life-form Species Species category * no. category * no.

5 9/; Alisma ptantago-aquatiaa 2 /;2 MyriophyZZym verrucosum 3 /; AzolZa fiZiauZoides 2 /;3 Nasturtium officinaZe 5 5 Baumea artiautata 2 /;/; Nitelta sp. 5 7 Baumea J:"!

3. Results

A. Data set after reCiuction by EIDENI'

Application of the Ratkowsky-Lance global comparison of cluster validity to the hierarchy generateCi by the program NIASM suggesteCi that a nine group solution was optUnal. Inspection of the infoDnation change associateCi with the fusion of sites indicated that the groups were less isolated than those produceCi by the first NIASM classification carried out on 1:..'1e complete data. set. The heterogeneity of the nine groups was' variable; group 101, a large grou~ of floristically rich sites had the highest infonnation level.

Groups of floristically poorer sites or snall groups, not unexpecteCily, containeCi less infoDnation.

At the five group level, where the scaleCi Cramer value was slightly lower than that calculated for nine groups, the heterogeneity associateCi with each group was similar, but quite high. The values for infonnation change indicated that the five groups were comparatively isolated. Large infonnation changes were associated with the saline and r::ennanent fresh, brown water groups, and a seven group solution, resulting in equally isolated groups, would split both. It was decideCi to truncate the classification at the nine group level although the infonnation level and infonnation change indicated

that these groups were not strongly isolated nor equally cohesive.

However, this solution would allow a comparison with the eight groups generateCi by the classification of the entire data set. 50207- - -­ ----r------~

4 44 Fig. 3.1 ~ U

~ n Dendrogram produced by NIASM classification of the~iden(value data 31 7 set listing species in order of their magnitude of contribution to

each dichotomy. Only speci~s whose contribution to group heterogeneity was >10% and <5% to the sibling group are shown.

I. Agrostis aemula 41. Myriophy llum propinqUlun 4. A.30lla [iliculoides 44. Nitella sp. 5. Baumeo articulata 48. Phragmites communis

____ 109 6. BalUnea juncea 50. Pot4mogeton tricarinatus 27634- --­ 33 7 7. Baumea rubiginosa 52. Ranuncu lus rl:vu laris 133 50 Carex nppressa Restio tetraplly Uus 40 63 10. 53. 34 18 12. Cllara sp. 54. Riccia sp. 72 14. Claytonia australasica 57. Scirpus fluitans 57 16. CrasBU l" /", lm.q i i 58. SCirpUB inundatus 17. I':l"oo"aris acuta 61. Spllagnum sp. 18259·· --­ @ - 18. I':leocharis sphacelata 63. Triglochin procera I 18 ) M ~ 20. EragroBtis austl'alasiC!Us 65. Typlla domingensis ) 5 21. Glycerin at.straUs 66. Typha orl:ental.is 1 26 II I 72 Ir 26. Jzmcus aJ,ticulatut) 67. u,-Hca incisa I ! 30. [Jemna m1:rlor 68. Utrl:cu7.aria australis

~ ~~ 57 31. Lemna tz-i,w lca 72. Villarsia rl'n):[ormis ~= 12 7735 --- -- Lepi la(',10 b i lOO1daris 78. Mueltlenbeckia cunninghamii 63 33. 14 34. Lepilnmlfl ('yl indrocarpa 96. Pseudoraphis paradoxa 35 20 )~~ ~~:~ ~! ,~4~1,05 'i~ IWJ~ ~~ 35. r,i laeopsi.q ['0 lya,",t/1O "7. Centella cordi[olia 11 ~ 10 72 65 40 7 96 I 48 57 53 99 44 3027 -- -- 58 39. MyriophyUum e7.aUn01:des "8. Juncus ingens 54 100 6 Rup['ia maritima Infolmaticn 26 40. My,>iophyUw" muell"ri 133. level " !i 94 97 69 92 78 81 85 98 93 q6 48.

The derrlrogram

Annotated dendrograms which indicate the species which make the largest attribute contributions to group dissimilarity and the envirormental relationships of the groups at each level of the hierarchy are shown in Figs. 3.1 and 3.2. The first division separates fresh, mostly pennanent, brown or turbid water groups (107), to which Azolla filiculoides, Lanna minor, r:..anna trisulca and

Eragrostis australasicus make large attribute contributions, from the saline groups arrl the fresh, clear or shallow brown water sites (108).

Group 108 is characterised by the high frequency of occurrence of

Chara, Lepilaena bilocularis, Nitella and Baumea rubiginosa. The

saline group (105) which has high attribute values for r:..epilaena bilocularis, Lepilaena glindrocarpa, Myriophyllum muelleri arrl Ruppia maritima, is then separated from the freshwater group (l06) which in

turn has high attribute contributions for Baumea rubiginosa,

Eleocharis sphacelata, potamogeton tricarinatus, Villarsia reniformis,

scirpus fluitans and Triglochin procera. The saline sites (group

105) are further divided, largely on the basis of the abundance of

r:..epilaena bilocularis and MYriophyllum propinquum (99), and Lepilaena

cylindrocarpa, Myriophyllum muelleri, Nitella and Ruppia maritima

(100) • Freshwater group 106 splits into a calcareous group, 101,

which has major attribute contributions fram Baumea rubiginosa and

MYriophyllum elatinoides, and the freshwater shallow group 102,

differentiated from 101 by the abundance of Eleocharis acuta,

Eleocharis sphacelata and potamogeton tricarinatus.

On the other side of the dendrogram, 107 is divided into group 103

canprising the turbid and brown water 5ites which have organic saitneor fresh shallow brown water or deep clear water

fresh turbid or deeper brown waters

109

fresh

saline

107 106

waters fresh mean pH =6,1 Shallow waters waters calcareous mean pH =78

102 104

1031 fresh 101 brown 105 turbId water specIes waters I sandy saline nch f h substrate very res saline specIes brown aCidIc poor water waters ~ organIc substrate I 911 195 94 I~ 691 \92 78 r1 .81 85 98 931 196

Fig. 3.2. Annotated hierarchy for the NIASM classi­ fication of the "eident" data set. 49.

substrates and the fresh, brown water sites wtlich have sandy substrates (group 104), where carmon species inclwe Baumea articulata, Eleocharis sphacelata, Juncus articulatus, Myriophyllum propinquum, Triglochin procera, utricularia australis and Villarsia reniformis. Group 103 is characterised by high attribute contributions for Azolla filiculoides, erassuIa helmsii, Eragrostis australasicus, Lerona minor, Lerona trisulca and Typha spp. Group 104 is subsequently separated into a group (97) of acidic water sites and a group (94) of deeper brown water sites (mean pH 6.6) • Group 103 is subdivided into the turbid water group (91) which has high attribute contributions fran Agrostis aanula, Eragrostis .?ustralasicus and

Juncus ingens, and group 95, whose manbers have brown water and organic substrates. crassula helmsii, Lerona trisulca and Triglochin procera make major contributions to group 95.

A ccmparison of these dendrograms with Figs 2.2 and 2.3 generated by the NIASM classification of the ccmplete data set indicates that there were two major charges introduced by the ranoval of species with low "eident" values. The initial division of the ccmplete data set separated groups with different geographic affinities. In contrast, the reJuceJ data set was initially split into two groups apparently on the basis of water depth, clarity and salinity. This change is thought to reflect the ranoval of many of the terrestrial species frcm the data, species which could be eocpecteJ to indicate geographical affinities. Data reJuction has also chargeJ the relative position of the turbid (91) and brown water (95) groups with respect to the saline groups (99 and 100). In the classification of the entire data set these were sibling groups; in the t"eJuceJ data set, they occupied opposite sides of the deoorogram. Table 3.4 Membership of groups generated by NIASM classification of species with high eident values.

Group Site Group Site number number

99 3 Tremaines Swamp 96 9 Lake Elusive 8 Deep Lake 12 Ewings Marsh 20 Lake Lookout 22 Bittern Lagoon 24 King River Meander 45 Brisbane Ranges 25 Lake Albacutya 28 Lake Wandella 91 18 Two Tree Swamp 38 Swan Lake (E.Vic.) 29 Loch Garry 41 Bleak House Lagoon 31 Middle Lake 53 Lake Struan 33 Lake Kanagulk 54 Lake Winchel sea 50 Little River

100 21 The Swannee 95 I Pal parra Settlement Swamp 27 Champion Lake 6 Floating Island 30 St. Mary's Lake 15 Gelliondale Swamp 37 Bitter Lake 40 Army Reserve Rockbank 94 2 Break Number 2 Swamp 42 Lake Tutchewop 4 Brown Reedy Swamp 51 Round Lake 11 Dock Inlet 55 Lake Terangpom 19 Cockpit Lagoon 23 Victoria Lagoon 101 5 Kanawinka Swamp 26 The Duck Holes 10 Lake Bong Bong 44 Lake Beadle 13 Swan Lake (W.Vic.) 49 Banyule Billabong 14 Little Lake Malseed 17 Dereel Lagoon 97 7 Cobrico Swamp 35 Lake Bridgewater 34 Heywood Golf Club Swamp 36 Cotters Lake 39 Rockbank Swamp 43 Lake Barracouta 47 Mackenzie Road 52 Bay of Islands Swamp

93 16 The Long Swamp 32 The Stony 46 Fernbank Swamp 48 Kangaroo Swamp 50.

Floristic groups

Membership of the nine final groups is shown in Table 3.4. The members of the first group to form, 91, all share Azolla filiculoides.

All the sites in this group have turbid water, and support largely helophytic and floating species. The few sutrnerged species recorded were restricted to transects located near inflow points, or grew on damp moo near the waters edge.

The wetlands in group 96 have brown waters over organic substrates and all contain Azolla filiculoides, Lerona minor, Lerona trisulca and Crassula helmsii. Palparra Settlement Swamp (1), a species-rich, shallow, freshwater site which joins group 95 late in the hierarchy, shares many species with members of the shallow freshwater group 96.

However, the presence and abundance of Azolla and Lerona spp. has resulted in its allocation to 95.

l'Dst of the deeper, fresh permanent water sites which are non-calcareous are in group 94, and are characterised by the preserce, and frequently abundance, of Triglochin procera. The shallow Victoria Lagoon (23) has floristic affinities with members of group 96, but contains species, notably Baumea articulata, not recorded for members of group 96 but carmon in 94. The Duck Holes (26) has similar floristic affinities wi th the shallow water group 96, but its abundance of utricularia australis has place::3. it in group 94. This site (26), although small, supporte::3. three different vegetation types, each sample::3. by a transect. Very shallow water areas supported dense stands of Typha orientalis, slightly deeper 'Haters were dominated by 51.

Villarsia renifor.rnis and MYriophyllum propinguum, whilst the deepest areas ~re fringed by Eleocharis sphacelata. Each transect shows floristic affinities wi th a different group. Such a site cannot readily be accommodated by a model within which sites are uniquely allocated to a group.

Group 97 contains the sites with very acidic waters, Heywood Golf

Club swamp (34) and MacKenzie Road SWamp (47) which are daninated by

Restio tetraphyllus and Sphagnum sp.. Rockbank (39) and Cobrico Swamp

(7), the other menbers of group 97, share one and two species respectively '\>'i th the acidic sites. Rockbank is a turbid water, species-poor ~tland, whilst Cobrico Swamp is floristically related to sites in group 95 which also have brown waters over organic substrates. These misclassifications may be attributed to the effective suppression of double zero matching in the case of Rockbank, and group size dependency in the case of Cobrico swamp. All the

~tlands located in calcareous coastal dunes or on other parent material of a calcareous nature have been placed in group 101, which also contains Lake Barracouta (43) and Kanawinka swamp (5), sites located in quartzose dunes in East Gippsland and south ~stern

Victorian respectively. All sites except Lake Barracouta and Dereel

Lagoon (17) have alkaline waters, but they have no single species in common. Ho~ver, all except SWan Lake (13) contain Baumea rubiginosa, and Triglochin procera has been recorded for all group members except

the frequently dry Cotter's Lake (36). Chara and Nitella fonn

frequently dense stands at all sites except the slightly turbid Lake

Barracouta, dry Cotter's Lake and shallow Bay of Islands SWamp (52).

The inclusion of Lake Barracouta in group 101 is due to the effects of double zero suppression (see Chapter 2). Table 3.5 Groups generated by classification of the eident value

data set showing diagnostic species (shared by all members of a group) at the 9 group level.

Group number Group type Shared species

99 Sa line *

100 Very saline *

101 Calcareous *

97 Acidic *

91 Turbid Azolla filiculoides

94 Fresh brown water Trigloahin proaera sandy substrate

95 . Fresh brown water Azolla filiculoides organic substrate Crassula helmsii Lemna minor Lemna trisulaa

93 Shallow fresh water basins ­ Myriophyllum propinquum floristically poor Potamogeton triaarinatus

96 Shallow fresh water basins ­ Eleoaharis sphaaelata floristically richer Myriophyllum propinquum

;~ No species common to al I group members 52.

GrOUpS 93 and 96 include most of the shallow freshwater basins sampled. The mEJl.1bers of group 96 share Eleocharis sphacelata and

Myriophyllum propinquum, and all except Lake Elusive (9) contain potamogeton tricarinatus, Scirpus fluitans, Triglochin pr.ocera and villarsia renifor.mis. The inclusion of Lake Elusive, a much deeper and floristically poorer site, can probably be attributed to the effects of double ~ero suppression. Group 93 is also characterised by the presence of Myriophyllum propinquum and potamogeton tricarinatus, but all the sites lack Triglochin procera, and are generally floristically poorer.

The saline sites form two groups. The generally more saline wetlands of group 100 (where only two sites had salinities less than 6

0/00) are characterised by the presence of Lepilaena cylind.rocarpa or

Ruppia maritima, both sanetimes fourrl in association with Myriophyllum muelleri. Group 99 includes two misclassif ied freshwater sites,

Tremaine's SWamp (3) and King River M:ander (24). The group also includes Swan Lake East (38) and Bleak House Lagoon (41), which by virtue of their mixture of freshwater and typically saline species cannot be uniquely assigned to any group satisfactorily. The latter two sites were classified with all the other sites containing

Lepilaena bilocularis.

At the nine group level, five groups can be characterised floristically (Fig 3.5). However, if b,e five misclassified sites,

l Tremaine s Swamp (3), Cobrico SWamp (7) I Lake Elusive (9) I King River

M:ander (24) and Rockbank Swamp (39), are removed from consideration, all t.l-}e manbers of each group except 100 (saline) and 101 (calcareous) Table 3.6 Membership of groups generated by NIASM classification of aquatic and semi-aquatic species.

Group Site Group Site number number 89 34 Heywood Golf Club Swamp 97 I Pal parra Settlement Swamp 39 Rockbank Swamp 6 Floating Island 47 Mackenzie Road Swamp 15 Gelliondale Swamp

92 7 Cobrico Swamp 91 8 Deep Lake 19 Cockpit Lagoon 20 Lake Lookout 49 Banyule Billabong 25 Lake Albacutya 28 Lake Wandella 94 2 Break Number 2 Swamp 38 Swan Lake (E.Vic.) 4 Brown Reedy Swamp 53 Lake Struan I 1 Doc kin let 54 Lake Winchel sea 23 Victoria Lagoon 26 The Duck Holes 95 3 Tremaines Swamp 44 Lake Beadle 22 Bittern Lagoon 48 Kangaroo Swamp 24 King River Meander 41 Bleak House Lagoon 99 5 Kanawinka Swamp 9 Lake Elusive 98 10 Lake Bong Bong 12 Ewing Marsh 14 Little Lake Malseed 13 Swan Lake (W.Vic.) 35 Lake Bridgewater 16 The Long Swamp 36 Cotter's Lake 17 Dereel Lagoon 43 Lake Barracouta 32 The Stony 52 Bay of Islands Swamp 45 Brisbane Range 46 Fernbank Swamp 100 21 The Swannee 27 Champion Lake 96 18 Two Tree Swamp 30 St. Mary's Lake 29 Loch Garry 37 Bitter Lake 31 Middle Lake 40 Army Reserve Rockbank 33 Lake Kanagulk 52 Lake Tutchewop 50 Little River 51 Round Lake 55 Lake Terangpom 53. share at least one species in cammon. All sites in group 100 contain

Lepilaena sylindrocarpa or Ruppia maritima, and MYriophyllum muelleri may be found with both these species. Wi thin group 101, SWan Lake west (13) is an outlier. Although it shares most of its species with ot..'1er manbers of the group, SWan Lake west supports no large sedge species. The absence of such species may be attributed to the lack of suitable habitat; Swan Lake lies in a steep-sided valley damned by coastal dunes, and its water level can fluctuate by more than two metres. The freshwater groups (93, 94, 95, 96) are characterised by species which are not uniquely assigned, and which frequently make significant attribute cc~tributions to a number of groups.

B. Results of classification of aquatic and semi-aquatic species

A ten group solution of the classification of aquatic and semi-aquatic species was suggested by the Ratkow~ky-Lance criterion.

At the ten group level, the degree of ccmpaction of the groups I as

indicated by their information level, was very variable. As in the previous classifications, small groups, or those which contained floristically poor sites, had low levels of heterogeneity. However, all the groups were ccmparatively isolated. Group membership is shown in Table 3.6.

The der.drogram

Ccmparison of Fig. 3.3 with the dendrogram generated by the

lI eident" classification (Fig. 3.2) indicates that similar group affinities were recognised, although the fresh brown water sites with sar.dy substrates (group 94) which were linked with the acidic sites in 108 1

fresh

saline iexcept 951

105 ! brown water

clear water I 1061 calcareous .~ I shallow fresh l water 1107

fresh brown water very sandy saline substrate 1104 turt~d I I fresh brown 1011 saline water I organic aCidiC substrate I ' 891 192 96 197 981 94 1991001 911 195

Fig. 3.3. Dendrogram indicating ecological affinities of the ten floristic groups produced by the NIASM classification of the aquatic-semi­ aquatic species. 54.

the "eident" classification had been fused with the shallow freshwater sites (99). An ecological basis for the separation of groups is suggested. However, the presence of groups 92 and 95, which contain environmentally and floristically dissimilar sites, makes the initial dichotomy, 106, 108, and subsequent separation of 106 into 101 and 103 difficult to interpret. Fig. 3.4 indicates the species which mcde the major attribute contributions at each dichotomy. The same species were important in both the "eident" and this classification except where the groups canpared differed. The high attribute contribution from MYriophyllum propinquum, generally a species of fresh water, to the saline groups resulted fran the fusion of "ragbag" group 95 with a saline group.

Floristic groups

Menbership of the saline groups 91 and 100, and the calcareous group 98 was identical to those groups (84, 96 and 97 respectively) generated by the classification of the entire data set. The group of turbid water sites was also recognised, all me:nbers sharing Azolla filiculoides. The misclassification of Rockbank Swamp (39) with the group of acidic water sites (89), with which it shares only one species, is probably due to the effect of group size dependency. The other acidic water sites are characterised by Restio tetraphyllus and

Sphagnum SPa Group 99 inclt:.des most of the shallow freshwater swamps, all of which share Myriophyllum propinquum and, exclt:.ding the species-poor deep Lake Elusive (9), Chara sp., Nitella spa and

Potamogeton tricarinatus. The deeper brown water sites with sandy substrates (Group 94) share no species in carmon be.:::ause of the

inclusion of Lake Beadle (44), another species-poor outlier. All the 46002­ --,

4 44 30 12 20 33 31 Fig. 3.4 16 72

Dendrogram produced by NIIISH class; fication of the aQu8ti<::-S""" -a'1t ,,'tic species data set showing species which made major contributions to each dichotomy. Species are listed in order of magnitude of contribution where their contribution to group heterogeneity was >10% and <5% to the ,ibllng group.

29334­ IO!L

~. /laotta fiti~u7.oides ~1. MyriophyUum prop1:nquwn 63 33 5. 8aumea Q1'ti~ulata ~3. Nasturtium offi~inale 18 1133 40 7. llaumea rubiginosa ~~. NiteUa sp. 72 34 12. Chara op. ~6. Ottelia ovalifolia 68 14. Claytonia t:!Ustralasi~a 48. Phrogmites australis 16. CraRsula hl'lmllii 50. Po tamogetml tri~arinatuo 17. Eteo~hariB a~ta 52. Rallwwulus rivulario 18. Eleo~hariG 8pha~e14ta 53. Reotio tetraphyllus 105 11086- - -~ 20. Eragrost1:S austraLas icus 55. Rumex biden8 18 21. r;ly~eri" austroU" 57. Soirpufl fluHanl) 50 ~I r68 26. ,luneus articuwtUG 61. Splwgnwn op. 57 5 L@mna mi110r 63. Trig~o~hin pl'Ooera 1102 30. 12297- --­ 4 31. ['e,",1;2 tl'isul~a 65. Typha domingemiis ~f~l- 30 4.1 l68 20 12 Lepilaena bilo~ularis 66. Typlw orientatiG 41 33. 31 llQJ I.cpi taenfl "Y 1indro~arl'a 68. Utri~utaria australis 66 34. 53 52 133 33 35. l.ilaeopsis polyantha 72. Villarsia renifof~ip lQ1 34 41 17 [,udtJigia 108. Marsilea angustifolia 20 31 40 38. 44 38 63 104 JunOUS ingens - 41 39 MyriophyHum etat in!'ide" lI8. 118 16 18 .7 52 33 40. /lyriapll!! llwq mtlP 1. tl'ri lB· Ruppia maritima 3554---­ 12 50 61 55 21 ~~n:! 39 35 53 11 48 39 108 14 14 Informfltion 41 :~ 57_ 46 ll?'v~t 89 92 96 97 98 94 99 100 91 95 55. other members, however, share Myriophyllum propinquum, Triglochin procera, utricularia australis and Villarsia reniformis. Group 97 inclooes sane of the sites v.hich have brown water over organic substrates. They are characterised by the occurrence of Azolla filiculoides, Crassula helmsii, Lerona minor and Lerona trisulca. The shallow Palparra Settlement swamp (1) shows sane floristic affinities with groups 94 and 99.

Within groups 92 and 95, the members share one or two species in cammon, (Rumex bidens, Triglochin procera and Myriophyllum propinguum respectively), but are not closely related floristically, and do not share similar environmental conditions. The sites were allocated to separate groups when the entire data set was classified. The deletion of species from these comparatively species-poor freshwater sites has resulted in the formation of groups whose members main shared property is their dissimilarity wib~ other groups.

At the ten group level, all except three groups can be characterised floristically (Table 3.7). Reallocation of t..'1e misclassified site 39 does not bnprove the floristic characterisation. Appropriate reallocation of members of groups 92 and 95 does not reduce the number of groups v.hose members share at least one species in cammon, since the sites to be reallocated contain at least one of the diagnostic species for their new groups.

The classification of aquatic and semi-aquatic species has produced more groups v.hich share at least one species in cammon than the classification of the entire data set, or the data set based on species "eident" values, but the diagnostic species are not uniquely Table 3.7 Groups generated by NIASM classification of the aquatic­ semi-aquatic species data set showing diagnostic

species (shared by all members of a group) at the

10 group level.

Group number Group type Shared species

91 Sa 1i ne LepiLaena biZocuZaris

100 Very sal ine 1:

98 Calcareous coastal Baumea rubiginosa

89 Aci di c water *

96 Turbid AzoZZa fiZicuZoides

94 Fresh brown water 1~ sandy substrate

97 Fresh brown water AzoZZa fiZiauZoides organic substrate CrassuZa heZmsii Lemna minor Lemna trisuZaa

99 Shallow fresh water basins MyriophyZZum propinquum

92 Ragbag Rumex bidens TrigZoahin proaera

95 Ragbag MyriophyZZum propinquum

* No species common to all group members 56.

assigned to groups.

4. Discussion

The effectiveness of the reduced data sets for classification may be evaluated by examining the groups generated, their canJ?a,ctness and

isolation, their floristic characterisation and their ecological

acceptability.

The canJ?a,ctness and isolation of groups is depend.ent on their

infonnation content and the change in infonnation associated with

fusion respectively. The infonnation levels of the groups generated

by both classifications using reduced data sets were variable. As

wi th the classification of the canplete data set, this was attributed

to the variation in floristic richness between sites, and the

variation in the size of the groups. Inspection of the infonnation

change ind.icated that the groups appeared canJ?a,ratively isolated,

although it was difficult to decide how big an infonnation change was

requirE!d for groups to be considered isolated. CanJ?a,risons of group

camJ?a,ction and isolation between the classifications are difficult to

make, since the deletion of species has resulted in different total

information contents for each data set.

Classification of the data set based on aquatic and semi-aquatic

species provided the better floristic characterisation of the wetland

groups, with seven of the ten groups sharing at least one species in

ccrnmon (Table 3.7). However, manbers of group 95 all shared

l'1yriophyllum propinquum, but were otherwise floristically and

environmentally dissimilar. Characterisation of a group by a single, 57.

ecologically wide-ranging sp;!Cies may not be very informative. Groups which shared more than one species in cannon terrled to be ecologically more coherent.

TO sane extent, the usefulness of a classification is dependent on how 'Well the classes can be characterised, since a clear cut definition will allow the best possible allocation of additional data. williams and Dale (1965) have suggested that all members of a class should have at least one attribute in cannon. Monothetic classifications, where the presence or absence of a single attribute is the necEssary criterion for class membership, and where thus no sample outside the class can contain the attribute, provide clear, unambiguous class definitions. Monothetic definition simplifies assigrment to classes and provides useful class labels, but often little more can be said about monothetically generated classes than can be said by the monothetic definition itself (Hoge

1981) • On the other hand, polythetic definition sacrifices the necessary criterion, which may result in classes whose members do not share enough properties to allow a ccmnon characterisation. The monothetic criterion where no sample outside the class can contain the attribute in question, is not always met by polythetic characterisation. A canpromise, oligothetic characterisation, which is expressed in terms of a small number of properties, may provide a more useful method of class definition, and would simplify

interpretation.

An ecologically acceptable classification generates groups which are recognised intuitively, and shows how the groups are related to each other. It provides an optimal assigrment of sites to groups 58.

wi thout misclassification, or the generation of a "ragbag" group. The grouping may also display previously unsuspected relationships. In

the "eident" classification, the allocation of Dereel Lagoon (17), located on Tertiary sediments, to the calcareous group of sites, pranpted re-examination of the geological map, which showed that the

Tertiary parent material was also calcareous in nature.

Both classifications using reduced data sets identified two saline groups, a calcareous group, a turbid group, a group of acidic water

sites, the shallow freshwater sites, brown freshwater sites wi th a

sandy substrate and a group with organic substrates. However, classification of the aquatic and semi-aquatic species produced two groups of floristically dissimilar sites 'Which v.ere ecologically

unacceptable, as well as misclassifying Rockbank SWamp (39).

Classification of the species with the highest "e identll values

produced ecologically acceptable groups, although the separation of

the shallow freshwater sites on the basis of their infoDnation content

into species-rich and species-poor sites may be difficult to justify

ecologically. Within these groups five sites were considered to have

been misclassified.

The term "misclassification" implies that the mathematical mcdel

adopted does not corresporrl to the biological mcdel it is intended to

copy (Lambert and Williams 1966). In the present study, sane

misclassifications have been attributed to the effect of double zero

suppression. The NIASM algorithm effectively suppresses double zero

occurrences since it only uses presence in the calculation of

heterogeneity (Dale and Arderson 1972). However, double zero

suppression has resulted in the fusion of same species-poor sites with Table 3.8 Sites allocated to the same group by classification of each of the three data sets, and * those considered misclassified in one data set which may be satisfactorily reallocated on the basis of their floristic affinity. Species in brackets were recorded at all the sites within a group.

Saline (LepiZaena biZoauZaPis) Calcareous (Baumea rubiginosa) 8 Deep Lake 10 Lake Bong Bong 20 Lake Lookout 14 Lake Malseed 25 Lake Albacutya 35 Lake Bridgewater 28 Lake Wande II a 36 Cotters Lake 38 Swan Lake East 43 Lake Barracouta 41 Bleak House Lagoon * 52 Bay of Islands 53 Lake Struan 54 Winchel sea Lake Turbid (AzoZZa fiZiauZoides, except 39) Very saline (LepiZaena cyZindrocarpa 18 Two Tree Swamp and/or Ruppia maritima) 29 Loch Garry 21 The Swannee 31 Mi dd I e Lake 27 Champion Lake 33 Lake Kanagulk 30 st. Mary's Lake 39 Rockbank Swamp 37 Bitter Lake 50 Little River 40 Army Reserve 42 Lake Tutchewop Fresh shallow water 51 Round Lake (MyriophyZZum propinquum~ 55 Lake Terangpom Potamogeton tricaPinatus) 12 Ewing Marsh Fresh brown water sandy substrate 16 The Long Swamp (TrigZochin procera) 22 Bittern Lagoon 2 Break Number Two 32 The Stony 3 Tremaine's Swamp * 45 Brisbane Range ~'t 4 Brown Reedy Swamp 46 Fernbank Swamp 9 Lake Elusive * I 1 Dock Inlet Fresh brown water organic substrate

19 Cockp i t Lagoon ~'~ (AzoZZa fiZicuZoides~ 23 Vi ctor i a Lagoon CrassuZa heZmsii) 16 The Duck Holes 1 Palparra Settlement Swamp 44 Lake Bead1e 6 Floating Islands 7 Cobrico Swamp Acidic (Restio tetraphyZZus~ Sphagnum sp.) 15 Gell iondale Swamp 34 Heywood Golf Club Swamp 47 Mackenzie Road Swamp 59.

unrelated and floristically much richer sites on the basis of a single shared species. These misclassifications may have been avoided by the use of an effective two parameter method, such as 'IWINSPAN (Hill

1979) or 'IWOPAR (Macnaughton-Smith 1965), which allows both sites and species to contribute differentially to classification.

A canparison of the groups generated by each of the three classifications indicates that, apart fram minor misclassification, the manbership of the saline groups, the acidic and the turbid water groups, and the brown water organic substrate group is canparatively stable. However, manbership of the shallow freshwater, calcareous and fresh brown water sandy substrate groups is dependent on the data set classified. EXamination of the tables of group manbership for each classification, (Tables 2.2, 3.4 and 3.6) indicates that these groups contain a stable "core" of sites whose allocation remains unchanged in all three classifications (see Table 3.8). Tremaine's

Swamp (3), Cockpit Lagoon (19), Bittern Lagoon (22) and Brisbane

Ranges (45) may be successfully allocated to one of these groups on the basis of floristic affinity and shared diagnostic species.

However, six sites, Kanawinka Swamp (5), swan Lake wast (13),

Dereel Lagoon (17), King River M::ander (24), Kangaroo SWamp (48) and

Banyule Billaboo:; (49), were allocated to different groups by each classification. The floristic canposidon of each of these sites indicates that they have affinities with two or more groups (Table

3.9), and that unique group assignment may not be the most appropriate way of classifying these individuals. The term "fuzzy set" has been proposed (Zadeh 1965) for a class of objects which does not have a precisely defined manbership criterion, but where degrees or grades of Table 3.9 Sites allocated to different groups by classification of the entire, "eident" and aquatic species data sets.

Site Name ent ire lIe ident" aquatic species number

5 Kanawinka Swamp fresh brown water calcareous shallow fresh water sandy substrate

13 Swan Lake West fresh brown water calcareous shallow fresh water sandy substrate

17 Dereel Lagoon fresh brown water calcareous shallow fresh water sandy substrate

24 King River Meander turbid sal i ne 'ragbag '

48 Kangaroo Swamp acidic shallow fresh water fresh brown water sandy substrate

49 Banyule Billabong brown water fresh brown water 'ragbag ' organic sandy substrate 60.

membership may be recognised.

Difficulties experienced in uniquely assigning sane of the wetlands sampled to a group suggests that wetlands may not have been the most appropriate unit to classify. Transects sampled wi thin the wetlands may have provided a more useful unit for classification since it is expected that the vegetation recorded along a single transect is responding to more localised factors, and thus may be markedly less heterogenous.

5. Cooclusions

The three classifications examined, i.e. those using the complete data set, deleting species with low "eidentU values, and deleting terrestrial species, effectively distinguished eight major groups of wetlands, although allocation of sites to groups was not always optimal. This was due to the intrinsic properties of the Infonnation statistic model used and to the nature of the data. The highly heterogenous data, the between site variation in floristic richness, and the ~esence of a number of species with ecologically wide ranges made appropriate allocation of all the sites difficult. Additionally, a number of sites appeared to have floristic affinities with more than one group. TO provide a better floristic discrimination between groups, it may be necessary to classify abundance data recorded at the transect or quadrat level. 61.

CHAPTER FOUR

Ordination of wetland sites

1. Introduction

Ordination techniques, methods which arrange samples and species in relation to one or more continous variab~es, have been used in the description and interpretation of vegetation to identify directions of community compositional variation which can be related directly to environmental gradients, and to identify and typify groups of related samples and species (NoY-Meir and Whittaker 1978). Approaches to ordination include methods which seek to order vegetation samples along recognised environmental gradients, and those which seek patterns in vegetation data directly.

However, axiomatic properties of the models from which many of the latter ordination techniques have been derived restrict the interpretation of the results. These properties imply that species distributions along environmental gradients are monotonic, independent of one another, and related linearly to causal functions (Whittaker 1967). These assumptions are untenable, since the response of species populations to environmental gradients are known to be Ilcomplex ~ individualistic, predominantly continuous, non-linear, and over sufficient ranges of gradients nonmonotonic (w"hi ttaker and Gauch

1978) • 62.

The linear treatment of non-linear relations of species populations results in distorted ordinations whose axes may not be interpretable in tenns of environmental factors. The nature of the species response curves along environmental gradients, and the non-linear decrease of sample similarity wi tl1 increasing sample separation, properties intrinsic to vegetation community data, result in curvilinear distortions to ordinations; a sequence of stands along a single environmental gradient will form a complex curve in three or more dimensions (Swan 1970, Noy-Meir and Austin 1970). In cases of severe distortion, involution or "folding in" of the stands near the ends 0= the axes may occur (Austin and Noy-Meir 1971). Such curvilinear distortions result in axes which are not optimal reductions of dimensionality, and may be misinterpreted by giving spurious axes biological meaning, often invoking causal factors, or by accepting a sequence of points along a real axis whose ends have been involuted as representing an ecological gradient (Austin and Noy-Meir 1971) •

However, van der Maarel (1980) has pointed out that some curvilinear configurations (produced by ordination of transformed data) are interpretable in tenns of ccmplex environmental gradients. According to Noy-Meir and Whittaker (1978), the residual non-linearity of field data which has a small range of sample variation, or has been transformed and a suitable sL~ilarity measure chosen will result in a simple curvature in relation to the second axis. Interpretation of second and higher axes, however, may be obscured by canplex cambinations of curvature effects with actual separation of entities along secondary ecological gradients and sample error (Gauch, Whittaker and Wentworth 1977) • 63.

A further limitation of ordination techniques is the lack of an appropriate test of the significance of eigenvalues Which may be used as a stopping rule. Application of a significance test invokes assumptions of multivariate nODnality of the data (Williams 1976b); such assumptions are unrealistic for ecolog ical data. Bartlett's

(1950) probablistic test of the hypothesis that all components after the p-th are not significantly different in their variances, is dependent on the assumption of nODnality (Noy-Meir 1971). Dale (1975) reviewed proposals for the deteDnination of the number of axes required for adequate redescription of data. In practice, empirical criteria may be employed to identify the number of meaningful axes.

An abrupt decrease in the magnitude of consecutive eigenvalues may provide a useful indicator (Dale 1975) •

In spite of the acknowledged limitations, ordination techniques provide a useful approach to the interpretation of vegetation data, particularly When it is applied to situations Which have a narrow range of vegetational and envirotnlental variation Where the assumptions of linearity hold as a first approximation (wl1ittaker

1967, Greig-Smith, Austin and Whi tmore 1967), and Where the detailed interpretations are limited and take account of the possibility of non-linearity (Austin and Noy-Meir 1971). The use of appropriate similarity measures and certain data transfoDnations have been shown to reduce curvilinear distortion (see SWan 1970, Hill 1973, Hill and

Gauch 1980), but no known transfoDnations completely solve the problems invoked by the assumption of linearity. An alternative approach is the developnent of non-linear ordination methods.

Noy-Meir and Whittaker (1978, Table III) provide a surrmary of such 64.

method.s, but all appear to have considerable limitations in application •

The data set to be analysed here has a wide range of vegetational am environnental variation. However, lithe problan of x;on-linearity is largely irrelevant if the purpose of the analysis is to provide a low-dimensional picture of variation, and in particular, to show and typify groups of related samples" (Noy-Meir and Whittaker 1978).

Ordination techniques have been successfully applied to data sets after classification in order to examine the floristic relationships between stands (del Moral 1975), and as a check on the classification of stams (Greig-Snith, Austin aoo Whi tmore 1967). For these purposes

Williams, Dale and Lance (1971) have suggested an extension of the

Principal Coordinate Analysis (PCOA) of G:>~r (1966, 1967), which allows examination of the relationships between the groups (and their members) obtained by prior classification.

2. Choice of strategy

The mod.el chosen for ordination of the ~t1and sites was principal

Coordinate Analysis (PCOA) of G:>~r (1966, 1967). FCOA utilises a dissimilari ty matrix which may be deri lied from most similari ty measures, incllrling non-Euclidean and after transformation, non-metric measures (Noy-Meir am Whittaker 1977). In addition, the choice of an appropriate similarity measure results in the transfer of some of the non-linearity to the coefficient (Dale 1975), thus minimising tlle distortion of the original distances. The lack of restriction on the choice of similarity measure also provides the opportunity to stress

features of the data thought to be of particular ecolog-ical importance 65.

(Clifford and Stephenson 1975). ECOA may also be used in situations where the matrix is not known to be positive semi-definite, although the resulting ordination space may not be everywhere "real" (Williams,

Dale and Lance 1971) •

A further major advantage of ECOA is that it can be applied to a similarity matrix without recourse to the original data. Suitable matrices should be symmetrical, with the principal diagonal elements set to unity or zero; however, diagonals may have any values in in

ECOA (Ross 1982), and a limited asymmetric extension has been proposed

(Williams, Dale and Lance 1971) •

A more recently developed technique, Detrended COrrespondence

Analysis (DCA) (Hill and Gauch 1980), which provides both species and sample ordinations, attempts to eliminate the curvilinearity inherent in most methods by adjusting the scores of segments of axes by subtracting a local mean value. In addition, the axes of species ordinations are rescaled to preserve ecological distances, so that distances in the ordination space have consistent meanings in tenns of cOllpositional change. The program is able to handle large data sets, and also includes an option for down-weighting of rare species. Tests using field data indicate that the technique can provide successful ordinations of cOllplex data sets.

Data transfonnations and similarity measures

Data transfonnations, nonnalisation techniques which change the effective weights of entities, are commonly applied to data prior to ordination. Choice of transfonnation is deJ.Jendent on the nature of 66.

the data and the purpose of the analysis (Austin and Greig-Smith

1968) • Further, the success of the ordination may depend on the appropriateness of the transformation (and the similarity measure) applied. Data centering, where the mean for the species or the sample is subtracted, removes the effects of contrast in mean abundance between species (Noy-Meir and Whittaker 1978). However, Noy-Meir

(1971) found that non-centera:l principal Ccmponents Analysis ordinations had SOlIe interesting properties, and provida:l SOlIe information which was lost in centered ordinations.

standardisation of species scores allows the specification of weights to be given to information frcm different species.

Non-standardisation weights the analysis in favour of the daninant species. This may be particularly desirable if it is thought that the major species were measured more effectively, and are thus more significant for the ordination than less abundant species (Noy-Meir and Wni ttaker 1978). Sample standardisation is used to remove the effect of different sample sizes on the distance calculations

(Whittaker and Gauch 1978). Various standardisations may be applied to species or samples, or both species and samples may be starrlardised •

In the following analyses, data were not initially centered or standardised, since any data centering used with FCOA is implicit in the dissimilarity measure, and it was desira:l to give the greatest weight to the daninant species. It was considered that the calculation of percentage frequency of occurrence of each species and subsequent scaling by 100 carria:l out by the prCXjram GPRES (Ross 1982) effectively normalised the data prior to classification. The choice 67. of similarity measure was pre-anpted by the use of the distance matrix generated by the NIASM classification program (see Chapter TwO), which allowed inspection of the relationships within and between the groups generated by this classification. The TAXON program PCOA (ROSS 1982) was used to provide ordinations of three data sets, the entire data set and the reduced data sets generated fran examination of the

"eident" values of the species recorded and the aquatic and semi-aquatic species (see Cl1apter Three). The diagnostic program BACRIV (Back Correlation of Individuals on Vectors, Ross 1982) was used to examine the attribute contributions to PCOA vectors by calculating a product manent correlation coefficient between each vector and individual attribute contributions.

Axes were selected for examination by plotting the eigenvalues for each eigenvector. The resulting graph was examined for a marked decrease in adjacent terms followed by a gradual decrease in the size of the remaining eigenvalues. Eigenvectors with eigenvalues below the gap distinguished were discarded.

The program DECOR (ROSS 1982) provided a Detrended Correspondence

Analysis ordination of the samples (wetlands) and species from the canplete data set. The option for down-weighting of rare species was employed since the presence of samples with rare species was likely to distort the analysis. DECOR nomalises data using both species and site totals as part of the canputation, and as r:ecarrnended by Hill and Gauch (1980), for the first ordination of field data the analyses were carried out on untransformed data. Rescaling was applied to maintain relative ecological distances between the sites. Table 4.1 Species which made large attribute contributions to PCOA vectors derived from ordination of the entire data set, the 'Ieidentll data set and the aquatic-semi-aquatic species data set. Values are product moment correlation coefficients between each vector and attribute calculated by the program BACRIV.

Entire Data'Set 'Eident' Data Set Aquatic-Semi Aquatic Data Set Vector No. 2 3 2 3 2 3 Species Azolla filiauloides .6079 -.5289 .5755 .5815 .5373 .6055 Eleoalurris sphaae~ta -.5707 -.5806 -.6005 Sairpus j1uitans -.5463 -.5395 -.5578 Villarsia reniformis -.5203 -.5179 -.5457 Potamogeton triaarinatu8 -.5099 -.5085 -.5303 Lemna minor .4733 -.5796 .5871 .6249 Trigloahin proaera -.4082 -.5689 .5427 .4952 -.6028 .4405 -.6045 Chara .5423 -.5709 -.6022 Eleoaharis aauta -.5205 .5286 .5381• Lemna trisulaa -.4729 -.4708 .5045 Myriophyllum propinquum -.4936 -.4797 .5166 68.

3. Results

Principal COordinates Analyses

The first four axes produced by Principal COordinates Analyses of the entire data set, the "eident" data set arrl the aquatic and semi-aquatic species data set recovered 35.4%, 40.7% and 42.6%

respectively of the variation. The slight increase in the proportion of variation recovered over the three ordinations was due to the decreases in heterogeneity which resulted fram application of the data

reduction techniques. The camparativdy low percentage of the variation recovered suggests that the ordinations were not able to adequately reflect the complexity of the compositional relationships within the three data sets. Low recovery of variation may thus

indicate that the data either contain a large proportion of randan

infoonation (noise) (Austin and Greig-smith 1968) or that there are a

large nunber of effective factors. It is not obvious how these

alternatives may be distinguished. Recovery of substantially similar

patterns using either all the data or a subset of less than 40 percent

of the species suggests that the data is in part "noisyll. However,

since the variation recovered was still low even for the reduced data

sets, it is likely that L~ere are a large number of effective factors.

The scatterplots produced for tte two reduced data sets were very

alike. As stated above, a similar pattern of relationships was

irrlicated by the plot produced for the entire data set, although

positions along L~e secor:rl axis were reversed. Table 4.1 shows the

species contributions to the first three vectors of each ordination.

The species and the order of magnitude of their contribution to the I Ll- I 2 ] ...... * ...... calcareous group 97 _____..../ .35 -36 ~

i~f-, • 10 ~O'\ " . • I \ / # '54\ .... --,': .....":./'/ Ii----LveryI saline group 96 _... tar basinS. '..... I 1"'9/.. I.: "'/'30 ·51 '.53 i 'ho"~ f._,w' ~ .. ,.., ..' " V i' 3/' ~'.Th '--- . group 102 ~.. .17.. . .. / ,...... /1 ,...... \ '."'. 55 + " ----sahne group 84 ./'" / i '14: / " ..' J ... ." '. ."" . ... .]!~• ---J f --2fJ...... \..'-/_i . . . ,..,...... ,../""...-:..' ... ,.,"! ',-38..-'~.+,._ / / ,,~ .., ~ / / . ""-". + . /------.... ,.,'" '. ~~;. t 21' .~;, g".p ~ . ....,....' ./ /".:-, .." ..4[:"'.-< ( i"46 32. ,." ! /'39 1() . "::-~'":.:--\ ,'; '\ turbid group 99 '5 45')_l \ r··· .....' 34. '24 \ 0 .. -" / '::.\E!..\. ~ '43./ .. _ ..... _ . ..../ 22 r ',.....' :.•7 . \.\ 1/-: \ .... _/ \ ". \ -...... __..... ____ ..... '­ ' "."'.... ------,'J '- \ ." --~ --...... ---- '" ..... """---. '. '. ----- ...... '31 .50 ~ ...... '. \ >'" -- ../ . °3 \ \ .33 I ...... 6)

"" '-- .\1 ","'19..I" "-"., / / l / .. " .4 ""': 0 49\, '\. / \ ·2 ·23 ..-".- ."-.,. '\...... '" '18 / ... fresh brown watar ~ ./. ". "'-. "/ : LfreSh brown water sandy substrata '\ /" '""'. . ... , ___ ./. organic substrate "~P "I>. .... '''' ".' ,ro", 11>1 \ I . ' \ , """\ .~ / / '. : " ./ \ .26 j ''''- .. ~1~...... - .. ' L '-

Fig. 4.1a. Scatterplot generated by PCOA ordination of the entire data set. Isolines indicate areas of ordination space occupied by members of the eight groups generated by classification of the data. Fig. 4. lb. Overlay showing group boundaries after reallocation of misclassified sites. * indicates sites identified as outliers from the NEAREST program. Circled sites are the IIfuzzy" sites allocated to different groups within each of the three classifications performed. IF """" " .",,~oo. 97=1:.----;1*35 *~ 'J i *10 j J~,-,i ~ ..:/ ,/'( ....,,""'/ ~l./) · -54 shallow freshwater basins /.... / ~j />7' i/ -".. \ y group 102 ~....~: ..../ / /1 t.:" ,!/Jo _51j:-:--ver saline group 96 ,.' 'Y 1 -14:' / lJ} Ii/ 37* it!.~'53 !I ,.' /.",r'" .. ' l""""/'" /' -;.~.. ~ ..~.. t;,'-!I... 'I 1,.-:--=1j}~ ~'1---~ 5 i-saline- group 84 :,/:,r-~_.;;;;.::.;--:...;;.._==,;;.:_"'/)..:~'52 If: '421~ ~'i38V • } I :( ~6*.\:' _ _ _ 32- I",; *16 ", f\ ~ ..... I _ acidic group 98 I ~ .. , I ------If c:yt·)-'I Ii: : "1"'/ I"::)~ turbid group 99 10 \. /~ _/ :~ >'" (I.43!J! " 'll12* _::--" ~" , \ ~'·7~. \. \. --- ;:,;:;=;.-;.:;;.- :?,,; ' .~------) ''J,''. . ;.;.o:.=--~.,. \. ''\. ,.'- , - . ... ". ~ '3 " \ \ '29 ~~"j'"~ '\.." ""-'''-'''~ r- \'" \ \' «," '''!33) .) "'•. 1 ·4 _11 44* '19 ; fresh brown water '1 ·2 '23 ".-.? '--". \ II J .:., .- jl sandy substrate':' .-'--;;;;'­ group 100\ ..·7 ·S / /' ,/'" _" w"., 't'>""'- ...."'- ... ~~'" ;f f- ....fre m•. substrate \ II ... " ".... "= - .. ... ",00, w, 'l 1/ ·····0\:'~/1 \ ." ) '15.-:;'/ "=J ....~ ~ -dJ

Fig. 4.la. Scatterplot generated by PCOA ordination of the entire data set. Isolines indicate areas of ordination space occupied by members of the eight groups generated by classification of the data. Fig. 4.lb. Overlay showing group boundaries after reallocation of misclassified sites. * indicates sites identified as outliers from the NEAREST program. Circled sites are the "fuzzy" sites allocated to different groups within each of the three classifications performed. 69.

vectors are similar for each data set, although ijYriophyllum propinquum and Lemna trisulca made a slightly larger contribution to the third vector from the aquatic and semi-aquatic species data set.

The scatterplots were examined for infoonation about the floristic affinities of the misclassified sites and the six sites Which could not be uniquely allocated to the groups generated by the NIASM classifications. The relative distinctiveness of the groups was also assessed. However, since the recovery of variation by the J?<::OA procedure was comparatively low, the conclusions must remain tentative.

Figs 4.1, 4.2 and 4.3 display the floristic relationships between the sites as indicated by the site loadings on the first two principel coordinate axes. Bleak House Lagoon (41) I misclassified in the entire and aquatic species data sets, clearly belongs to the saline group of sites. Cockpit Lagoon (19), considered misclassitied wi thin the aquatic species data set, and Tremaine's SWamp (3), misclassitied in this and the "eident" data set, appear to be most closely related to members of the brown freshwater sandy substrate group. The Long

SWamp (16), and Bittern Lagoon (22) share the two dimensional ordination space of the shallow freshwater group to Which ~~ey were allocated in two classifications. The floristic affinities of the remaining misclassified sites, Cobrico Swamp (7), Lake Elusive (9) and

Rockbank (39) are less obvious. Cobrico and Rockbank SWamps occupy the same quadrants as the brown water organic substrate and turbid groups respectively, but appear not to be closely related to members of these groups. A data artefact, the result of noonalisation of the data frcm the two quadrats sampled, has resulted in the inclusion of 70.

Lake Elusive (9) in the shallow freshwater group in two of the three classifications. It could be more appropriately allocated to the fresh brown water sandy substrate group.

The wetland sites mich were regarded as having floristic affinities with more than one group (Table 3.9) occupy positions on the margins of the groups to mich they were allocated by each classification. In the scatterplots for the entire data set (Fig 4.1) am the "eident" data set (Fig 4.2) the position of these sites causes an overlap in the freshwater groups. Ordination of the data set containing only aquatic and se:ni-aquatic species suggests that the classification of this data set providoo more distinctive groups.

Examination of the water chemistry, geology and substrate data suggests that the complex floristic affinities expressed by the sites may be relatoo to habitat factors. This point 'Hill be pursued in later chapters.

Scatterplots for all three data sets suggest that the two saline groups recognisoo in the classifications overlap, at least in the two dimensional representation, am that the distinction made between than may not be useful. The turbid group of sites appears to be relatively compact apart fran species-poor outlier Roclroank (39). 'rhe sites grouped as brown water wetlands with organic substrates may not be very closely relatoo floristically. The e.vo acidic water sites,

Heywood Golf Club SWamp (34) and f-iackenzie Road Swamp (49), do not form a compact group. Although they share the same dominant species,

Restic tetraphyllus and Sphagnum sp., which were not recordoo for any other sites, site 34 is floristically much more heterogenous. I fresh brown water '-"'" b wn water / .r ""'. fresh ro bstrate ...... 2 / '15..' ~ "./organic substrate sandy su ______/ ·26 '" to / group 95 group 94 __: '. . /' . y '" :. .1 ". . . /' ~--.'~ ""'.\ I '-.... / '18 ~. . / "\. / '" \ \ ,49 \ / \ \ \ I 23. // ·31 ~.' ·6 .: '2 '29 "~' ·11 '33 shallow freshwater basins ·4 44­ 50. ) species rich -.:.:=.. :::..--c-../ __ ~ __ __ turbid group 91 ("3-~i~·-( "'-- --- :::-­ group 96 ~ _- .... - . , , " , " ·12 \ ,..,..---, I \ " .22 \ -'-., '\ \ I ...... I.. \. aCidic group 97 " \ / .::. ·13 \ \ , ,1 ;,...... , " ' ... -"" ,./' 48. '43 ~ "\. \ \ ___ ------j>,.45 ~,/ ,16 ."----'\ \ shallow freshwater basi~ ,------32' 1[) species poor ~. ;',' 5- ~~.46 .,-,...... ,.._1_"', :',.. ),.,\\ '\ .-'"21· / n 1'_.,. ____ saline group 100 group 93 '. -., _,...... ' "": \\ l<' \ ~ y-­ '. '.'~"-I--I ::/ . _.­ ... ····..• 9}\ .. '42 \ \. .38"--'41 .28'. ) "\ ... ' 52' ""'( ,._._._ 20' '/'55 + -'" ...... " (7} ._._. ..,;

calcareous group--->'" 101 ·17 .";- ". '. 99 ". \ _30 51' \~."\\.8 \ very sa ,;~ .~"' "', '.... J",- ~ ...... ":.~ \. 3G- ", .. \ • '\ "10 /!. '54 . ·10 ~. I '-.... / ,.­ / ./ ('25',._...... /' 35.

L ~

Fig. 4.2a. Scatterplot generated by PCOA ordination of the "e ident" data set. Isolines indicate areas of ordination space occupied by members of the nine groups generated by classification of the data.

Fig,4.2b. Overlay showing group boundaries after real1ocation of misclassified sites. )'( indicates sites identified as outliers from the NEAREST program. Circled sites are the Ilfuzzy" sites allocated to different groups within each of the three classifications performed.

__:~~_••_ ~ ,~c~~~~ ib_~" '1'''''~t' ~ ';;4;;~':~'~:L~-'-'-:-:'~'~? F -. group 94 ______.26 ~~ 6 I~'/ .- Vfreshorgantc b~own substrate water f;:~~:':;:t;::;er r;;;,;;.---...... ~~ t2 ~., I ",.. \ .. ..,* '\. ••••, 9'

\'" /' '18 \

23.., ) ~~) -31 \ shallowspecies freshwaterrich basins \.'- -4 -11 44' -19 /.//" *29 J'.'~' ."." .. ______, _ ._. _. ___ ~\\ ____.I;;_.C~:( "'" (f.:.) i J ~ --.,." /turbidI group 91 \ ;":,1':'\ ,,-\ i -­ . \ ;1 C~~..·.... ·,··...\ \\ : *:(34 ~ ~.;, ,.00' ;;-­ - ./ ______~). e ~...... /' ,,~• ... \JQI43 . P\' \ . -39 sha~~O:Ci::e~~~ ~4~ ____ --/ 32. .1'l--' i ttr4.2,\~...... "':'~ "00' OJ \eh'''* .~...----...J j t l!\\ 1C> "'" ~~-';"~ ...··1 \.. ,\ ..... \ .14 ------> ... calcareous group 101 . very saline group 99 ". e \ , ..~ .... ". \ .m 3 )'\:.~ \ I ~ \~( b ,.J

Fig.4.2a. Scatterplot generated by PCOA ordination of the "eidentll data set. Isolines indicate areas of ordination space occupied by members of the nine groups generated by classification of the data. Fig. 4.2b. Overlay showing group boundaries after reallocation of misclassified sites. * indicates sites identified as outliers from the NEAREST program. Circled sites are the "fuzzy" sites allocated to different groups within each of the three classifications performed. 71.

The isolines drawn on the scatterplots indicate the extent of the two dL"TIensiona1 space occupied by the groups generated by each classification. They show that within the classification of the entire data set there was no clear discrlinination between the shallow freshwater and the fresh brown water sandy substrate groups.

Ordination of the same sites using only those species with high

"eidene' values sho\Ed that these two groups could be separated, but that there was a degree of overlap between the shallow freshwater and calcareous groups and the fresh brown water sandy substrate and turbid groups (Fig 4.2). In both classifications the pattern of overlapping groups is largely attributable to the six sites which had floristic affinities with more than one group.

DetrendErl Correspondence Analysis

The scatterplot producErl from the DECOR ordination is shown in

Fig. 4.4a. The shallow freshwater, fresh brown water sandy substrate and calcareous groups are not clearly separated (Fig 4.4b). This may be due to the presence of extreme outl iers (Rockbank SWamp 39 and r-'ackenzie Road SWamp 47) in the data which may have resulted in distortion of the ordination space. Hill and Gauch (1980) noted that

Detrended Correspondence Analysis sho\Ed "persistent difficulties in copiIl3' with outliers". An examination of the seven nearest neighbours of the 55 sites indicated that 13 of the sites in the entire data set were not listErl as nearest neighbours of any site. These sites can be considerErl to be outliers. It is thought that an ordination of only those sites wi thin the freshwater complex of groups would have providErl a more informative result. .,..--... ./····~15·" r .../" \ I : ""-'" fresh brown water fresh brown water ( ... -.., 6:/ / . \...... ---- organic substrate sandy substrate______. ! ·26 \ 2 (.1 " \. '\ group 97 group 94 ~" ...... -/ \ '. : \ /~-~ . \ / .18 \ . -6 \: .' : \ // \ \ ..... / J \ ..'...... "/ .31 \ / '23 '---- . ___ '" •••• '49 \ 50' ·33 ~bid group 96 ,2.11 .44)..··...... \ .29 __ ./ 1 -4 .' • -19.·... _ :.'-. .. -'" '-" 1':;-:~·······1 ....'\------7: - -, ragbag grOUP~95'\ '1\I: #'1- ---_I ...... ' __ragbag group 92 / ·12 '\ 24' , '\ .\'~.... ( '~.34 \', ". 48 ....1... ~ '39 \ ", ----,\ • :;"43 ". \ '\ '-~------'16 1\.J)... ". \ '- acidic group 89 " '46 13. ~\ ". "'" _ " ------1\ /A'7=.-, ,,---.,. "'\" ("--..... , I - 22 / .9 I '. _ • shallow freshwater basins \ ','. _// I .... _\ '\.. .21 11)

I ~, # I. • "~~ group 99 ~ I '45 " " - -" _" / ...... ' '; -, _ ,.47 very saline group 100 -' '32. " / ... ."'52 .36...; -" ''>- / ". ,. , ....., ~.... ~:It ./ .5 -- ... '" /. . ..7.42-.... -,_41.), ._". saline group 91 I\ '" ...'\ I""-,.. ..' I• '"._ - ...... --::::- ,-',,/.• 28.lV ) \ \. I :' ;' \ ('38-"-'- '20 /.55 \ .1/ /.... '. k ,,_._.-., /'27 " '''':'' . ~ 37' \ ~3;' . '14 .. \ 8 \ . ".30 51." \ ., ~ \ .... " ,.. \ "... ".\" \ -.. ". "\ 54. ') ·10 ",~"i. / i /~ ./ ...... •36.. /1" calcareous ( '25 ,/./ ...... group 98 ...... -. L -' Fig. 4.3a. Scatterplot generated by PCOA ordination of the aquatic-semi-aquatic data set. Isolines indicate areas of ordination space occupied by members of the ten groups generated by classification of this data set. Fig. 4.3b. Overlay showing group boundaries after reallocation of misclassified sites. * indicates sites identified as outliers from the NEAREST program. Circled sites are the "fuzzy" sites allocated to different groups within each of the three classifications performed.

ff ~ ~ ~ ~---. 'N.h b_o /".-::;-;;:;" F w~" fresh brown water .;/A5 -=, ~""y ~b""'...~ ~. organic substrate ~_94 ~.~\ 6 ,of \ group 97 .\ \ 2 e,/? 1/h \ /~:;;-~~- 1/'" \ 1/./ '\ \-6 i.:' ." ,,=-~ ...... ·····i/ 03' \j\ ~ ... _ .. ~ ',~.4 :':~ ...... \ 50- ·33 ~ . •;';;"', " ...... p<..... ti'~...::..' ...l ". .29 turbld.roup 96 (".'" , ~_~,.3 {._____Y.'/\_~~-=-~:4 / I "0 / ;.<.... - J shallow f __'t. -46 ~~______--~.'-----:\~~.~.~.,\,...... - ~~'- ' ..... " ...... ' ,Mh...... · ,... '. group_ 99 - ~ ~slns--....,./ .... \tf. .::::..~\.._ / (,-;y....,...... '\,{.,~ \ ,--....;: "=, ...." --'.,." 89 /~ '\. 0 /'1<,<) '\-, -'-,.~. ------­ 32 \ V I....·oI!J'l'"' " 7 11'\ ,~ ,./~ ~\~ \_./'/ /"~ .i/.~...~,""036.)"~" • "'_ - /-y ""0' "''''00 y "e,'__ ,' (...I '\r (_~~~~,~~ .-:< /~ ..... 9' \ \~~.,o~l:-\~r~ . \. ~ '. ,\, \ '\\. ." I'. ""."".<--. ~'\\' ,,\\ '-:". -35 {/~ ~~ll.J ,~/ ::::";~. =p,.~/ // b ~ Fig. 4.3a. Scatterplot generated by PCOA ordination of the aquatic-semi-aquatic data set. Isol ines indicate areas of ordination space occupied by members of the ten groups generated by classification of this data set. Fig. 4.3b. Overlay showing group boundaries after reallocation of misclassified sites. * indicates sites identified as outliers from the NEAREST program. Circled sites are the "fuzzy" sites allocated to different groups within each of the three classifications performed. 72.

4. Discussion

Methods proposed for the assess:nent of the perfoIJl1ance of ordination techniques include the percentage variation explained by each axis, with success being dependent on the amount of infoonation concentrated on the first few axes. However, the distribution of infoIJl1ation across the axes is heavily dependent on the structure of the data; Noy-Meir (1971) found that twenty axes were required for the adequate redescription of his data, and these axes only accounted for

63% of the variation.

Comparison of the results of several different ordination techniques has been suggested as a method of evaluation, with agree:nent amongst several techniques providing a measure of perfOIJl1ance. However, since many carmon ordination methods have similar mathematical assunptions, several different techniques may agree on the wrong answer (Wilson 1981). Unfortunately, such comparisons provide no measure of how well the underlying structure of

the data is reflected. Whittaker (1978) has suggested that a comparison of the variance of species distributions, or the similarity

of adjacent samples on ordination axes produced by different

ordination techniques may indicate ordination effectiveness, the

ordination producing the narrowest species distributions and the

greatest adjacent sample similarities being the most successful.

Austin (1976) suggested that prediction is an appropriate test of

an ordination whose patterns have been interpreted in terms of

environmental gradients. From a successful ordination, it should be

p:!ssible to predict environmental characteristics given canmunity I

*47

34* *37 48* *16 *30 *21 43* *42 *35 36* 11* *10 17* 2* 44* *4 5* *9 19**23 3* *45 *52 *14 55* 12* *46 *13 *38 25* 7**26 51* 31* *32 *40 24**49 8**20 *27 53* 15* *28 54* *22 *41 *1

29**6 *18

*TI ~ l100------*=·39~------

Fig. 4. 4a. Scatterplot produced by DECOR ordination of wetland sites using entire floristic data set.

Fig. 4.4b. Overlay showing isol ines which indicate areas of ordination space occupied by members of the floristic groups after reallocation of misclassified sites. Circled sites were not satisfactorily classified by NIASM. :1 I [

I: ~ ! i: G! , I t i /icgroUP ~ i

.;:a/,careous group verZysaline group

brown " ,--, "..­ !:I sandy swatubster.- 8 ,43*'".Cj!:~\ ././ *37' ,I __...... /... '...... r *",/'-' " *21\ r I -r "~" ..\ .• ~ ...*;~~:!--/ "'. '.1 . . '17" *'0 .. . ~ "1\ brown wet 2* \...... ' \ \ " ~ ~! organic subs-"-"'" 44*'?f-~~. *4 .. ' 5 ;:'Itri!"...1~:::-·*9)/ ,.- . '\ ',I r . " 1(';*"- *' ,*" ...... /' '-._-.- \ " 0'" 26... ,-\ ' I• *38 " . . . ;"* *' \ *32! " .'lY·---....·../ 55* \ f" ~ : I ". \ I '\ ~ *40 / - , ',*22 '/ \ 53* ,..,...",~8* *""". * \$~*, . ,,' " (29 ~):) \ "",-/y ",.... \ '.,*" .-' 54* *""/ ,-./ Ij 0::;;' *" \ -. A,\ _._.-­ \.... fresh shallow water-saUne

\ (I *33 .---_ '-­ --, t b' 6 .....*5;( ___ ,~d group 2 1 ----*39)

Fig. 4.4a. Scatterplot produced by DECOR ordination of wetland sites using entire floristic data set.

Fig. 4.4b. Overlay showing iso1 ines which indicate areas of ordination space occupied by members of the floristic groups after reallocation of misclassified sites. Circled sites were not satisfactorily classified by NIASM. 73.

ccmposition, and fran the position of a site on the axes, the probable ccmposition of the camm.mity at that site. This test cannot be applied to the current problem, since it was decided at the outset

that the assunptions of linearity of spacies response over the wide envirormental range samplerl were not justifierl, arrl that

interpretation of the results in terms of envirormental gradients may

not be infonnative.

However, the ecological efficiency of the ordinations may be assessed by the degree of success in correlating the site distributions in ordination space with other ecological information

(Austin and Greig-Smith 1968). Examination of the scatterplots produced by ordination of the three wetland data sets (Figs 4.1-4.3)

indicated that same groups of samples could be identified with areas of ordination space. The members of these groups share similar water chemistry and substrate characteristics. However, the relationships of spacies-poor sites such as Lake Elu..c;ive (9) and Rockbank Swamp

(39), and same sites considererl to be outliers, Palparra settlement

Swamp (1), Cobrico SWamp (7) and Mackenzie Road Swamp (47), to t.'1e major groups recogniserl were not clear.

Further criteria proposed for the evaluation of ordination success

are accuracy arrl consistency (Wilson 1981). Accuracy implies adequacy

of fit - does the model reflect the underlying structure of the data?

Where the aL.'tl of the ordination procerlure is sequential arrangement of

spacies and samples along environmental gradients, the accuracy of an

ordination may be tested by application of a non-parametric test

(Wilson 1981) designerl to estimate a spacies maximun likelihood

ranking as the best statistical estimate of the true ranking. An 74.

ordination that is consistent will provide the same amOl.lIlt of information about the real structure of the data \IA1en it is performed on each of several replicate subsamples fram a data set (Wilson 1981) •

In the present study, PCQA ordinations were performed on the entire data set and two reduced versions. These latter data sets could be regarded as replicates, and recovery of substantially the same patterns by PCOA for each set suggests that the method performed consistently.

How well an ordination method performs is often de.r;:endent on the characteristics of the data set (Gauch, Whittaker and Wentworth 1977) •

Beta diversity, the degree of floristic difference among samples of a set, has major consequences for ordination performance. High beta diversity results in increased distortion (Whittaker and Gauch 1978) and sample displacement into higher dimensions. AI though beta diversity has not been calculated for the data sets used, it is eXJ?eCted to be high because of the wide envirormental range of the aquatic habitats sampled.

The effect of noise, introduced by sample error, on ordinations is to scatter points fram their original positions (Gauch and Whittaker

1972), reverse consecutive sample positions and scatter samples to second and higher axes (Gauch, Whittaker and 'Wentworth 1977). The increase in the proportion of variation recovered by ordinations of

38% and 45% of the sJ?eCies fram the entire data set indicated that the sample error may have been quite high. Additional sample error was introduced by normalisation of the data prior to classification.

Calculation of the probability of occurrence of each sJ?eCies within each wetland site represented Lake Elusive as having eight species, 75.

each with a 50% frequency of occurrence, although the eight species were recorde::l fran the only two vegetate::l quadrats sample::l.

When sample choice is detennined by the availability of

undisturbed sites, samples may be irregularly spaCe::l wi thin the range of cannunity variation. '!be data collecte::l may then incltrle clusters of similar samples, outliers (samples with unusual floristic canposition), and there may be gaps in the sample representation

(Whittaker and Gauch 1978). In same ordination techniques, clusters cause proble:ns because of the axes' response to variance. The

aoristic similarity of sane of the saline sites sampled may have

resulte::l in the fonnation of clusters in the wetland data.

outliers may be moderately deviant sites from atypical habitats or

stron;Jly deviant disturbe::l sites. Their effect on ordinations is dependent on the type, the nunber present, the beta diversi ty of the data set and the ordination technique used (Gauch, Whittaker and

W:ntworth 1977). In general, moderately deviant sites sean to be

ordinated centrally to other sites and may cause same displacanent of

site positions, whilst stron;Jly deviant sites may cause canpression of

other sites into a small region of the ordination field. In a large

data set identification of outliers prior to analysis may be

difficult. Black swamp (56) was removed following preliminary

analyses of part of the data. The program NEAREST identified

thirteen, six and four of the sites as outliers in the entire,

"eidentl1 am aquatic species data sets respectively. Figs. 4.1 4.2 am

4.3 show the outlier sites for each of these data sets. 76.

Gaps in site representation may result in complete disjunction of the data matrix when groups of sites share no species in carmon. The presence of Chara and Nitella, genera which were recorded for a wide range of sites, probably provided connections between groups of sites fran different environnents. Identification of these plants to species level probably would have effected complete disjunction. In this case the unrelated subsets should be ordinated separately to avoid the effects of high intrinsic dimensionality (Gauch and Whittaker 1978). Partial disjunctions within a data set may result in lack of differentiation between subsets by the axes of eigenvector ordinations, and cause difficulties in the interp:!.:etation of axes. Examination of the floristic composition of the groups generated by classification suggests that partial disjunctions occurred in the data sets, and that they may have been exacerbated by the data reduction techniques employed.

When species richness and abundance is variable within a data set axes tend to separate well-represented parts, while species-poor sites

(such as Rockbank swamp 39) are placed near the origin (Austin and Noy-Meir 1971). However, the PCOA ordinations have placed species-poor sites in the appropriate quadrants of ordination space.

Field data sets differ in their combinations of beta diversity, sample error, presence of clusters, outliers and disjunctions, and species amplitude. These characteristics will affect ordination t;:erforroance, but may be difficult to recognise prior to analysis. Addi tionally, the effects of different combinations of these characteristics are relatively unpredictable (Gauch, whittaker and Wentworth 1977) . 77.

Consideration of sample set properties may also enable choice of a mathematically appropriate ordination mcdel. FeCA, given samples representing a gradient sufficiently long enough to inclwe both slopes of a bell-shaped species distribution curve, is not able to produce an optimal representation of compositional data in ecological space (Austin 1976). However, groups can be identified with areas in ordination space, although they may not be related in any simple way to irrlividual axes. The distortion imposed by curvilinearity may mean that ecological distances bet'f.leen samples may not be preserved, resulting in distortion of the distances between sites near the ends of the axes. OCA is able to maintain the relative ecological distances between samples by rescaling the axes, and is able to estimate successfully most partial disjunctions (Hill arrl Gauch 1980) • HO'f.lever, its inability to deal with outliers resulted in a less than infonnative ordination of the entire data set.

5. Conc:l usions

In spite of the difficulty experienced in assessing how 'f.Iell the ordinations reflected the underlying structure of the data, FeOA was able to provide infonnation on the appropriate allocation of sites considered to be misclassified. It enabled identification of the six "fuzzy" sites as possible causes of the overlap observed between the fresh brown water sandy substrate, fresh shallow water and calcareous groups of sites, and provided some indication of the discreteness of the groups generated by classification. 78.

6. Summary of results of floristic analyses

Classification of the entire, "eident" and aquatic life-form data sets recCX}nised eight, nine and ten groups of wetlands respectively.

Allocation of sites was not always optimal. Misclassification of sites resulted from axiomatic properties of the NIASM algorithm. These sites were reallocated intuitively on the basis of their floristic composition. However, six sites had floristic affinities with more than one group of sites and their satisfactory unique allocation was difficult.

Inspection of the PCOA ordinations indicated that seven discrete groups of sites could be recCX}nised. These groups may be typified as the shallow freshwater, fresh brown water sandy substrate, turbid I calcareous I acidic water, brown water organic and saline groups of sites. The distinction between the saline and very saline groups of sites suggested by the results of the classifications was not maintained. Comparison of tr.e scatterplots for each of the three data sets showed that classification of the aquatic and semi-aquatic species data set provided the clearest discrlinination between groups. 79.

CHAPl'ER FIVE

Physical and chemical characteristics of aquatic

macrophyte habitats in victoria

1. Introduction

The distribution of species between water bodies has been related

to a large number of factors which include: water depth and amount of water level fluctuation (Walker and Coupland 1968, 1970, stewart and

Kantrud 1969, 1972, Walker and wehrhahn 1971, Brock 1981, Glenn-Lewin and Crist 1981, Kirkpatrick and Harwood 1981), salinity (Walker and

Coupland 1968, 1970, Walker and Wehrhahn 1971, Brock 1981, Kirkpatrick and Harwood 1981) and conductivity (Swindale and CUrtis 1957, Seddon 1967, 1972; Crow:1er, Bristow, King and vanderkloet 1977, Felzines

1977). Additionally, disturbance by alteration of water table levels (Haslam 1965), cultivation, grazing and mowing (Stewart and Kantrud 1969, Walker and Coupland 1970, Walker and Wehrhahn 1971) may also affect species distributions.

other chemical parameters which have been positively correlated with aquatic species distributions include alkalinity (Spence 1964, 1967), total alkalinity and sulphate concentration (Moyle 1945), total alkalinity and total filterable residue (Pip 1979), and concentrations of calcium and bicarbonate (Lundh 1951), inorganic carbon and calcium (Wiegleb 1978), phosphate (Jones and CUllbnore 1973) and the hardness ratio, ca+Mg / Na+K (Seddon 1972) • 80.

In peatlands, which are wetlands frequently dominated by Sphagnum spp_, the distribution of species and ccrrmuni ty types has been correlated with pH (Sjors 1950, Heinselman 1970, vitt and Slack 1975,

Schwintzer 1978, Brown, Crov.d.en and Jarman 1982), calcium and magnesiun concentrations (Heinselman 1970, Vi tt and Slack 1975,

Schwintzer 1978) and light (Vitt and Slack 1975). Studies by Tyler (1980) on Schoenus dominated ccrrmunities also suggest that hydrological conditions, soil nutrient status, soil carbonate content and the calciun-magnesiun ratio of the soils can be related to observed variations in species composition.

within lakes, water depth (Sheldon and Boylen 1977), water level fluctuations (Quennerstedt 1958, Hejny 1971, Segal 1971) wave action

(Raspopov, Slepukhina, Vorontsov and Rychkova 1978), and the physical and chemical nature of sedlinents (Misra 1938) have been related to the distribution of aquatic species. Same investigators have suggested that the local distribution of aquatic species is influenced by interaction of a number of factors, including sedlinent type, light and shelter (Pearsall 1920), water depth and sedlinents (Higginson 1965), water activity and sedlinent (Nicholson, IRNey and Clute 1975), and sedlinent, wave action and light (Spence 1982) •

Field studies suggest that the distribution of same suJ::rnerged aquatic taxa may be influenced by certain water chemistry parameters, and, within water bodies, correlations between species distributions and the nature of the substrate have been demonstrated. CUrrent evidence suggests that rooted, submerged aquatic angiosperms are able to obtain solutes through both their roots and shoots, although the behaviour of taxa is likely to be dependent on anatomy, morphology and 81.

current physiological state (Denny 1980), and solute availability

(Spence 1982). In well mixed waters of closed basins, where solute concentration might be expected to be relatively homogenous, it is suggested that water chanistry parameters are the daninant influence in between-site distribution, whilst sediment-wave-light interactions influence distribution at the local level. However, the paucity of aquatic macrophytes in sane Western Victorian lakes examined can probably be explained by the effects of sediment-wave-light interactions, and it seans likely that between-site species distribution is affected by a canplex of interacting envirormental factors.

In the present study, major ionic constituents of wetland waters were measured, water transparency was estimated am substrate texture, pH and percentage salts were determined. Determination of the levels of nitrogen and phosphorus canpounds in waters at anyone site requires frequent sampling for adequate characterisation, (Bowner

1981) am was beyom the scope of this sttrly.

2. M2thods

A. Sample collection and analyses

Most of the wetland sites sampled were canparatively shallow and

in open situations. under these conditions waters are likely to be

isothermal and well-mixed, chanical stratification is unlikely and surface samples are considered to be representative of the water body.

Sample points were located in open water (where possible) near vegetation transects. Three surface samples were collected for each Table 5.1 Summary of methods employed for analysis of water samples

Parameter Container Preservat ion Analytical Method

Calcium 0.45~m filtration: Atomic absorption Magnes ium acidification to spectrophotometry * pH 2.0 with

Potassium polyethylene H2 SO 4 j Flame emission Sodium storage 4°c spectrophotometry

Chloride Corning Chloride Meter,Model 921

Carbonate glass stoppered placed in insulated Potentiometric titration Bicarbonate bottle container, (Golterman, Clymo and Ohnstad, 1978) determined within to standard endpointj 6 hours pH 4.5 using O.IOM HCl

Sulphate plastic whirlpack snap frozen using Ion chromatography dry ice; (Small, Stevens and Bauman, 1975) storage O°C ------.. --.­ Conduct i v i ty determined in the Yellow Springs Conductivity Meter field Model 31 pH de tet II! i Ilcd ill t Ministaph Portable pH Meter Model PT35 field

,': Lanthanum chloride sup:->ressant (United St.ltes Environmental !'rotectioll Agency 19lJ) 82.

wetland. Sample containers were well rinsed prior to collection, and

polyethylene storage bottles were cleaned using nitric or chranic

acid. Sample preservation, storage and methods of determination are

shown in Table 5.1.

Substrate samples were collected by hand augering fran each of the major vegetation zones identified along the transect lines. Substrate

texture was estL~ted, the pH was measured using a Ministaph portable

pH meter, and the samples were stored in polythene bags for laboratory determination of soil water conductivity to enable calculation of

J;ercentage salts according to t.'1e method of Jackson (1962). The same

sample preparations were used to determine chloride levels using a

Corning Chloride meter.

At each wetland sampled, the maximun depth of the basin when

flooded, the geolCX3Y of the catchnent area and the daninant substrate

texture were recorded. water transparency was scored as clear, brown

or turbid on the basis of Secchi disc measurerne.l"')ts of light

J;enetration taken at each quadrat sampled. water regime was recorded

as J;ermanent if the wetland basin had held water continuously for at

least ten years, serni-J;ermanent if the wetland dried during years of

low rainfall, and annual filling if the site dried out each year.

Average annual rainfall (rnm) was ascertained for the recording station

nearest each site and labtude and longitude of the sites were determined fran 1:63360 map sheets. 83.

B. Nunerical analyses of data

The water chemistry data, following averaging of the deteoninations for each site by the program GPRES, were classified to see whether there was any correspondence between groups generated by classification of the floristic data, and groups classified using water chemistry parameters. The program MCAN utilised the canberra

Metric to calculate dissimilari ty coefficients between the sites. canberra Metric is self-standardising (Lance and Williams 1977) and sensitive to proportional rather than absolute differences between attributes, because an attribute with a dramatically high value can only dcminate one of the series of fractions that are surn:nated to yield the coefficient (Clifford and Stephenson 1975). It is particularly suitable for use with the nunerical data collected, since same large outlying values were recorded, and these would otherwise have required same considerable transfoonation prior to analysis, an option not presently available within the TAXON package (Ross 1982) •

The dissimilarity matrix generated by the program MCAN was classified lJSing a hierarchical agglcmerative method similar to that used in earlier chapters. The program SAHN of the TAXON package accepts a matrix fran M:AN. The sorting strategy, which defines a distance measure between groups of elements, was the Flexible sorting strategy, with beta, the "cluster intenSity coefficiene' (Williams

1971) set to -0.25 to sharp:n the resulting classification and aid interpretation. The program RATLAN provided the Ratkowsky-Lance global ccmparison of cluster validity, TABOUT produced two-way tables for each group generated, NEAREST listed the seven nearest neighbours of each wetland, and the programs crOM and CRAMER indicated the SlI-10F NO. SITE SAMPLE C03 Hc03 CL S04 CA MG NA K ANIONS CATIONS DIFF 'tERROR SALINITY pH CONDUCT.

46. FERNBANK SWAMP A 0.00 0.30 0.20 0.01 0.10 0.26 1.63 0.07 0.50 2.07 -1.57 -67.47 0.07 6.00 87 B 0.00 0.27 0.06 nd. 0.10 0.25 1.63 0.08 0.31 2.06 0.06 5.80 78 C 0.00 0.28 0.25 nd. 0.11 0.23 1.63 0.07 0.51 2.04 0.07 6.10 77

47. MACKENZIE ROAD A 0.00 0.00 0.65 0.02 0.01 0.17 1.96 0.00 0.67 2.14 -1.47 -59.45 0.07 4.20 159 B 0.00 0.00 0.65 0.01 0.01 0.18 1.96 0.00 0.66 2.14 -1.48 -60.00 0.07 4.00 150 C 0.00 0.00 0.23 0.02 0.01 0.18 1.96 0.00 0.25 2.14 -1.90 -83.56 0.06 4.00 148

48. KANGAROO SWAMP A 0.00 0.31 2.34 0.01 0.20 0.74 2.72 0.14 2.66 3.80 -1.14 -22.27 0.18 6.20 280 B 0.00 0.31 2.40 0.02 0.16 0.77 2.83 0.09 2.73 3.84 -1.11 -21.32 0.19 6.20 267 C 0.00 0.31 1.50 0.01 0.14 0.78 2.83 0.13 1.82 3.88 -2.07 -43.12 0.15 6.20 193

49. BANYULE BILL. A 0.00 2.71 3.67 0.15 0.41 1.58 3.70 0.19 6.53 5.88 0.65 7.15 0.42 6.90 478 B 0.00 2.71 3.64 0.15 0.38 1.62 3.70 0.20 6.50 5.90 0.60 6.58 0.42 6.90 485 C 0.00 2.71 2.96 0.16 0.38 1.58 3.81 0.21 5.83 5.98 -0.15 -1.69 0.40 6.90 486

50. LITTLE RIVER A 0.00 10.82 22.13 0.02 0.28 1.04 25.40 0.05 32.97 26.77 6.20 14.33 2.05 8.30 2435

51. ROUND LAKE A 3.19 10.66 177.40 1.34 1.14 36.87 144.41 0.91 192.59 183.33 9.26 3.31 10.92 8.80 15500 B 3.19 10.66 183.10 1.29 1.24 37.94 163.11 0.91 198.24 203.20 -4.96 -1.64 11.57 8.80 15000 C 3.19 10.66 185.21 1.35 1.20 38.02 150.07 0.90 200.42 190.20 10.22 3.52 11.35 8.80 15000

52. BAY OF ISLANDS A nd. nd. 18.00 4.00 1.03 3.91 19.20 0.38 21.95 24.53 1.35 8.80 1780 B 0.06 0.59 16.14 4.00 0.99 3.84 18.92 0.38 20.79 24.13 -3.34 -9.68 1.32 8.60 1720 C nd. nd. 17.86 4.00 1.05 3.87 20.33 0.42 21.81 25.68 1.38 9.60 1800

53. LAKE STRUAN A 0.72 3.00 19.27 1.29 1. 74 7.38 16.88 0.16 24.29 26.16 -1.87 -4.89 1.47 9.00 2076 B 1.20 2.82 19.70 1.35 1.91 7.56 17.40 0.17 25.07 27.03 -1.96 -4.96 1.51 9.10 2095 C 1.52 2.73 19.84 1.35 1.74 7.43 17.40 0.18 25.44 26.76 -1.31 -3.33 1.51 9.40 2124

54. LAKE WINCHELSEA A 2.06 8.21 220.54 nd. 3.33 4.68 189.21 0.86 230.79 198.09 12.88 9.00 17081

B 2.06 8.21 199.66 nd e< 3.90 4.66 181.82 0.86 209.91 191.24 11.99 9.00 17297 C 2.06 8.21 178.16 nd. 3.73 4.70 173.99 0.84 188.41 183.26 11.04 9.10 16757

55. LAKE TERANGPOM A 4.28 3.55 19.84 1.18 0.97 6.75 21. 75 0.83 28.85 30.30 -1.45 -3.23 1. 74 9.90 2300 B 4.28 3.55 19.98 1.05 0.99 7.01 21.05 0.82 28.86 29.88 -1.02 -2.30 1.72 9.90 2300 C 4.28 3.55 19.98 1.20 0.93 6.58 20.18 0.81 29.01 28.50 0.51 1.18 1. 70 9.90 2300 'I ',h I " r ') C' " , dId

-< .­ SUM OF NO. SITE SAMPLE C03 HC03 CL S04 CA MG NA K ANIONS CATIONS DIFF.%ERROR SALINITY pH CONDUCT.

37. BITTER LAKE A nd. nd. 280.06 nd. 10.01 66.09 247.93 2.81 279.99 326.85 16.74 9.00 22431 B nd. nd. 274.36 nd. 9.74 65.21 241.93 3.09 214.29 326.03 16.53 9.00 22462 C nd. nd. 280.06 nd. 10.41 66.83 276.21 2.53 279.99 355.98 17.40 8.70 23530

38. SWAN LAKE (E) A 0.00 0.65 111.62 12.99 2.78 19.34 102.22 3.32 131.26 127.66 3.60 1.86 7.60 7.10 13333 B 0.00 0.65 129.41 13.37 3.27 30.78 115.27 2.81 143.43 152.14 -8.71 -3.89 8.47 8.00 13939 C 0.00 0.65 120.89 13.66 3.08 27.16 110.92 3.00 135.20 144.16 -8.96 -4.23 8.04 8.10 13939

39. ROCK BANK A nd. nd. 198.15 4.11 2.59 52.76 156.59 1.12 202.21 213.06 11.56 8.60 17609 B nd. nd. 200.97 4.35 2.94 54.98 169.64 1.18 205.27 228.74 12.01 8.70 17514 C nd. nd. 205.91 6.05 3.19 48.31 176.16 1.28 211.91 228.95 12.34 8.90 18143

40. ARMY RESERVE A 8.92 11.48 142.50 12.47 0.40 3.52 113.09 1.20 175.37 118.21 57.15 27.76 9.32 9.80 10894 B 11.32 10.06 124.73 12.47 0.46 3.38 97.87 0.97 158.57 102.69 55.88 30.71 8.31 10.00 11319 C 11.32 10.06 127.58 11.83 0.48 3.34 100.04 1.27 160.79 105.14 55.65 30.00 8.44 10.00 11319

41. BLEAK HOUSE A 0.20 4.51 12.30 0.61 0.79 2.62 15.44 0.21 17.62 19.06 -1.44 -5.17 1.16 8.40 1356 B 0.20 4.50 10.05 0.62 0.79 2.60 12.07 0.21 15.38 15.67 -0.29 -1.25 1.00 8.40 1333 C 0.20 4.70 10.01 nd. 0.72 2.62 11.96 0.21 14.89 15.51 -0.62 -2.70 0.98 8.30 1356

42. LAKE TUTCHEWOP A 0.00 3.75 754.04 72 .27 1.50 37.04 632.88 4.60 830.06 676.03 154.03 14.12 45.64 8.20 52250 B 0.00 3.79 668.77 57.96 1.06 43.62 571.99 4.76 730.53 621.43 109.10 11.06 40.61 8.30 50450 C 0.13 4.32 725.64 9.56 0.77 44.44 582.86 4.53 739.64 632.60 107.04 10.68 40.58 8.40 nd.

43. LAKE BARRACOUTA A 0.00 0.40 3.22 0.14 0.08 0.74 3.35 0.10 3.75 4.27 -0.52 -8.40 0.24 6.50 439 B 0.00 0.39 3.19 0.14 0.08 0.74 3.13 0.10 3.72 4.05 -0.33 -5.62 0.23 6.50 439 C 0.00 0.38 2.96 0.13 0.06 0.74 3.52 0.11 3.48 4.43 -0.96 -15.52 0.23 6.20 441

44. LAKE BEADLE A 0.00 0.34 2.20 0.05 0.13 0.49 2.04 0.23 2.59 2.90 -0.31 - 7.37 0.17 6.50 280 B 0.00 0.34 2.20 0.05 0.13 0.49 2.00 0.23 2.59 2.85 -0.27 -6.40 0.16 6.60 281 C 0.00 0.34 2.28 0.05 0.13 0.41 2.09 0.23 2.67 2.87 -0.20 -4.65 0.17 6.60 281

45. BRISBANE RANGES A 0.00 0.43 2.26 nd. 0.17 0.81 2.61 0.10 2.67 3.70 0.18 6.70 317 B 0.00 0.43 2.20 nd. 0.18 0.79 2.72 0.05 2.61 3.74 0.18 6.70 276 c 0.00 0.44 2.20 nd. 0.18 0.77 2.50 0.11 2.62 3.56 0.18 7.30 257 TabLe 5.2 Cont'd

~ SUM OF NO. SITE SAMPLE C03 HC03 CL S04 CA MG NA K ANIONS CATIONS DIFF ERROR% SALINITY pH CONDUCT

27. CHAMPION LAKE A nd. nd. 81.09 nd. 1.44 12.86 112.22 0.99 81.02 127.51 5.68 8.70 8889 B nd. nd. 83.94 nd. 1.39 13 .09 113 .09 1.00 83.87 128.57 5.80 8.60 9130 C nd. nd. 83.94 nd. 1.32 12.54 100.04 0.98 83.87 114.89 5.49 8.40 8869

28. LAKE WANDELLA A 0.00 2.73 924.58 29.48 67.47 54.32 767.73 5.42 956.79 894.93 61.86 4.50 54.24 8.10 7684{) B 0.00 2.73 103.8* 3.80 67.47 39.51 719.88 4.14 106.5* 830.99 233.81 17.15 56.66 8.10 75790

29. WCB GARRY A 0.00 0.37 0.08 0.05 0.09 0.26 1. 74 0.05 0.50 2.14 0.07 6.60 71 B 0.00 0.38 0.08 0.08 0.10 0.30 1.63 0.05 0.54 2.08 0.08 6.80 63

30. ST. MARY'S LAKE A nd. nd. 268.27 16.27 5.04 29.96 176.16 3.94 284.49 215.10 14.96 8.70 17826 B nd. nd. 203.59 16.27 4.96 26.17 167.46 3.61 219.82 202.20 12.41 8.80 16087

31. MIDDLE LAKE A 0.00 0.63 0.95 0.14 0.21 0.64 2.39 0.07 1.72 3.31 -1.59 -38.17 0.15 7.60 200 B 0.00 0.58 7.39 0.20 0.50 2.59 6.63 0.11 8.17 9.84 -1.66 -11.94 0.51 7.10 188

32. THE STONY A nd. nd. 1.02 0.01 0.24 0.40 2.83 0.10 0.98 3.56 0.11 7.40 237 B nd. nd. 0.85 0.01 0.16 0.37 2.83 0.10 0.80 3.46 0.11 7.30 233

33. LAKE KANAGULK A 0.00 nd. 3.07 0.33 0.31 0.92 10.29 0.22 3.38 11. 74 0.39 8.10 872 B 0.00 nd. 5.22 0.33 0.25 0.71 11.09 0.25 5.53 12.30 0.48 7.80 870 C 0.00 0.02 4.34 0.34 0.27 0.57 11.09 0.17 4.70 12.11 -7.40 -51.20 0.45 8.10 870

34. HEYWOOD G.C.S. A 0.00 0.00 10.89 nd. 0.44 2.10 9.57 0.28 10.87 12.39 0.65 4.30 nd. B 0.00 0.00 9.87 nd. 0.25 2.05 8.92 0.26 9.85 11.48 0.59 4.40 nd.

35. LAKE BRIDGEWATERA 0.00 2.62 4.91 nd. 1.02 2.21 5.22 0.15 7.51 8.60 0.51 8.40 789 B 0.00 2.58 5.61 0.51 1.03 2.21 5.11 0.20 8.71 8.55 0.15 1.20 0.55 8.50 789 C 0.14 2.60 6.46 0.36 1.05 2.24 5.22 0.13 9.56 8.64 0.91 6.81 0.58 8.40 779

36. COTTER'S LAKE A nd. nd. 10.01 0.77 4.62 2.83 7.61 0.14 10.74 15.20 0.70 nd. 1494 B 0.00 5.80 14.53 1.39 6.39 3.18 12.07 0.22 21.71 21.86 -0.14 -0.44 1.39 7.50 1952 C 0.00 4.55 7.70 0.59 3.57 1.83 6.96 0.14 12.84 12.49 0.34 1.81 0.84 7.20 1070

Table 5.2 Cont'd SUM OF NO. SITE SAMPLE c03 HC03 CL S04 CA MG NA K ANIONS CATIONS DIFF.%ERROR SALINITY pH CONDUCT.

18. TWO TREE SWAMP A nd. nd. 2.34 0.08 0.28 1.34 3.26 nd. 2.37 4.86 0.18 8.80 374 B nd. nd. 3.05 0.30 0.29 0.91 2.83 0.22 3.30 4.24 -0.94 -15.94 0.21 9.00 251

19. COCKPIT LAGOON A 0.38 1.98 3.50 0.05 0.43 1.54 3.81 0.08 5.91 5.86 0.05 0.52 0.38 8.90 520 B 1.20 2.08 3.50 0.05 0.51 1.53 3.81 0.08 6.83 5.93 0.90 9.63 0.41 8.50 537 C 1.08 1.98 4.34 0.05 0.46 1.53 3.81 0.07 7.45 5.87 1.59 16.53 0.43 9.30 507

20. LAKE LOOKOUT A 1.80 4.54 39.35 1.44 1.04 9.14 4:J.50 0.56 47.13 54.24 -7.11 -9.14 2.95 9.20 4078 B 1.80 4.47 38.30 1.50 0.99 10.45 45.67 0.53 46.07 57.64 -11.57 -14.34 2.97 9.30 3922 C 1.80 4.47 38.25 1.50 1.01 9.79 45.67 0.56 46.02 57.03 -11.02 -13.76 2.97 9.30 3098

21. TIlE SWANNEE A nd. nd. 29.98 0.56 1.22 3.34 45.67 1.16 30.49 51.39 2.25 8.60 3130 B nd. nd. 28.29 0.52 1.37 3.47 43.50 1.18 28.76 49.52 2.14 8.70 3200 C nd. nd. 28.01 0.67 1.39 3.49 43.50 1.20 28.63 49.57 2.14 8.70 3087

22. BITTEP~ LAGOON A 0.00 1.89 0.42 0.01 0.23 0.43 2.39 0.14 2.33 3.20 -0.87 -20.02 0.20 6.40 228 B 0.00 1.40 0.85 0.01 0.16 0.43 3.04 0.07 2.26 3.71 -1.45 -29.95 0.20 6.90 248 C 0.00 0.59 0.08 0.01 0.07 0.12 1.85 0.03 0.68 2.08 -1.39 -57.72 0.09 6.50 82

23. VICTORIA LAGOON A 0.00 0.52 3.07 0.02 0.22 0.66 2.70 0.24 3.62 3.82 -0.20 -3.51 0.23 6.80 332 B 0.00 nd. 3.19 0.01 0.12 0.67 2.52 0.10 3.18 3.41 0.18 4.80 324 C 0.00 0.70 2.00 0.01 0.21 0.66 2.26 0.28 2.72 3.41 -0.69 -14.51 0.19 6.60 298

24. KING RIVER A 0.00 0.35 0.00 0.01 0.00 0.94 1.52 0.06 0.36 2.52 -2.16 -80.13 0.07 5.90 44 B 0.00 0.32 0.00 0.01 0.01 0.94 1.63 0.07 0.33 2.65 -2.33 -82.63 0.07 6.20 44

25. LAKE ALBACUTYA A 0.48 8.09 29.50 1.58 1.11 7.87 29.93 0.88 39.66 39.79 -0.13 -0.22 2.47 8.80 3289 B 0.75 7.87 29.05 1.58 1.16 7.90 28.53 0.91 39.25 38.50 0.75 1.29 2.42 8.80 3200 C 0.68 8.04 29.36 1.60 1.14 7.74 29.06 0.92 39.68 38.85 0.83 1.42 2.45 8.80 3262

26. TIlE DUCK HOLES A 0.00 nd. 3.50 0.02 0.14 0.65 4.57 0.79 3.50 6.15 0.27 6.50 591 B 0.00 nd. 3.78 0.03 0.17 0.45 4.68 0.18 3.79 5.48 0.26 5.70 500 C 0.00 nd. 4.77 0.01 0.20 0.81 5.87 0.81 4.77 7.69 0.35 6.20 681

Tahl.e S.2 ConL'cJ SUM OF NO. SITE SAMPLE c03 HC03 CL S04 CA MG NA K ANIONS CATIONS DIFF ERROR% SALINITY pH CONDUCT.

9. LAKE ELUSIVE A 0.00 nd. 3.07 0.19 0.09 0.77 3.37 0.09 3.25 4.33 0.21 6.20 526 B 0.00 0.12 3.10 0.19 0.09 0.83 3.26 0.09 3.41 4.28 -0.87 -14.47 0.22 6.20 551

10. LAKE BONG BONG A 0.00 2.44 3.07 0.39 1.44 1.28 3.91 0.08 5.90 6.71 -0.80 -8.33 0.41 8.40 648 B 0.00 2.49 4.77 0.39 1.47 1.30 3.72 0.08 7.64 6.57 1.07 10.32 0.47 8.30 650 C 0.00 2.40 3.92 0.39 1.43 1.25 3.87 0.09 6.71 6.64 0.06 0.63 0.44 8.30 610

11. DOCK INLET A 0.00 0.13 3.64 0.22 nd. nd. nd. nd. 3.99 33.30 0.84 6.10 414 B 0.00 0.19 4.77 0.31 0.11 1.07 4.87 0.12 5.27 6.17 -0.90 -10.23 0.33 6.80 540

12. EW ING MARSH A 0.00 1.63 3.92 0.10 0.58 0.99 3.81 0.13 5.65 5.51 0.15 1.77 0.36 7.00 780 B 0.00 0.72 5.75 0.04 0.73 1.21 5.55 0.12 6.52 7.61 -1.09 -10.03 0.41 6.60 780 C 0.00 0.80 5.47 0.20 1.11 1. 76 5.22 0.10 6.47 8.19 -1. 72 -15.05 0.42 7.70 560

13. SWAN LAKE W. A 0.00 nd. 3.10 0.64 0.88 0.77 3.15 0.11 3.73 4.91 0.24 6.70 426 B 0.00 nd. 2.85 0.90 0.88 0.69 3.04 0.07 3.73 4.68 0.24 6.90 410 C 0.00 1.26 2.91 0.95 0.86 0.75 3.04 0.07 5.11 4.73 0.39 5.29 0.32 7.60 415

14 . 1.1'•LAKE r-IALSEED A 0.00 3.50 3.50 0.43 1.57 2.67 4.13 0.04 7.43 8.41 -0.98 -8.11 0.52 8.10 720 B 0.00 3.00 3.64 0.43 1.46 2.26 4.02 0.06 7.07 7.81 -0.73 -6.47 0.48 8.30 700 C 0.00 3.16 3.36 0.48 1.42 2.50 4.13 0.09 7.00 8.14 -1.14 -9.79 0.49 8.40 700

15. GELLIONDALE A 0.00 1.02 6.91 nd. 0.41 1.41 6.31 0.23 7.91 8.35 0.49 8.20 810 B 0.00 0.67 7.76 nd. 0.40 1.39 7.39 0.23 8.41 9.41 0.52 7.80 927 C 0.00 0.80 7.59 0.16 0.46 1.51 7.61 0.23 8.54 9.80 -1.26 -8.94 0.54 6.40 761

16. THE LONG SWAMP A 0.00 0.28 2.40 0.05 0.09 0.47 2.61 C.12 2.72 3.28 -0.56 -12.13 0.18 5.60 328 B 0.00 0.29 2.68 0.08 0.10 0.57 3.04 0.12 3.05 3.83 -0.78 -14.55 0.20 4.70 287

17. DEREEL LAGOON A 0.00 0.30 4.06 0.01 0.16 0.67 3.70 0.08 4.38 4.62 -0.25 -3.60 0.26 7.20 490 B 0.00 0.26 5.22 0.01 0.16 0.95 4.35 0.13 5.49 5.59 -0.11 -1.28 0.32 6.80 576 C 0.00 0.20 2.91 0.02 0.09 0.54 3.70 0.14 3.12 4.47 -1.35 -22.35 0.21 5.80 314

Table 5.2 Cont'd SUM OF NO. SITE SAMPLE C03 HC03 CL S04 CA MG NA K ANIONS CATIONS DIFF. tERROR SALINITY pH CONDUCT

1. PALPARRA SWAMP A 0.88 0.50 1.81 nd. 0.55 0.38 2.83 0.01 3.16 3.78 0.20 9.40 294 B 0.68 0.46 1.89 nd. 0.47 0.34 2.72 0.01 3.01 3.54 0.19 9.80 297 C 0.84 0.55 1.86 0.08 0.54 0.40 2.83 0.01 3.33 3.78 -0.45 -6.29 0.21 9.60 299

2. BREAK NO. 2 A 0.00 1.89 12.01 0.54 0.59 3.47 11.09 0.15 14.44 15.31 -0.87 -2.91 0.88 7.00 1422 B 0.00 1.72 11.59 nd. 0.58 3.50 10.55 0.14 13.29 14.76 0.82 7.20 1405 C 0.00 0.00 11.87 nd. 0.59 3.52 11.09 0.15 11.85 15.36 0.74 7.20 1272

3. TREMAINES SWAMP A 0.00 0.06 1.97 nd. 0.06 0.41 2.61 0.03 2.01 3.10 0.14 4.90 243 B 0.00 0.03 1.92 nd. 0.05 0.4G 2.50 0.02 1.94 2.98 0.13 5.10 233 C 0.00 0.02 1.64 nd. 0.06 0.42 2.50 0.02 1.64 3.00 0.12 5.00 228

4. BROWN REEDY A 0.00 0.03 1.10 0.03 0.12 0.26 2.07 0.11 1.16 2.55 -1.39 -37.37 0.10 4.90 286 B 0.00 0.22 0.62 nd. 0.13 0.24 1. 74 0.03 0.82 2.14 0.08 5.40 133 C 0.00 0.18 0.56 0.02 0.15 0.24 1.74 0.02 0.76 2.16 -1.39 -47.70 0.08 5.50 114

5. KANAWINKA SWAMP A 0.00 2.02 9.99 0.05 0.64 2.71 10.00 0.22 12.05 13.57 -1.52 -5.93 0.76 7.30 1244 B 0.00 1.94 10.88 0.10 0.59 2.57 9.57 0.21 12.91 12.94 -0.03 -0.12 0.78 7.00 1235 C 0.00 2.71 10.01 nd. 0.44 2.71 9.57 0.00 12.70 12.72 0.78 7.20 1219

6. FLOATING ISLAND A 0.00 2.53 4.65 0.01 0.53 3.19 5.22 0.10 7.19 9.05 -1.85 -11.41 0.49 6.70 743 B 0.00 2.88 5.19 nd. 0.53 3.51 5.33 0.10 8.05 9.47 0.54 7.10 750 C 0.00 2.90 5.50 0.02 0.62 3.70 5.22 0.10 8.42 9.64 -1.22 -6.77 0.55 7.10 823

7. COBRICO SWAMP A 0.00 2.07 12.97 1. 74 2.34 4.79 9.79 0.13 16.79 17 .05 -0.26 -0.77 1.01 7.60 1625 B 0.00 2.03 12.97 1.81 2.34 4.81 9.79 0.13 16.81 17.06 -0.25 -0.74 1.01 7.40 1515 C 0.00 2.02 13.12 nd. 2.41 4.77 10.22 0.13 15.12 17.54 0.93 7.50 1515

8. DEEP LAKE A 0.65 6.83 14.07 0.33 1.21 6.91 13.05 0.23 21.88 21.40 0.48 1.11 1.37 8.40 1632 B 0.36 6.52 14.18 0.33 1.22 6.42 12.72 0.22 21.39 20.58 0.81 1.93 1.33 8.40 1663 C 0.60 6.25 13.91 0.34 1.23 6.58 13.16 0.22 21.10 21.19 -0.09 -0.30 1.32 8.40 1663

1 Table 5.2 Concentration of major ions (m-equiv/I- ) sal inlty, pH and conductivity for 55 Victorian wetlands. Sal inity 1 (0/00) was calculated as the sum of the ions determined. pH and conductivity (reported as pS cm- at 18°e) were measured in the field. Percentage error was calculated as !(sum of anions - sum of cations) /!(sum of anions + sum of cations) x 100. Slight discrepancies in the figures are due to the rounding of figures to two decimal places. 1 nd. ::::: not determined. -J, =x 10- •

., 84.

attribute contributions to each dichotomy and the relative linportance of attributes over the canplete set of classificatory groups respectively • The sites \Vere ordinated using PCOA (Principal

Coordinates Analysis) to examine the relationships between the groups generated by classification. '!he program BACRIV examined attribute contributions to the PCOA vectors.

3. Results

A. Laboratory analyses

Table 5.2 shows the concentrations of the major constituents in

the water samples analysed. Total ionic concentrations of the waters

ranged fran 2.5 to more than 1500 m-equiv/l-l , although total concentration for the majority of the samples was less than 100 m-equiv/l-l • For samples of up to 200 m-equiv/1-1 total ionic concentration, Johnson, Cole, Johnson, McPherson, Muir and Szczepanski

(1979) have suggested that under optimal analytical corrlibons the

accuracy of the total ions should be wi thin the range 0.1 to 0.5 m-equiv/l-l • Ionic balances (where the sun of the cations is canpared

with the sun of the anions) calculated for such samples in the present data show differences of between 0.03 and 2.16 m-equiv/1-l , arrl generally an excess of cations. The occasional anion excesses

recorded in the present study \Vere almost invariably associated with

alkaline waters, am sanetimes with rather low values for calciun

which may be due to calciun carbonate precipitation (Maddocks 1967) .

Johnson et al. (1979) have shown that apparent excesses of cations

in ion balances may be due to silicate anions, whilst the 85.

presence of silicic acid or silic gel results in an excess of anions.

The latter condition is associated with alkaline waters containing measurable amounts of silica. These ionic imbalances are due to reactions between soluble and insoluble silica with carbonate/bicarbonate 10ns which cannot be detected by the alkalinity titration or by the detennination of soluable silica.

Lack of ionic balance is cammon in acidic waters and those with low «5.00 m-equiv/l-l ) ionic concentrations (Han 1970). In the present study, eleven water bodies wi th ionic concentrations <7 m-equiv/1-1 showed an apparent excess of cations. The anion concentrations (especially that of chloride) may have been too low to be detennined accurately by the methods anployed. Underestimaticn of bicarbonate may have resulted fran usin:; pH 4.5 as a stamard endpoint in alkalinity titrations. The pH meter could not be read wi th sufficient accuracy to canplete the titration at the actual end point, and brown-coloured or turbid water prevented accurate indicator endpoint detection.

In accordance with Hem (1959) results with an ion balance accuracy of 10% were accepted for subsequent numerical analysis. Sites with greater discrepancies between anion and cation concentrations were included where the total ion concentration was low. The take

TUtchewop (42) and Little River (50) samples, for which surprisingly low levels of calcium, and calcium and magnesium respectively were recorded, were inclooed to ensure representation of all 55 wetlands samples. 86.

The low level of variation in the calculated salinity of the

samples collected within the same waterbody (Table 5.2) suggests that

there was generally little horizontal variation in the chemical parameters measured. However, the pH and bicarbonate of waters

samplai fran densely vegetated areas tendai to be slightly lower than

those recordai for open waters in the same site, e.g., Dereel Lagoon

(17), and minor horizontal variations occurred across sites vtlich

received significant inflow via natural or artifical channels (e.g.,

Cockpit Lagoon 19, Middle Lake 31, Swan Lake East 38) •

Ratios of cationic dominance for most vf the waters examined are

Na > Mg > Ca > K, the major exceptions being Palparra Settlement

Swamp (l), Lake Borg Borg (10) , Swan Lake ~st (13) and Cotter I sLake

(36), all of which are waterbodies associated with calcareous sands

arrl where Ca > tvg. The dominant anion is almost invariably Cl, arrl the most cammon pattern of dominance is Cl HC0 S04' frequently with > 3 >

C03 absent, and occasionally Cl > HC03 > C03 > S04 (sites 8, 19, 20, 51). At Lake Elusive (9) and Dock Inlet (11) and the more saline

sites 28, 33, 38, 40, 42, 52, Cl S04 HC0 C0 • At Bittern > > 3 <> 3 Lagoon (22) , King River ~ander (24) and Fernbank Swamp (46) HC0 > Cl 3 > SO4' but these results are equivocal since the cation excess and low corrluctivities recorded for these sites suggest that the anion concentrations may have been too low for accurate detection by the methods employed. salinity salinity

~O·5(%a ~O 70/""

108 3 ______•__ ~r~u'p ..!.e~~

salinity I <3%0 >6%0

107

salinity >01%0 <01%0 Na>28 Na<28 (except site 35) 106 5 g.!:..o~_I~e.! sallnltv 104 105 103 <20%0 >40%0 carbonate <053 >06 Isalinltvl pH I ,.12%0 >13%0 ~89 :$89 (except 50)

pH >5 <5

92 01100 i97 102 86 99 78

Fig. 5.1. Dendrogram produced by classification of the water chemistry data indicating relationships between the groups at the eight group level. Broken 1ines indicate the group compositions at the three and five group levels (see text). 87.

B. NlInerical analyses

Classification

Application of the Ratkowsky-Lance criterion calculated by the program RATLAN to the hierarchy of groups produced by the SAHN classification shows that the ratio of between to total group variance

(the scaled Cramer value) is maximised at the three group level, and is only slightly lower at the five group level. The groups 'Which would result from acceptance of a three or five group solution are indicated in Fig. 5.1. At the eight group level the scaled Cramer value is significantly lower. However, the eight group solution is examined for evidence of correspondence between the water chemistry group membership and the membership of the eight groups generated by classification of the floristic data.

The dendrogram

Fig. 5.2 shows the dendrogram produced by the classification of the water chemistry data and the group average values for the attributes 'Which made greater than ten percent contribution to the dichotomies. As before, the convention of reading the hierarchy downwards has been adopted.

The first dichotomy separated group 107, which had comparatively low average values for sulphate, chloride, salinity, sodium, calcium and carbonate, fran group 108 which had an average salinity of 4.8

0/00 and corresponding high average values for these ions. The freshwater group 107 was further divided on the basis of average 0·12 504 7-49

3·00 CI 138'46

0·1 5%. 4·8

4-00---­ -3-78 Na 121·44

390 K18 11680

0·07 C03 1·3

108

3-00- -­ -- 1'06 504 : 8

18-05 CI 03

32'58 Na 25

2036 K18 79

1·4 5%.

1·35 Ca 107 4·61 Mg 2-00­ 0,38 K 8

1 C03 0

: 5%. 0·003 .....-Wo=

o 9 504 0,02

~'- ~- r--~ 0·1 C03 1'97 0·18 K 0·72 '-00- -0·83 C03 0 14,4 Na 32-45 0'05 5%. 0·24 12'89 CI 27·39 0·04 K 0·18 0·06 504 0·21 3·93 CDJ o 3·44 Ca 34·33 0·003 5%. 0 9·6 504 47·28 0·44 tiCOJ 0 179,28 CI 817'98 1'02 CI 5·5 156·55 Na 685·09 15451 K18 64095

0-00- - 92 101100 97102 86 99 78

Fig. 5.2. Dendrogram produced by SAHN classification of the water chemistry data showing attribute~ in order of their magnitude of contribution to each dichotomy. Figures are group means for each attribute; carbonate (C03), bicarbonate (HC03), chloride (Cl), sulphate (S04). calcium (Ca), magnesium (Mg). sodium (Na), and 1 potassium (K) in m-equlv/l- , conductivity (KI8) measured as ~S cm at 18°e, salinity (s%o) in parts per thousand. 88.

values for carbonate, salinity and sulphate into group 103, which contained no carbonate and had extremely low values for salinity and sulphate, and group 104 which had slightly higher average values for these parameters. Group 104 was subdivided into final group 92 whose members had an average carbonate content of 0.84 ~equiv/l-l, and an

average salinity of 0.5 0/00, and group 101 which contained no

carbonate and had an average salinity of 0.24 0/00.

The more saline group 108 was divided into groups 105 and 106 on the basis of average values for sulphate, chloride, sodiun, comuctivity, salinity, calciun, magnesiun am potassiun. The average

salinity of group 105 was 1.4 0/00, whilst the members of group 106

had an average salinity of 10.9 0/00 and correspomingly higher values

for cations, chloride and sulphate. Group 105 was split into final

groups 102 and 86 on the basis of carbonate, potassiuu, sodiun and chloride concentrations. Group 106 was divided to produce final groups

99 and 78. Group 78 had substantially higher values for calciuu,

sulphate, chloride, sodiun and comuctivity, but contained no carbonate.

Fig. 5.2 shows the attribute contributions to the classification as average values. Canparison of the ranges of values for these chemical parameters wi thin and between t.~e groups indicates a degree of overlap between groups. Fig. 5.1 indicates the parameters which best discriminate between groups at each level of the hierarchy. Table 5.3 Membership of groups generated by classification of water chemistry data.

Group Site Group Site number number 101 6 Floating Island 102 2 Break No.2 Swamp 9 Lake Elusive 5 Kanawinka Swamp 10 Lake Bong Bong 7 Cobrico Swamp 11 Dock Inlet 8 Deep Lake 12 Ewing Marsh 35 Lake Bridgewater 13 Swan Lake (W.Vic.) 36 Cotters Lake 14 Little Lake Malseed 41 Bleak House Lagoon IS Gelliondale Swamp 50 Little River 16 The Long Swamp 52 Bay of Islands Swamp 17 Dereel Lagoon 18 Two Tree Swamp 99 27 Champion Lake 23 Victoria Lagoon 30 St. Mary's Lake 26 The Duck Holes 37 Bitter Lake 31 Middle Lake 38 Swan Lake (E.Vic.) 33 Lake Kanagulk 39 Rockbank Swamp 43 Lake Barracouta 40 Army Reserve Swamp 49 Banyule Billabong 51 Round Lake 54 Lake Winchel sea 100 3 Tremaines Swamp 4 Brown Reedy Swamp 86 20 Lake Lookout 22 Bittern Lagoon 21 The Swannee 24 King River Meander 25 Lake Albacutya 29 Loch Garry 53 Lake Struan 32 The Stony 55 Lake Terangpom 44 Lake Beadle 45 Brisbane Ranges Swamp 78 28 Lake Wandella 46 Fernbank Swamp 42 Lake Tutchewop 48 Kangaroo Swamp 97 34 Heywood Golf Club Swamp 92 1 Pal parra Settlement Swamp 47 Mackenzie Road Swamp 19 Cockpit Lagoon 89.

Water Chemistry Groups

Membership of the eight final groups is shown in Table 5.3. Table

5.4 shows the range of values for water chemistry parameters for each group. The manbers of group 78, Lake warrlella (28) arrl Lake Tutchewop

(42) are sometimes used to store saline waters which would otherwise reach the Murray River. '!be salinities recorded at the time of sampling exceeded 40 0/00 arrl the sparse Lepilaena cylindocarpa was almost dead. TOtal ionic concentration as evidenced by corrluctivities of 51400 and 76800 ps respectively YJere the highest recorded in any vegetated water bodies sampled, as YJere the levels of chloride and sodiun.

The salinity of the waters of menbers of group 99 ranged fran 6.27 to 17.1 0/00, and corrluctivities YJere in the range 8960 to 22800 pS.

This group also had much higher concentrations of sodiun arrl chloride than any other group except 78. Group average values for carbonate, sulphate, calci1.m, magnesiun arrl potassiun YJere higher than those for all groups except 78, but the 10YJest values for each of trese ions fell wi thin the rarge of values recorded for other groups. The waters

YJere generally alkaline with pH exceeding 8.5 and slightly turbid, arrl highly turbid at Rockbank Swamp (39). All sites except Swan Lake East

(38) were on the basalt plains of western Victoria or in the Wilnnera region. SWan Lake East (38), located in coastal East Gippsland, has brown waters with a pH of 7.7 and no carbonate. The ionic concentration of this lake, which receives significant fresh water

inflow fran a mrnber of small streams, is attributed to the influx of tidal water fran Sydenham Inlet via a connectirg channel. Table 5.4 Ranges of values for water chemistry parameters of the eight groups generated by classification of

1 1 water chemistry data. Salinity as 0/00 , conductivity as ~S cm- at 18°C, major ions in m-equiv/l- •

1 nd. = not determined, * = x 10- •

Group number 97 100 92 101 102 86 99 78

Salinity 0.07- 0.6 o -0. I 0.2 -0.4 0.2 - 0.5 0.7 - 2.05 1.3 - 2.8 6.3 -17.1 >40

Conduct i v i ty >150 44-283 279-521 191 -871 786-1230 2100-4080 8960-22800 51400-76800 Carbonate o o 0.80-0.88 o o - 0.53 0.64- 4.28 0 -10.52 o Bicarbonate o 0.03-1.29 0.50-2.01 0.12- 3.33 0.59-10.82 2.85- 8.00 0.65 -10.66 2.73- 3.95 Chloride 0.51-10.38 0 -2.33 1.86-3.78 2.54- 7.45 5.66-22.13 19.60-29.30 82.99 -278.0 716-981 Sulphate nd. 0.01-0.06 0.05-0.08 0.33- 0.33 0.02- 4.00 0.58- 1.59 1.32 -16.27 16.44-46.6

Ca I dum 0.01- 0.35 0.01-0.18 0.49-0.52 0.08- 1.46 0.28- 4.86 0.98- 1.79 0.45 -10.50 1.11-67.46

Magnesium 0.18- 2.08 0.24-0.94 0.37-1.53 0.52- 3.47 1.04- 6.63 3.43- 9.79 3.41 -66.03 41.70-46.9 Sodium 1.96- 9.29 1.57-2.83 2.82-3.81 2.82-10.82 5.18-25.4 17.22-42.22 10.37*-25.73* 59.6*-74.4* Potassium o - 0.27 0.02-0.23 0.01-0.08 0.06- 0.59 0.05- 0.39 0.17- 1.18 0.85 - 3.77 4.6 - 4.7 pH 4.1 - 4.3 5.0-7.4 8.9 - 9.6 5.6 - 8.9 7.1 -9.0 8.7 - 9.9 7.7 - 9.9 8.1 - 8.2 90.

Waters of the wetlands in group 86 have salinities ranging from

1.3 to 2.8 0/00, whilst values recorded for conductivity range from

2100 to 4080 fS. The waters are alkaline (pH 8.6-9.9) and slightly brown or slightly turbid depending on wind conditions and the density of floating macrophyte growth. Chloride and sodiun (except for Lake Struan 53) values are intermediate between group 102 and group 99.

All the sites have sandy or clayey substrates, are located in the Mallee region or in the Western District on Tertiary outliers, and except for Lake Albacutya which fills episodically, all have permanent waters.

Gro~~ 102 includes Lake Bridgewater (35), a very clear lake on calcareous sands, Cotter's Lake (36) and Bay of Islands Swamp (52), very shallow swamps on calcareous substrates, Break Number TwO SWamp (2) and Kanawinka SWamp (5) in south-western Victoria, Cobrico Swamp (7) a peat-filled maar, and Bleak House Lagoon (41) a shallow lagoon and turbid Deep Lake (8) on the basalt plains in the Western District.

The salinity of the waters of these sites ranges from 0.4 0/00 to 1.15

0/00 and corrluctivities are between 786 and 1230 pS. Salinity, conductivity, chloride and sodium concentrations for Lake Bridgewater (35) are significantly lower than those for other manbers of group

102. They fall wi thin the range of values for group 101, and it is suggested that Lake Bridgewater has been misclassified. A more appropriate allocation would place it with Lake Bong Bong (10) and Lake Malseed (14), other coastal sites on calcareous sediments in group 101. Cotter's Lake (36) and Bay of Islands Swamp (52), other coastal sites on calcareous sediments are located on the western side of Wilson's Promontory and on the clifftops at Port Campbell respectively. The higher values for salinity, conductivity, chloride 91.

and sod.iun recorded for these sites is probably due to seaspray.

Little River (50) joins group 102 very late in the hierarchy. It has

higher levels of sodiun, chloride and thus higher salinity than members of this group, but its very low level of cations has placed it

in group 102 rather than group 86. The ionic balance for this site

(Table 5.2) indicates a cation insufficiency Which may be due to

analytical error.

Group 101 inclooes: coastal lakes and swamps in eastern and

western Victoria such as turbid water Middle Lake (31), Lake Kanagulk

(33), Two Tree SWamp (18) and Floating Island (6, which is in the

stony rises region of the western District basalt plains); and inland

swamps on sandy terrain such as Dereel Lagoon (17), The Long Swamp

(16), Victoria Lagoon (23), The Duck Holes (26), all of which have

brown waters, and turbid Banyule Billabong (49). The salinity ot

these sites ranges frcrn 0.18 0/00 to 0.5 0/00 , whilst conductivity is

between 191 and 871 pS. The range of ionic concentrations recorded for members of this group overlaps with those for groups 102 and 92,

and in most cases with group 100. lwDst of the members of this group

have peDllanent brown waters and pH ranges fran 5.6 to 8.9.

Palparra Settlanent SWamp (1) and Cockpit Lagoon (19), members of

group 92, have fresh waters (salinity 0-0.2 0/00) which are highly

alkaline (pH> 8.9) with carbonate concentrations similar to sites

whose salinities exceed 2 0/00. Calciun levels are similar to those

for freshwater sites on calcareous substrates but magnesium, sodium,

potassiun, sulphate and bicarbonate levels are wi thin the ranges

recorded for groups 100 and 101. ·41 I 102 2. .35

7. ·36 86

·6 49. 101 ·10 .12

.18 ·31

9· .43

.23 17· 10

.27

99 100

78 97

·30 .37

L ~

Fig. 5.3a. Scatterplot produced by PCOA ordination of water chemistry data for 55 Victorian wetlands. Isolines indicate ordination space shared by the eight groups generated by classification of the water chemistry data. Group numbers are shown outside isolines.

Fig. 5.3b. Overlay showing ordination space shared by water chemistry groups if the five group level is accepted.

""- ~~ -, F ",,'41---~ ~2 I.U 102 2. '35

7. '36 86

~~ "~ ~~ .50 ~

1[)

'40

99 100

78 '39 97

.30 .37 ~ lc '" -- -.lI

Fig. 5.3a. Scatterplot produced by PCOA ordination of water chemistry data for 55 Victorian wetlands. Isolines indicate ordination space shared by the eight groups generated by classification of the water chemistry data. Group numbers are shown outside isolines.

Fig. 5.3b. Overlay showing ordination space shared by water chemistry groups if the five group level is accepted. 92.

The waters of members of group 100 are shallow (except for Lake

Beadle 44), brown and generally acidic (mean pH 6.3) with low conductivities and very low salinities « 0.1 0/00). No carbonate was recorded. TOtal ionic concentration for each site did not exceed 7

~equiv/l-l; however the values recorded for individual ions generally overlapped with the lower limits for groups 92 and 101.

Members of group 97 have very acidic waters (mean pH 4.2). The significantly higher ionic concentrations of the waters of Heywood

Golf Club SWamp (35) are thought to be due to the geology of p:!.rt of its catchment which lies ~n basalt terrain. Additional input may came fran drainage of nearby pasture. l"Iackenzie Road SWamp (47) is situated on deeply weathered Tertiary sands.

principal Coordinates Analysis

Fig. 5.3a shows the scatterplot produced by FCOA of the water chemistry data shown in Table 5.2. A camp:!.ratively low proportion of the variation (55.02%) was recovered by the four princip:!.l axes.The horseshoe shaped disposition of sites represented in two dimensional space is due to distortion rather than any real structure in the data.

Such "archil distortions have been attributed to a lack of independence of the axes (Hill 1973, Gauch, 'Whittaker and ~ntworth 1977, Hill and

Gauch 1980) I which in most indirect ordination techniques are orthogonal and therefore uncorrelated but not necessarily independent. on the first and second axis the sites are generally arranged according to their position along the salinity gradient. Same involution has occurred and the most saline sites I Lake Wandella (28) and. Lake Tutche\vop (42) are located inside the arch. The more central 93.

position of Little River (50), which joined the classificatory hierarchy at a late stage, indicates that its water chemistry is ananalous. Sites 34 (Heywood Golf Club Swamp) and 47 (Mackenzie Road) which were grouped together by the classification, are not closely related in two dimensional space. Palparra Settlement (1) and Cockpit

Lagoon (19) show sane affinities with members of group 101.

Isolines indicating the shared two dimensional ordination space of the eight groups identified by classification suggest that eight groups may be distinguished on the basis of water chemistry parameters since the groups do not overlap. Fig. 5.3b indicates the relationship between the groups at the five group level. At this level the groups are less canpact, and more heterogenous, but no more isolated than at the eight group level.

4. Discussion

The ionic daninance ratios reported in this study are characteristic for Australian waters, the usual order of cation daninance for saline waters being Na > Mg > Ca > K, and for fresh waters Na > Mg > ca > K or Na > Ca > Mg > K (Williams 1967). The order of anionic daninance is more variable; chloride usually daninates, although same exceptions were recorded where HC0 > Cl in 3 waters with very low cooouctivity. Tim:ns recorded HC0 > Cl for 3 on the floodplain (1973)! and in an abandoned course of the (south western Victoria) (1977). Williams

(1967) has suggested that the usual order of anionic daninance for fresh waters (i.e. < 3 0/00 salinity) is Cl > HC0 + c0 > 804' 3 3 although Buckney (1980) suggests that calciun, and to a lesser extent 94.

magnesium and bicarbonate, may be of greater imp:>rtance in some areas than early data indicated. In the present study Cl > HC0 + C0 > S04' 3 3 The expected order of dominance for saline athalassic waters is Cl >

SO4 > HC03 + C03, which was recorded for the more saline sites investigated and same freshwater coastal sites (Lake Elusive 9, Dock

Inlet II, arrl Bay of Islarrls 52) •

Detenninations of the major ionic constituents published for Deep Lake (8) (Maddocks 1967), Lake Elusive (9), Dock Inlet (11), Lake Barracouta (43), Lake Be!adle (44) (Timms 1973), Lake .Bong .Bong (10),

SWan Lake Wi::st (13) and Lake Bridgewater (44) (Tirrrns 1977) show the same orders of ionic dominance recorded in the present stooy.

Mechanisms controlling the chemical comp:>sition of surface waters inclooe abnospheric precipitation, the nature of the lithology arrl the evap:>ration-fractional crystallisation process (Gibbs 1970). In the present study, same general relationships between total ionic concentration and geology, and water transparency and the nature of the substrate are suggested. Wetlands with catcrments predominantly in areas of the basalt plains covered by heavy clay soils usually have total ionic concentrations> 350 m-equiv/l-l , with salinities> 10

0/00 and generally turbid waters. The waters of sites sampled within highly calcareous substrates are very clear arrl alkaline (pH usually> 8.1) • Often calcium concentrations (m-equiv/l-l ) exceed those of magnesium, and evidence of calcium carbonate precipitation is cammon,

On Tertiary sediments (which frequently inclooe calcareous sands) waters are generally alkaline, with salinities ranging fram 0.2 to 2.0

0/00. With the exception of Lake Kanagulk (33) where suspension of clay particles results in "milkyll turbid waters, waters are generally 95.

only slightly coloured.

In contrast, wetlands sampled on siliceous coastal sands are characterised by brown waters with pH ranging from 6.0 to 7.0, and salinities of 0.2 - 0.4 0/00 • The size and open situation of Lake

Barracouta (43) makes its brown waters slightly turbid. Sites located on inland siliceous sands in south western Victoria and East Gippsland also have brown waters. pH ranges from 4.9 to 7 and salinity is

<1 0/00•

Sites sampled in north western Victoria near BJenhope, Douglas and

Kerang lie largely in aeolian sands and lacustrine se:J.iments. The waters of these sites are slightly turbid-brown or turbid depending on the sandiness of the substrate, and usually alkaline, but the salinity

is very variable and is probably dependent on the nature of the aquifer in which the basin lies.

The sampling program did not pennit examination of p:>ssible

temp:>ral variation in the water chemistry of the wetlands sampled.

However, observations suggest that for many sites there is considerable seasonal fluctuation in water level, and that long-tenn

variation may be irregular. Mark....:::.d variation in water level has

particularly been noted in sites with stream inflow, such as Swan Lake

west (13), swan Lake East (38) and Lake Elusive (9). Williams (1976,

1981) and Williams and Buckney (1976) have suggested that while marked

fluctuations in salinit"J occur in saline lakes, ionic prop:>rtions

remain relatively constant. Little is known about variation in total

ionic concentration or ionic prop:>rtions for fresh or slightly saline

Victorian waters. Comparison of values detennined in the present study 96. with published results (TllnmS 1973, 1977) suggests that there is same seasonal variation in total ionic concentration•

.Criteria· which may be used to assess the cd.equacy of a classification have been discussed in Olapter Two. principal coordinate ordination suggested that classification of the water chanistry data at the eight group level proouced relatively hanogenous groups which were sufficiently isolated not to overlap, although the ranges of values for many specific attributes may overlap (Table 5.4) •

The groups may best be distinguished on the basis of salinity. Exceptions include group 97, which falls into the range of salinity values recorded for several groups; and group 92 which was separated on the basis of low pH values which may be due to the presence of

Sphagnum sp. at these sites (Clymo 1967). Salinity values for group 92 and a single member of group 102, Little River (50), overlap with values for groups 101 and 86 respectively. The foonation of group 92 is a result of the use of a space-dilating sorting strategy which is

"inherently likely to produce one or more non-conformist groups whose members have little in carmon beyorrl the fact that they are rather unlike everything else" (Lance arrl Williams 1967). ECOA ordination suggests that Little River is an outlier in this data set, and thus difficult to classify satisfactorily.

Groups which were recognised intuitively can be characterised as

1. saline waters (salinity> 3 0/00) ,

2. slightly saline (salinity 1-3 0/00),

3. fresh (salinity < 3 0/00 Williams 1964) turbid waters Secchi disc depth < 20 am, 97.

4. fresh very clear calcareous waters pH > 8.00, 5. fresh acidic waters pH < 4.5, l 6. fresh waters, salinity < 1 0/00, corrluctivity < 300 psan- , l 7. fresh waters, salinity < 1 0/00, corrluctiv i ty > 300 )lSan- ,

Groups recognised by mmerical classification of the water chemistry data at the eight group level inclooe: group 99 where salinity > 3 0/00, group 86 where salinity 1-3 0/00, group 97, with fresh water pH 4.5. Group 78 inclooes highly saline sites (> 40

0/00). Groups 100, 101 and 102 all inclooe freshwater sites Which were regarded as having turbid or clear calcareo'.JS waters, and those

Where salinity < 1 0/00, and corrluctivities 40-1300 psan-l • These freshwater groups can be best separated on the basis of their salinity

(ie. total ionic concentration) •

The turbid and clear-water calcareous groups were not recognised within the classification because the water chemistry parameters, salinity, corrluctivity, pH and the concentrations of major ions, did not reflect the characteristics of such waters. The chemistry of t.."le turbid and clear calcareous waters sampled (as described by the attributes listed above) does not appear to be sufficiently different

from other freshwate~ sites sampled to warrant recognition as separate groups. In the Victorian wetlands sampled, turbidity is due to the presence of suspended silts or clays in the water, and may be more closely related to the texture of the bottom sediments than to any particular water chemistry parameter. While some of the fresh calcareous waters sampled had distinctive cationic dominance ratios where Na > Ca > Mg > K, calcium concentrations were not higher than

t..'ose for magnesium in all of the fresh, calcareous waters sampled. 98.

This may have been due to the observed precipitation of calcium carbonate. Calcareous dune lakes have been recognised as a hanogenous group by Timns (1977) who differentiated them from lakes found on siliceous dune sands on the basis of their genesis, water chemistry and fauna. lldditionally both fresh, clear calcareous waters and turbid waters have distinctive macrophyte floras.

5. Conclusions

The 55 aquatic macrophyte habitats sampled occur on a range of lithologies, Tertiary sands and clays, Quaternary basalts, Recent alluvial, lacustrine and inland aeolian deposits and Recent coastal dune deposits, both siliceous and calcareous. water transparency varied fran very clear on calcareous dune sands, to brown on other sandy substrates, whilst water over clay substrates was turbid and relatively opaque. Considerable fluctuations of water level occur in many of the sites sampled, especially those which receive stream flow.

Much of this fluctuation is seasonal, how.:ver, longer term patterns of variation sean to be irregular. Water level fluctuations are probably accanpanied by changes in salinity. How.:ver evidence suggests, at least for saline lakes, (Williams and Buckney 1976) that changes in total ionic concentration are not accanpanied by proportional changes in ionic concentration.

Cllemical analyses of the waters sampled sho~ that macrophytes grow in a wide range of salinities, although species diversity is low where salinity> 1 0/00, and only species such as Lepilaena bilocularis, Lepilaena cylindrocarpa, Myriophyllum muelleri and Ruppia 99.

maritima carmonly occur in waters whose salinity> 3 0/00 • M:>st of these saline (>3 0/00) sites occur on the basalt plains of W;stern victoria, or in the Mallee and Wimnera regions of north ~stern

Victoria. Yezrlani (1970) ha.s suggested that the inland saline lakes in W;stern Victoria are largely the result of the closed and ephemeral drainage system, the low rainfall and high evaporation.

Salinity exerts "a critical and probably canplex influence"

(Sculthorpe 1967) on the distribution of aquatic vegetation. '!be orders of ionic daninance and total ionic concentrations reported in the present study show that chloride and sodiun daninate most

Victorian lentic waters, and support Williams (1981) contention tb~t

Australian lentic waters do not conform to the worldwide standard canposition proposed by Conway (1942), Rodhe (1949) and Benoit (1969).

Thus it is suggested that salinity may be a major influence affecting the distribution of aquatic macrophyte vegetation in lentic waters in Victoria. 100.

CHAPI'ER S I X

vegetation - environment relationships

1. Introduction

In the present study classification and ordination of the vegetation of the 55 wetlands identified same distinctive groups of wetland plant communities. Investigation of same of the physical and chemical characteristics of these cannunities, and subsequent classification and ordination of these data suggested that a similar number of aquatic environments could be distinguished. However, the groups identified by numerical analyses of the floristic data (Table

3.8) did not have the same membership as those distinguished on the basis of water chemistry parameters. It was suggested in Chapter Five that this may have been due in part to the choice of water chemistry attributes.

It was decided to further investigate the nature of the hypothesized joint pattern between the floristic and environmental attribute sets in order to find how much of the observed variance in the vegetation could be related to the environmental parameters measured, which of these para~eters were most highly correlated with the variance and whether any underlying causes of community distribution were indicated.

A major limitation of most currently available strategies for the analysis of patterns shared by two data sets is their derivation from un. models which imply that attribute distributions are independent of one another and linearly or at least monotonically related to causal functions. In methods which use significance tests to evaluate the separation of species groups on environmental variables, normality of within group distributions and hanogeneity of within group variances or variance-covariance matrices are also assumed (Green and Vascotto 1978). Additionally, all statistical methods require some element of randomness with regard to the selection of the sample data (Scott

1974) •

The nature of s~ies distributions along environmental gradients was discusserl in Chapter Four; clearly the assumptions of independence, and linear or monotonic relations to causal functions are untenable, and Williams and Lance (1968) have suggested that the distribution of some attributes may be so highly ske-weCi or discontinuous that no transformations can provide even approximate linearity. Many ecological data sets may not meet the required assumptions of univariate or multivariate normality, and thus any significance tests must be interpreted very cautiously.

2. Choice of strategy

A. Canonical Correlation Analysis Canonical Correlation Analysis can be used to ascertain the extent to which one set of measurements is correlated with another and the particular attributes which are responsible for these correlations. The basic strategy of Canonical Correlation Analysis is to find an axis fram each data matrix ~,at is most highly correlated with the corresponding axis fram the other matrix, but orthogonal to all other 102.

axes from both data sets (campbell 1982). However, this method has been little used in the comparison of biological data sets, since such data often violate the method's assumption of linearity (Gauch and

Wentworth 1976) and, the resultant vectors are often very difficult to interpret (Williams 1976c) •

However, a modified method of Canonical Analysis, Canonical Coordinate Analysis, has frequently proved informative. This version

utilises the vectors produced by principal Coordinate Analysis (PCOA) which, by definition, are orthogonal. Some of the noise inherent in most data sets will have been discarded with the PCOA vectors which explained only a small percentage of the observed variation. The PCOA axes are themselves linear combinations, and there is at least same

hope that any non-linearities will be partitioned into somewhat . similar linear components in the two data sets. The use of the PCOA vectors also overcomes any problem of missing values.

In the present study, the first four vectors from the TAXON program PCOA (used to ordinate both the floristic and water chemistry data sets) were input to CAOCORT (ROSS 1982) which carried out the canonical correlation analysis on these orthogonal vectors. CAOCORT provides two sets of canonical vectors with corresponding roots and canonical correlations and the coordinates of the original attributes

on the new vectors. The program BACRIV (ROSS 1982) was subsequently used to provide correlations with the original data attributes for the

canonical vectors. Each of the entire, "eident" and aquatic-semi­

aquatic floristic data sets were analysed using C~ORT. 103.

B. Analysis of variance Although correspondence between the groups generated by floristic classification and those identified by classification of the water chEmistry data appeared poor, it was decided to evaluate the separation, if any, of the floristic groups on the basis of the water chemistry parameters.

To date, examination of the wetland data sets has utilised pattern analysis techniques in order to search for any structure in the data which would allow econanical description and suggest hypotheses to account for the observed floristic variation. However, classical statistical techniques are required to test the hypothesis that the floristic groups have significantly different water chEmistry attributes.

Analysis of variance provides a suitable technique to test whether subgroups within a data set are significantly different from one another. The underlying assumptions of analysis of variance include randan sampling of individuals and independence, equal variance and a nonnal distribution for the error tenns (Sokal and Rohlf 1969) •

A one way analysis of variance was carried out for each water chEmistry variable using the ANOVA directive fran the GENSTAT program (Alvey et al. 1979). The apparent heteroscedasticity in the data set was removed by application of a log transfonnation (log (y + 0.1» for all variables, except pH, prior to analysis. Protected Least Significant Differences between the means of groups were calculated (Snedecor and Cochran 1980) • 104.

Although univariate analyses can provide valuable infoonation, they may not discriminate between groups where differences are not marked. They also ignore the dependence existing between variables which may well be a source of differentiation. Williams and Lance (1968) have suggested that the responses which exist between the dependent and independent variables must always be in the same sense, and that where this is not the case, a negative result is not evidence of lack of relationship.

C. Discriminant analysis

Where a univariate approach does not adequately separate groups, another statistical technique, Discriminant Analysis (also calle::1 Canonical variate Analysis) may be used to find linear cornbinat:ons of variables (discriminant functions) which maximise group differences. The discriminant functions are define::1 by coefficients derived by

solving the general eigenvector problen WA = A Ba, where B and Ware,

respectively, the between and within-group sums of squares and cross products matrices.

The discriminant functions are of the form

Di = dil Zl + di 2 Z2 + ••• + dip 2p

where Di is the score on the discriminant function i, the d's are the weighting coefficients and the Z's are the standardised values of the p discriminating variables used in the analysis (Klecka 1975). For a complete description of the method, see Cooley and Lohnes (1971). 105.

A geometric interpretation of Discrlininant Analysis is given by plotting each groups as a point in space where each discriminant function is a unique dimension describing the relative location of that group. The points are projected onto a plane or hyperplane to accentuate the separation between the groups, and thus provide a plot of the canonical variates (Dixon 1975). These canonical variates are related to Canonical Correlation Analysis, which as previously discussed, finds the linear canbinations of two sets of variables which are most highly correlated. If one set of variables is a dummy set indicating group membership, then Canonical Correlation Analysis becomes Discriminant Analysis and equivalently, multivariate analysis of variance (Muller 1982).

The computation procedures for Discrlininant Analysis include an initial separatory function wherein a set of variables which provides satisfactory discrimination betweeen groups of known membership is

identified, and a classificatory function which provides an assessment of the adequacy of the discriminant function by classifying the original set of individuals to see how many are correctly assigned by the variables being used. B. Williams (1981) has demonstrated that there is a statistical relationship between Discriminant Analysis and minlinum error classification procedures (assuming nODmality and equal covariance relationships) •

The statistical theory behind Discriminant Analysis assumes that the discriminanting variables have a multivariate nODmal distribution

and equal variance-covariance matrices for each group. Klecka (1975) has pointed out that the technique is very robust, and that these assumptions need not be strongly adhered to. Other characteristics of 1@6.

the data set which may lead to difficulty in the application of

Discriminant Analysis techniques can incllrle missing values, possible singularity, and lack of linearity between the environmental variables

(ie water chemistry variables). Since the program automatically deleted sites with missing values, two analyses were carried out, one wi th all the water chemistry variables incllrled (14 sites were thus deleted) , and one without bicarbonate, carbonate and sulphate to allow inclusion of all 54 sites.

Discriminant analysis may be performed either by entering all discriminating variables directly into the analysis, or through a variety of stepwise procedures which select the optimal set of discriminating variables. Utilisation of a stepwise procedure provides the most parsimonious analysis and is mcst appropriate when it is thought that the full set of independent variables may contain excess information about the group differences, or that same of the variables may not be very useful discriminators (Klecka 1975).

However, Lachenbruch (1975) points out that the use of stepwise programs may be hazardous, since it is possible that same of the nbest" variables selected may merely be noise. Green and Vascotto

(1978) suggest that stepwise procedures might be appropriate if it is possible to assune that the original environmental variables (water chemistry) are themselves the independent environmental factors controlling species distributions. Stepwise discriminant analysis may provide an effective subset of discriminant functions, but there is no guarantee that these are the controlling factors. In the interests of canparison, both stepwise and direct methods were carried out. un.

The Program BMDP7 (Dixon 1975) was used to carry out the discriminant analyses. In the direct method, all variables are entered into the analysis concurrently and the discriminant functions are calculated directly fran the entire set of variables. When the steP'Wise option is invoked, variables are entered into the classification function one at a time until the group separation ceases to improve noticeably. Initially, the standard univariate

analysis of variance test is made for each of the variables, and the variable for which the means differ most is the first to enter. In subsequent steps the computed F-to-enter values are conditioned on the variables already present in the function. Thus variables are selected on the basis of their power to add to the discrimination, which may not necessarily be the same as the order of their individual usefulness. Previously selected variables are tested at each step to determine whether they still make a sufficient contribution; any redundant variables are removed.

The output produced by the program at each step includes F statistics which can be used to test the equality of means between each pair of groups and provide an indication of the relative distances between group means, a classification table based on prior probabilities, and a "pseudo jacknife" classification table which can be used for cross-validation purposes. The jacknife procedure reduces bias inherent in the group classification by computing a classification function with each case omitted in turn, and then using the function to classify the left-out case (Dixon 1975). After all the variables have been entered, the program lists the Mahalanobis 02, the difference between the group means of the discriminant function (Lachenbruch 1975) from each case to the centre of each group, and the 108. posterior probability of the case being assigned to each group. The posterior probabilities provide an idea of how well each case is classified. A high percentage of correct classifications indicates that group differences do exist and that the variables selected display these differences.

The output also provides the percentage of correct classifications, the eigenvalues, which are a measure of the relative importance of the functions, and their associated canonical correlations which are measures of association betweeen each discriminant function and the set of dtnmy variables which define the group memberships (Klecka 1975) •

The assistance of Dr. D. Ratcliff, CSIRO Division of Mathematics and statistics, St. Lucia, Queensland, who carried out the analysis of variance and discriminant analyses on my behalf is gratefully acknowledged.

3. Results

A. canonical Correlation Analysis

Table 6.1 shows the canonical correlations for the first three vectors from each of the floristic data sets with the water chemistry vectors, and suggests that there are significant relationships between the first two vectors for the water chemistry and the floristic data sets. However, the probabilities given must be treated with caution, since they assune that the data are multivariate normal in distribution (ROSS 1982). As anticipated in Chapter Three, the first Table 6.1 Canonical correlations for the first four floristic vectors with the first four water chemistry vectors from the analyses using the entire, "e ident" and aquatic-semi-aquatic species data sets. Probabilities are shown in brackets.

Floristic data Canonical correlations set Vec 1 Vec 2 Vec 3 Vec 4 ent ire 0.715 (.000) 0.583 (.005) 0.215 (.616) 0.072 (.606)

"eident" 0.786 (.000) 0.569 (.008) 0.225 (.612) 0.033 (.815) aquatic- 0.794 (.000) 0.576 (.008) 0.185 (.773) 0.025 (.857) semi-aquatic 1139.

vector for the reduced floristic data set containing only aquatic and semi-aquatic species shows the best correlation with the first water chemistry vector, although the differences between the correlation coefficients for the data sets are not large. Only the results for the aquatic-semi-aquatic species data set will be further presented.

Table 6.2 displays the results of the CANCORT analysis of the above data set. The coefficients for each of the floristic and water chemistry vectors shows the correlation between the input PCOA vectors and the canonical vectors, and suggests that in the case of the first floristic canonical vector and the second wdter chemistry canonical vector, there are two dominant elements which define the direction of these vectors. Thus PCOA results will not be useful in interpreting the results of the canonical correlation analysis.

However, the program BACRIV provided correlations for the canonical vectors with the original data attributes; the attributes

fran each data set which are most closely correlated with the canonical vectors are also most closely correlated with each other

(Clifford and Stephenson 1975). Tables 6.3 and 6.4 show the back correlations of the original data with scores for the first and second canonical vectors respectively. Back correlation with the masked

attributes, data not used in the calculation of the respective canonical vectors, provides an additional aid to interpretation,

particularly since two-tailed significance indicators are given for

them (although their use is subject to the usual caution). The masked

attributes listed all had probabilities < 13.13131; whilst only those

attributes remaining after masking which had correlations> 13.5 with

the respective canonical vector, are shown. Table 6.2 Results of the canonical correlation analysis of the aquatic­ semi-aquatic species data set with the water chemistry

attributes.

Vec 1 Vec 2 Vec 3 Vec 4 Canonical Correlations 0.794 0.576 0.185 0.025

Probability 0.000 0.008 0.773 0.857

Coefficients for VI -0.622 0.216 o. 156 -0.737 floristic vectors V2 O. ]20 o. 115 0.501 -0.467 V3 0.000 -0.955 -0.053 -0.291

v4 0.310 0.167 -0.850 -0.393

Coefficients for VI 0.977 0.132 -0. 168 0.000 water chemistry V2 -0.016 -0.728 -0.664 O. 171

vectors V3 -0.206 0.672 -0.668 0.244 v4 -0.056 0.041 -0.290 -0.954 110.

The first canonical axis indicates that there is a relationship between the first and secorrl plant vectors and the first water chemistry vector. EXamination of Table 6.3 suggests that the plant vectors are composed of species associated with saline and freshwater environnents respectively, whilst the water chemistry vector is composed of variables associated with saline waters. The second canonical axis indicates a negative relationship between the third floristic vector, to which species generally associated with fresh shallow brown waters make the greatest attribute contributions, arrl the second and third (not shown) water chemistry vectors which are associated with high pH and HC03 and high levels of most of the major ions respectively.

B. Analysis of Variance

The analysis of variance indicated that there were significant differences (p < 0.(01) between the eight floristic groups generated by classification of the entire data set for all the water chemistry variables measured. Table 6.5 shows, for each variable, which groups had significantly different means. Groups 84 and 96, the saline groups, were significantly different from almost all the other groups on most variables, but did not differ significantly from each other on any variable. Group 94 (acidic waters) was significantly different from the saline groups on all variables, and on the basis of mean values for pH arrl bicarbonate, differed significantly fram all other groups. The group of calcareous water sites (97) was Significantly different fram group 99 (turbid waters) on the basis of conductivity and chloride, and fram groups 100 and 102 (fresh brown water sandy Table 6.3 Back correlation of the aquatic-semi-aquatic species and

water chemistry data and the masked attributes (see text)

with scores for the first canonical vector.

Floristic attributes Water chemistry

LepiZaena biZoauZaris -0.6249 HC0 3 -0.7423 EZeooharis sphaoeZata 0.5395 pH -0.7289 TrigZoohin pPOoera 0.5323 Sa lini ty -0.7064

MyriophyZZum propinquum 0.5239 Magnesium -0.5526

C03 -0.5190

------masked attributes ------­

Water chemistry Floristic attributes

HC0 3 -0.6696 MYPiophyZZum propinquum 0.5166 pH -0.5999 LepiZaena biZoauZaris -0.4788

Sa I in i ty -0.5017 Ruppia maritima -0.4768

C03 -0.4675

Floristic dominance* Floristic dominance* LepiZaena biZoauZaris -0.6375 LepiLaena biZoauZaris -0.5217 EZeooharis pusiZZa 0.4889

Substrate and water clarity Substrate and water clarity pH 20 - 30+cm -0.6171 pH o - 10 cm -0.7203 pH o - 10 cm -0.6111 pH 20 - 30 cm -0.7050 clay s'Jbstrate -0.5848 clay substrate -0.6271 brown water 0.5645 %salts 20 - 30 cm -0.4519 brown water 0.4336

* species recorded as dominant at sample site + depth at which sample was collected from substrate Ill.

substrate and shallow freshwater respectively) on the basis of mean values for sulphate and calciun, but there were no significant differences between groups 97 and un. However, there were no significant differences between the freshwater groups 100, 101 and

102, and while group 99 could be distinguished fram group 100 (pH and chloride) and group 101 (conductivity and chloride), it was not significantly different from group 102.

c. Discriminant analyses

A surmary of the results of all the discriminant analyses undertaken is given in Table 6.6. Results for t.l-)e direct method will not be further discussed since, although the percentage of sites correctly classified on the basis of all the water chemistry '7ariables entered was relatively high, the percentage correctly classified following the application of the jacknife procedure dropped markedly.

It is suggested that this was due to the inclusion of variables which did not contribute to the separation of the groups. In the stepwise procedure only those variables which contributed to a useful separation were utilised.

However, for the stepwise procedures the level of agreement between the water chemistry variables and the floristic classification of the wetlands was not very high. 54.7% of the 54 wetlands were correctly classified on the basis of the three discriminant functions computed from the nine water chemistry variables, whilst only 14.6% of the 41 sites input were correctly classified by the single discriminant function calculated fram the twelve water chemistry variables (Table 6.7). These results indicate that the eight Table 6.4 Back correlation of the aquatic-semi-aquatic species data

and the water chemistry data and masked attributes (see text) with scores for the second canonical vector.

Floristic attribute~ Water chemistry

Myriophyllum propinquum -0.5716 no attributes correlated

Eleocharis acuta -0.5383 ~ 0.5000 Potamogeton tricarinatus -0.5156

------masked attributes ------­

Water chemistry Floristic

no attributes with p ~ 0.001 no attributes with p ~ 0.001

Substrate and water clarity Substrate and water clarity

no attributes with p ~ 0.001 no attributes with p ~ 0.001 112. floristic groups were generally poorly discrlininated amongst by the water chemistry variables.

Table 6.8 shows the coefficients of the discrlininant functions for these two analyses. The variables shown are those which did provide significant (p~ 0.05) separation of the groups of wetland sites. For the analysis utilising all twelve water chemistry variables, no further significant discrlinination was achieved after the entry of the first variable, magnesium. It discrlininated pJorly between the groups, essentially separating only groups 84 and 96 from the rest

(Table 6.9) •

In the second analysis, which utilised only nine water chemistry variables, the best separation of the ~~tland floristic groups WdS achieved after entry of the first three variables. The first discrlininant function accounted for almost 84% of the variation in the water chemistry data and was highly correlated with all three variables, sodium (r = 0.89), pH (r = 0.81) and chloride (r = 0.71), (p < 0.01) (Table 6.9). The second function was responding to both chloride (r = 0.67) and sodium (r = 0.41) and the third largely to pH

(r = 0.57). All three flIDctions contributed to the separation of groups 84 and 96 from all other groups, although they 'Were not separated from each other; group 94 was distinguished from the other freshwater groups on the basis of pH and chloride, and group 99 was separated fram the remaining freshwater groups 100, 101 and 102 on the basis of chloride. Group 97 (calcareous water sites) was not separated fram groups 101 and 102. T"ble 6.5 Results of the analysis of variance fOI' the >later chemistry v~riabl<-s mensured for "etlands of the eight floristic groups# (see Table 2.2) Identified by classification of the entire data set. ,', indicates a sign! flcant di fference betwe"n means (p <0.05)

Salrnlty Conduc t iv i t y pH 'the 8 floristic groups were characterised as 84 94 96 97 99 100 101 84 94 96 97 99 100 101 84 94 96 97 99 100 10 B4 sal ine

94 1, 94 1< 94 ,., 94 acidic water 96 ve ry sa line 96 96 '* 96 97 calcareous coastal 97 +- 1. * 97 '" ,., " 97 It '" 99 ,., 1< 99 1< * ~'r. 99 ,\ * ,\ 99 turbid ;, ,., 100 fresh brm.n water sandy 100 1< ,\ 100 * 100 ~, * '* 101 ,.. ,., 101 fresh brown water organic 101 1< 1, ,~ 101 ,', * 102 {, * 102 " * 102 * '" ,., 102 shallow fresh water

Carbonate Bicarbonate Sulphate Chloride B4 94 96 97 99 100 101 84 94 96 97 99 100 101 84 94 96 97 99 100 101 B4 94 96 97 99 100 101 94 ,., 94 ,., 94 * 94 * ,., 96 ,\ 96 It 96 " 96 ,., 1: 97 ,\ ~': 97 '" 97 '" 97 * 1; 9~ ;, 1: '* 99 ,', 99 '" * 99 '" 1; I< 100 1: 100 ,., .. 100 '" 100 " 1< '" '* * 101 ,., 1:. }~ 101 '* '1r 101 * 101 ;, 102 " '" 102 " '" 102 ,.. '* +: 102 " *

Calcium Magnesium Sodium PotassIum 84 94 96 97 99 100 101 84 94 96 97 99 100 101 84 94 96 97 99 100 101 84 94 96 97 99 loa 101 94 ,\ 94 *' 9/t '" 94 '" ,., 96 " 96 It 96 '" 96 97 1; 97 97 If " " 97 '" " " '" ,., 99 ,~ 99 ,., 99 " 99 It {, 100 ,.. '" " 100 '" " 100 " * 100 " " ,., 101 101 '* '" * 101 '" * 101 * ,., ,., :\ 102 102 ,\ It 102 " '" 102 ,\ " '" 113.

Discussion

A. Canonical Correlation Analysis

The results of the canonical correlation analysis suggest that there is a significant relationship between the distribution of aquatic species and the water chemistry parameters. The BACRIV analysis revealed that salinity and associated high concentrations of bicarbonate, carbonate and magnesium, and high values for pH are dominant factors affecting the distribution of Victorian aquatic species, relationships which were suggested in Chapter Five. However, it is suggested that since two of the original floristic PCOA vectors and two of the water chemistry PCOA vectors contributed substantially to canonical vectors one and two respectively, the relationships between species distributions and water chemistry could be more complex than those revealed by canonical correlation analysis.

While salinity is most likely a dominant influence over the whole range of wetland vegetation examined, it may overshadow more subtle relationships between other water chemistry attributes and species distributions. This may have been in part due to the choice of attributes; other parameters such as turbidity, water depth and water level fluctuation which may affect species distribution were not included in this analysis.

B. Analysis of variance

The results of the analyses of variance show that the eight floristic groups cannot be distinguished on the basis of any single Table 6.6 Summary of the results of discriminant analyses carried out

using stepwise and direct methods (see text). * Values for bicarbonate, carbonate and sulphate were deleted from these analyses due to the number of missing values in , the data. Where all 12 water chemistry variables were input,

the 13 sites mis~ing data were excluded from the analyses.

Method of No. of water Water chemistry %of sites Jacknifed analysis chemi stry var i ab 1es correct Iy classification variables entered classjfied %correct

Stepwi se 12 Mg 34. 1 14.6

9* Na, pH, C1 56.6 54.7

Direct 12 all 80.5 34. 1

9* 9 64.2 41.5 114. variable. However, they indicate that at the eight group level, groups 84 and 96 can be distinguished from all others on the basis of their salinity, but not separated from each other. The distinction between these two groups, made initially on the basis of classification of the floristic data, had not been maintained in the ordination of the floristic data either. At the eight group level, classification of the water Chemistry data resulted in two saline groups but their membership was not identical to that of the floristic groups. Group 94 can be distinguished by its pH, its waters being significantly more acidic than those of any other group, whilst group 97 (clear calcareous waters) can be separated from most other freshwater groups (except 101) by its relatively high calcium concentrations. There are no significant differences between any of the fresh brown water groups (100, 101 and 102) for any of the variables measured.

Analysis of variance of the floristic groups following reallocation of the misclassified sites (see Table 3.8) may have provided a basis for the separation of the freshwater groups 100, 101 and 102, particularly if the six sites which could not finally be satisfactorily allocated to any floristic group (Table 3.9) were removed from the analysis.

C. Discriminant analyses

The results of the discriminant analysis indicated that the eight groups identified through classification of the floristic data in Chapter Two cannot completely be discriminated between on the basis of the water chemistry variables. However, the analysis which utilised Table 6.7 The jacknifed classification of wetlands by the discriminant

analyses for the eight wetland floristic groups.

A, util ising 12 water chemistry variables, B, util ising

9 water chemistry variables.

A Group Percent Number of cases classified into group ­ correct 84 94 96 97 99 100 101 102 84 0.0 0 0 6 0 0 0 0 0

94 66.7 0 2 0 0 0 0 0

96 0.0 3 0 0 0 0 0 0

97 20.0 0 0 0 0 0 3

99 0.0 0 0 0 2 0 0

100 11. I 0 2 a 2 0 3 101 40.0 0 a 2 a a 2 0

102 0.0 a 2 a 0 a

Total 14.6 3 8 6 8 4 6 5

B Group Percent Number of cases classified into group correct 84 94 96 97 99 100 101 102

84 42.9 3 a 2 2 a 0 0 a

94 75.0 0 3 a 0 a 0 a

96 75.0 2 a 6 0 a a a a

97 60.0 a 0 3 0 a a

99 50.0 0 0 3 a a

100 50.0 a 3 0 a 6 2 0 101 40.0 0 a a 0 2 2 0

102 50.0 0 a 0 0 2 a 3

Total 54.7 7 6 8 9 3 12 4 4 llS.

nine variables effectively separated four groups: a saline group of sites (84 + 96), the acidic water sites (group 94), the turbid water sites (group 99), and a large freshwater group which included groups

100 (fresh brown water sandy substrate), 101 (fresh brown water organic substrate) and 102 (fresh brown shallow water) •

A fifth group of calcareous sites (97) could not be distinguished fran two of the freshwater groups, 101 and 102. Group 97 contained

Lake Barracouta (43), which was considered to be misclassified (see

Chapter Two), but did not include the calcareous Cotter's Lake (36) which was dropped fram the discrlininant analysis because of missing data. However, results of the analysis of variance (in which all the calcareous sites were included) indicated that the calcareous group could be separated from groups 100 and 102 on the basis of calcium and sulphate concentrations. It is suggested that the presence of the anomalous calcareous site palparra settlement Swamp (1) resulted in an overlap between group 101 and the calcareous group.

If the eight floristic groups had been clearly separable, the discrlininating chemical variables, sodium, pH and chloride, could have been considered to be closely associated with the variance observed in the wetlands vegetation. It is expected that a significant amount of the variation in the vegetation at the proposed five group level

(saline, turbid, calcareous, acidic and freshwater groups) could be related to the water chemistry variables contributing to the discrlininant functions. However, it is clear that factors other than those included in the analysis are also influencing the distribution of species, particularly within the freshwater sites of groups 100,

101 and 102. Table 6.8 Coefficients for the discriminant functions and the

correlation coefficients (r, in brackets) of the variables with the discriminant scores in the nine variable stepwise discriminant analysis. The percentage of the total variation accounted for by each discriminant function is shown. For 51 degrees of freedom, the critical values of r are 0.27 (p <0.05) and 0.35 (p <0.01).

Coefficients for discriminant functions

(83.79%) I I (11.05%) III (5.16%) Variable Sodium 1.46558 (0.89) -0.55887 ( 0.41) -1.10518 (-0.15) pH 0.67878 (0.81) -0.52680 (-0.13) 0.65735 ( 0.5]) Chloride -0.94625 (0.71) 1.47442 ( 0.67) 0.94237 ( 0.14)

Constant 10.9475 -2.2444 2. 102.0 116.

Re-examination of the results of the NIASM floristic classification (see Chapter TWo) showed that at the five group level, groups which could be identified include a saline group (84 + 96) and a calcareous group of sites (97), but the acidic water sites (94) join group un. The very species rich freshwater group (102) ranains separate. However, identification of a single freshwater group (100 +

101 + 1(2) as proposed above, w~uld not have been possible since the infotmation statistic anployed in the classification is known to be group size dependent (Clifford and Stephenson 1975); combination of groups 100, 101 and 102 would have resulted in a group of 23 sites, 42% of the number of sites classified.

Floristically the freshwater group (100 + 101 + 1(2) is relatively heterogenous. Ninety six species -were recorded across all three groups with only 15.6% of t.l£ese being shared by all three (although not by all their manhers). Thirty five percent of the species -were relatively uncamnon, being recorded fran a single wetland only, although they may have been locally important. The freshwater group as a whole could not be characterised floristically on the basis of shared species.

Discriminant analyses have proved useful in relating the distribution of aquatic fauna groups to environmental variables (Green and Vascotto 1978, Marchant, Mitchell and Norris 1984) and in predicting the water quality of unsampled streams from the presence of macrophyte indicator species (Wiegleb 1981c). In tJ."le present study it is suggested that the relatively poor discrimination of the eight floristic groups achieved is due, at least in part, to the use of a Table 6.9 Summary table for the stepwise variable selection procedure for discriminant analyses of the water chemistry data. Refer to text for explanation of headings.

A. Stepwise discriminant analysis with 12 variables input (41 wetlands)

Step Variable entered Approx. F df. p. Sig. btn. groups I Magnesium 10.229 7, 33 * * * 84 & 94, 97, 99, 100, 101, 102 94 & 96. 97, I01 96 & 97, 99, 100, 101, 102

B. Stepwise discriminant analysis with 9 variables input (53 wetlands)

Step Variable entered Approx. F df. p. Sig. btn. groups Sodium 18.064 7, 45 * * * 84 & 94, 97, 99, laO, 101, 102 94 & 96 96 & 97, 99, 100. 101, 102

2 pH 9.980 14, 88 * * * 84 & 94, 97, 99, 100. 10 1, 102 94 & 96, 99. 100, 101, 102 96 & 97, 99, 100 , 101, 102 97 & 100

3 Ch lori de 8.149 21, 124 * * 1: 84 & 94, 97, 99. 100, 101, 102 94 & 96, 97, 99, 101, 102 96 & 97, 99, 100. 101, 102 97&99,100 99 & 100. 101, 102 117. less than optimal classification of the floristic data.

5. Conclusions

The analyses performed indicated that overall, the variance observed in the vegetation of the wetlands sampled was significantly influenced by sane of the water chemistry variables measured. However, the level of vegetational variation identified as a result of classification of the floristic data (eight groups) did not correspond well with the measured differences in the water chemistry variables.

Vegetation differences which could be attributed to differences in water chemistry are those between the saline groups (84 + 96), t.'1e turbid group (99), the acidic water group (94), the freshwater group

(100 + 101 + 102) and possjbly the calcareous group (97). Methodological factors which may have contributed to the lack of correspondence between the floristic groups and the water chemistry are examined in Chapter Seven.

The water chemistry variables which appeared to be most highly correlated with these differences included salinity, pH, bicarbonate (fran canonical co-ordinates analysis), salinity, pH, bicarbonate, conductivity, chloride, calcium (analysis of variance) and sodium, pH and chloride (stepwise discriminant analysis). The variables most highly correlated with floristic variance were not identical for all the analyses, since each analysis utilised a slightly different set of wetlands (due to missing data), and in the case of the stepwise discriminant analysis, the contribution of all the variables was not considered. It is suggested that these results largely reflect vegetation response to salinity and pH, factors with which the other 118.

important variables are thought to be highly correlated (calcium being an exception). Salinity largely separated groups 84 and 96 fram the rest, while pH separated the acidic water (94) and calcareous water

(97) groups fran the other freshwater groups 100, 101 and 102. The turbid water group (99) was separated by a canbination of pH and salinity variables.

These results are in accordance with the findings of Kirkpatrick am Harwood (1983) who sampled 530 Tasmanian wetlands. They concluded that a canbination of salinity and permanence indices explained over a third of the floristic variation between wetlam cammunities, with pH being more important in separating freshwater wetland carmunities.

TUssock sedge1and and marginal herbfie1ds, the plant canmunities identified by Kirkpatrick am Harwood (1983) as being characteristic of wetlands wi th prolonged substrate exposure, and thus influenced by water permanence, were not carmonly encountered in this study.

However, it is suggested that water depth may have been an important factor in differentiating freshwater sites. 119.

CHAPTER SEVEN

Conclusions

1. Summary of results

A. Floristic analyses

Classifications of the entire, "eident" and aquatic-sani-aquatic floristic data sets recognised eight, nine and ten groups of wetlands respectively. However, allocation of sites to groups was not always optimal, in part due to the axiomatic properties of the infounation statistic algorithm employed. Within the eight group solution for the entire data set it was suggested that six sites could be satisfactorily reallocated on the basis of their floristic affinity am shared diagnostic species (Table 3.8). However, six other, "fuzzy' sites appeared to have floristic affinities with more than one group of sites (Table 3.9), and could not be allocated satisfactorily. Eleven of the twelve aberrant sites supported species generally cammon to members of the freshwater complex (groups 100, 101 and 102). The classification also identified two groups of saline sites, both species poor and relatively compact, although there were some shared species.

Ordination of the floristic data (Fig. 4.1) indicated that the floristic groups were not well isolated and suggested which sites were responsible for overlaps. The ordination showed that the two saline groups 84 and 96 were very similar floristically, again largely 120.

because they WE!re very sr:ecies poor. Wi thin the freshwater groups

(100, 101, 102) the "fuzzyll sites 5, 13 aoo 17 were responsible for the overlap of groups 100 arXI 102, whilst overlap with the calcareous group of sites (97) was due to the misclassification of sites 9 and

43.

It is considered that classification of the entire floristic data set produced a useful classification of victorian WE!tlaoos. Attempts to improve on this result by deleting species wi th low "e ident" values or terrestrial species in order to reduce the noise inherent in the data set did not lead to marked improvement. The choice of iooividual wetlaoo as the unit for analysis resulted in an acceptable classification amenable to improvement through the reallocation of some misclassified sites. Analysis of the quadrat and transect data

(not reported in detail) did not provide more informative results.

HOWE!ver, simultaneous classification of all the WE!tland individuals may not have been the best approach. Sampling of sites across such a wide envirormental range provided a highly heterogeneous data set which, as a result of cumulation of the quadrat data, contained many heterogeneous individuals. In retrosr:ect, sane partitioning of the data prior to analysis may have provided an even better result.

B. Water chemistry analyses

Classification of sites using water chemistry data grouped the wetlarXls largely on the basis of their salinity levels. At the eight group level, which was utilised for comparison with the eight group floristic classification, three groups of saline sites could be identified. At the five group level there WE!re two groups of saline 121.

sites. Whilst almost all the sites identified as saline following the floristic classification (members of grops 84 and 96) were included within the groups classified as saline on the basis of water chemistry (groups 78, 86, 99 in Table 5.3), the membership of the individual groups was not identical. Although there was a general tendency for

Lepilaena bilocularis and Lepilaena cylindrocarpa to be dominant in saline and very saline waters respectively, the salinity ranges for these species, as suggested by Snoeijs and van der Ster (1980) overlap, and thus factors other than salinity may be important in effecting species distribution.

In the classification based on water chanistry the freshwater sites including the acidic, calcareous and turbid waters ~re also grouped according to their relative salinities.

The water samples collected at each transect provided an adequate description of the water chanistry of the ~tlands at the time of sampling. However, they provided no indication of the tanporal variability in ionic concentrations which might affect species distributions. Since other environmental factors such as turbidity, water depth and water level fluctuations affect species distribution, it would have been useful to include them in this analysis.

C. Vegetation - environment relationships

The results of the canonical Correlation Analysis suggested that there was a significant relationship between the distribution of aquatic species and the water chanistry parameters. The Analysis of Variance indicated that, in spite of the apparently poor match of the 122.

wetland groups identified through classification of the floristic and water chemistry data, there were significant differences in the water chemistry of the floristic groups. The saline group (84 + 96), the freshwater complex (100 + 101 + 102), the turbid water group (99), the acidic water group (94) and the calcareous water group (97) could be characterised using a combination of water chemistry characteristics.

Discriminant Analysis confirmed that the first four of the above groups could be separated on the basis of a combination of water chemistry characteristics, but that the calcareous group of sites was difficult to separate from elements of the freshwater complex (101 and 102) • It is conclLrled that while the water chemistry variables, particularly salinity and pH, significantly influenced serne of the broader patterns of vegetational variance, the water chemistry variables utilised in the analyses could not be related to the floristic heterogeneity of the saline (84 + 96) and freshwater (100 +

101 + 102) complexes.

2. Evaluation of the methods and results of the study

This study aimed to document the variability of wetlands vegetation in victoria, to produce a classification for remaining

Victorian wetlands, to examine the relationships between species distribution and environmental factors and to assess the performance of available numerical methods to a heterogenous and large but sparse data set.

It is considered that this study has adequately documented the variability of Victorian wetlands vegetation, and produced a useful 123.

classification. However, it has been more difficult to establish the relationships between species distributions and environmental factors, since factors other than those measured also appear to be important.

A. Variability of wetlands ve

While most non-tidal lowland wetland types were sampled, the chenopod-fringed saline lakes of north western Victoria were not included in the analyses since they contained no aquatic ve

The floristic canposition of the wetlands sampled was highly variable. A total of 143 species was recorded for the 55 sites examined, but only 45% of these species were considered to be aquatic or semi-aquatic and thus confined to wetland camnunities. The ranainder were graminoids or snall leaved herbs recorded at the margins of wetlands. Within the 1277 quadrats sampled 74 species were recorded as daninants, 26 of these species being dominant over one or more of the wetlands sampled. Thirty one species were recorded fran a single wetland only, although sane were locally camnon.

Species richness of the sites was also highly variable. Only a single species was recorded for the very saline Lake wandella whilst the large, shallow freshwater Ewing Marsh supported at least 29 124. species. Species richness was generally observed to increase with shallowness and size of the wetland and to decrease with increasing salinity.

The distribution of daninant species across the sites led to difficulties in adequately characteristing the eight wetland groups identified through classification of the floristic data; the "necessary" criterion wherein all sites must have at least one attribute (species) in cammon could only be satisfied for groups 84,

97, HH~, 101 and 102 (Table 2.3), although reallocation of the misclassified sites would have improved this to sane extent (Table 3.8). The "sufficient" criterion for group membership, no site outside the group possessing that attribute, could not be utilised since the diagnostic species for group membership were invariably widespread species. Tasmanian wetlands have similarly been described as having few species both highly constant and highly faithful (Kirkpatrick and Harwood 1983) •

B. Assessment of the classification of wetlands vegetation

The classification produced (Table 2.2) provides a typology for Victorian wetland sites and a framework within which more detailed studies can be undertaken. However, the eight group solution proposed is not ideal, since several groups do not have a diagnostic species

(Table 2.3). Within the freshwater canplex (groups 100, 101 and 102) in particular, the diagnostic species have a wide ecological amplitude and are not necessarily exclusive to a particular group. This lack of clear cut membership criteria may make allocation of additional freshwater wetlands not included in this study difficult. Although 125.

the proposed reallocation of apparently misclassified sites results in all the groups having a diagnostic species, the "fuzzyll or intermediate sites present a further problem.

The classification of the wetlands data is a function of the nature of the data and the decisions taken regarding sampling, the choice of descriptor, the unit to be described, the scale of measurements applied and the choice of analytical strategy. Dale

(1982) has suggested that a reconsideration of these components may be useful where the results of a numerical analysis have been less than optimal.

Since little was known about the variability of Victorian wetlands prior to this study, selection of sites for sampling was based on an attempt to inclLrle wetlands fran all major lithological provinces represented on Victorian lowlands. This approach almost certainly resulted in a variable sampling intensity, with sane types being over represented, whilst others were almost certainly underrepresented.

This uneven representation of types probably accentuated the effects of the NIASM algorithm1s group size dependency.

The sampling unit utilised was a one metre square quadrat.

However, for reasons discussed in Chapter TWo, the individual wetlands were the units chosen for analysis. This approach ignored the often quite high spatial heterogeneity observed within the wetlands sampled.

An alternative procedure would have been to classify quadrat data into community types, and to classify wetlands on the basis of the presence or absence of these types. In practice, use of an agglomerative procedure would have entailed an initial splitting of the data set. 126.

Jensen and van der Maarel (1980) used a similar procedure with the preliminary classification carried out on quadrat data from each site.

They found it difficult to identify lIexact species combinations

(ca:rmunities) tI from the results.

It was anticipated that classification of the data at the transect level (produced by amalgamation of quadrat data for each transect) might take into account same of the broad scale heterogeneity due to possible within-site differences in shoreline exposure and substrate type. However, the results of such an analysis indicated that within­ site variation was less than expected and smaller than between site variation so that most of the three transects sampled within sites were classified together.

The difficulties in identifying suitable diagnostic or faithful species for wetland community types suggest that species may be an inappropriate choice of descriptor for classification of these communities. The use of species as descriptors may result in overdefinition of the situation. Examination of patterning in functionally equivalent units, ego lifeforrns, may result in a reduction of the noise contributed by the high levels of heterogeneity and concomitantly produce same useful results. To this end some preliminary analyses were carried out utilising several life form systems and higher taxa (Famil ies and Orders). While same general patterns emerged, the results were disappointing.

The reasons for the use of frequency data were established in

Chapter Two. It is possible that results may have been more informative if cover abundance had been recorded. In particular, 127. although many species were shared by members of the freshwater groups 100, 101 and 102, the relative abundance of these species was very variable; estimates of abundance may have provided a clearer separation of these groups.

Choice of the classification strategy was largely determined by the type of data generated. The axianatic properties of the infoz:mation statistic were discussed in Chapter Two. In this study group size dependency effects and the interaction between presence and abundance were thought to have resulted in a number of misclassifications. A change of search strategy from agglomerative to divisive would at least have had the advantage of providing diagnostic species for almost all the groups generated. However, a preliminary analysis of the quadrat data utilising the monothetic, divisive program DIVINF (Ross 1983) produced results which were not readily interpretable. Dale (1982) has noted that such a change is a weak choice since it addresses a nUllerica1 rather than an ecological problem.

The centroid sorting strategy utilised by the NIASM algorithm results in a strong group size dependency effect. wllile such strategies may be useful in accentuating the differences between potential groups, in the present study it resu1ted in a number of misc1assifications because sane wetland types (eg. acidic waters) were clearly underrepresented. Use of another sorting strategy may have reduced sane of the effects of group size dependency.

Re-examination of the notion of similarity (or dissimilarity) may result in a more useful change. In the present case, the choice of 128. dissimilarity measure was constrained by the use of frequency data. Whilst the algorittm anployed was asyrrmetric, and double zero matches were not included in the calculation of heterogeneity, a more explicit suppression of double zero matches may have reduced the number of misclassified sites. If abundance data had been used, the choice of dissimilarity measure would have been wider and problan of presence/abundance interactions perhaps avoided.

Alternative approaches to the analysis of p3ttern in wetland carmunities include changing the method of structuring the data. Ordination methods were utilised to provide same additional insights into the relationships between the groups generated by classification, but the level of heterogeneity was too high for the method to provide

a clear cut solution. Ordination of the freshwater sites alone may have provided more information.

Another approach to restructuring the wetland floristic data for analysis utilises the quadrat data and the sp3tial relationships of

these quadrats wi thin the transects. Since the quadrat data contains a high level of noise, it is necessary to start by classifying the quadrats into groups. These groups are then used to label every quadrat with its appropriate group letter. This enables transects to

be redescribed as a sequence of quadrat groups, eg. aabbbbcccccccdddc.

Pattern analysis methods can be used to classify such sequences providing an appropriate similarity measure can be defined. The Levenshtein distance used in spelling correction (Hall and Dowling 1980), nODnalised to account for different sequence lengths is one such method. Transects which are sp3tial sequences of coded quadrats 129. can then be classified using an appropriate method. If the three transects sampled at each wetland are thought of as replicatE's, the

Sandland-Young test (1979) may be used as a stopping rule to establish the number of groups. A preliminary analysis of this kind has been undertaken, and will be reported elsewhere.

C. Vegetation - environment relationships

The methods utilised, canonical Correlations Analysis, Analysis of Variance and Discriminant Analysis provided same interpretable results although they did not establish relationships between water chemistry parameters and all the floristic groups identified. The results of the canonical Correlation Analysis, which are often not very infonnative when sample diversity is high, suggested same oZ the major factors affecting species distributions. Analysis of Variance and

Discriminant Analysis, which were used to examine whether there were any significant differences in water chemistry between the eight floristic groups, establish~ differences between the saline complex, the turbid group, the acidic waters, the freshwater complex and possibly the calcareous water group. However, the methods were unable to discriminate between members of the saline complex or members of the freshwater complex. It is not known whether this is due to a lack of difference or whether it can be attributed to the nature of the data set which may have affected the power of the tests. The considerable variance within the groups and the presence of several small groups for which estimates of variance may have been wide could have led to a type B error - acceptance of the null hypothesis that there was no significant difference in ~~ese cases. However, other factors may be important in explaining the vegetational heterogeneity ------

130.

of the freshwater groups 100, 101 and 102, and the saline groups 84 and 96.

A primary consideration in understanding the distribution of species is the availability of propagules for colonisation. wnile dissemination methods for aquatic and semi-aquatic species are poorly docunented, it seems likely that some are wind distributed, and

Vlaming and Proctor (1968) have suggested that a nunber of widespread

aquatic species are adapted to internal transport by waterbirds.

Differences in the effectiveness of dissemination mechanisms may in

part explain the wide distribution of same cammon freshwater species

such as Myriophyllum propinquum, potamogeton tricarinatus and

Triglochin procera whilst many other species were recorded relatively

infrequently.

Following arrival at a site, the fate of a propagule will depend on its ability to establish and grow to maturity. Van der Va1k (1981)

has suggested that the floristic camposihon of freshwater wetlands may be largely attributed to the life history characteristics of

species present in wetland seedbanks, especially their propagule

longevity, establisl1:nent requirements, lifespan, and their response to

environnental conditions, in particular the presence or absence of

standing water. This model suggests that at any single point in time

the species present in freshwater wetlands are a function of the

contents of the saedbank and recent water level fluctuations more than

a response to any particular selective mechanism in the chemical

environnent. 131.

While changes in the flora of Victorian wetlands due to fluctuations in water level were not studied systematically, it was noted that stunted forms of species such as Myriophyllum propinquum, potamogeton tricarinatus, Scirpus fluitans am Triglochin procera persisted on drying muds, and rhizomatous helophytes persisted even when the watertable had fallen to more than a metre below the surface of the substrate. Dry wetland substrates were frequently colonised by pasture weeds and grasses. It is not known whether their propagules were present in the wetlands seedbank or if their invasion was opportunistic.

The possible role of nutrients, in particular phosphorus and nitrogen, in influencing the distribution of wetland species was not examined in this study since it was decided at the outset that the nutrient status of sites could not be characterised aciequately by samples collected at a single point in time. Little is known about the concentrations of these nutrients in Australian inland waters, although it has been suggested (Williams and Wan 1972) that the levels in Victorian standing waters are relatively high. It is expected that there will be same differences in the nutrient status of the wetlar~s sampled since there were major lithological differences between the sites, and 1ithology is thought to be an important mechanign controlling the chemical composition of surface waters (Gibbs 1970). In particular, nutrients may limit the growth of some species of aquatic macrophytes in wetlands located on deeply leached acid sands, while some sites surrounded by agricultural land showed evidence of cultural eutrophication. 132.

other factors which may affect the species composition of wetlands inclooe species interactions such as allelopathy (Wi un-Anclersen ,

Anthoni , Christophersen and Houen 1982), competition (Yen 1980) and clonal growth, herbivory and the presence of pathogens. However, relatively little work has been done in these areas, and the significance for between site distribution of aquatic species is not known.

Buckley (1983) has suggested that vegetation in general contains patterns at a variety of scales, and that in complex systems, P£ticularly those with a substantial noise component, the patterns need to be appreciated through several successive stages of approximation. In this study sane strong first order patterns, probably associated with salinity and pH have been identified. However, understanding of sane of the relationships between sites and groups, especially those of the freshwater complex, probably requires a more detailed consideration of species dynamics in freshwater

wetlands.

D. Relationships with o~,er Australian wetlands

The previously mentioned study of Tasnanian wetland plant communities by Kirkpatrick and Harwood (1983) is the only sufficiently detailed work on Australian wetlands which can be used for comparative

purposes.

A comparison of the quadrat data collected for this study with the 16 Tasnanian wetland communities identified by Kirkpatrick and Harwood (1983) suggests that 12 of the latter communities occur within the 133.

victorian wetlands sampled, although their species complements are not necessarily identical. Those canmunities not noted for Victoria inclooe Baumea arthrophy11a sedgelam, Lamprothamnium aquatic herbfield, Wilsonia rotundifolia marginal herbland am Selliera radicans marginal herbland. Baumea arthrophylla has rarely been recorded for Victoria, whilst Lamprothamnium, considered only doubtfully distinct fram Chara at the generic level (Aston 1973), was subsuned urrler that genus for the purpose of this stooy. Wilsonia am

Selliera herblams are more commonly associated with tidal wetlands in

Victoria.

Victorian communities conspicuously absent from Tasmanian wet1ams inc100e those dominated by Baumea rubiginosa, Eragrostis austra1asicus, Lepilaena bilocularis and Restio tetraphy11us ­ sphagnum spp.

Of the 118 species recorded for Tasmanian wetla"1d communi ties

(Kirkpatrick and Hardwood 1983) 107 are recorded for Victoria (Willis

1970, 1972). Fifty four of these species were recorded fram the

Victorian sites scmpled. Many of those not recorded ccmnonly occur in

Juncus kraussii and Lepidosperma longitudena1e sedgelands, communities which are fourrl in Victoria but which were not extensively sampled in this stooy.

Marginal herblands appear to be less widespread in Victoria than in Tasmania. However, communi ties similar to those described for

Tasmania were recorded fram shallow 'N'et1ands in relatively high rainfall areas such as palparra settlement Swamp, Fernbank, Cotter's

Lake, Ewing Marsh and Victoria Lagoon, and in a slightly saline 134.

situation at Lake Terangpc:rn. The margins of wetlands in areas of lower rainfall, especially in north west and western Victoria, often consist of bare clay or sand, sometimes supporting widely scattered herbs. This may be due to the relatively unstable nature of the environment; considerable fluctuations in water level are cammon and at deeper more exposed sites the shoreline may be subject to wave action.

3. Recommendations for conservation

Calculations by Corrick and Norman (1980) and Corrick (1981, 1982) indicate the extent of wetland loss in victoria. In the present study, wetlands sampled were predcminantly located wi thin National

Parks and Fisheries and Wildlife Reserves or on Public Land. There were relatively few undisturbed freshwater sites on private land, althou;Jh sane of the larger and deeper slightly saline and saline wetlands were surrounded by pri vate land. Wetland types which are considered to be inadequately protected within the existing system of reserves include shallow freshwater basins, particularly t..'1ose associated with flood plains and prior stream systems, and the

Muehlenbeckia dcminated wetlands, which while represented elsewhere, have almost disappeared frcm the plains west of Melbourne. Acidic water sites dcminated by Sphagnum are also underrepresented.

Wi thin Victoria the Land Conservation Council has prirnary responsibility for making recommendations with respect to the use of public land. It is charged wi th providing for the balanced use of land wi thin victoria. The Land Conservation Council has reccmnended the inclusion of many wetlands in t..'1e Victorian system of National 135.

Parks, Forest Parks, Scientific Reference Areas and wildlife Reserves.

However, sane areas of public land have deliberately been left uncarmitted, and investigations are required to ascertain whether these areas contain wetlands of the types considered to be under conserved in Victoria, or wetlands with particularly interesting features which may merit conservation.

Features of particular interest which should be considered when assessing the conservation status of wetlands include whether the site supports representative remnants of wetland ccmnunities originally thought to be widespread, or undisturbed characteristic eKamples of wetland vegetation or plant communities displaying unusual features.

These may include uncarmon floristic associations, structurally unusual forms of a ccmnunity, or rare plants; species seldan encountered and those with disjunct occurrences. The fauna supported by wetlands also requires a similar assessment, whilst unusual features of the hydrology, geanorphology and water chemistry of wetlands may also make sites worthy of conservation. Wetlands may be of special interest for educational and research purposes where they can be used to demonstrate ecological relationships or carmunity dynamics, or where they provide an opportunity to monitor changes over time and space. 136.

It is hoped that the abol i tion of taxation incentives for the draining of swamps and clearing of land announced in the 1983-84

Federal Budget will discourage unnecessary destruction of the renaining wetlands on private property. Certainly more scope exists for the managenent of wetlands for conservation purposes on private land. The provision of managenent advice to landholders and continued public education canpaigns such as those initiated by the "Western

Australian Department of Conservation and Environment (1980) will help to ensure the protection of surviving wetlands. 137.

BIBLIOGRAPHY Adam, P., Birks, H.J., Huntley, B. and prentice, I.C. (1975). Phytosocioloc:Jical studies at Malham Tarn :Moss and Fen, Yorkshire, England. Vegetatio 30: 117-132. Alvey, N.G. et ale (canpiled by) (1979). "Genstat: A general statistical proc:Jram". The Statistics Department, Rothamstead Experlinental Station, Harpenden. Almquist, E. (1929). Upplands vegetation och flora. Acta Phytoc:Jeogr. Suec. 1: 1-624. Anon. (1983). The Victorian Yearbook No. 97. (Victorian Office of the Australian Bureau of Statistics: Melbourne). Aston, H.I. (1973). "Aquatic plants of Australia". (Melbourne University Press: Melbourne). Aston, H.I. (1979). Aquatic plants: new Victorian records. vict. Nat. 96: 67-69. Austin, M.P. (1976). Perfonnance of four ordination techniques assuming three different non-linear species response models. Vegetatio 33: 43-49. Austin, M.P. and Greig-Smith, P. (1968). The application of quantitative methods to vegetation survey. I I Sane methodoloc:Jical problems of data fran rain forest. J. Ecol. 56: 827-844. Austin, M.P. and Noy-Meir, I. (1971) • The problem of non-linearity in ordination: experiments with two-gradient models. J. Ecol. 59: 763-773. Australia, Department of Foreign Affairs (1976). Convention on wetlands of international linportance especially as waterfowl habitat. Australia, Department of Foreign Affairs, Treaty Series 1975 no. 48 (AGPS: canberra).

Bailey, T.A. and Dubes, R. (1982). Cluster validity profiles. Pattern Recognition 15 (2): 61-83. Bartlett, M.S. (1950). Tests of significance in factor analysis. Brit. ~ psychol. Statist. Sect. 3: 77-85. Beadle, N.C.W. (1981). "The vegetation of Australia". (Gustav Fischer Verlag: Stuttgart).

Benoit, R.J. (1969). Geochemistry of eutrophication. In "Eutrophication, causes, consequences, correctives". International symposiun on Eutrophication, University of Wisconsin. pp. 614-630. (National Academy of Sciences: Washi03ton) • 138.

Boissonneau, A.N. and Pala, S. (1979). An ecological classification project for the Ontario portion of the Hudson Bay - James Bay region. In "Applications of ecological (biophysical) land classification in Canada". ed. C.D.A. Rubec pp. 65-71. Environment Canada, Lands Directorate Ecological Land Classif. Sere 7.

Bridgewater, P.B. (1975). Peripheral vegetation of westernport Bay. Proc. Roy. Soc. Vict. 87: 69-78. Briggs, S.V. (1981). Freshwater wetlands. In "Australian Vegetation", ed. R.H. Groves. pp. 335-360. (cambridge university Press: Melbourne). Britton, R.H. and POdlejski, V.D. (1981). Inventory and classification of the wetlands of the camargue (France). Aquatic Botany 10: 195-228. Brock, M.A. (1981). The ecology of halophytes in the south east of South Australia. Hydrobiologia 81: 23-32.

Brown, M.J., Cro\oiden, R.K. and Jarman, S.J. (1982). The vegetation of an alkaline pan-acidic peat mosaic in the Hardwood River Valley, Tasmania. Aust. J. Ecol. 7: 3-12.

Brown, P. (1973). The vital wetlands. Victoria's Resources 15: 29-32.

Brundin, L. (1958). The bottom faunistical lake type system and its application to the southern hemisphere. Moreover a theory of glacial erosion as a factor of productivity in lakes and oceans. Verh. into Ver. Limnol. 13: 288-297. Buckley, R.C. (1983). Successive approximation in pattern analysis. Aust. ~ Ecol. 8: 333-337. Buckney, R.T. (1980). Chemistry of Australian waters: the basic pattern, with comments on same ecological implications. In "An ecological basis for water resource management". ed. W.O. Williams pp. 12-22 (ANU Press: Canberra). campbell, P. (1982). Association between small marrmals and vegetation on Fraser Island. Ph.D Thesis, University of Queensland.

Clifford, H.T. and Stephenson, W. (1975) "An Introduction to Nunerical Classification". (Academic Press: New York).

Clifford, H.T. and Williams, W.T. (1966). Similarity measures. In "Pattern Analysis in Agricultural Science". ed. W.T. Williams, pp. 37-46 (CSIRO: Melbourne and Elsevier Scientific Publishing Campany: Amsterdam). 139.

Clyrno, R.S. (1967). Control of cation concentrations, and in particular of pH in Sphagnum daninated carmunities. In "Chanical envirornlent in the aquatic habitat". ed. H.L. Gelterman and R.S. Clyrno, pp.273-284. Proceedings of an IBP Syrnposiun held in .Amsterdam and Nieuwersluis, 1966. (N. V. NOord Hollandsche Uitgevers Maatschappij: .Amsterdam).

Committee of Inquiry into the National Estate (1974). Report of the National Estate. (AGPS: Canberra) • congdon, R.A. and McCanb, A.J. (1976). The nutrients and plants of Lake Joondelup, a mildly eutrophic lake experiencing large seasonal changes in volune. ;!:.. ROY. Soc. W. Aust. 59: 14-23.

Conway, E.J. (1942). Mean geochanical data in relation to oceanic evolution. Proc. R. Ir. Acad. (Ser. B). 48: 119-159.

Cooley, W.W. and Lohnes, P.R. (1971). "Multivariate data analysis". (Wiley: New York).

Corrick, A.H. (1981). wetlands of victoria II. wetlands and waterbirds of South Gippsland. Proc. ROY. Soc. vict. 92: 187-198.

Corrick, A.H. (1982). Wetlands of Victoria III. Wetlands and waterbirds between Bay and . Proc. Roy. Soc. vict. 94: 69-87.

Corrick, A.H. and Norman, F.I. (1980). wetlands of Victoria I. wetlands and waterbirds of the Snowy River and Gippsland Lakes catctment. Proc. ROy. Soc. Vict. 91: 1-15.

Corrick, A.J. and Cowling, S.J. (1975). A survey of the wetlands of Kerang, Victoria. Fish. Wildl. Pap., vict. 5.

Corrick, A.J. and Cowling, S.J. (1978). A survey of the wetlands in the Lake Cooper area, Victoria. Fish. Wildl. Pap., vict. 17.

Cowardin, L.M., Carter, V., Gelet, F.C. and LaRoe, E.T. (1979). Classification of wetlands and deepwater habitats of the Uni ted States. U. S. Fish and Wildlife Services Program; FWS/OBS - 79/31.

Crocker, R.L. and Eardley, C.M. (1939). A South Australian Sphagnum bog. Trans. Roy. Soc. S.A. 63: 210-214.

Crowder, A.A., Bristow, J.M., King, M.R. and Vanderkloet, S. (1977) • Distribution, seasonality and biomass of aquatic macrophytes in Lake Opinicon (eastern Ontario). Nat. Can. (Que) 104 (5): 441-456.

Cullen, P. and Rosich, R. (1980). Managing urban lakes. In "An ecological basis for water resource management". ed. W.O. Williams pp. 90-99. (AND Press: Canberra). 140.

CUllen, P., Rosich, R. and Bek, P. (1978). A phosphorus budget for Lake Burley Griffin and management implications for urban lakes. Australian Water Resources Council Technical Paper No. 31 (AGPS: canberra).

Dale, M.B. (1971). Information analysis of quantitative data. In "Many Species, Populations, Ecosystems and Systems Analysis". eds. G.P. Patil, E.C. Pielou and W.E. Waters pp. 133-147. Statistical Ecology Series Volume 3. (Pennsylvania State University press) •

Dale, M.B. (1975). On objectives of methods of ordination. Vegetatio 30: 15-32.

Dale, M.B. (1982). Strategy and tactics in pattern analysis: a response to Harrington, Dawes and Ludwig. Aust.:!..:.. Ecol. 7: 411-414. Dale, M.B. and Anderson, D.J. (1972). Qualitative and quantitative information analysis. :!..:.. Ecol. 60: 639-653. Dale, M.B. and Anderson, D.J. (1973). Inosculate analysis of vegetation data. Aust.:!..:.. Bot. 21: 253-276.

Dale, M.B. and Clifford, H.T. (1976). On the effectiveness of higher taxoncrnic ranks for vegetation analysis. Aust. J. Ecol. 1: 37-62. Dale, M.B. and Williams, W.T. (1978). A new method of species reduction for ecological data. Aust. J. Ecol. 3: 1-5. Daniels, R.E. (1978). Floristic analyses of British mires and mire communities. J. of Ecol. 66: 773-803. oansereau, P., Buell, P.F. and Dagon, R. (1966). A universal system for recording vegetation: II A methodological critique and an experiment. Sarracenia 10: 1-64. Denny, P. (1980). Solute movement in sutmerged angiosperms. Biol. Rev. 55: 65-92. Department of Conservation and Envirorment (1980). wetlands guidelines for protection and management. Department of Conservation and Envirorment, . Bulletin NO. 79. Dixon, W.J. (1975). "Bicrnedical ccrnputer programs". (University of California press: Berkeley).

Dodson, J .R. and Wilson, LB. (1975). Past and present vegetation of Marshes Swamp in south eastern South Australia. Aust. J. Bot. 23: 123-150. Ducker, S.C., Brown, V.B. and calder, D.M. (1977). An identification of the aquatic vegetation in the Gippsland Lakes. Report prepared for the Ministry of Conservation, Victoria. Envirormental Studies Prograrrrne. (School of Botany: university of Melbourne) • 141.

Du Rietz, G.E. (1921). "Zur methodologischen grundlage der modernen pflanzensoziologie" (l>..dolf Holzhausen: Upsala). Du Rietz, G.E. (1930) • vegetationsforschung auf soziationsanalytischer. Grundlage. Abderhalden. Handb. bioI. Arb. Meth. 11: 5.s. pp 293-480 (Berlin and wein) • Eardley, C.M. (1943). An ecological study of the vegetation of Eight Mile Creek Swamp, a natural south Australian coastal fen formation. Trans. ROY. Soc. ~ Aust. 67: 200-223. East, L.R. (1935). Swamp reclamation in Victoria. J. Instit. Engineers Aust. 7: 77-91. Felzines, C.J. (1977). Analysis of the relationship between the mineral content of the stagnant fresh waters and the distribution of aquatic macrophytes. Ann. Sci. Nat. Bot. BioI. ~ 18(3): 221-249. Frankel, O.H. and Bennett, E. (eds) (1970). "Genetic resources in plants-their exploration and conservation". IBP Handbook No. 11 (Blackwell Scientific Publications: Oxford).

Frenkel, R.E. and Harrison, C.M. (1974). An assessment of the usefulness of phytosociological and numerical classificatory methods for the ccmnunity biogeographer. ;G. Biogeog. 1: 27-56. Gauch, H.G. and wentworth, T.R. (1976). Canonical correlation analysis as an ordination technique. Vegetatio 33: 17-22. Gauch, H.G. and Whittaker, R.H. (1972). Comparison of ordination techniques. Ecology 53: 868-875. Gauch, H.G., Whittaker, R.H. and wentworth, T.R. (1977). A comparative study of reciprocal averaging and ob'er ordination techniques. J. Ecol. 65: 157-174. Gibbs, R.J. (1970). Mechanisms controlling world water chemistry. Science 170: 1088-1090. Glenn-Lewin, D.C. and Crist, A.M. (1981). The fine structure of a prairie pothole and pothole border. Stuckey and Reese. Ohio BioI. Surv. BioI. Notes No. 15.

Golet, F.C. and Larson, J.S. (1974). Classification of freshwater wetlands in the glaciated north-east. U.S. Fish Wildlife Service Resour. Publ. 116. Goodrick, G.N. (1970). A survey of the wetlands of coastal New South Wales. CSIRO Div. Wildl. Res. Tech. Memo No.5. Gower, J.C. (1966). Same distance properties of latent root and vector methods used in multivariate analysis. Biometrika 53: 325-338. 142.

Gower, J.C. (1967). Multivariate analysis and multi-dbnensional geometry. Statistician 17: 13-28.

Graham, W.A.E. (1973). Perennial submersed and floating aquatic weeds in south eastern Australia. 2nd Victorian weeds Conf., Melbourne. Green, R.H. and Vascotto, G.L. (1978). A method for the analysis of envirorroental factors controlling patterns of species canposition in aquatic cannunities. Water Res. 12(8): 583-590. -­ Greig-Smith, P. (1980). The developnent of numerical classification and ordination. Vegetatio 42: 1-9.

Greig-smith, P., Austin, M.P. and Whitmore, T.C. (1967). The application of quantitative methods to vegetation survey. I Association-analysis and principal canponent ordination of rain forest. J. Ecol. 55: 483-503. Hall, P.A.V. aoo Dowling, G.R. (1980). Approxbnate string matching. Comput. Surv. 12(4): 381-402. Hansen, K. (1961). Lake types and lake sediments. Verh. into Ver. Lirnnol. 14: 285-290. Hartog, C. den and Segal, S. (1964). A new classification of the water plant cannunities. Acta Bot. Neerl. 13: 367-393. Heinselman, M.L. (1970). Landscape evolution, peatlarrl types arrl the envirorroent of the Lake Agassiz Peatlands Natural Area, Minnesota. Ecol. Monogr. 40: 235-261.

Hejny, S. (1971). The dynamic characteristic of littoral vegetation with respect to changes in water level. Hidrobiologia Bucur. 12: 71-85. Hem, J.D. (1970). Study and interpretation of the chemical characteristics of natural waters. U.S. Geol. Surv. water Supply Pap. No 1473. 2nd edition. HeyYt'Ood, R.B., Dartnall, H.J.G. and priddle, J. (1980). Characteristics and classification of the lakes of Signy Island, South orkney Islands, Antarctica. Freshwater BioI. 10: 47-59. Higginson, F.R. (1965). The distribution of submerged aquatic angiosperms in the Tllggerah Lakes system. Proc. Linn. Soc. N.S.W. 90(3): 328-334. Hill, M.O. (1973). Reciprocal averaging: an eigenvector method of ordination. J. Ecol. 61: 237-249.

Hill, M.O. (1979). ~flINSPAN. A Fortran program for arranging multivariate data in an ordered two way table by classification of the irrlividuals and attributes. Program CEP-4l, Department of Ecology and Systematics. (Cornell University: New York) • 143.

Hill, M.O. and Gauch, H.G. (1980). Detrended correspomence analysis: an improved ordination technique. vegetatio 42: 47-58. Hill, R.S. (1980). A stopping rule for partitioning demrograms. Botanical Gazette 141(3): 321-324. Hills, E.S. (1975). "The physiography of victoria". (Whitccmbe am Tanbes: Christchurch). Hogeweg, P. and Hesper, B. (1981). Oligothetic characterisation of clusters. Pattern Recognition 14(1-6): 131-136. Howard-Williams, C. am Walker, B.H. (1974). The vegetation of a tropical African lake, classification am ordination of the vegetation of Lake Chilwa, Malawi. ~ Ecol. 62(3): 831-854. Hutchinson, G. (1957). itA treatise on limnology". Vol. 1. (Wiley: New York) • Iversen, J. (1936). "Biologische Pflanzentypen als Hilfsmittel in der vegetationsforschung". (Lev in am Munksgaard: Copenhagen) •

Jackson, M.L. (1962). "Soil chemical analysis". (Constable: Lomon) • Jardine, N. and Sibson, R. (1971). "Mathematical Taxonany". (Wiley: New York) • Jenkins, R.E. and Bedford, W.B. (1973). The use of natural areas to establish environnental baselines. BioI. Conserv. 5: 168-174.

Jensen, S. (1977). An objective method for sampling the macrophyte vegetation in lakes. Vegetatio 33: 107-118. Jensen, S. (1979). Classification of lakes in southern Sweden on the basis of their macrophyte canposition by means of multivariate methods. Vegetatio 39: 129-146. Jensen, S. and Maarel, E. van der (1980). Numerical approaches to lake classification with special reference to macrophyte communities. Vegetatio 42: 117-128. Johnson, W.D., Cole, T.J., Johnson, M., McPherson, W.P., Muir, G.L. am Szczepanski, R. (1979). Ion balance in water analyses the effect of added silica on the carbonate-bicarbonate titration. Aust. J. Mar. Freshwat. Res. 30: 315-323. --- Jones, G. and CUllimore, D.R. (1973). Influence of macro-nutrients on the relative growth of water plants in the Qu'Appelle Lakes, canada. Environmental Pollution 4: 283-290. 144.

Kershaw, A.P. (1978). The analysis of aquatic vegetation on the Atherton Tableland, north-east Queensland, Australia. Aust. J. Ecol. 3: 23-42.

Key, L.M. (1967). Historical geography of the KooWl5!erup district. M.A. Thesis, University of Melbourne.

Kirkpatrick, J.B. and Harwood, C.E. (1981). The conservation of Tasmanian wetland macrophytic species and canmunities. A report to the Australian Heritage Canmission fran the Tasmanian Conservation Trust Inc.

Kirkpatrick, J.B. and Harwood, C.E. (1983). plant canmunities of Tasmanian wetlands. Aust. J. Bot. 31: 437-451.

Klecka, W.R. (1975). Discriminant analysis. In "SPSS Statistical Package for Social Sciences". eds. N.H. Nie, C.H. Hull, J .G. Jenkins, K. Steinbrenner and D.H. Bent. pp. 434-467 (McGraw Hill: New York) •

Kusler, J.A. (1979). Wetland protection: is Science meeting the challenge? In "Wetlands functions and values: the state of our understanding". eds. P.E. Greeson, J.R. Clark and J.E. Clark pp. 31-42. Proceedings of b~e National ~posium on Wetlands. (American Water Resources Association: Minneapolis) •

Lachenbruch, P.A. (1975). "Discriminant analysis". (Hafer: New York) •

Ladd, P.G. (1979). Past and present vegetation on the in the Highlands of Eastern Victoria. I Present Vegetation. Aust. J. Bot. 27: 167-184.

Lake, P.S. (1980). Conservation. In "An ecological basis for water resources management". ed. W.O. Williams pp. 163-173 (AND Press: Canberra).

Lambert, J .M. (1971). Theoretical models for large scale vegetation survey. In" Mathematical Models in Ecology". The 12th ~posiun of the British Ecological Society. ed. J.N.R. Jeffers, pp. 87-100. (Blackwell: Oxford).

Lambert, J.M. and Williams, W.T. (1966). Multivariate methods in plant ecology. IV Ccrnparison of information analysis and association-analysis. J. Ecol. 54: 635-664.

Lance, G.N. and Williams, W.T. (1967). A general theory of classificatory sorting strategies. I Hierarchical systems. Ccrnput. ~ 9: 373-380.

Lance, G.N. and Williams, W.T. (1977). Hierarchical classification methods. In "Mathematical Methods for Digital Canputers III Statistical Methods for Digital Canputers". eds. K. Enslein, A. Ralston and H.S. Wilf. pp. 269-295. (Wiley: New York) • 145.

Leech, L. (1973). The sand dunes on the Yanakie Isthmus, Wilson's Pramontory. M.Sc. Thesis, University of Melbourne.

Legendre, P., Long, F. and Beauvais, A. (1980). Ecological typology of a group of lakes in northwestern Quebec. Ann. L~ol. 16: 135-158. Lillieroth, S. (1950). Uber die Folgen Kulturbedingker Wasserstandssenkungen fUr Macrophyten-und Planktongemeinschaften in seichten seen des slk1schwedischen 01 igotrophiegebiekes. Acta Limnologica 3 (Lund).

Lundh, A. (195l). Studies on the vegetation and hydrochemistry of Scanian lakes. III Distribution of macrophytes and same algal groups. Botaniska Notiser Supplement 3{1): 1-138.

Maarel, E. van der (1972). Ordination of plant communities on the basis of their plant genus, family and order relationships. In "Grundfragen und Methoden in der Pflanzensoziologie". eds. E. van der Maarel and R. TUxen. pp. 183-192 (Junk: The Hague) •

Maarel, E. van der (1980). On the interpretability of ordination diagrams. Vegetatio 42: 43-45.

Maarel, E. van der (1982). On the manipulation and editing of phytosociological data. Vegetatio 50: 71-76.

Macnaughton-Snith, P. (1965). Same statistical and other nunerical methods for classifying individuals. Hane Office Research unit Report No. 6 (HMSO: London).

Maddocks, G.E. (1967). The geochemistry of surface waters of the Western District of Victoria. Aust. J. Mar. Freshwat. Res. 18: 35-52. ------

Maemets, A. and Raitviir, A. (1977). On the classification of Estonian lakes based on the analysis of principal camponents and co-ordinates. Eesti Nsv. Tead. Akad. Tom. Biol. 26: 138-148.

~kirinta, U. (1978a). A phytosociological classification of the aquatic vegetation of Lake Kukkia, Southern Finland. Acta Univ. Oul. A 75 Biol. 5.

~kirinta, U. (1978b). A new ecamorphologic system of lifefonns of the aquatic macrophytes. Phytocoenologia 4: 446-470. Malme, L. (1975). Phytosociological studies of aquatic and marsh vegetation in More og Ransdal, Western Norway. ~ Nor. Vidensk Seisk. Mus. Misc. 22: 1-30.

Marchant, R., Mi tchell, P. and Norris, R. (1984). The distribution of benthic invertebrates along a disturbed section of the , Victoria: An analysis based on nunerical classification. Aust. J. Mar. Freshwat. Res. 35: (in press) • --- 146.

Maristo, L. (1941). Die Seetypen Finnlands auf floristischer und vegetationsfysionomischer Grundlage. Ann. Bot. Soc. zool. ~ Fenn. Vanamo 15(5) 1-34.

Martin, C., Hotchkiss, N., Uhler, F.M. and Bourn, W.A. (1953). Classification of the wetlands of the United States. Special Scientific Report NO. 29 U.S. Fish and Wildlife Service.

Mathiesen, H. (1969). S¢ernes planter. Oanmarks Natur, Bam 5: De ferske Vame pp. 237-280. Politikens Forlag.

Millar, J.B. (1976). wetlam classification in western Canada: A guide to marshes am shallow open water wetlams in the grasslands am parklams of the prairie Provinces. Can. Wildl. Servo Rep Sere 37: 1-37.

Misra, R.D. (1938). E}:japhic factors in the distribution of aquatic plants in the English Lakes. J. Ecol. 26: 411-451.

Mi tchell, D.S. (1977). Aquatic weeds in Australian inland waters. Department of Envirorment, Housing am Cc:mnunity Developnent (AGPS: Canberra).

Mitchell, D.S. (1980). Aquatic weeds. In "An ecological basis for water resource management". ed. W.o. Williams pp. 80-89 (ANU Press: Canberra).

Moral, R. del (1975). Vegetation clustering by means of Isodata: revision by multiple discriminant analysis. Vegetatio 29: 179-190.

Moral, R. del am Denton, M.F. (1977). Analysis and classification of vegetation based on family composition. Vegetatio 34: 155-165.

Morgan, N.C. am Boy, V. (1982). An ecological survey of standing waters in north west Africa: I Rapid survey am classification. BioI. Conserve 24: 5-44.

Moyle, J .B. (1945). Some chemical factors influencing the distribution of aquatic plants in Minnesota. Am. Midl. Nat. 34: 402-420.

Muller, K.E. (1982). Understanding canonical correlation through the general linear model am principal ccmponents. The American Statistician 36: 342-354.

Mueller-oombois, D. am Ellenberg, H. (1974). "Aims and Methods of vegetation Ecology". (Wiley International: New York) •

Musil, C.F., Grunow, J.O. and Bornman, C.H. (1973). Classification am ordination of aquatic macrophytes in the pongolo River Pans, Natal. Bothalia 11: 181-190.

Nicholson, S.A., Levey, R.A. and Clute, P.R. (1975). Macrophyte-sediment relationships in Chautaugua Lake. Verh. into Ver. Limnol. 19: 2758-2764. 147.

Noy-Meir, 1. (1971). Multivariate analysis of the semi-arid vegetation in south eastern Australia: nodal ordination by component analysis. Proc. Ecol. Soc. Aust. 6: 159-193.

Noy-Meir, 1. and Austin, M.P. (1970). principal component ordination and simulated. vegetational data. Ecology 51: 551-552. Noy-Meir, I. and Whittaker, R.H. (1977). Continuous multivariate methods in ccmnunity analysis: sane problens and developments. Vegetatio 33: 79-88. Noy-Meir, I. and Whittaker, R.H. (1978). Recent developments in continuous multivariate techniques. In "Ordination of Plant Ccmnunities". ed.. R.H. Whittaker pp. 339-378. (Junk: The Hague) • O'Connell, M. (1981). The phytosociology and ecology of Scragh Bog, Co. Westmeath. New Phytol. 87: 139-187.

O:1uro, E.P. (1979). The value of ~dands: a hierarchical approach. In "wetlands functions and values: the state of our understanding". Etls. P.E. Greeson, J.R. Clark and J.E. Clark. pp. 16-25. proceed.ings of the National ~posium on wetlands (American Water Resources Association: Minneapolis). Orloci, L. (1978). "Mul tivariate analysis in vegetation research". 2nd edition (Junk: The Hague) •

Paijmans, K. (1981). The of inland northern New South wales, Australia. CSIRO Division of Land Use Research Technical Paper No. 41: 22p.

Pakarinen, P. (1976). Agglomerative clustering and factor analysis of South Finnish mire types. Ann. Bot. Fenn. 13: 35-41. Pakarinen, P. and Ruuhijarvi, R. (1978). Ordination of northern Finnish peatland vegetation with factor analysis and reciprocal averaging. Ann. Bot. Fenn. 15: 147-157. Pearsall, W.H. (1920). The aquatic vegetation of the English Lakes. J. Ecol. 8: 163-199. pietsch, W. (1977). Beitrag zur Soziologie und Okologie der europaischen Littorelletea-und Utricularietea-Gesellschaften. Feddes Repertorium Band 88, Heft 3: 141-245. pietsch, W. (1978). Zur Sociologie, Okologie und Bioindikation der Eleocharis multicaulis - Best~nde der Lausitz. Gleditschia Band 6: 209-264. Pip, E. (1979). Survey of the ecology of sul::Inerged. aquatic macrophytes in central Canada. Aquatic Botany 7: 339-357.

Popma, J., Mucina L., TOngeren, O. van and Maarel, E. van der (1983) • On the determination of optimal levels in phytosociological classification. Vegetatio 52: 65-75. 148.

QUennerstedt, N. (1958). Effect of water level fluctuation on lake vegetation. Verh. into Ver. Lirnnol. 13: 901-906.

Rai, H. and Hill, G. (1980). Classification of central Amazon lakes on the basis of their microbiological and physico-chemical characteristics. Hydrobiologia 72: 85-99.

Raspopov, I.M., Slepukhina, T.D., Voronksov, F.F. and Rychkova, M.A. (1978). Dynamics of water and fODnation of biocenoses in the littoral zone (using Lake Kubenskoye as an example). Soviet Journal of Ecology 9(b): 557-560.

Ratkowsky, D.A. and Lance, G.N. (1978). A criterion for determining the number of groups in a classification. Aust. Compo ~ 19(3): 115-117. Raunkiaer, C. (1907). Planterigets Livsformer og deres Betydning for Geografien (~enhaven). In "The Li fe forms of plants and Statistical Plant Geography". C. Raunkiaer 1934 (Oxford University Press) •

Rodhe, W. (1949). The ionic composition of lake waters. Verh. into Vera Lirnnol. 19: 377-386. RosS, D. (1982). The TAXON User's Manual. Edition P3. (CSIRO Division of Computing Research: Brisbane). samuelsson, G. (1925). Untersuchungen tllier die honere Wasserflora von Dalarne. SVenska Vaxtsociologiska Sallskapets Handlingar IX.

Sandland, R.L. and Young, P.C. (1979). probablistic tests and stopping rules associated with hierarchical classification techniques. Aust. J. Ecol. 4: 399-406.

Schwintzer, C.R. (1978). vegetation and nutrient status of Northern Michigan fens. can. ~ Bot. 56(24): 3044-3051.

Scott, J.T. (1974). Correlation of vegetation with environment: a test of the continuum and community type hypotheses. In "Handbook of Vegetation Science, 6: vegetation and Environment". eds. B.R. Strain and W.O. Billings. pp. 89-109. (Junk: The Hague) •

Sculthorpe, C.D. (1967). "The biology of aquatic vascular plants". (Edward Arnold: London).

Seddon, B. (1967). The lacustrine environment in relation to macrophytic vegetation. In IlQuaternary paleoecologyll. eds. E.J. CUshing and H.E. Wright pp. 205-215 (Yale University Press: London and New Haven) •

Seddon, B. (1972). Aquatic macrophytes as limnological indicators. Freshwater Biology 2(2): 107-130. 149.

Segal, S. (1966). Ecological studies of peat-bog vegetation in the northwestern part of the province of Overijsel (The Netherlands). Wentia 15: 109-141.

Segal, S. (1971). principles on structure, zonation and succession of aquatic macrophytes. Hidrobiologia Bucur. 12: 89-95.

Shaw, S.P. and Fredine, G.C. (1956). Wetlands of the United states. U.S. Fish. Wildl. Servo Circ. 39.

Sheldon, A.L. (1973). A quantitative approach to the classification of inland waters. In "Natural envirorments. Stu:iies in theoretical and applied analysis". ed. J.V. Krutilla pp. 205-261. (Johns Hopkins university Press: Baltimore) .

Sheldon, R.B. and Boylen, C.W. (1977). Maximum depth inhabited by aquatic vascular plants. Am. MidI. Nat. 97(1): 248-254.

Shiel, R.J. (1980). Billabongs of the Murray-Darling system. In "An ecological basis for water resource management". ed. w.O. Williams pp. 376-390. (ANU press: Canberra).

Silvestri, L. and Hill, I.R. (1964). Sane problems of the taxanetric approach. In IIPhenetic anJ Phylogenetic Classification". eds. V.H. Heywood and J. M:Neill. pp. 87-104. (Systematics Association: London).

Sjtks, H. (1950). On the relation between vegetation and electrolytes in north Swedish mire waters. Oikos 2: 241-258.

Sloey, W.E., Spangler, F.L. and Fetter, C.W. (1978). Management of freshwater wetlands for nutrient assimilation. In "Freshwater wetlands ecological processes and management ll potentiai • eds. R.E. Good, D.F. Whigham and R.L. Simpson. pp. 321-340. (New York: Academic pres) •

Snartt, P.F.M. (1978). Sampling for vegetation survey: a flexible systematic model for sample location. ~ of Biogeog. 5: 43-56.

Snartt, P.F.M., Meacock, S.E. and Lambert, J.M. (1974). Investigations into the properties of quantitative vegetational data II Further data type canparisons. J. Ecol. 64: 41-78.

Snedecor, G.W. and COChran, W.G. (1980). "Statisticalmethods". Seventh edition p 234. (IOwa State university Press: Ames, Iowa) • snoeijs, P. and van dar Ster, H. (1981). An integrated hydrobiological-geobotanical stu:iy of the Lepilaeno-Ruppion in South Australia. Laboratorium voor Aquatische Oecologie, Botanisch Laboratorium, Afdeling Geobotanie Katholieke Universiteit Tbernooiveld, Nijmegen No. 134 74 pp. 150.

Sokal, R.S. and Rohlf, F.J. (1969). "Bianetry". (W.H. Freanan: San Francisco) •

Spence, D.H.N. (1964). The macrophytic vegetation of freshwater lochs, swamps and associated fens. In "The vegetation of Scotland". ed. J.H. Burnett pp. 306-425 (Oliver and Boyd: Etlinburgh) • spence, D.H.N. (1967). Factors controlling the distribution of freshwater macrophytes with particular reference to the lochs of Scotland. J. Ecol. 55: 147-170. spence, D.H.N. (1982). The zonation of plants in freshwater lakes. Advances in Ecological Research 12: 37-125.

Stephenson, W. and Cook, S.D. (1980). Elimination of species before cluster analysis. Aust. J. Ecol. 5: 263-274.

Stephenson, W., Williams, W.T. and Lance, G.N. (1970). The macrobenthos of MJreton Bay. Ecol. Monogr. 4g: 459-494.

Stewart, R.E. and Kantrud, H.A. (1969). proposed classification of potholes in the glaciated prairie region. Saskatoon wetlands Saninar. canadian Wildlife Service Report NO.6. (Ottawa) •

Stewart, R.E. and Kantrud, H.A. (1972). Vegetation of prairie potholes, NOrth Da.kota, in relation to qual i ty of water and other envirormental factors. U.S. Geol. Surv. Prof. Paper 585-0 36p.

SWan, J.M.A. (1970). An examination of sane ordination problans by use of simulated vegetational data. Ecology 51: 89-102. swindale, D.N. and Curtis, J.T. (1957). Phytosociology of the larger sul::merged plants in Wisconsin Lakes. Ecology 38: 397-409.

Theinemann, A. (1925). Die BinnengewMsser Mitteleuropas. Eine 1imnologische Einfllirrung. Die BinnengewMsser 1 (Stuttgart).

Thunmark, S. (1931). Der see Fiolen und Vegetation. Acta Phytogeogr. Suec. 2: 1-198.

Tirrms, B.V. (1973). A limnological survey of the freshwater coastal lakes of East Gippsland, Victoria. Aust. J. Mar. Freshwat. Res. 24: 1-20. ------

Tirrms, B. V. (1977). A study of sane coastal dune lakes in western Victoria. Proc. ROY. Soc. vict. 89: 167-172.

Topping, M.S. and Scudder, G.G.E. (1977). Scme physical and chemical features of saline lakes in central British Columbia. Syesis 19: 145-166.

Tyler, C. (1980). Schoenus vegetation and env ironmental conditions in south and south east SWeden. Vegetatio 41: 155-170. 151.

Van der Valk, A.G. (1981). Succession in wetlands: A Gleasonian approach. Ecology 62: 688-696.

Verhoeven, J.T.A. (19813). The ecology of Ruppia daninated carmunities in western Europe. II Synecological classification. structure and dynamics of the macroflora and macrofauna communities. Aquatic Botany 8: 1-86. vitt, D.H. and Slack, N.G. (1975). An analysis of the Sphagnum daninated kettle-hole bogs in relation to envirormental gradients. can. ~ Bot. 53: 332-359. vlaming, V. de and proctor, V.W. (1968). Dispersal of aquatic organhills: viability of seeds recovered fran the droppings of captive killdeer and mallard ducks. Amer. ~ Bot. 55: 213-26.

Walker, B.H. and Coupland, R.T. (1968). An analysis of vegetation - envirorment relationships in Saskatchewan sloughs. can. ~ Bot. 46: 5139-22.

Walker, B.H. and Coupland, R.T. (19713). Herbaceous wetland vegetation in the aspen grove and grassland reg ions of saskatchewan. can. J. Bot. 48: 1861-1878.

Walker, B.H. and wehrhan, C.F. (1971). Relationship between derived vegetation gradients and measured envirormental variables in Saskatchewan wetlands. Ecology 52: 85-95. weaver, J.E. and Clements, F.E. (1938). "Plant Ecology". (McGraw Hill: New York) •

~Vebb, L.J., Tracey, G.J., Williams, t"i.T. and Lance, G.N. (1967). Stt.rlies in the Ol..merical analysis of canplex rainforest communi ties II The problem of species sampling. J. Ecol. 55: 525-538.

Webb, L.J., Tracey, G.J., Williams, W.T. and Lance, G.N. (19713). Stt.rlies in the Ol..merical analysis of canplex rainforest carmunities. V A canparison of the properties of floristic and physignanic-structural data. ~ Ecol. 58: 2133-232.

Welch, D.M. (1978). Land/water classification. A review of water classifications and proposals for water integration into ecological land classification. Envirorment canada, Ottawa (canada) Lands Directorate. Ecol. Land Classif. Ser., 5 (Supply and Services canada: Hull, Quebec). -­

Wheeler, B.D. (198@a). plant communities of the rich-fen systems in England and Wales, UK. I Introduction: Tall sedge and reed communities. J. Ecol. 68: 365-396. wneeler, B.D. (198@b). Plant communities of rich fen systems in England and Wales, UK. II Communities of calcareous mires. J. Ecol. 68: 4135-4213.

TNhittaker, R.H. (1967). Gradient analysis of vegetation. Biol. Rev. 42: 2137-264. 152.

Whittaker, R.H. and Gauch, H.G. (1978). Evaluation of ordination techniques. In "Ordination of Plant Ccmnunities". ed. R.H. Whittaker pp. 277-336. (Junk: The Hague) • wiegleb, G. (1978). Investigations of the relationship between hydrochemical enviromental factors and the macrophytic vegetation in standing waters. Arch. Hydrobiol. 83: 443-484. Wiegleb, G. (198la). Probleme der Syntaxonanischen Gliederung der Potametea. Berichte der Internationalen Symposien der Internationalen Vereinigung fUr vegetationskunde. pp. 207-249. (Cramer: Vaduz).

Wiegleb, G. (198lb). Struktur, Verbrietung und Bewertung von Makrophytengesellshaften nieders~chsischer FliessgewSsser. Lllnnologia (Berlin) 13: 427-448. Wiegleb, G. (198lc). Application of multiple discriminant analysis on the analysis of the correlation between macrophyte vegetation and water quality in running waters of central Europe. Hydrobiologia 79: 91-100. Williams, B.K. (198l). A simple demonstration of the relationship between classification and canonical variates analysis. The American Statistician 36: 363-365. Williams, M. (1973). Gecrnorphology of Sydenham Inlet. Honours Thesis, univerity of Melbourne. Williams, W.o. (1964). A contribution to lake typology in Victoria, Australia. Verh. into Vera Lllnnol. 15: 158-168. Williams, W.o. (1967). The chemical characteristics of lentic surface waters in Australia: a review. In "Australian Inland Waters and their Fauna: Eleven Stooies" ed. A.H. Weatherly. pp. 18-77 (AND Press: Canberra). Williams, W.o. (1980). Distinctive features of Australian water resources. In "An ecological basis for water resource management" • ed W.O. Williams pp. 6-11. (AND Press: Canberra) • Williams, W.O. (1981). The limnology of saline lakes in western Victoria, Australia: a review of recent stooies. In"Sal t Lakes: Proceedings of an International Symposium on Athalassic (Inland) Salt Lakes". ed. W.O. Williams. pp. 233-259 (Junk: The Hague). Williams, W.O. and Buckney, R.T. (1976). Stability of ionic proportions in five salt lakes in Victoria, Australia. Aust. J. Mar. Freshwat. Res. 27: 367-377. Williams, W.O. and Wan, H.F. (1972). Sane distinctive features of Australian inland waters. Water Res. 6: 829-836. Williams, W.T. (1971). principles of clustering. Ann. Rev. of Ecology and systematics 2: 303-326. 153.

Williams, W.T. (1973). Partition of infoonation. The centperc problem. Aust. ~ Bot. 21: 277-281.

Williams, W.T. (1976a). Types of classification. In "Pattern Analysis in Agricultural Science". ed. W.T. Williams pp. 76-83 (CSIRO: Melbourne and Elsevier Scientific Publishing Canpany: Amsterdam).

Williams, W.T. (1976b). Ordination: principal canponent analysis. In IIPattern Analysis in Agricultural Science". ed. W.T. Williams pp. 47-58. (CSIRO: Melbourne and Elsevier Scientific Publishi03 Canpany: Amsterdam).

Williams, W.T. (1976c). Other ordination procedures. In l1 I1Pattern Analysis in Agricultural Science • ed. W.T. Williams pp. 59-69. (CSIRO: Melbourne and Elsevier Scientific Publishi03 Canpany: Amsterdam).

Williams, W.T. (1981). Underlying assumptions in mmerical classification. In I1vegetation Classification in Australia". eds. A.N. Gillison and D.J. Anderson. pp. 117-119 (CSIRO and AND Press: canberra).

Williams, W.T. and Dale, M.B. (1965). Fundamental problems in nuner ical taxonomy. Adv. Bot. Res. 2: 35-68.

Williams, W.T., Dale, M.B. and Lance, G.N. (1971). Two outstanding ordination problems. Aust. ~ Bot. 19: 251-258.

Williams, W.T., Lambert, J.M. and Lance, G.N. (1966). Multivariate methods in plant ecology V Similarity analysis and infoonation analysis. ~ Ecol. 54: 437-445.

Williams, W.T. and Lance, G.N. (1968). Choice of strategy in analysis of canplex data. Statistician 18: 31-44.

Williams, W.T., Lance, G.N., Webb, L.J. and Tracey, J.G. (1973). Studies in the numerical analysis of canplex rainforest communities IV Models for the classification of quantitative data. J. Ecol. 61: 47-70.

ll Willis, J.H. (1970). "A Handbook to Plants in victoria • vol. 1 (Melbourne University Press: Melbourne).

ll Willis, J.H. (1972). I1A Handbook to Plants in Victoria • Vol. 2 (Melbourne University Press: Melbourne).

Wilson, M.V. (1981). A statistical test of the accuracy and consistency of ordinations. Ecology 62: 8-12.

Winter, T.C. (1977). Classification of the hydrologic settings of lakes in the north central united states. Water Resources Research 13: 753-767. 154.

Wiun-Andersen, S., Anthoni, U., Christophersen, C. and Houen, G. (1982) • Allelopathic effects on phytoplankton by substances isolated from aquatic macrophytes (Charales). Oikos 39: 187-190.

Yen, S.T. (1980). Comparative ecology of three species of amphibious freshwater plants. Ph.D Thesis, University of Sydney. yeroani, G.H. (1970). A study of the Quaternary vegetation history in the volcanic lakes region of Western victoria. Ph.D Thesis, Monash University.

Zadeh, L. (1965). Fuzzy sets. Information and Control 8: 338-353.

Zoltai, S.C. (1979). Wetland classification. In "Ecological (biophysical) land classification in canadaII • Proceedings of the first meeting, May 1976 Petawa, Ontario, ed. J. Thie and G. Ironside, Environment Canada, Lands Directorate Ecological Land Classif. Ser., 1.

Zutshi, D.P. (1975). Associations of macrophytic vegetation in Kashmir Lakes. Vegetatio 39: 61-66. E 5

( II.; ' . ., ~ . , ll...~ ~. l.) i4 ,- 1 " ,,' _1 .. ~4 25 26 27 28 2~ 3M 31 32 ,,3 34 35 36 37 38 39 48 41 42 43 44 45 46 47 43 49 5u 51 52 53 5·\ 5'0 " " Hlllpril brOlilHS rlei:S I I Amphlbromus r'\:'curvatus Aster subel.tus Azull. filicuie.de5 Baume. arliculata " " " " " .. Baumea Juncea Baumea rublgl~osa " B'lJmea tetcagona " Ek.ch)'come 0_5alllea C.llltr ict.e hamul.l. " " Call1tl'lehe slagnallS Car•• appress. " .. Care, faseicularls " Centella cordiflora " " [entiped. minima " * Cnara spp. • Char I z andi' Ii c/frlbar I d. " " " " " " " " C'adiurn procE'rurn • " " Claytoni. austral •• ,ca II " lotula coronopifol i" " " " " " CI'as.ul. h.lm,ii I Damasonium minus " " Echillochloa sp. Elaline 9raliolol015 " Eleocharis acula '" II II " " Eleocharis pusilia I Eleochal'is spnacelata II I Epilobiurn sp. " " " " " " " " * * * Epi lobium sp. • " Erogrostis australaslc•• * • * Gann i .. f I I tirn G.I ium y'Udlct"U"!1 i " * Inus opposi tifol ius I. austral I!> * iagr.cdls ~ " " humlll S pedunculata iola peruoi.r,. oragis brownil * * aragis mlcr.nth. " Il~drucotyle hirt. lIydrocolyle ",uscas. " Hydrocotyle .p. * " I-' Ii,pochoeri. radlcat. * U1 Junc.s art,cul.tus * U1 * * • "

Appendix I Species recorded from Victorian ~tlands S I I t; s "h'eIE;; 2 .:l 5 6 0 9 10 11 12 13 14 15 16 1/ lil 19 i:B <1 22 23 24 25 26 27 2& 29 38 31 32 33 34 35 36 37 38 39 43 41 42 43 44 45 46 47 48 49 50 51 S2 53 S4 55 focK. i I gregi f 1oru, h01 o~choenlls .. .. Ingens JlJf1CljS mal" I t imus Juncu. pailldu,; .. Juncus pauciflorus * " * Juneus planifoliu5 ,Iuncus procerus * * Juneus subsecundus * * Lemna minor • .. • • .. L .mlla tn sulca .. .. * .. .. " Lepldusperma longituJ.nal~ ...... * .. .. L.pidosperma neesi i " " • • * * L.pila.oa bllocularis " .. * * • * Lepi laena c;tl indl'Oc4f'pa ...... lilaeapsis polyantha • * * * * • .. .. • • • .. Lobel ia alata * • * • Lobel ia pratioide. It • * Lud<.igia palustr,s * Lud>,.igia peplo,des Marsiln angustifolla ..* * * Ileialeun Hicifalia * .. Helaleuea sqarros. • .. Her,tha diemenica • .. * ~Iimulus re~.ns .. Muehhnbec ia cunnin\lhami i * .. .. tlyriaphylluffi elat,noldes .. .. * .. * II H;riophyllum muelleri " .. Myriophyllum propinquum .. • • * * *' " • " " " " * " " " " " " " • " " " " " " Myriophyllum verruco~um Nasturtium otficlnal. " " tHtella spp. * .. Nymphoides crenata " " " * " " ".. " " • Ot tel iJ. oual ifol i. " Paspalum dlstichum it Phragmit.", australis " "'0 I ygonum minus " " * Polygonum prostratu", " " Polygonum strigosum " " Polypogon mons~el lensis Potamogeton oc reatu5 " Pot.mog.ton tricarinatus * " * "ratia concolor " " "* " " .. " " 'ratia pedunculata " " 'seudorhapl. paradoxa 'seudcrhapis splneseens ,anunculus inundatu5 " I-' lanunculus rlvularis " U1 lanuncu 1 us trichophyllus " " en " Appendix I continued Appendix I continued

I-' 11l ;-J

Minerva Access is the Institutional Repository of The University of Melbourne

Author/s: Barson, Michele Mary

Title: Numerical analyses of macrophyte vegetation in Victorian wetlands in relation to environmental factors

Date: 1984

Citation: Barson, M. M. (1984). Numerical analyses of macrophyte vegetation in Victorian wetlands in relation to environmental factors. PhD thesis, School of Botany, Faculty of Science, The University of Melbourne.

Publication Status: Unpublished

Persistent Link: http://hdl.handle.net/11343/38252

File Description: Numerical analyses of macrophyte vegetation in Victorian wetlands in relation to environmental factors

Terms and Conditions: Terms and Conditions: Copyright in works deposited in Minerva Access is retained by the copyright owner. The work may not be altered without permission from the copyright owner. Readers may only download, print and save electronic copies of whole works for their own personal non-commercial use. Any use that exceeds these limits requires permission from the copyright owner. Attribution is essential when quoting or paraphrasing from these works.