ChapterPDF Available

Floristic-ecological mapping in the Northern Kimberley: field survey methods and mapping protocols

January 2013

In book: Vegetation mapping in Northern Kimberley, Western Australia (pp.26–87)
Chapter: Floristic-ecological mapping in the Northern Kimberley: field survey methods and mapping protocols
Publisher: Curtin University, Perth
Editors: Ladislav Mucina, Glen Daniel

Authors:

Ladislav Mucina

Murdoch University

Glen Daniel

Department of Parks and Wildlife

Show all 9 authorsHide

This chapter describes the procedures and outcomes of the floristic-ecological mapping of the Mitchell Plateau and Pantijan areas of the Northern Kimberley region, Western Australia. A protocol is presented comprising a seven step process of: setting mapping goals, collecting and collating data, identifying the major ecological drivers of vegetation patterns, setting classification and mapping rules, modelling procedures, verification and presentation. The mapping protocol provides a definitive approach to vegetation mapping at intermediate scales (1:50,000 to 1:250,000) where the availability of ground data is constrained by limited access to the mapping area. This is the first time that formal vegetation data-analytical tools have been used to classify the vegetation of Western Australia’s savanna biome. The study is also unique in linking the resultant vegetation classification system to formal mapping modelling, to produce a map of the potential natural vegetation of part of the State. The field-survey, data analytical tools and methods of vegetation This chapter describes the procedures and outcomes of the floristic-ecological mapping of the Mitchell Plateau and Pantijan areas of the Northern Kimberley region, Western Australia. A protocol is presented comprising a seven step process of: setting mapping goals, collecting and collating data, identifying the major ecological drivers of vegetation patterns, setting classification and mapping rules, modelling procedures, verification and presentation. The mapping protocol provides a definitive approach to vegetation mapping at intermediate scales (1:50,000 to 1:250,000) where the availability of ground data is constrained by limited access to the mapping area. This is the first time that formal vegetation data-analytical tools have been used to classify the vegetation of Western Australia’s savanna biome. The study is also unique in linking the resultant vegetation classification system to formal mapping modelling, to produce a map of the potential natural vegetation of part of the State. The field-survey, data analytical tools and methods of vegetation

Survey sites showing the actual positions of the plots sampled in the newly delimited Pantijan mapping area in 2011.

…

An example of the relative error curve produced by CART modelling.

…

Survey sites showing the actual positions of plots sampled in the Mitchell Plateau mapping area in 2011. Validation sites, represented by yellow circles, indicate the position of independent verification points collected during the 2012 field campaign.

…

Final CART diagram for the Pantijan zonal vegetation units (detailed classification). The alpha-numerical (and colour) coding of the classes is as follows: B (incl. C, D), E, G (incl. H), I, j, K, L. (Communities A, F, M were not modelled.). Insert: Simplified CART classification tree showing the hierarchy of classifiers (splitters).

…

Figures - uploaded by Russell L Barrett

Content may be subject to copyright.

Content uploaded by Russell L Barrett

Content may be subject to copyright.

Curtin University, Perth, WA

Editors: Ladislav Mucina and Glen Daniel

VEGETATION MAPPING IN

THE NORTHERN KIMBERLEY,

WESTERN AUSTRALIA

2013

Vegetation Mapping in the

Northern Kimberley,

Western Australia

Editors

Ladislav Mucina and Glen Daniel

Curtin University, Perth

2013

by the Copyright Act 1968, this material may not be

reproduced, stored or transmitted without the permission

of the copyright owner. Information and other material

included in the Publication is made available on the

understanding that the Publication expresses the views of

the researchers and are published in good faith by Curtin

University of Technology (“Curtin”). The Publication can

include views or recommendations of third parties and such

material does not necessarily reﬂect the view of Curtin or

indicate a commitment to follow a particular course of action.

The Publication is provided on an “as is” basis. Curtin

does not make any warranty or representation as to the

accuracy, adequacy, reliability or completeness of the

information and data contents in the Publication. Any

material or data obtained through use of this Publication

is at your, the User’s, sole discretion and risk and that you,

the User, will be solely responsible for any resulting loss

or damage that you may incur or sustain from the use

or reliance upon any material, products or other data or

otherwise referred to in the Publication. Neither Curtin, its

ocers, employees or agents, shall be liable for any loss

or damage of any kind however caused (whether by any

negligent or other unlawful act or omission of, by or on the

part of the Curtin or otherwise) arising from the use of or

reliance upon any information data or advice expressed or

implied in this Publication.

Published in July 2013 by:

Curtin University

Kent Street

Bentley 6102

Australia

ISBN 13: 978-0-646-90719-2

ISBN 10: 0646907190

Printed by: Advance Press

This a scientiﬁc peer-reviewed document.

Suggested citation formats:

The Book:

Mucina, L. & Daniel, G. (eds) 2013. Vegetation Mapping

in the Northern Kimberley, Western Australia. Curtin

University, Perth, WA. ISBN 13: 978-0-646-90719-2,

ISBN 10: 0646907190

A chapter:

Mucina, L., Daniel, G., Stephenson, G., Boonzaaier, I., van

Niekerk, A., Barrett, M., Barrett, R., Tichý, L. & Valachovič,

M. 2013. Floristic-ecological mapping in the Northern

Kimberley: Field survey methods and mapping protocols.

In: Mucina, L. & Daniel, G. (eds), Vegetation mapping in the

Northern Kimberley, Western Australia. pp. 26-87.

Curtin University, Perth, WA. ISBN 13: 978-0-646-90719-2,

ISBN 10:0646907190

Photographic credits:

All photographs in this book have been made by the members of the ﬁeld

teams associated with the project Northern Kimberley Vegetation Mapping,

except those where authorship have been speciﬁed explicitly and the photo

on page 2 (upper) by F.G.T. Radlo, Stellenbosch, South Africa, ﬁg. no. 2 on

page 7 and ﬁg. no. 4 on page 27 (images from Digital Globe, google earth).

Glen Daniel

Glen Daniel is a Senior Fire Operations Ocer with the

Western Australian Department of Parks and Wildlife

(DPaW). His qualiﬁcations are in Environmental Science,

with an emphasis on land management. He has previously

worked in vegetation and wetland management.

His primary focus is currently the development of ﬁre

planning instruments for Western Australia’s varied natural

and cultural environments. Other interests include studies

of ﬁre behaviour and post ﬁre recovery of natural systems.

Glen undertook his work on the Northern Kimberley

Vegetation Mapping Project as a two year secondment

to Curtin University.

Contact address:

Fire Management Services Branch

Department of Parks and Wildlife (DPaW)

Locked Bag 104

Bentley Delivery Centre 6983

Perth, Australia

W/Professor Ladislav Mucina

Laco Mucina is a Winthrop Professor of Vegetation

Science and Biogeography, Iluka Chair, at the School of

Plant Biology, The University of Western Australia, Perth,

Australia. He has worked in a number of universities on four

continents (Europe, Asia, Africa and Australia), published

more than 300 scientiﬁc papers (23 books), served as editor

of several international scientiﬁc journals, and supervised

more than 30 PhD and MSc student projects. His global

scientiﬁc interests span descriptive vegetation science

(especially vegetation surveys, classiﬁcation and mapping),

biosystematics, molecular phylogeny, evolutionary

biology, biogeography, palaeoecology, biodiversity science,

environmental management, plant community restoration,

and conservation biology.

Contact address:

School of Plant Biology

The University of Western Australia

35 Stirling Highway

Crawley 6009

Perth, Australia

Email: Laco.Mucina@uwa.edu.au

Ladislav Mucina1, Glen Daniel2, Garth Stephenson3,

Ilze Boonzaaier3, Adriaan van Niekerk3, Matthew Barrett4,

Russell Barrett4, Lubomír Tichý5 and Milan Valachovič6

1 Department of Environment & Agriculture, Curtin University, GPO Box U1987,

Perth 6845, Australia; School of Plant Biology M084, The University of Western

Australia, 35 Stirling Highway, Crawley 6009, Australia;

Email: Laco.Mucina@uwa.edu.au

2 Fire Management Services Branch, Department of Parks and Wildlife (DPaW),

Perth, Locked Bag 104, Bentley Delivery Centre, WA 6983, Bentley, Australia.

3 Centre for Geographical Analysis, Department of Geography & Environmental

Studies, Stellenbosch University, Private Bag X1, Matieland 7602, Stellenbosch,

South Africa.

4 Botanic Gardens & Parks Authority, Kings Park & Botanic Garden, Fraser Avenue,

WA 6005, West Perth, Australia; School of Plant Biology, M084. The University of

Western Australia, WA 6009, Crawley, Perth, Australia.

5 Department of Botany & Zoology, Masaryk University, Kotlářská 2, CZ-61137

Brno, Czech Republic.

6 Department of Geobotany, Institute of Botany, Slovak Academy of Sciences,

Dúbravská cesta 9, SK-84523 Bratislava, Slovakia.

ABSTRACT

This chapter describes the procedures and outcomes of

the ﬂoristic-ecological mapping of the Mitchell Plateau and

Pantijan areas of the Northern Kimberley region, Western

Australia. A protocol is presented comprising a seven step

process of: setting mapping goals, collecting and collating

data, identifying the major ecological drivers of vegetation

patterns, setting classiﬁcation and mapping rules, modelling

procedures, veriﬁcation and presentation. The mapping

protocol provides a deﬁnitive approach to vegetation mapping

at intermediate scales (1:50,000 to 1:250,000) where the

availability of ground data is constrained by limited access to

the mapping area. This is the ﬁrst time that formal vegetation

data-analytical tools have been used to classify the

vegetation of Western Australia’s savanna biome. The study

is also unique in linking the resultant vegetation classiﬁcation

system to formal mapping modelling, to produce a map of

the potential natural vegetation of part of the State. The

ﬁeld-survey, data analytical tools and methods of vegetation

Mucina, L., Daniel, G., Stephenson, G., Boonzaaier, I., van Niekerk, A., Barrett, M.,

Barrett, R., Tichý, L. & Valachovič, M. 2013. Floristic-Ecological Mapping in the

Northern Kimberley: Field Survey Methods and Mapping Protocols. In: Mucina, L.

& Daniel, G. (eds), Vegetation Mapping in the Northern Kimberley, Western Australia.

pp. 26-87. Curtin University, Perth, WA.

Chapter 2

Floristic-Ecological Mapping

in the Northern Kimberley:

Field Survey Methods and

Mapping Protocols

modelling proved very successful and both maps are useful

depictions of the patterning of vegetation at the targeted

scale. The validity of these maps is evident from both formal

veriﬁcation and interpretation (the message conveyed by

the maps about ecological drivers and previous

ﬂora-assembly processes).

INTRODUCTION

1.1 Motivation

Vegetation mapping utilises technical protocols and tools

to translate vegetation-ecological theory into a model

known as a ‘vegetation map’. This is a very complex process,

but there are some basic steps that may be followed to

construct an ecologically informative and scientiﬁcally

sound map. The ﬁrst step is to gather the pertinent data

sets. These are then used to classify the vegetation, with the

aim of simplifying the vegetation complexity and deﬁning

operational units called plant communities or vegetation

mapping units. The third step is to identify the major

ecological drivers, or environmental fabric, underpinning the

vegetation complexity. This allows the classiﬁcation to be

linked to the environmental fabric. Finally the vegetation-

environment relationship is translated into a spatial model

or vegetation map.

The basic steps described above are common to all

vegetation-mapping protocols (see for instance Keith

2002, Brocklehurst et al. 2004, Sivertsen 2009 and

Neldner et al. 2012 for some representative examples from

Australia). Each dierent mapping area, however, will have

idiosyncratic features that must be considered. Examples

of such features are the availability of geographic data

sets, their quality and spatial resolution, accessibility of the

mapping area and taxonomic challenges.

The Northern Kimberley is a challenging region in which to

undertake vegetation mapping. It does, however, provide a

good opportunity to develop and test vegetation-mapping

strategies for areas where data availability is less than

optimal. With this goal in mind, two areas within the

Northern Kimberley were chosen as locations to construct

vegetation maps that will inform future biodiversity and

social benchmarking research. This was also used as an

opportunity to develop mapping protocols for future large-

scale ﬂoristic-ecological mapping of the remainder of the

Kimberley and the State of Western Australia.

1.2 Operational Targets

The broad goals described in the preceding paragraph were

reﬁned to the following operational targets:

(1) Test the applicability of the basic vegetation-mapping

protocols under development (Mucina in prep.) for

WA and further develop them for areas with limited

accessibility and scarce ground data (sampling plots)

coverage;

(2) Formulate ﬂoristic-sociological vegetation

classiﬁcation systems (based on ﬂoristic co-occurrence

data) for the Mitchell Plateau and Pantijan areas, using

contemporary ﬁeld-survey procedures and numerical-

analytical tools;

(3) Construct ecologically informative vegetation maps of

the Mitchell Plateau and Pantijan areas, at a mapping

scale of 1:50,000, that inform biodiversity and social

benchmarking; and

(4) Draw lessons from this classiﬁcation and mapping

exercise to facilitate Kimberley-wide and state-wide

vegetation mapping programmes.

1.3 Challenges

The Kimberley is situated in the northernmost part of

Western Australia. The region features extensive tracts

of wilderness, where human activity is limited and

infrastructure sparse. There are few roads and access is

further limited by the topographically highly articulated

landscape and extreme climatic conditions during the wet

cyclone season.

A vegetation survey in such a remote and poorly accessible

area faces a number of formidable challenges, such as:

• Highcostofeldsurveyworkduetothelargedistances

travelled, need for specialised 4WD vehicles, dependence

on signiﬁcant logistical support and low quality of the

road network oen resulting in damage to vehicles.

• Highlevelofrisktoparticipantsduetohealthhazardsand

a lack of medical emergency infrastructure.

CHAPTER 2 FIELD SURVEY AND PROTOCOLS

APPROACH

The approach used in the current study follows the

vegetation-environmental paradigm. That is, it relies on

the high predictive value of the link between the current

vegetation patterns and the environmental matrix. It

designates the environmental matrix as the current

ecological driver of the biotic patterns at the regional,

landscape and habitat scales. This approach rests on seven

basic postulates of vegetation mapping as deﬁned by

Mucina et al. (2006):

Postulate 1: Vegetation is a real, tangible object expressed

in the form of recognizable patches. In other words:

vegetation is a real phenomenon and can be studied.

Postulate 2: The dierences between the vegetation

patches in terms of structure, texture (ﬂoristic composition)

as well as in terms of environmental composition of the

habitats supporting the vegetation, make the classiﬁcation

of vegetation (or conceptualisation of theoretical constructs

called ‘vegetation types’) possible. In other words: we can

classify vegetation patches into vegetation types.

Postulate 3: The great complexity of vegetation, both of a

discrete and continuous nature, makes the classiﬁcation

of vegetation (or the reduction of information content to a

simpliﬁed system) necessary. In other words: classiﬁcation

is one (and a very eective) way of simplifying the

complexity of vegetation.

Postulate 4: The levels of dierence between vegetation

types make building a hierarchical system (comprising

a series of nested vegetation types and their groups)

possible. The hierarchical system is another tool for further

simpliﬁcation of vegetation complexity. In other words:

the hierarchical system is another eective way to view

important emergent properties of the major patterns of

vegetation.

Postulate 5: The structure and dynamics of vegetation is

a result of properties of its constituent plant populations

and their response to the nature and dynamics of the

environment which can aid classiﬁcation and mapping

(‘vegetation-environment axiom’). In other words:

environmental conditions determine (together with the

properties of vegetation itself) the complexity of vegetation.

Postulate 6: Vegetation is composed of populations of

plant species (representing taxa). Each taxon oen shows

an individual response to ecological factors and hence

serves as an important ecological indicator. Major eorts to

devise an alternative classiﬁcation of functional types have

yet to yield a viable system for widespread application. In

other words: we use ﬂoristic composition as the primary

entity for the conceptualisation of mapping units.

• Lowqualityofspatialdata,particularlythosethatare

based on ﬁeld observations. For example, geological

mapping may be of poor quality and the precision of

bioclimatic modelling may be low due to the paucity of

climatic stations collecting baseline data.

• Qualityofecologicalandvegetationelddata

compromised due to a lack of suitable spatial data and

challenges imposed by climatic seasonality. For example,

ﬁeld sampling is usually limited to the phenologically

suboptimal dry season.

• Impededprogressoftaxonomicknowledgeoftheora

and a lack of taxonomic expertise due to the diculty of

obtaining comparative ﬂoristic material.

• Restrictionsonaccesstosomeareasforculturalreasons.

In addition to the above, budgetary limitations and time

constraints are omnipresent challenges to any ﬁeld

survey work.

Expectations deﬁned by the Challenges

The challenges described above are likely to have a

deleterious eect on the process of developing a vegetation

map for a remote area. Vegetation mappers in such areas

should, therefore, enter into the process with the following

expectations:

(1) The costs of the ﬁeld survey work will be high hence the

quantity and quality of data may be low. Alternative

ways of collecting ground data should be sought in

order to meet the minimum requirements.

(2) The volume of ground data will be small and the

spatial dispersion of the data points will be suboptimal

(spatially biased).

(3) Plant specimen identiﬁcation will be dicult as many

of the specimens are likely to be collected out of their

optimal phenological phase. It may be necessary to

engage external contractors with particular expertise

in the local ﬂora. Short-lived plants and grasses should

be excluded from sampling due to the high risk of

ﬁeld sampling bias, especially amongst inadequately

trained or less experienced ﬁeld sta. Such workers

may overlook species that are in poor phenological

condition. Collections of ephemeral taxa in poor

condition would also necessitate forensic identiﬁcation

hence incurring disproportionate eort and cost.

(4) Qualitative ecological data will be used in the analyses

to a greater extent than would be the case in regions

where high-quality GIS and remote-sensed data are

available. This will not necessarily impede the analyses.

(5) Attention must be paid to the selection of a modelling

(mapping) approach that is suitable for sparse data

and low quality GIS coverage.

29CHAPTER 2 FIELD SURVEY AND PROTOCOLS

4) the repeatability of the mapping process, i.e. whether

it is a one-o mapping exercise or part of a monitoring

program; and

5) whether there are any constraints that might

prevent project goals from being achieved e.g.

access limitations, disturbance regimes or technical

constraints.

Decisions and Outcomes

1) The Mitchell Plateau mapping area was selected

because it is relatively accessible, has previously been

mapped, has high landscape and geological diversity,

and therefore presumably also has high biotic

variability, and is culturally important to Traditional

Owners. The Pantijan mapping area was selected

because it receives less rainfall than the Mitchell

Plateau, has high landscape and geological diversity,

and therefore presumably also high biotic variability,

and is culturally important to Traditional Owners.

2) Both maps have a ﬂoristic-ecological theme. They

depict major biodiversity (vegetation) patterns

that reﬂect the current ecological fabric and past

evolutionary history (biogeographic eects) of the

mapping area.

3) The acceptable mapping scale is between 1:50,000

and 1:250,000. The precision of most of the available

spatial data for these areas is within this range.

4) Neither map will serve a targeted monitoring

programme. As such, it is reasonable to incorporate

mapping tools that are not fully automated.

5) The following additional constraints were considered:

a. There is a strong inﬂuence of ﬁre in the mapping

areas that aects the utility of some remotely-

sensed data sets, such as NDVI.

b. The mapping areas are characterised by strong

climatic seasonality that restricts the identiﬁcation

of some elements of the ﬂora and the use of some

remotely-sensed datasets.

c. There is very high grazing pressure on nutrient-

rich soils derived from igneous rocks. The grazing

pressure in environments characterised by low

nutrient availability, such as on sandstone and

laterite is low.

d. Access to the mapping areas is restricted by a lack

of roads and highly articulated topography. There

is particularly poor access to sandstone-dominated

landscapes.

e. Permission for access to the mapping areas was

granted by the relevant Traditional Owners; these

processes were facilitated by the Kimberley Land

Council.

Postulate 7: Vegetation patches occur in space hence

they can be mapped in spatial models. In other words:

complexity of vegetation can be shown on a map.

The concepts of zonality and azonality (Walter 1964,

1973) also underlie the approach to mapping adopted

in the current project. For details on these and related

concepts, consult Box 2.1.

The theme of this mapping exercise is colloquially known

as a ﬂoristic-ecological map. The reason for this is that

the map is based on the relationship between the present

plant species (ﬂora) and the environment.

The maps produced during the current project feature

potential natural vegetation (see Tüxen 1956 for basic

concepts and Mucina 2010 for a recent review). This

reﬂects the use of the current ﬂora/ecology relationship to

predict the distribution of vegetation patterns in areas not

sampled. Some of the units will, however, be delineated by

the mapping of actual patches. This applies especially to

azonal and relict vegetation such as Kimberley monsoonal

rainforests (see Chapters 3 and 4 for details).

VEGETATION SURVEY AND MAPPING PROTOCOL

Seven basic steps summarise the entire survey and

mapping exercise, from the deﬁnition of goals to the ﬁnal

presentation. These steps may also be seen as the basis of

the state vegetation mapping protocols. The steps are:

Step 1: Primary decisions: setting goals and parameters.

Step 2: Collecting and collating mapping data sets.

Step 3: Identifying major drivers of vegetation patterns.

Step 4: Setting mapping rules and modelling.

Step 5: Making post-modelling adjustments.

Step 6: Veriﬁcation.

Step 7: Presentation.

Each of these steps is explained in detail below.

3.1. Step 1: Primary Decisions: Setting Goals

and Basic Parameters

The ﬁrst step in the mapping process is of utmost

importance. It aims to increase the speciﬁcity of the

client’s description of the desired product by identifying

the theme/topic, mapping scale, mapping frequency and

any constraints on the process. In particular, the following

must be decided:

1) the exact delimitation of the mapping area(s);

2) the theme or topic of the map, i.e. whether real

vegetation or modelled vegetation will be mapped;

3) the mapping scale, although this may be revised in

Step 2 as the quality of the available data sets may

inﬂuence the mapping targets;

seven bands per Landsat scene. The stacked images were

radiometrically corrected to reﬂectance using ATCOR2

(Richter 2004) in PCI Geomatica (www.pcigeomatics.

com/). This improved the compatibility between scenes.

ATCOR3, the version of ATCOR that includes topographical

corrections, was not used, since a digital elevation model

with suitable quality was not available for the study area.

The Landsat scenes were visually inspected to determine

their suitability for further analyses. Images showing

cloud cover or large recent ﬁre scars were discarded.

The following images were selected for use in the

modelling process:

Panjitan: L5109071_20100331 (dated 31 March 2010).

Mitchell Plateau: L510970_20100603 (dated 3 June 2010).

The scenes were clipped to the extent of the pre-deﬁned

Mitchell Plateau and Pantijan mapping areas.

Image Transformations

An image transformation is any operation that

re-expresses the information content of an image

for the purpose of deriving usable data not apparent

from the individual bands (Mather 2011). Two image

transformations were applied to the clipped Landsat

scenes, Principal Components Analysis (PCA) and the

Normalised Dierence Wetness Index (NDWI).

There is oen a high degree of correlation between the

individual bands of a multispectral image. PCA (for general

algorithms and explanation in ecological context see Orlóci

1978, Legendre & Legendre 1998) reduces the redundancy

of the data by identifying the optimal linear combinations

of the original channels and altering the dimensional

axes in such a manner that correlation is minimized. The

outcome of PCA is a series of coecients, or eigenvectors,

that align the principal axis along the strongest degree

of correlation in dimensional space. This concentrates

the maximum amount of information possible into one

band, called the ﬁrst principal component (PC1). The

second largest axis, mutually perpendicular to (and hence

independent from) the ﬁrst axis in dimensional space is

called the second principal component (PC2). This contains

the maximum amount of information from that which

remains aer the creation of PC1. This process is repeated

to form the same number of principal components as the

original number of bands (or otherwise speciﬁed), resulting

in a set of layers of decreasing variability (Liang 2004,

Campbell 2006, Mather 2011).

In the Kimberley mapping, a six-band PCA was undertaken

on bands 1–5 and 7 of the Landsat images. Band 6 is a

thermal band and was excluded from all analyses due to its

lower resolution.

3.2 Step 2: Collecting and Collating Mapping

Data Sets

The availability of spatial data has a major impact on

the quality of vegetation mapping. Published literature

and non-published scientiﬁc reports are both important

sources of data. Spatial data sets may also be derived

from a variety of sources, but should be carefully validated

before use.

Decisions and Outcomes

1) We obtained all accessible published and unpublished

(grey) literature that is pertinent to vegetation survey

and mapping in the Kimberley. The literature studies

reinforced our primary assessment that geology is

one of the major drivers of vegetation patterns in the

region. The best available geological mapping of the

Mitchell Plateau and Pantijan mapping areas is the

Australian 1:250,000 Geological Map Series. These

were available as a hard-copy map or as a scanned,

geo-referenced raster image. We digitised the relevant

map sheets (see also Chapter 1) to obtain a dataset

that could be manipulated with GIS. The Mitchell

Plateau mapping area coincided with the Montague

Sound and Drysdale-Londonderry geology map sheets

while the Pantijan mapping area incorporated the

Charnley and Prince Regent-Camden Sound map

sheets.

2) We obtained numerous spatial datasets as GIS layers

or rasters. A full list of these is provided in Appendix

1.1. The 35 low-resolution climatic variables derived

from the BioClim mean monthly climate estimates (for

details see Nix 1986) were of particular importance.

These were used in modelling the map units while the

19 basic WorldClim variables (Hijmans et al. 2005)

were used in the numerical (ordination) analyses.

Landsat Imagery Selection and Preprocessing

Landsat satellite imagery was recognised as a dataset

that is crucial to mapping in the Kimberley. This imagery

is freely available from the Earth Resources Observation

and Science Centre of the United States Geological Survey

(http://glovis.usgs.gov/). The Landsat scenes used in the

current project covered the period from 1999 to 2010;

these were derived from Landsat 5 and Landsat 7. Both of

these platforms possess very similar multispectral sensors

(Turner et al. 2003). Preference was given to the Landsat

5 imagery because the correction of a fault inherent in

the Landsat 7 imagery aects the quality of the images

supplied by that platform.

The Landsat imagery was supplied in a Standard Terrain

Correction Level 1T format that included systematic

radiometric and geometric corrections. A separate image

for each sensor spectral band was supplied. These band

images were then stacked, resulting in one image with

31CHAPTER 2 FIELD SURVEY AND PROTOCOLS

devised for the areas by Beard (1978, 1979, 1988) and

Hnatiuk & Kenneally (1981). The methods used to produce

the Mitchell Plateau map are summarised in Box 2.2.

A similar approach was adopted to produce the preliminary

vegetation map of the Pantijan area.

The vegetation classiﬁcation system of Hnatiuk &

Kenneally (1981) recognises 26 habitat/vegetation types.

The classiﬁcation and mapping by Beard (1978, 1979) is

simpler in both of the mapping areas as it clearly reﬂects

the driving inﬂuence of geology. The later Beard map

(1988) of the Mitchell Plateau also reﬂects the importance

of geology as a driver of vegetation patterns.

The preliminary map of the Mitchell Plateau was used to

identify suitable locations to situate sample plots during

the survey. This selection was designed to incorporate

the full range of vegetation variability depicted on the

preliminary map while allowing for considerations

of access. We refrained from pre-selecting sampling

locations in the Pantijan mapping area because of a lack of

information about the nature of that area.

Validating Previous Maps

The two classiﬁcation systems previously developed for

the mapping areas (Beard 1978, 1979, 1988, Hnatiuk &

Kenneally 1981) were tested in the ﬁeld to determine if

they adequately represented the vegetation. We assigned

vegetation patches studied in the ﬁeld to communities

using the characteristics of the plant communities

described in the source publications. The results of this

process were compared to the ﬁndings of an expert

classiﬁcation undertaken in the ﬁeld. In some cases,

these comparisons assisted in selection of sampling plots

(see below for more detail). We have established that the

vegetation classiﬁcation system of Hnatiuk & Kenneally

(1981) for the Mitchell Plateau was very logical.

Establishing Survey Plots

It was necessary to modify the pre-selected set of plots

in the Mitchell Plateau mapping area (Fig. 2.1). When the

pre-selection process was undertaken, it was thought that

a helicopter would be used to access some of the more

remote sampling locations. However, since an aircra was

not available at the time of the survey, such sites could

not be reached. This particularly aected sampling of the

dissected sandstone landscapes where the nature of the

terrain makes overland travel very dicult. As a result

of this constraint, plot sampling was limited to areas in

the proximity of roads. The extent of the road network in

the Mitchell Plateau mapping area allowed a reasonably

representative set of sampling plots to be established. The

vegetation of the lateritic plateau and associated low-lying

basalt landscapes was sampled more intensively, however,

than that of the sandstone landscapes (Fig. 2.3).

Spectral indices were not used in mapping as it was

thought that the individual spectral bands and six principal

components would encompass all of the spectral variation

within the Landsat images. The only exception was an

attempt to use the Normalized Dierence Wetness Index

(NDWI) to create a dataset informative of the distribution

of water in the landscape. It was hoped this would assist

with the identiﬁcation of azonal riparian vegetation. The

NDWI was therefore calculated for both areas according to

the following formula by Gao (1996): NDWI = [(Band 4) -

(Band 5)] / [(Band 4) + (Band 5)].

3.3 Step 3: Identifying Major Drivers of

Vegetation Patterns

A vegetation map is a model of vegetation that expresses a

theory about the pattern of vegetation and the logic of its

distribution in space and time. A scientiﬁcally plausible and

informative map relies on the mapper’s ability to recognise

the processes that play a role in shaping the vegetation.

These may be acting in the present at an ecological scale or

may be an historical artefact considered at an evolutionary

scale. Identifying the drivers of vegetation patterning is

crucial to modelling the vegetation patterns themselves.

In the preliminary steps of the Kimberley mapping project

we learnt about the drivers of vegetation patterns from the

observations of previous workers in the area. In particular,

the work of Beard (1978, 1979, 1988) and Hnatiuk &

Kenneally (1981) provided valuable information about the

processes that shape the vegetation of the Mitchell Plateau

and Pantijan areas. We supplemented this information,

however, by collecting ﬁeld data from the mapping areas.

We analysed these data sets and drew conclusions about

the role of environmental factors that might have driven

the current as well as relict vegetation patterns observed in

the mapping areas.

3.3.1. Field Sampling

Two vegetation surveys were conducted during the

Kimberley mapping project. The Mitchell Plateau mapping

area was sampled between 20 July 2011 and 26 July 2011

and the Pantijan mapping area was sampled between 1

August 2011 and 6 August 2011. These surveys collected

vegetation-plot material for modelling and identiﬁed the

major ecological drivers of vegetation patterns in the

Mitchell Plateau and Pantijan areas. The Mitchell Plateau

survey also tested the validity of the previous vegetation

classiﬁcation systems applied to that area (Beard 1979,

1988, Hnatiuk & Kenneally 1981).

Survey Preparation

The ﬁrst task in preparing for the survey campaign was

to prepare preliminary vegetation maps for the Mitchell

Plateau (Fig. 2.1) and the Pantijan area (Fig. 2.2). These

were based on the vegetation classiﬁcation systems

FIG. 2.3

Survey sites

showing the

actual positions

of plots sampled

in the Mitchell

Plateau

mapping

area in 2011.

Validation sites,

represented by

yellow circles,

indicate the

position of

independent

veriﬁcation

points collected

during the 2012

ﬁeld campaign.

FIG. 2.4

Survey sites

showing the

actual positions

of the plots

sampled in the

newly delimited

Pantijan mapping

area in 2011.

33CHAPTER 2 FIELD SURVEY AND PROTOCOLS

3) The classiﬁcation of tree and shrub savanna plant

communities using only the tree component has

been shown to be highly informative (Nežerková-

Hejcmanová et al. 2005 and ample literature cited

therein). Arndt & Norman (1959) also found little

interdependence between tree and ground ﬂora.

On the basis of the information given above, we concluded

that the woody component carried sucient ecological

information to allow an eective classiﬁcation of the

vegetation. Indeed, as the herbaceous layer is prone

to bias, woody vegetation is the preferred source of

classiﬁcation variables.

3.3.2 Collation of the Field Data

Plant Identiﬁcations

The survey team included few individuals with sucient

taxonomic-ﬂoristic knowledge to undertake accurate ﬁeld

identiﬁcation of specimens. As such, almost all woody

plant species were collected, marked and preserved by

pressing. Very few plants were identiﬁed with conﬁdence

in the ﬁeld and hence not collected. Where appropriate,

photographs were also taken of the plant from which a

specimen was collected. Information about the habit, bark,

leaves, fruit and ﬂowers was thus recorded and proved very

useful during the identiﬁcation process.

Most of the collected plant material was identiﬁed using

the facilities of the State Herbarium (PERTH) located at

the Western Australian Biodiversity Conservation Centre

(DPaW), Kensington. The identiﬁcations were aided by:

• Flora of the Kimberley Region (Wheeler 1992);

• Field Guide to Eucalyptus (Brooker & Kleinig 1994);

• EUCLID (Centre for Plant Biodiversity Research 2006) was

used for Eucalyptus;

• WATTLE (Maslin 2001) was used for Acacia;

• Plants of the Kimberley Rangelands by Petheram et al. (2003);

• WAorachecklistbyPaczkowska&Chapman(2000);

• FloraBaseidenticationtools(www.orabase.dec.wa.gov.au;

and

• thelatestrevisionsofsomegeneraencounteredinthe

collected material.

Despite the sub-optimal phenological states and poor

condition of many of the specimens, more than 90% were

identiﬁed unequivocally to species (and in many cases

subspecies) level.

Collation of the Environmental and Geographical Data

There was some inconsistency in the way that

environmental and geographical data were collected

during sampling. This resulted from the large number of

participants in the survey and the limited opportunity

to provide training in survey techniques. The lack of

completeness and consistency in the environmental and

geographical datasets limited their usefulness as a source

of data for modelling. In particular:

1) The variables describing the topographic position

and landform (see Tables 3 and 4 in Appendix 2.1

The sandstone vegetation in the Pantijan mapping area

was sampled more intensively than in the Mitchell Plateau

area. This was possible because the access route, the

Munja Track, passed through extensive areas of sandstone.

Overall, however, the track network in the Pantijan mapping

area was very limited and a lack of access constrained the

placement of sampling plots. In fact, a signiﬁcant proportion

of the sampling plots in the Pantijan area were placed

outside the originally deﬁned mapping area (Fig. 2.4). This

should not be seen as detrimental to the mapping exercise,

however, as the strong link between major vegetation

patterns and geology in the Kimberley (Gardner 1942, Miles

& Burbridge 1975, Kabay & Burbridge 1977, George 1978,

Beard 1978, 1979, Fox et al. 2001, Kenneally et al. 2008)

allows extrapolation (modelling) over large distances.

Field Sampling Protocols were designed prior to the

commencement of the survey campaign. These are

presented in Appendix 2.1, along with justiﬁcation of the

survey design. Sampling commenced much later than

expected, however, aer unusually high rainfall in the

2010/2011 wet season caused extensive damage to the

road network. This prevented access to the mapping areas

until late in the dry season and meant that herbs and

grasses were in very poor condition at the time of sampling.

It was necessary, therefore, to alter the protocols to allow

for the prevailing conditions.

Sampling was limited to woody vegetation (trees, shrubs,

lianas) and perennial herbs. We recorded the total cover

(in %) of the tussock (e.g. Themeda, Heteropogon,

Cymbopogon etc.) and hemisphaeric grasses (Triodia) as

separate variables. Grasses and short-lived herbs were

excluded from sampling for the following reasons:

1) Part of the grassy component of the vegetation is

short-lived and can only be found while in an optimal

phenological state (ﬂowering and seeding) during, and

shortly aer, the wet season. At the time of sampling,

these grasses had deteriorated signiﬁcantly and were

found in the post-reproductive stage of development.

The progressed phenology made ﬁeld identiﬁcation

extremely dicult and would have impeded the

laboratory identiﬁcation of plant material.

2) The grassy understorey in savanna vegetation is highly

dynamic and is characterised by rapidly changing

dominance relationships in response to ﬁre and

precipitation. Hüttich (2011) and Hüttich et al. (2011)

showed that woody vegetation in savannas provides

more stable variables for use in vegetation modelling/

mapping than herbaceous vegetation. Also, that woody

vegetation is less likely to present errors of omission

and commission compared with herbaceous vegetation.

Woody vegetation is essentially less sensitive to inter-

annual variability than herbaceous life forms (Archibald

& Scholes 2007).

FIG. 2.5

Example of the OptimClass analyses for vegetation classiﬁcations of the Mitchell Plateau plot data (for 3-layer data

matrix only) using various combinations of the Unweighted Pair Group Method Using Arithmetic Averages. X axis

denotes the number of cluster solutions tested and the Y axis represents the OptimClass value.

FIG. 2.6

Example of the OptimClass analyses for the vegetation classiﬁcations of the Pantijan plot data (for 3-layer data matrix

only) using various combinations of the Unweighted Pair-Group Method Using Arithmetic Averages. The X axis denotes

the number of cluster solutions tested and the Y axis represents the OptimClass value.

35CHAPTER 2 FIELD SURVEY AND PROTOCOLS

FIG. 2.7

A dendrogram of the UPGMA based on Similarity Ratio (= quantitative form of Jaccard’s Similarity Index)

and logarithmic transformation of the percentage data of the Mitchell Plateau sample plots. Capital letters

code the plant community as recognised in the ﬁnal classiﬁcation system for the region (see Table 2.4).

FIG. 2.8

A dendrogram of the UPGMA based on Similarity Ratio (= quantitative form of Jaccard’s Similarity Index) and

logarithmic transformation of the percentage data of the Pantijan sample plots. Capital letters code the plant

community as recognised in the ﬁnal classiﬁcation system for the region (see Table 2.6). The plots engulfed in a red

rectangle are representatives of Community J that were misclassiﬁed as Community C (see text for more detail).

data set. The diagnostic value of each species can be

ascertained using various statistical criteria; Fisher’s exact

test is used in OptimClass (see Tichý & Chytrý 2006).

If a classiﬁcation produces well-deﬁned, interpretable

clusters, then these clusters should have a high number of

diagnostic species.

We performed an OptimClass search on a selection of

the most widely-used techniques in community ecology,

including Unweighted Pair-Group Method using Arithmetic

Averages (UPGMA), Beta Flexible Clustering, Ward’s Method

in combination with a Bray-Curtis Index, Similarity Ratio,

Chord Distance and none or logarithmic and/or power

transformation. The cluster analyses used to calculate the

OptimClass were performed using the soware package

PC-ORD 5.20 (McCune & Meord 2011). The full list of the

tested combinations is shown in Table 2.2.

We tested all three data matrix options (all layers, 3-layers,

1-layer) on the Mitchell Plateau and Pantijan vegetation

data matrices. Examples of selected OptimClass analyses

are presented in Figs 2.5 and 2.6. The information provided

by OptimClass inﬂuenced us to treat our data as follows:

1) We selected the 3-layer data matrix for the

classiﬁcation. The inclusion of more detailed vertical

layering, such as two to three sub-layers in the

tree layer and two sub-layers in the shrub layer

(see Appendix 2.1), is valuable in the description of

vegetation structure. It may, however, introduce ‘noise’

to the analysis by introducing variables with little

ecological information value. This may result in an

overzealous fragmentation of the ﬂoristic information.

Flattening the data by collapsing all structural layers

into one, however, provides information only on species

(not species per layer) presence and abundance.

Valuable, ecologically relevant information about

vegetation structure may be lost. The use of a three

layer data matrix is a suitable compromise between

these extremes.

2) We performed a log-transformation on the data

prior to clustering. This was necessary because the

vegetation projected cover measurements collected

in the ﬁeld were highly biased towards low cover

values. In most plots, more than ½ of the species were

scored as 1% cover. On the other hand, in most of

our sampled plots there were one or two pronounced

dominants (scoring high cover percentage estimates).

The log-transformation balances the % (scale 0–100)

data by recalculating them into smaller span, hence

underweighting the dominants slightly and giving

more weight to the low-score species. The use of the

log-transformation is hence useful when there is high

variation among attributes (estimated cover values of

the recorded species) within the sampled data set (see

for example McCune et al. 2002, McCune & Meord

2011 for more details).

for deﬁnitions) were not used in the analyses. Field

descriptions of these variables appeared biased

toward particular categories, suggesting survey team

members had diculty categorising them in the ﬁeld.

2) It was necessary to supplement the ﬁeld-collected

descriptions of geology prior to analyses. We used

a simpliﬁed system of rock types (see Table 5 in

Appendix 2.1), with surveyors designating the

sampling site as being on sandstone, laterite, basalt

or dolerite. It was apparent from an examination of the

ﬁeld datasheets, however, that this information was

only recorded accurately where outcrop was visible

at the sampling location. There was considerable

ambiguity in situations where deep soils covered the

parent rock. In order to correct this, the description of

geology collected in ﬁeld was cross-referenced against

the descriptions of habitat and soil texture, geological

maps and satellite imagery and amended where

required. Since vegetation on dolerite not sampled

during the 2011 ﬁeld trip, one plot from dolerite was

sampled during the 2012 veriﬁcation exercise and was

added to the modelling data set (see below).

Preparation of the Final Data Matrices

The vegetation data from each plot were entered into an

MS Excel spreadsheet and all taxonomy was checked for

currency. A MS Excel macro (developed by W/Prof. Phil

Withers, UWA, Perth) was then used to combine the plot

vegetation data into a single comprehensive vegetation

data matrix. The environmental and geographical data were

also entered in an MS Excel spreadsheet and the variables to

be used in numerical analyses (see below) were coded (for

details of the variable and the coding see Table 2.1).

The collated and corrected versions of the original plot

data are provided as Electronic Appendices on the CD

accompanying this report. The relevant ﬁles are Kimberley

Vegetation Plot Data and Kimberley Header Data. The

vegetation data matrices used in the classiﬁcation are also

provided on the accompanying CD. The relevant ﬁles are

VegClassMitchell and VegClassPantijan.

3.3.3 Classiﬁcation of the Field Plot Data

OptimClass: Search for the Optimal Classiﬁcation Tools

There are many classiﬁcation analytic options available

that represent dierent combinations of clustering

techniques, resemblance, and data transformation (see for

instance Legendre & Legendre 1998, Podani 2000, 2001,

McCune et al. 2002). We undertook an OptimClass analysis

to determine the most suitable approach to classiﬁcation

based on the available data. OptimClass (Tichý et al. 2010)

is a test of the classiﬁcation quality using the total count

of diagnostic species as the major criterion. Diagnostic

species can be deﬁned as a species that would characterise

a cluster (community or group of communities) in the

37CHAPTER 2 FIELD SURVEY AND PROTOCOLS

4) The choice of Similarity Ratio (quantitative form of the

well-known Jaccard’s Similarity Index; see for instance

Tamás et al. 2001) was motivated by its simplicity

and emphasis on species co-occurrence patterns.

Wishart’s similarity ratio has been widely used in/for

resemblance in numerical syntaxonomy (see Mucina

& van der Maarel 1989) and remains currently useful

since it has produced very informative (i.e. having

high interpretative value) results in combination

with common cluster techniques, particularly those

preserving the resemblance space (Hajdu 1981,

Hubálek 1982). Also, the OptimClass analyses

suggested such combinations to be highly informative.

Clustering, Deﬁnition of Clusters and Tabular Sorting

The log-transformed three-layer vegetation data matrices

for the Mitchell Plateau and Pantijan mapping areas were

clustered with UPGMA using the Similarity Ratio as the

resemblance. The ﬁnal clustering analyses were run in

the SYNTAX 2000 package (Podani 2001) because of the

favourable quality of the graphical outputs in this program.

The dendrograms showing the results of the clustering

analyses are shown in Figs 2.7 and 2.8.

3) We selected UPGMA as the clustering tool. The analysis

of the OptimClass results revealed that UPGMA, Beta

Flexible Clustering (beta: -0.25) and Ward’s Method

were nearly equally powerful in revealing classiﬁcation

structure in the analysed data sets. We chose UPGMA

because of its simple philosophy and wide use but

this choice was also governed by consideration

of the nature of the resemblance space distortion

imposed by some techniques. As succinctly captured

by the classic textbook by Sneath & Sokal (1973, p.

214–245), the dimensions of the resemblance space

become distorted as new resemblances are computed

between the growing clusters during the agglomeration

(amalgamation) process. Some of the clustering

techniques preserve the original resemblance space,

while other techniques either dilate or contract the

space (Lance & Williams 1967). In order to preserve the

original resemblances, we chose UPGMA because this

technique is known for its space-preservation property.

TABLE 2.1

List of environmental variables used in the CCA analyses as well as CART modelling.

Name Code Scale States Unit Source Note

Geology Geol Nominal/categorical San (sandstone)

Lat (laterite)

Bas (basalt)

Dol (dolerite)

1/0

ﬁeld data

ﬁeld data only in Pantijan data set

Altitude Alt Ratio n/a meter GPS device

Slope Slope Ratio n/a degree ﬁeld estimates

Aspect Asp Ratio/Circular n/a degree compas measurement

Disturbance Dist Nominal/categorical No-Dist

Graz

Fire

1/0

no obvious disturbance

grazing disturbance

light ﬁre disturbance

Annual Mean Temperature BC01 Ratio n/a ° C BioClim (modelled)

Mean Diurnal Range BC02 Ratio n/a ° C BioClim (modelled) mean of monthly (max temp - min temp)

Isothermality BC03 Ratio n/a %BioClim (modelled) (BIO 2/BIO7) *100

Temperature Seasonality BC04 Ratio n/a BioClim (modelled) (standard deviation * 100)

Max Temperature of Warmest Month BC05 Ratio n/a ° C BioClim (modelled)

Min Temperature of Warmest Month BC06 Ratio n/a ° C BioClim (modelled)

Temperature Annual Range BC07 Ratio n/a ° C BioClim (modelled) BIO5 - BIO6

Mean Temperature of Wettest Quarter BC08 Ratio n/a ° C BioClim (modelled)

Mean Temperature of Driest Quarter BC09 Ratio n/a ° C BioClim (modelled)

Mean Temperature of Warmest Quarter BC10 Ratio n/a ° C BioClim (modelled)

Mean Temperature of Coldest Quarter BC11 Ratio n/a ° C BioClim (modelled)

Annual Precipitation BC12 Ratio n/a mm BioClim (modelled)

Precipitation of Wettest Month BC13 Ratio n/a mm BioClim (modelled)

Precipitation of Driest Month BC14 Ratio n/a mm BioClim (modelled)

Precipitation Seasonality BC15 Ratio n/a mm BioClim (modelled) coecient of variation (CV)

Precipitation of Wettest Quarter BC16 Ratio n/a mm BioClim (modelled)

Precipitation of Driest Quarter BC17 Ratio n/a mm BioClim (modelled)

Precipitation of Warmest Quarter BC18 Ratio n/a mm BioClim (modelled)

Precipitation of Coldest Quarter BC19 Ratio n/a mm BioClim (modelled)

TABLE 2.2

List of OptimClass options.

Clustering Resemblance Transformation

Average linkage (UPGMA) Bray-Curtis none

Average linkage (UPGMA) Bray-Curtis log (2)

Average linkage (UPGMA) Bray-Curtis power 0.333

Average linkage (UPGMA) Bray-Curtis power 0

Average linkage (UPGMA) Jaccard none

Average linkage (UPGMA) Jaccard log (2)

Average linkage (UPGMA) Jaccard power 0.333

Average linkage (UPGMA) Jaccard power 0

Beta ﬂexible (β= -0.25) Bray-Curtis none

Beta ﬂexible (β= -0.25) Bray-Curtis log (2)

Beta ﬂexible (β= -0.25) Bray-Curtis power 0.333

Beta ﬂexible (β= -0.25) Bray-Curtis power 0

Beta ﬂexible (β= -0.25) Jaccard none

Beta ﬂexible (β= -0.25) Jaccard log (2)

Beta ﬂexible (β= -0.25) Jaccard power 0.333

Beta ﬂexible (β= -0.25) Jaccard power 0

Beta ﬂexible (β= -0.1) Bray-Curtis none

Beta ﬂexible (β= -0.1) Bray-Curtis log (2)

Beta ﬂexible (β= -0.1) Bray-Curtis power 0.333

Beta ﬂexible (β= -0.1) Bray-Curtis power 0

Beta ﬂexible (β= -0.1) Jaccard none

Beta ﬂexible (β= -0.1) Jaccard log (2)

Beta ﬂexible (β= -0.1) Jaccard power 0.333

Beta ﬂexible (β= -0.1) Jaccard power 0

Beta ﬂexible (β= -0.4) Bray-Curtis none

Beta ﬂexible (β= -0.4) Bray-Curtis log (2)

Beta ﬂexible (β= -0.4) Bray-Curtis power 0.333

Beta ﬂexible (β= -0.4) Bray-Curtis power 0

Beta ﬂexible (β= -0.4) Jaccard none

Beta ﬂexible (β= -0.4) Jaccard log (2)

Beta ﬂexible (β= -0.4) Jaccard power 0.333

Beta ﬂexible (β= -0.4) Jaccard power 0

Ward's method (= ISS) Chord (= normalised ED) none

Ward's method (= ISS) Chord (= normalised ED) log (2)

Ward's method (= ISS) Chord (= normalised ED) power 0.333

Ward's method (= ISS) Chord (= normalised ED) power 0

Ward's method (= ISS) Euclid none

Ward's method (= ISS) Euclid log (2)

Ward's method (= ISS) Euclid power 0.333

Ward's method (= ISS) Euclid power 0

Furthest neighbour (CLC) Bray-Curtis none

Furthest neighbour (CLC) Bray-Curtis log (2)

Furthest neighbour (CLC) Bray-Curtis power 0.333

Furthest neighbour (CLC) Bray-Curtis power 0

Furthest neighbour (CLC) Jaccard none

Furthest neighbour (CLC) Jaccard log (2)

Furthest neighbour (CLC) Jaccard power 0.333

Furthest neighbour (CLC) Jaccard power 0

39CHAPTER 2 FIELD SURVEY AND PROTOCOLS

to be straightforward except for the classiﬁcation of Plots

A108, A115 and B151. These were all sampled on basalt

and are indicated by the red square in the dendrogram (Fig.

2.5). The clustering process classiﬁed these three plots as

Community C (Terminalia canescens Comm.), despite the

occurrence of basalt indicators such as Eucalyptus tectiﬁca,

Corymbia greeniana, C. disjuncta, Hakea arborescens and

Grevillea mimosoides. This anomaly may be ascribed to

the dominance of Terminalia canescens in the three plots,

one of a few tree species showing apparent indierence to

substrate. The three anomalous plots were reclassiﬁed as

Community J (Eucalyptus tectiﬁca−Corymbia greeniana

Comm.). The ﬁnal non-hierarchical classiﬁcation vegetation

system of the Pantijan data is presented in Table 2.6.

3.3.4 Ordination of the Field Plot Data

Data Collation and Analytical Parameters

The Mitchell Plateau and Pantijan plot data were subjected

to separate ordination analyses with the aim of revealing

and quantifying the major ecological drivers of vegetation

patterns. We used Canonical Correlation Analysis (CCA:

ter Braak 1986) and the soware package CANOCO 4.5

(ter Braak & Šmilauer 2002) to achieve this. The choice

of CCA as the ordination model was motivated by a high

probability that species would show a unimodal rather than

a linear response along putative environmental gradients.

We used a simple CCA analysis with no transformation of

the percentage cover data prior to analysis, with scaling

focused on interspecies distances, biplot scaling applied

and no forward selection of environmental variables. Visual

representations of the ordination results were prepared

using the CanoDraw option of the CANOCO 4.5 soware

package. These show ordination planes of Axes 1 and 2.

The CCA analysis uses two data matrices. The ﬁrst matrix

is the vegetation data matrix featuring the occurrence

patterns of species in plots with percentage cover estimates

used as the original input values. The second matrix is

the environmental data matrix. This contains selected

environmental variables (scored in all studied plots)

expected to control the vegetation patterning. Both matrices

have the same number of plots, although the number of

variables (species variables and environmental variables)

may dier.

The CCA analysis of the Mitchell Plateau data set involved

355 species variables shared between 67 plots. The CCA

analysis of the Pantijan data set involved 300 species

variables shared by another set of 67 plots. A species

variable represents a taxon recorded in a particular vertical

vegetation layer; this implies that each taxon can occur in

the matrix up to three times if it occurs in each of the three

predeﬁned vertical vegetation layers considered (E3, E2, E1;

see Appendix 2.1 or Mucina et al. 2000 for the deﬁnition of

this layering system).

Aer clustering, the data were ordered. The procedure below

describes the ordering of the data matrix of the Mitchell

Plateau. An analogous procedure was applied to achieve the

ordering of the Pantijan matrix.

1) The plots of the original data matrix were ordered using

the new order of plots as suggested by the dendrogram

(Fig. 2.7) from le to right.

2) The ordered table was sequentially subdivided from the

top down using the amalgamation pattern depicted by

the dendrogram. The four outlying clusters (O1–O4)

were recognised ﬁrst, followed by clusters A and B (Fig.

2.7). The A and B clusters were subdivided into A1, A2,

B1 and B2.

3) The species limited to particular clusters were identiﬁed,

along with species groups shared by combinations of

clusters.

4) A preliminary interpretation of the clusters was

undertaken to assess the plausibility of the ecological

message conveyed by the recognised cluster versus

species group combinations.

5) The clusters A1, A2, B1 and B2 were further sequentially

subdivided by recognising the major gaps between

the sub-clusters. Steps 3 and 4 were repeated on the

subdivided clusters.

6) The process of subdividing clusters and repeating steps

3 and 4 continued until we reached the approximate

optimum number of clusters as suggested by the

UPGMA/SR/log algorithm. The resultant clusters were

recognised as ecologically interpretable and ﬂoristically

well deﬁned, each having their own diagnostic species

group.

Deﬁnition of the Final Vegetation Classiﬁcation System for

the Mitchell Plateau

OptimClass identiﬁed 22 clusters as the optimal solution

when applying the UPGMA/SR/log algorithm to the Mitchell

Plateau data. We iteratively applied the six-step sorting

algorithm described above to order the Mitchell Plateau

vegetation table and achieve this optimal result (Table

2.3). Plot A28 had been misclassiﬁed in the resultant

dendrogram as it had very low similarity to its allocated

Cluster (C+D) and should have been placed in Cluster E.

This misclassiﬁcation was ascribed to the inﬂuence of

Terminalia canescens in plot A28. This is a dominant species

of Clusters C and D and appeared dominant in plot A28 only

because the plot was extremely species-poor. The ﬁnal non-

hierarchical classiﬁcation vegetation system of the Mitchell

Plateau data is shown in Table 2.4.

Deﬁnition of the Final Vegetation Classiﬁcation System

for Pantijan

We used the same analytical classiﬁcation combination

(UPGMA/SR/log) and six-step algorithm to structure the

Pantijan vegetation table as used for the Mitchell Plateau

data (Table 2.5). The deﬁnition of communities appeared

TABLE 2.3

Phytosociological data of the Mitchell Plateau that served as the basis of the

vegetation classiﬁcation system for the region (see Table 2.4 on page 44).

Running number LH12345678910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

Field Code

A01

B69

A72

E63

A74

B50

A77

A73

B14

B61

E45

E08

B70

B71

A20

B99

A32

B60

A76

E44

B67

B13

B57

E19

B16

A28

A52

E58

B36

Community A A A A A A A A B B B B B B C C C C D D D D D D D E E E E

Corymbia nesophila E3 10 30 520 15 40 ...........2....5......

Corymbia nesophila E2 .1.10 3..1.........2...........

Corymbia nesophila E1 .1.10 .2.......................

Livistona eastonii E3 52515 1 . 10 5 . . 10 15 .............2.5.

Livistona eastonii E2 10 5 2 10 1.310 5 5 20 1.....2.....1.110 5 1

Livistona eastonii E1 2 . 1 3 3 . 5 1 . 1 1 1 . . . . . . . . . 1 . 1 . . . . 1

Gossypium rotundifolium E2 H......1......................

Gossypium rotundifolium E1 H1.1.....11...................

Merremia quinata E1 H......1..1..1................

Eucalyptus miniata E3 . . . 5 . . 1 . . 30 40 30 10 25 5 2 1 . . ...35.....

Eucalyptus miniata E2 . . . 3 . . . . 5 5 . 10 10 111.1...........

Eucalyptus miniata E1 5....23..1.2..1..........1...

Acacia oligoneura E2 ........1.1..................

Acacia oligoneura E1 ...........................2.

Eucalyptus tetrodonta E3 20 510 5 . . . . . 5 5 3 . . 20 20 10 325 30 510 530 25 5 5 . .

Eucalyptus tetrodonta E2 . . 1 10 . . . . 2 1 2 2 . . . 5 . 1 . 20 5 4 5 . 10 ....

Eucalyptus tetrodonta E1 . . . 3 . . . . 1 1 1 . . . . . 1 2 . 3 1 . 1 . . . . . .

Petalostigma pubescens E2 ....12........3.1....523.....

Petalostigma pubescens E1 .1..............1....1.......

Tricoryne sp. Kimberley (K.F.

Kenneally 4857) E1 H..1.1...........1............

Hibbertia hooglandii E1 1.............20 50 10 25 .1....1....

Bossiaea bossiaeoides E2 ..............510 2............

Bossiaea bossiaeoides E1 ...............21............

Acacia areolata E2 ..............2.5.1..........

Acacia areolata E1 ................1............

Stenocarpus acacioides E2 ...............11............

Stenocarpus acacioides E1 ................2............

Bossiaea sp. West Kimberley

(R.L. Barrett & M.D. Barrett RLB

4045)

E2 ..................15 20 1........

Bossiaea sp. West Kimberley

(R.L. Barrett & M.D. Barrett RLB

4045)

E1 ..................1..........

Corymbia latifolia E3 .........................320 20 5

Corymbia latifolia E2 ..................1.2.......1

Corymbia latifolia E1 .....1....................111

Eucalyptus apodophylla E3 .............................

Eucalyptus alba var.

australasica E3 .............................

Eucalyptus alba var.

australasica E1 .............................

Eriocaulon tortuosum E1 H . . . . .........................

Melaleuca leucadendra E1 .............................

Melaleuca nervosa subsp.

nervosa E2 .............................

Melaleuca nervosa subsp.

nervosa E1 .............................

Acacia retinervis E2 ...............4...12........

Acacia retinervis E1 .............................

Gonocarpus leptothecus E1 ...............1...1.........

Terminalia carpentariae E3 .............................

Terminalia carpentariae E2 .............................

Acacia gonocarpa E2 .............................

Corymbia torta E3 .............................

Corymbia torta E2 .............................

Corymbia torta E1 .............................

Acacia sp. Kununurra (Lullﬁtz

6195) E2 .............................

Acacia delibrata E3 .............................

Acacia delibrata E2 .............................

Tephrosia rosea E2 .............................

Tephrosia rosea E1 .............................

41CHAPTER 2 FIELD SURVEY AND PROTOCOLS

Each column in the main body of the table represents a relevé (plot) and has the ﬁeld code of the plot at the point of sampling.

The codes in Column L indicate the vertical layers (E3: tree layer, E2: shrub layer, E1: ground layer). The code “H” in Column H indicates

which of the taxa are herbs (all others are either trees, shrubs, saplings or woody seedlings). The species/taxa (per layer) are grouped

in such a manner as to delimit a community or groups of communities. Only diagnostic species are shown (the species with low or

no diagnostic value were omitted). Codes for the Communities match those in Table 2.4. The colour coding of the background of the

Community Codes matches the colour coding of major mapping units in the maps (Green: Sandy Woodlands; Yellow: Clayey Woodlands;

Grey: Sandy Shrublands). See the text for detail of table sorting.

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

E27

A02

Z11

A56

A55

B86

B97

B33

A62

E41

B18

E12

A06

E47

E07

E40

A79

A04

A54

A30

A22

B35

B31

A03

B15

B30

E42

A29

B17

E06

A05

A75

B65

B68

E10

B64

E43

B66

E F F G H H H H I I J J K L M M M N O O O O O P P P P R R R S T T T U V V V

......................................

..............1.......................

.5..........1..........31.............

. . 1 . . . . . . . 1 5 . . 2 5 . 20 .1....1.............

.1.........1..21.......1....11........

......................................

.................1....................

.......................1..............

......................................

......1...............................

......3........10 ......................

........1.............................

......221.............................

......1...............................

......................................

......1...............................

......................................

......1...............................

......................................

..............................70 .......

..............................1.......

20 5....................................

5....1................................

15 ....1................................

.10 20 ...................................

..5...................................

..1...................................

. . 10 ...................................

..1...................................

.1....................................

.31...................................

2 . . . 10 875......................1.......

....111...............................

....10 .................................

....1..1...1..........................

........10 1.........................1..

......3.510 ..................7.........

......1..10 ............................

.......2.1..................1.........

.........5............................

.........5...... ......................

.......1.1............................

........21............................

........1.............................

TABLE 2.3 Continued

Running number LH12345678910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

Field Code

A01

B69

A72

E63

A74

B50

A77

A73

B14

B61

E45

E08

B70

B71

A20

B99

A32

B60

A76

E44

B67

B13

B57

E19

B16

A28

A52

E58

B36

Community A A A A A A A A B B B B B B C C C C D D D D D D D E E E E

Pandanus spiralis E2 .............................

Pandanus spiralis E1 .............................

Antidesma ghaesembilla E2 .......1...............1.....

Antidesma ghaesembilla E1 .............................

Corymbia bella E2 .............................

Acacia stigmatophylla E2 .............................

Xanthostemon eucalyptoides E3 . ............................

Corymbia bella E3 .............5...............

Melaleuca viridiﬂora E3 .............................

Melaleuca viridiﬂora E2 ....1........................

Melaleuca viridiﬂora E1 ....2........................

Erythrophleum chlorostachys E3 3. ..1..1.....................

Erythrophleum chlorostachys E2 1 . . 20 1 1 . 5 . . . . 2 . 1 . . . . . . . . 2 10 ....

Erythrophleum chlorostachys E1 ...1.................1...3...

Sauropus torridus E1 H . . . . . 1 1 . 1 . 2 1 1 . . . . 1 . . . . . . 1 . . . .

Eucalyptus bigalerita E3 .............................

Eucalyptus bigalerita E2 .............................

Eucalyptus bigalerita E1 .............................

Lophostemon grandiﬂorus

subsp. riparius E2 .............................

Eucalyptus tectiﬁca E3 .........2..............5....

Eucalyptus tectiﬁca E2 ...1.....1...................

Eucalyptus tectiﬁca E1 .............................

Corymbia disjuncta E3 .................3.......3...

Corymbia disjuncta E2 ........................21...

Corymbia disjuncta E1 .............................

Corymbia greeniana E3 ........................1....

Corymbia greeniana E2 ............21...............

Corymbia greeniana E1 .............1...............

Dendrolobium cheelii E1 .............................

Gossypium pilosum E1 ......................1......

Terminalia ﬁtzgeraldii E3 .............................

Terminalia ﬁtzgeraldii E2 .............................

Terminalia ﬁtzgeraldii E1 .............................

Calytrix exstipulata E2 ...........................1.

Calytrix exstipulata E1 .............................

Calytrix achaeta E2 ..............15....5......1.

Calytrix achaeta E1 ...............11..........2.

Acacia kelleri E2 .............................

Grevillea agrifolia subsp.

microcarpa E2 .............................

Calytrix brownii E2 .............................

Calytrix brownii E1 .............................

Acacia arida E2 .............................

Terminalia canescens E3 .............................

Terminalia canescens E2 .2....1.....2.....110 3 . . 2 1 1 1 . 1

Terminalia canescens E1 ....................5........

Buchanania obovata E3 .........................1...

Buchanania obovata E2 . . . 1 . 1 . . . . . . . . 3 . 1 . . 2 2 . . 1 1 1 . . .

Buchanania obovata E1 ......1.......1.1....1.1.....

Grewia retusifolia E2 .............1.........21....

Grewia retusifolia E1 ....11.......2...1......1....

Hakea arborescens E3 .............................

Hakea arborescens E2 .......................1.2...

Hakea arborescens E1 .......................1.....

Grevillea mimosoides E2 ...1.............1.....2.1...

Grevillea mimosoides E1 ........1....................

Planchonia careya E3 .......................3.....

Planchonia careya E2 . . . 1 . . . 1 . . . . 1 . 1 1 . 1 . . . . . 1 . . . . .

Planchonia careya E1 .1.1..1.1.......1............

Gardenia resinosa subsp.

kimberleyensis E3 .............................

Gardenia resinosa subsp.

kimberleyensis E2 ...............1....1........

Gardenia resinosa subsp.

kimberleyensis E1 ................1............

Psydrax pendulina E2 1...... ...........1..........

Psydrax pendulina E1 .11.....1.....2.......1......

Phyllanthus aridus E1 ..1..1...........1...1......1

Dolichandrone heterophylla E2 ..1..........................

Dolichandrone heterophylla E1 ..1....1.........1......1....

Ficus aculeata E2 .......1...............2.1...

Ficus aculeata E1 ......1......................

Passiﬂora foetida var. hispida E3 .............5...............

Passiﬂora foetida var. hispida E2 .............2.........1.....

Passiﬂora foetida var. hispida E1 .............2.........221...

Vachellia valida E1 .............................

Vachellia valida E2 ........................23...

43CHAPTER 2 FIELD SURVEY AND PROTOCOLS

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

E27

A02

Z11

A56

A55

B86

B97

B33

A62

E41

B18

E12

A06

E47

E07

E40

A79

A04

A54

A30

A22

B35

B31

A03

B15

B30

E42

A29

B17

E06

A05

A75

B65

B68

E10

B64

E43

B66

E F F G H H H H I I J J K L M M M N O O O O O P P P P R R R S T T T U V V V

..........10 510 2........................

............11........................

..........15.10 1 1 . 1 . . . 1 . . . 1 . 1 1 1 . . . . . . . .

...........2..1.......................

..........11.1........................

............15 .........................

............5.........................

.............20 ........................

.............50 ........................

..........2.110 ........................

.............1........................

..............25 20 5....1........3.......

....2.........515 5....1................

...............13.....................

................1.....................

...............10 .20 ....................

...............1.1....................

.................1....................

. . ...............10 ....................

................5.520 10 40 2...............

. . . . . . . . . . 1 . . . 2 1 1 . 3 . . . 2 1 1 5 20 2..........

........................1.............

...................1......553.........

..........2...12.1.1...10 215 1...........

...1..........1..1.......5..2.........

....................10 5................

...1............5.1.2211.10 5..5........

..........................1...........

.................1.... . . 2 5 5 . 1 2 . . . . . . . .

......................131.............

...........................10 .5........

...........1.4...2.........3520 ........

...........1................215 ........

....1..1.......................551....

.................................1....

......................2........2.1....

.................................1....

..................................45 ...

....5.............................15 ...

....................................54

...................................2.1

...................................520 .

..................1.....1.............

...10 151.1.1...5......22.1.............

....13..1.....4................1......

..............................5.......

1..2....1......1...111................

1..1......................1...........

.............1.21..1.......3..........

..........12..11....2..1. 2 3 2 1 2 . . . . . . . .

........................1.............

..............1.1.....5...1...........

................1.....112.............

..............1...11.11......1........

..............1......11......1........

...................1.......1..........

..........11.....1......1...12........

1.1...........1..1....................

......................2...............

1...............3.....2.2.............

......1.........1.....................

......................................

...1...........11...11..1....1........

...1......1...1.1....1................

........................1.............

......................................

...1................12...1.11.........

................1....2....2...........

................1....21..13...........

TABLE 2.4

Vegetation system for the Mitchell Plateau as interpreted on the basis of numerical

classiﬁcation.

A: Corymbia nesophila−Livistona eastonii Comm.

B: Eucalyptus miniata−Livistona eastonii Comm.

C: Eucalyptus tetrodonta−Hibbertia hooglandii Comm.

D: Eucalyptus tetrodonta−Bossiaea sp. West Kimberley Comm.

E: Corymbia latifolia Comm.

F: Eucalyptus apodophylla−Eucalyptus alba var. australasica Comm.

G: Terminalia canescens Comm. (NOT MODELLED)

H: Acacia retinervis−Gonocarpus leptothecus Comm.

I: Acacia gonocarpa−Corymbia torta Comm.

J: Pandanus spiralis−Melaleuca nervosa Comm. (NOT MODELLED)

K: Pandanus spiralis−Acacia stigmatophylla Comm. (NOT MODELLED)

L: Melaleuca viridiﬂora−Antidesma ghaesembilla Comm. (NOT MODELLED)

M: Erythrophleum chlorostachys−Livistona eastonii Comm.

N: Eucalyptus bigalerita Comm.

O: Eucalyptus tectiﬁca−Corymbia greeniana Comm.

P: Eucalyptus tectiﬁca−Corymbia disjuncta Comm.

R: Terminalia ﬁtzgeraldii Comm.

S: Bossiaea sp. West Kimberley−Corymbia polycarpa Comm.

T: Calytrix exstipulata−Calytrix achaeta Comm.

U: Acacia kelleri−Grevillea agrifolia Comm.

V: Acacia arida−Calytrix brownii Comm.

45CHAPTER 2 FIELD SURVEY AND PROTOCOLS

TABLE 2.5

Phytosociological data of the Pantijan area that served as the basis of the vegetation classiﬁcation system for the

region (see Table 2.6 on page 50). For other legends and explanations of the codes, see the header of Table 2.3.

Running number LH12345678910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

Field Code

A37

A114

A81

A109

A86

B89

C88

B90

C85

C255

A101

B158

C25

B159

A108

A115

B151

B152

B38

C258

A102

C84

A110

A111

B160

C259

B165

C82

C93

Community A A B B B B B B B B C C C C J J J C D D E E E E E E E E E

Melaleuca leucadendra E3 10 70 ...........................

Pandanus cf. darwinensis E2 10 20 ...........................

Pandanus cf. darwinensis E1 .1...........................

Timonius timon E3 5............................

Timonius timon E2 10 ............................

Acacia plectocarpa subsp.

plectocarpa E3 .............................

Acacia plectocarpa subsp.

plectocarpa E2 11...........................

Corymbia torta E3 . . 2 3 10 5 5 20 15 10 .3.................

Corymbia torta E2 . . . . 1 1 . . 1 . 1 1 3 1 . . . . . 1 . . . . 1 . . . 1

Corymbia torta E1 ..111.....1...........1......

Owenia vernicosa E3 . . . 1 2 2 10 . . 10 .................5.

Owenia vernicosa E2 ...1.1...1.........1......1..

Owenia vernicosa E1 ...11....1..................1

Planchonella arnhemica E3 ......32.....................

Planchonella arnhemica E2 ..11.1.....................1.

Tephrosia rosea E1 ......1.1.1......1...........

Terminalia canescens E3 ........5......2...5.......55

Terminalia canescens E2 . . 2 1 . 1 1 1 10 . 5535510 5 2 10 4 . . . . . . 1 . 1

Terminalia canescens E1 . . 1 1 . . 1 1 . . 1 . 1 2 2 . . . . . . . . . . . . 1 .

Acacia gonocarpa E2 . . 2 . . . . 1 . . . 2 15 .....10 ..........

Acacia gonocarpa E1 ..1...............1..........

Canarium australianum var.

glabrum E3 .........10 ...................

Acacia platycarpa E3 .........5...................

Acacia platycarpa E2 .........3..................1

Acacia platycarpa E1 ...1.....1...................

Calytrix exstipulata E2 ..........52.................

Calytrix exstipulata E1 ..........1..................

Acacia dunnii E2 ..........5..................

Terminalia carpentariae E3 .......2.5........10 10 .........

Terminalia carpentariae E2 ...21........................

Vitex glabrata E2 .............1....10 ..........

Eucalyptus brachyandra E3 ...................10 .........

Eucalyptus miniata E3 .....25......1......10 20 20 25 55519 5

Eucalyptus miniata E2 ....................1........

Eucalyptus miniata E1 ....................11...2...

Petalostigma pubescens E3 .........................10 ...

Petalostigma pubescens E2 ....22............10 . 531112252

Petalostigma pubescens E1 ....................1.11.2...

Bossiaea bossiaeoides E2 ..........2.........52......3

Bossiaea bossiaeoides E1 ....................13.1.....

Acacia nuperrima E2 ....................2....3...

Acacia nuperrima E1 ....................32.1.....

Corymbia polycarpa E3 .......................2.....

Corymbia polycarpa E2 .......................11....

Corymbia dichromophloia or

Corymbia polycarpa E3 ...........................10 .

Callitris columellaris E3 .............2...............

Callitris columellaris E1 ...1.........................

Eucalyptus tetrodonta E3 .............1...............

Eucalyptus tetrodonta E2 .............1...............

Eucalyptus tetrodonta E1 .............................

Buchanania obovata E3 .....................5.......

Buchanania obovata E2 . . 251 . 1 . 112112 . . . . . . 551251121

Buchanania obovata E1 . . . 2 1 . 1 1 1 . . . . . . . . . 1 . 2 . 2 2 . . . 1 1

Acacia dissimilis E2 .......................1.....

Acacia dissimilis E1 .............................

Acacia arida E2 ........2....................

Acacia arida E1 .............................

Acacia delibrata E3 .............................

Acacia delibrata E2 .........3..............1..1.

Acacia delibrata E1 .............................

Celtis philippensis E2 .............................

Capparis jacobsii E1 .............................

Bombax ceiba E2 .............................

47CHAPTER 2 FIELD SURVEY AND PROTOCOLS

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

C34

B150

B162

B96

A116

A100

A113

C83

A87

C257

A82

C24

A104

A91

A105

B154

A103

B78

C251

B164

C253

C254

A106

C252

B39

B155

C94

C250

A92

B153

A107

B156

B157

C80

A112

B161

B163

B95

E E E E F G G G H H I I I J J J J J J J J J J J J J J J J J K K K K L L L M

......................................

2..2.................................1

1.......1.............................

......................................

1..1..................................

......................................

.......5...................5..........

...1...1.............10 ..1..1..........

. . .....1.............2................

......................................

........1.............................

......................................

.......1..............................

......................................

5.....................................

1.1...................................

......................................

10 252..................................

1.....................................

1..2..................................

......................................

.10 21....1.............................

......................................

...10 ................................2.

. . . 1 ................................1.

......................................

.15 5.................................5.

......................................

....30 .................................

....1.................................

..5.2.................................

....5.................................

....5................... ..............

...................................1..

221.22..1..........................11.

......................................

....1.................................

..2.1.................................

.....325 20 ..............................

.....1................................

........5.............................

........55............................

........11............................

........1.............................

........2.............................

........1.............................

TABLE 2.5 Continued

Running number LH12345678910 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

Field Code

A37

A114

A81

A109

A86

B89

C88

B90

C85

C255

A101

B158

C25

B159

A108

A115

B151

B152

B38

C258

A102

C84

A110

A111

B160

C259

B165

C82

C93

Community A A B B B B B B B B C C C C J J J C D D E E E E E E E E E

Brachychiton viridiﬂorus E2 ...........1.......2....1....

Ficus virens E2 .............................

Flagellaria indica E2 H...................1.........

Flagellaria indica E1 H............................1

Exocarpos latifolius E2 .............1...............

Hypoestes ﬂoribunda var. a.

suaveolens E2 .............................

Hypoestes ﬂoribunda var. a.

suaveolens E1 .............................

Plectranthus congestus E2 .............................

Ficus brachypoda E3 .............................

Ficus brachypoda E2 .............................

Eucalyptus bigalerita E3 .............................

Eucalyptus bigalerita E2 ................1............

Eucalyptus bigalerita E1 .............................

Eucalyptus tectiﬁca E3 .........................32..

Eucalyptus tectiﬁca E2 ..............12..........1..

Eucalyptus tectiﬁca E1 ...............5.............

Corymbia greeniana E3 ................1............

Corymbia greeniana E2 ................1............

Corymbia greeniana E1 ................1............

Corymbia disjuncta E3 ...............1.............

Corymbia disjuncta E2 ..............32.............

Corymbia disjuncta E1 .............................

Grevillea mimosoides E2 ................1. ...1....1..

Grevillea mimosoides E1 .............................

Hakea arborescens E3 .............................

Hakea arborescens E2 ..............23.............

Hakea arborescens E1 ..............12.............

Cullen badocanum E2 .. ...........................

Cullen badocanum E1 .. ...........................

Decaschistia occidentalis E2 .............................

Decaschistia occidentalis E1 .........................1...

Dendrolobium cheelii E1 .............................

Corymbia cf. latifolia E2 .............................

Corymbia cf. latifolia E1 .............................

Terminalia ﬁtzgeraldii E3 .............................

Terminalia ﬁtzgeraldii E2 ............... ..............

Terminalia ﬁtzgeraldii E1 .............................

Corymbia bella E3 .............................

Vachellia suberosa E2 .............................

Vachellia suberosa E1 .............................

Banksia dentata E3 .............................

Banksia dentata E2 .............................

Banksia dentata E1 .............................

Grevillea pteridifolia E3 .............................

Grevillea pteridifolia E2 .............................

Grevillea pteridifolia E1 .............................

Eucalyptus alba var.

australasica E3 .............................

Eucalyptus alba var.

australasica E2 .............................

Eucalyptus alba var.

australasica E1 .............................

Melaleuca viridiﬂora E3 .............................

Melaleuca viridiﬂora E2 .............................

Eucalyptus houseana E3 .............................

Eucalyptus houseana E1 .............................

Pandanus spiralis E2 1....................1.......

Pandanus spiralis E1 .............................

Osbeckia australiana E2 .............................

Osbeckia australiana E1 .............................

Antidesma ghaesembilla E3 .............................

Alphitonia excelsa E3 .............................

Alphitonia excelsa E2 .............................

Myrtaceae E2 .............................

Cajanus hirtopilosus E2 . . . . . . .......................

Cajanus hirtopilosus E1 .............................

Carallia brachiata E2 .............................

Corymbia ptychocarpa subsp.

ptychocarpa or Corymbia

polycarpa

E2 .............................

Lygodium microphyllum E2 H.............................

Lygodium microphyllum E1 H.............................

Terminalia hadleyana or

Terminalia ferdinandiana E3 .............................

49CHAPTER 2 FIELD SURVEY AND PROTOCOLS

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

C34

B150

B162

B96

A116

A100

A113

C83

A87

C257

A82

C24

A104

A91

A105

B154

A103

B78

C251

B164

C253

C254

A106

C252

B39

B155

C94

C250

A92

B153

A107

B156

B157

C80

A112

B161

B163

B95

E E E E F G G G H H I I I J J J J J J J J J J J J J J J J J K K K K L L L M

........1.............................

........2.............................

......................................

........1.............................

........5.............................

.........70 ............................

.........15 ............................

.........5............................

..........50 20 5.........................

..........2.1.........................

..........1...........................

.............2310 5 2 10 10 20 15 515 5 5 10 10 1.........

. . . . . . . . . . 1 3 . 7 5 2 1 2 5 1 1 1 1 1 . 1 . . 1 3 1 . 2 . . . . .

...............1515.3153..1...........

..............1...5210 5......1.........

.............1551.11.1.11..11.........

. . ............11.....1.1.1............

.................1......10 2............

................11....1..1...1......2.

...............111....1...............

................211...1....1..........

................211...1.....21........

............................1.........

...............1...1111.2.3.1.........

..............11....1..1....1.........

.............1..1.......1.5.1.........

................1..... ......11........

............................1.........

....................1..1..............

..............1.......35.2............

....................5.................

....................1.................

..........1......................5....

..........1...........53......5555....

.......................1......1.......

...................2..........55......

...................2..........2.1.....

..............................1.1.....

...................................22.

.5................................60 2 1 .

..................................5...

.1................................1.1.

..................................255.

..................................1...

...................................10 . .

..................................1...

..................................2...

..................................1.10 .

...................................2..

.5........1.........................10 15

...................................1..

..................................15215

.....................................2

....................................1.

.....................................1

.....................................20

.....................................4

.....................................10

.....................................5

.....................................3

.....................................5

.....................................10

.....................................1

.....................................10

TABLE 2.6

Vegetation system for Pantijan as interpreted on the basis of numerical classiﬁcation

and table sorting analyses.

A: Melaleuca leucandendra Comm. (NOT MODELLED)

B: Corymbia torta−Terminalia canescens Comm.

C: Terminalia canescens Comm.

D: Terminalia carpentariae Comm.

E: Eucalyptus miniata−Petalostigma pubescens Comm.

F: Callitris columellaris Comm. (NOT MODELLED)

G: Acacia arida Comm.

H: Acacia delibrata−Ficus brachypoda Comm.

I: Eucalyptus bigalerita Comm.

J: Eucalyptus tectiﬁca−Corymbia greeniana Comm.

K: Terminalia ﬁtzgeraldii−Corymbia bella Comm.

L: Banksia dentata−Grevillea pteridifolia Comm.

M: Alphitonia excelsa Comm. (NOT MODELLED)

51CHAPTER 2 FIELD SURVEY AND PROTOCOLS

FIG. 2.9

Ordination plot (Axes 1 and 2) of the canonical correspondence analysis for Mitchell Plateau sample plots. The red

circles indicate the position of the plots and the blue arrows indicate the axes deﬁned by the environmental data. The

nominal variables (all except the climatic data, aspect and slope) are also shown as vectors (arrows) to allow quicker

orientation. Actually the nominal variables that show a large spread of positions of the particular states (position is

deﬁned by the tip of the arrow) are supposed to have a stronger inﬂuence on the vegetation patterning than those

located close to each other.

53CHAPTER 2 FIELD SURVEY AND PROTOCOLS

FIG. 2.10

Ordination plot (Axes 1

and 2) of the canonical

correspondence analysis

for Pantijan sample plots.

For other explanations see

caption of Fig. 2.9.

Disturbance is less informative. Most of the plots scored the

state Non-Dist indicating that no disturbance was detected

in the plot. The states Fire and Graz were shown to play a

role only in the basalt-dominated landscapes. Slope and

Aspect do not play a major role in structuring the data.

Direct Gradient Analysis of the Pantijan Data: Interpretation

The results of the CCA ordination of the Pantijan data are

shown as a biplot in Fig. 2.10. The combination of the CCA

Axes 1 and 2 explains 7.6% and 17.1% of the variance of

species data and species-environment relations respectively.

The same environmental factors were used in the canonical

correspondence analyses of the Pantijan data as in the

Mitchell Plateau data set. The nominal variable Geology

consists of states San, Bas and Dol.

Three plots (39, 64 and 65) were identiﬁed as clear ﬂoristic

outliers. Plot 39 is classiﬁed as Plectranthus congestus-

dominated sandstone woodland (Community H), while

the other two plots are representatives of Community L

(seasonally wet sandstone woodlands). Plot 34 is also a

slight outlier as it represents ecologically odd, ﬁre-protected

Callitris columellaris woodland on dolerite. The BioClim

variables again show a contracting grouping pattern, with

one group characterising high-altitude, high precipitation

sandstone landscapes and the other being characteristic

of low-altitude, warmer basalt-dominated landscapes.

Geology again plays an important role as a driver and data-

structuring variable (see the broad span of the states of the

variables on the CCA ordination plane deﬁned by Axes 1 and

2 in Fig. 2.10).

The environmental data sets used in the CCA analyses for

the Mitchell Plateau and Pantijan are shown in Appendices

2.2 and 2.3 respectively. A list of the environmental

parameters, including the measurement scale and the unit

of the measurement, is given in Table 2.1.

Direct Gradient Analysis of the Mitchell Plateau Data:

Interpretation

The results of the CCA ordination of the Mitchell Plateau

data are given in Fig. 2.9 as a biplot showing the position

of all 67 plots and the visualised importance of the studied

environmental factors. The combination of the CCA Axes

1 and 2 explains 5.9% and 16% of the variance of species

data and species-environment relations respectively.

Two of the environmental factors are nominal variables:

Geology, containing states San, Lat, Bas and Disturbance,

containing states Graz, Fire, No-Dist. The remaining

variables are in ratio scale. The climatic (BioClim) variables

show an interesting polarised pattern. Two distinct groups

of variables emerge with one indicating high precipitation

for the vegetation on the elevated laterite plateau. The

ratio variable ‘Alt’ also groups with this ‘high-precipitation’

group of variables. The other group of BioClim variables is

informative of higher temperature being characteristic of

parts of the low-altitude basalt woodlands and outlying

sandstone shrublands (Plots 53, 63–66; Fig. 2.9). Geology

appears to play an important role in structuring the data.

All three character states (connected by the pink triangle

in Fig. 2.9) are remote and show distance from the centre

of the coordinate system, indicating their exclusive role in

separate parts of the ordination space. The nominal variable

regarding predictor variable or target class modiﬁcation

to be implemented at any stage of the process. For these

reasons, CART analysis was selected as the most suitable

classiﬁcation methodology for this project.

In practicality, a CART analysis models a target

variable according to a set of predictor variables. In this

project, this meant establishing which environmental

variables (predictors) were of highest importance when

determining splitting threshold values for vegetation

type (target) classiﬁcation, and how these ‘splitters’

should be structured in a logical sequence. An alternative

interpretation would be that the CART analysis determined

the optimal set of rules for vegetation classiﬁcation.

CART 6.0 (Steinberg & Golovnya 2006), the soware used

in this project, is based on the original CART algorithm

developed by Breiman et al. (1984). In order to determine

an optimal decision tree, the soware runs thousands

of random splitting iterations resulting in hundreds of

dierent trees of varying classiﬁcation accuracy. The ﬁnal

return is a proﬁle of the ‘best’ trees, determined by the

number of nodes (tree size) against the calculated relative

cost of that tree (total classiﬁcation error divided by the

number of nodes). Relative cost is scaled between 0 and 1,

where a value of 0 would indicate no error and a value of 1

would represent the performance of random guessing. Fig.

2.11 gives an example of such a proﬁle, where each square

represents one tree, plotted according to relative cost and

tree size. The 10-node tree highlighted by the horizontal

line is considered the optimal tree, as the relative cost is

lowest of all of the trees displayed.

However, since the resulting trees determined through CART

analysis are not deﬁned according to a stopping criterion,

the tree with the lowest relative cost will not always model

every class of the target variable. In such cases, the optimal

tree can be ‘grown’ (increasing the number of branches) in

the hope that the CART algorithm will ﬁnd a splitter for a

3.4 Step 4: Setting Mapping Rules and Modelling

3.4.1 Classiﬁcation and Regression Trees

Over the past decade-and-a-half, the use of classiﬁcation

trees has become steadily more prevalent for autonomously

creating rulesets (Lawrence & Wright 2001). Classiﬁcation

trees, also known as ‘decision trees’ (or DTs), are constructed

by recursive division of training data into increasingly

homogeneous subsets. A branch or split in the classiﬁcation

tree is a threshold value for the image feature that produces

the most deviance in the data set. Subsequent subsets are

subject to further division, perhaps using a dierent feature

showing high heterogeneity, until either a pre-set variance or

classiﬁcation tree level is reached. The result is a hierarchical

ruleset used for digital image classiﬁcation (Hansen et al.

1996, Lawrence & Wright 2001).

Classiﬁcation and Regression Trees (CART; see Breiman

et al. 1984, Fielding 2007 for detailed descriptions of

the technique and Bayes & Mackey 1991, Michaelsen

et al. 1994, Cains 2001, Miller & Franklin 2002 for selected

vegetation-related reviews and applications) is a robust

decision-tree tool for data mining, predictive modelling and

data processing. It can be used to generate accurate and

reliable predictive models for a wide range of applications

using both discrete and continuous data sets. CART diers

from other decision-tree methodologies in several aspects.

Firstly, the iterative data splitting (recursive division)

is binary rather than multi-way and this avoids data

fragmentation and a consequent accompanying decrease

in pattern detection. Secondly, the size of a decision tree

developed by CART is not determined by a stopping rule,

e.g. a given degree of homogeneity on the terminal nodes.

Rather, the tree is over-developed, so as not to miss any

structure in the data, and then pruned as necessary.

Thirdly, and of particular importance for our study, CART

implements cross validation self-testing, resulting in more

robust and transferable trees when utilizing smaller data

sets (Steinberg & Golovnya 2006). Finally, unlike several

other classiﬁcation methods, the CART process is both

transparent and interpretable, allowing for expert decisions

FIG. 2.11

An example of the relative error curve produced by CART modelling.

0.70

0.60

0.50

0.40

0 10 20 30 40 50 60 70

Number of Nodes

Relative Cost

0.488

55CHAPTER 2 FIELD SURVEY AND PROTOCOLS

FIG. 2.12

Final CART diagram for the Mitchell Plateau zonal vegetation units (detailed classiﬁcation). The alpha-numerical

(and colour) coding of the classes is as follows: A, B, C (incl. D), E, F, H, I, M (incl. N), O (incl. P), R, S (incl. T, U, V).

(Communities G, J, K and L were not modelled.) Insert: Simpliﬁed CART classiﬁcation tree showing the hierarchy of

classiﬁers (splitters).

57CHAPTER 2 FIELD SURVEY AND PROTOCOLS

FIG. 2.13

The mapping rule set for the Mitchell Plateau zonal vegetation data set.

/*Terminal Node 1*/

GEOLOGY$ == Bas ||

GEOLOGY$ == Dol

PC6 <= -3.87293

{

terminalNode = 1;

class = R;

}

/*Terminal Node 2*/

GEOLOGY$ == Bas ||

GEOLOGY$ == Dol

PC6 > -3.87293 &&

PC4 <= 6.54848 &&

PC2 <= -13.3268

{

terminalNode = 2;

class = M;

}

/*Terminal Node 3*/

GEOLOGY$ == Bas ||

GEOLOGY$ == Dol

PC6 > -3.87293 &&

PC4 <= 6.54848 &&

PC2 > -13.3268 &&

B7 <= 62.5

{

terminalNode = 3;

class = O;

}

/*Terminal Node 4*/

GEOLOGY$ == Bas ||

GEOLOGY$ == Dol

PC6 > -3.87293 &&

PC4 <= 6.54848 &&

PC2 > -13.3268 &&

B7 > 62.5

{

terminalNode = 4;

class = M;

}

/*Terminal Node 5*/

GEOLOGY$ == Bas ||

GEOLOGY$ == Dol

PC6 > -3.87293 &&

PC4 > 6.54848

{

terminalNode = 5;

class = R;

}

/*Terminal Node 6*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 <= 15.5 &&

B2 <= 28 &&

NDWI <= 0.371848 &&

B5 <= 34.5

{

terminalNode = 6;

class = E;

}

/*Terminal Node 7*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 <= 15.5 &&

B2 <= 28 &&

NDWI <= 0.371848 &&

B5 > 34.5

{

terminalNode = 7;

class = C;

}

/*Terminal Node 8*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 <= 15.5 &&

B2 <= 28 &&

NDWI > 0.371848 &&

NDWI <= 0.410774

{

terminalNode = 8;

class = A;

}

/*Terminal Node 9*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 <= 15.5 &&

B2 <= 28 &&

NDWI > 0.410774 &&

PC4 <= 0.661713 &&

PC5 <= 0.875853

{

terminalNode = 9;

class = F;

}

/*Terminal Node 10*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 <= 15.5 &&

B2 <= 28 &&

NDWI > 0.410774 &&

PC4 <= 0.661713 &&

PC5 > 0.875853

{

terminalNode = 10;

class = B;

}

/*Terminal Node 11*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 <= 15.5 &&

B2 <= 28 &&

NDWI > 0.410774 &&

PC4 > 0.661713

{

terminalNode = 11;

class = B;

}

/*Terminal Node 12*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 <= 15.5 &&

B2 > 28

{

terminalNode = 12;

class = H;

}

/*Terminal Node 13*/

GEOLOGY$ == Lat ||

GEOLOGY$ == Sand

B1 > 15.5

{

terminalNode = 13;

class = I;

}

59CHAPTER 2 FIELD SURVEY AND PROTOCOLS

4) The communities of the sandstone and laterite shrublands

were a priori placed into one mapping unit because of their

uniqueness and very high ﬂoristic variability.

5) Using the selected set of environmental predictors (see

above), CART analyses were run and the result was

evaluated according to ability of the CART to separate

(on predictors) the given mapping classes. In cases

where we deemed this separation weak or non-existent,

we further amalgamated the mapping classes.

6) The ﬁnal step of this iterative process is depicted by

the ﬁnal CART analysis of the zonal vegetation of

the Mitchell Plateau (Figure 2.12). In this decision

tree, Node 1 commences with all 65 sample points

representing the ﬁeld observations and extracted

spectral values. The CART analysis determined that

Geology type was the best initial splitter, suggesting a

class of samples characteristic for basalt and dolerite.

This class consisted of 17 sample points, while the

remaining data (48 sample points) represented other

types. The further splits were based on PC6 and the B1

(the blue band of the Landsat image). Some terminal

nodes (leaves) of this tree failed to show pure pattern

(one class only). An example of this is the terminal

node 4, which shows a 50% split between classes

M and O (2 sample points each). Nevertheless this

tree was selected as the one with the best trade-o

between low relative cost and the highest number of

classes modelled.

terminal node that encompasses more than one class. This

is not advisable, however, as this increases the relative cost

of the CART and therefore decreases the accuracy of the

classiﬁcation. There is then a trade-o between the number

of classes to be modelled and the accuracy of the modelled

classes, which requires a judgement call from the operator.

Another alternative is to merge the ambiguous classes,

should that be logical in the context of the analysis.

Ruleset for the Mitchell Plateau Region:

Iterations and Outcomes

The creation of the ruleset for the Mitchell Plateau region

was the critical part of the vegetation mapping process.

This comprised several iterations characterised by varying

spectrum and number of environmental variables. Initially

the number of predictors tested for the Kimberley CART

analyses was high—geology, elevation, slope, aspect,

NDWI and 19 of the BioClim climate variables were all

added to early CART analyses for each area. Additionally,

the values of the six Landsat bands and the six PCA

components were extracted to the ﬁeld site points to

complement the environmental variables. Early CART

analyses showed a high informative (classiﬁcation) value

of elevation, aspect and slope. These variables were,

however, deemed less reliable due to scale discrepancies

between the ﬁeld estimates/measurements of the aspect

and slope and other GIS data used. The BioClim data are a

result of modelling based on interpolation between climatic

stations in the modelling area. Since the density of climatic

stations delivering real data for the Kimberley is known to

be extremely low, the modelled climatic surfaces have to

be used with utmost caution. We consequentially decided

to exclude the BioClim climatic variables from the CART

analyses due to their comparably low spatial resolution

(and reliability).

CART analysis was undertaken in two phases of modelling:

for zonal mapping units (see Box 2.1) and separately for

azonal vegetation.

The decision (CART) trees were derived using the following

predictors: geology type, the six Landsat bands, the six

principal components of the Landsat bands and NDWI.

The selection of classes (mapping units) was undertaken

according to the following procedure:

1) The complete vegetation classiﬁcation system, as

revealed by the numerical-classiﬁcation analyses (see

Table 2.4) was taken as the starting base.

2) The azonal and relic units were identiﬁed ﬁrst and

excluded from the modelling.

3) Furthermore, the units represented by only one relevé

(such as G Terminalia canescens Community in the

Mitchell Plateau data) were also excluded.

FIG. 2.14

Final CART diagram for the Pantijan zonal vegetation units (detailed classiﬁcation). The alpha-numerical (and colour)

coding of the classes is as follows: B (incl. C, D), E, G (incl. H), I, J, K, L. (Communities A, F, M were not modelled.).

Insert: Simpliﬁed CART classiﬁcation tree showing the hierarchy of classiﬁers (splitters).

61CHAPTER 2 FIELD SURVEY AND PROTOCOLS

FIG. 2.15

The mapping rule set for the Pantijan zonal vegetation data set.

/*Terminal Node 1*/

GEOLOGY == 1

) &&

PC1 <= 71.7311

{

terminalNode = 1;

class = I;

}

/*Terminal Node 2*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI <= 0.553033 &&

B2 <= 19.5

{

terminalNode = 2;

class = K;

}

/*Terminal Node 3*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI <= 0.553033 &&

B2 > 19.5

{

terminalNode = 3;

class = J;

}

/*Terminal Node 4*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI > 0.553033

{

terminalNode = 4;

class = K;

}

/*Terminal Node 5*/

GEOLOGY == 4

) &&

B5 <= 28.5

{

terminalNode = 5;

class = L;

}

/*Terminal Node 6*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 <= 57.5

{

terminalNode = 6;

class = E;

}

/*Terminal Node 7*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 > 57.5 &&

PC3 <= 103.507

{

terminalNode = 7;

class = G;

}

/*Terminal Node 8*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 > 57.5 &&

PC3 > 103.507

{

terminalNode = 8;

class = B;

}

/*Terminal Node 1*/

GEOLOGY == 1

) &&

PC1 <= 71.7311

{

terminalNode = 1;

class = I;

}

/*Terminal Node 2*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI <= 0.553033 &&

B2 <= 19.5

{

terminalNode = 2;

class = K;

}

/*Terminal Node 3*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI <= 0.553033 &&

B2 > 19.5

{

terminalNode = 3;

class = J;

}

/*Terminal Node 4*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI > 0.553033

{

terminalNode = 4;

class = K;

}

/*Terminal Node 5*/

GEOLOGY == 4

) &&

B5 <= 28.5

{

terminalNode = 5;

class = L;

}

/*Terminal Node 6*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 <= 57.5

{

terminalNode = 6;

class = E;

}

/*Terminal Node 7*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 > 57.5 &&

PC3 <= 103.507

{

terminalNode = 7;

class = G;

}

/*Terminal Node 8*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 > 57.5 &&

PC3 > 103.507

{

terminalNode = 8;

class = B;

}

/*Terminal Node 1*/

GEOLOGY == 1

) &&

PC1 <= 71.7311

{

terminalNode = 1;

class = I;

}

/*Terminal Node 2*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI <= 0.553033 &&

B2 <= 19.5

{

terminalNode = 2;

class = K;

}

/*Terminal Node 3*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI <= 0.553033 &&

B2 > 19.5

{

terminalNode = 3;

class = J;

}

/*Terminal Node 4*/

GEOLOGY == 1

) &&

PC1 > 71.7311 &&

NDWI > 0.553033

{

terminalNode = 4;

class = K;

}

/*Terminal Node 5*/

GEOLOGY == 4

) &&

B5 <= 28.5

{

terminalNode = 5;

class = L;

}

/*Terminal Node 6*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 <= 57.5

{

terminalNode = 6;

class = E;

}

/*Terminal Node 7*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 > 57.5 &&

PC3 <= 103.507

{

terminalNode = 7;

class = G;

}

/*Terminal Node 8*/

GEOLOGY == 4

) &&

B5 > 28.5 &&

B7 > 57.5 &&

PC3 > 103.507

{

terminalNode = 8;

class = B;

}

the modelled coverage of the azonal vegetation based on

high environmental wetness did not reveal interpretable

patterns. Many patches of azonal vegetation were

suggested outside the inﬂuence of riparian (riverine) zones,

possibly linked to landscapes experiencing temporary

wetness. We concluded that we might not have optimal

NDWI coverage and abandoned the attempt to use NDWI to

model azonal vegetation.

As the ﬁnal step in creating a ruleset, a sequence of ‘if-

then’ rules was created to classify the zonal vegetation

(Fig. 2.13). These rules were used to produce the vegetation

map by applying them to the segmented image of the

mapping area (see below) using Object-Oriented Image

Analysis (OBIA) in eCognition 8.0.

We admit that selecting a CART analysis was a risky

strategy as we had only 65 plots and created 10−15

classes. It appears, however, that the modelled zonal

vegetation patterns are entirely plausible. The apparent

success of the modelling exercise may be ascribed to the

strong link between the presumed ecological drivers and

the vegetation patterns.

A separate CART analysis was undertaken for the azonal

vegetation type Riparian Thickets. Only two classes were

required for this classiﬁcation (zonal and azonal) and only

one predictor variable (NDWI) was used. The CART analysis,

therefore, essentially determined the optimum threshold

value of NDWI for the classiﬁcation of azonal vegetation.

This value was then used to reclassify the NDWI layer in

ArcMap to create a Riparian Thickets layer. Inspection of

63CHAPTER 2 FIELD SURVEY AND PROTOCOLS

The advantages of OBIA over pixel-based classiﬁcation

methods include the following (Benz et al. 2004, Bock et al.

2005, Hay et al. 2005):

• Moremeaningfulstatisticalcalculationofspectraland

textural qualities;

• Theavailabilitytotheclassicationprocessofobject

feature qualities such as shape and topology;

• Thecapacitytoconsiderintuitivespatialrelations

between real-world objects and image objects; and

• TheeaseofintegrationbetweenGISandremotesensing

environments and ﬂexibility among dierent soware

platforms.

Additionally, the hierarchical nature of OBIA is ideally

suited to implement rulesets, a feature strongly supported

by eCognition 8.0.

Segmentation of the images into objects was carried out

using a multi-resolution segmentation algorithm on the

bands with the most spectral information: PC1 and PC2.

The CART ruleset was then applied to the image objects

as a set of ‘if-then’ rules, resulting in the sequential

classiﬁcation of the image into the modelled target classes.

3.5 Step 5: Post-Modelling Adjustments to the Maps

3.5.1 Digitising Non-modelled Units

CART modelling was unsuccessful in depicting the extent

of azonal vegetation types. To correct for this, it was

necessary to implement a number of post-modelling

cartographic modiﬁcations. These adjustments involved

digitising the spatial extent of units which:

1) we were not able to model successfully using NDWI

(azonal vegetation);

2) we have not suciently studied in the ﬁeld (neither by

our predecessors nor us) to yield the volume of spatial

data required to model their occurrence (freshwater

wetland vegetation, mangroves, saline ﬂats, basalt

rocky outcrops, beaches and clis etc.); and

3) were of relic origin and hence their occurrence was

dicult to predict by using modern spatial data layers

(monsoonal rainforests).

The following units of azonal vegetation were digitised

from high-resolution aerial photographs or satellite

images:

a) riparian thickets, recognisable as dense, dark-coloured

patches fringing waterways and wetlands;

b) non-wooded freshwater wetlands, usually occurring

as river headwater grasslands or as belts around

backwaters and billabongs;

c) mangroves (no attempt was made to recognise

separate vegetation zones within mangrove

communities);

Ruleset for the Pantijan Region: Iterations and Outcomes

The experience gained performing CART analyses for

the Mitchell Plateau mapping area was valuable for the

analyses undertaken in the Pantijan area. As in the case

of the Mitchell Plateau map, the predictors for the Pantijan

area were limited to geology, the six Landsat bands, the six

principal components and NDWI. Several iterations were

modelled, starting with the selection of zonal mapping

units (for the full vegetation system see Table 2.6). Aer

the ﬁnal iteration we settled on seven mapping units,

informed by the power of separation of the units in CART

and its ecological interpretability.

The ﬁnal CART for the model of the Pantijan mapping area

is shown in Figure 2.14. As with the Mitchell Plateau region,

geology is the primary driver of vegetation patterns and it

was therefore identiﬁed as the ﬁrst splitter in the CART. The

le branch, from Node 2, deﬁnes all the vegetation classes

found on either basalt or dolerite (Geology = 1). PC1, NDWI

and Band 2 (green band of the Landsat image) are used

as the splitters for Classes I, J and K. The right branch,

from Node 5, deﬁnes all the vegetation classes found on

either sandstone or laterite (Geology = 4). Bands 5 and 7

of the Landsat image and PC3 are used as the splitters for

Classes B, E, G and L. A noteworthy feature of the Pantijan

CART, as compared to the Mitchell Plateau equivalent, is

the high level of purity (one class only) of the terminal

nodes. Only Terminal Nodes 6 and 8 predict more than one

class and even in those instances, the overlap is minimal.

This is a remarkable result, given the constraints of the low

number of data points and high number of classes.

As in the Mitchell Plateau ruleset, segmentation of the

objects in the Pantijan ruleset was carried out using a

multi-resolution segmentation algorithm on the bands

with the most spectral information: the ﬁrst two principal

components (PC1 and PC2). The CART ruleset was then

applied to the image as a set of ‘if-then’ rules, resulting in

the sequential classiﬁcation of the image into the modelled

target classes. The Pantijan ruleset, as implemented in

eCognition 8.7, is shown in Figure 2.15. The application

of these rules resulted in a classiﬁcation that was then

exported as a ti image and imported into ArcMap for

further treatment.

3.4.2 Image Segmentation

Object-Oriented Image Analysis (OBIA) is the delineation

and classiﬁcation of homogeneous image segments in

much the same way as the human eye perceives and

identiﬁes objects in the real world. The primary beneﬁt of

this approach is the ability to analyse image features as an

intuitively-determined and meaningful unit, rather than an

arbitrarily deﬁned image pixel.

3.6 Step 6: Veriﬁcation

The validation of the Mitchell Plateau maps was undertaken

by comparing a validation dataset to the modeled map data.

Ninety-ﬁve validation data points were collected during a

ﬁeld trip in August 2012. Selection of the points in the ﬁeld

was semi-random, being constrained by proximity to roads

but placed randomly along these routes. At each sampling

point (see Fig. 2.3 for their position), a GPS reading was

taken and the identity of the sampled vegetation patch

established using the classiﬁcation system deﬁned for the

Mitchell Plateau mapping area (see Tables 2.3 and 2.4).

The classiﬁed validation dataset was then overlain on

the modeled map in ArcMap. A comparison between the

validation and modeled data sets was undertaken at both

the detailed (ML1) and simpliﬁed (ML2) mapping levels. The

result was that the modeled data agreed with the validation

data point at 43 of the 95 (45%) validation points at the

ML1 mapping level and at 75 of the 95 (79%) validation

points at the ML2 mapping level.

A second validation process was undertaken with a 100 m

buer applied to the validation data set. This determined

if a validation data point lay within 100 m, or about three

Landsat pixels, of a correctly modeled location. A 100 m

margin of error was deemed appropriate because the 30 m

pixel size of the Landsat images used in modeling means

that the location of a community may be more precisely

determined on the ground than it can be on Landsat images.

The result of the second round of validation was that

modeling agreed with the validation data point at 71 of the

95 (74%) locations at ML 1 and 84 of the 95 (88%) locations

at ML2.

3.7 Step 7: Presentation

The purpose of our maps of the Mitchell Plateau and Pantijan

areas is to illustrate the results of a modelling analysis,

highlight spatial relationships between vegetation mapping

units and facilitate understanding of the vegetation patterns.

We have aimed at maximising the ease of user interpretation

by following the principles of good cartography and

complying with accepted mapping conventions.

The Mitchell Plateau and Pantijan vegetation maps are

thematic in nature. This is usual for vegetation maps, as

information about vegetation is qualitative in nature and

as such is best communicated in a thematic manner. A

thematic map focuses on the geographic distribution and

spatial variation of a certain feature or theme; in this case

vegetation units. This is in contrast to general-purpose

maps, such as topographic maps, that communicate speciﬁc

quantitative data.

In symbolising the vegetation data on our maps, we

considered the visual distinctiveness of the colours used

and commonly accepted conventions for vegetation colour

d) mud (saline) ﬂats;

e) beaches and coastal clis;

f) basalt rocky outcrops; and

g) monsoonal rainforests (vine thickets).

3.5.2 Further Adjustments of the

CART Modelling Outcomes

In addition to the amendments for azonal mapping units,

the following post-modelling adjustments were made to

the Mitchell Plateau map:

1) The CART analyses could not separate classes I and S

(Fig. 2.12). These two classes, which represent dierent

forms of Kimberley shrubland, were merged and

mapped as a single unit.

2) Class B (Eucalyptus miniata-Livistona eastonii

Community) was reclassiﬁed as Class A (Corymbia

nesophila-Livistona eastonii Community) where it

occurs on laterite.

3) Classes A and B were reclassiﬁed as Class C

(Eucalyptus tetrodonta-Hibbertia hooglandii

Community) where these clases occurred on

sandstone. This meant that Class B does not appear on

the ﬁnal map.

The ﬁnal vegetation maps of the Mitchell Plateau

and Pantijan areas are presented in Figs 16 and 17

respectively. They are also included in PDF format in the

Electronic Appendices of this report with the ﬁle names

Mitchell Plateau Vegetation Map_basic version and Pantijan

Vegetation Map_basic version.

3.5.3 Production of Simpliﬁed Versions of the

Vegetation Maps

We prepared simpliﬁed versions of the vegetation maps of

both the Mitchell Plateau and Pantijan areas, which map

the major vegetation groups. They are shown in Figs 18

(Mitchell Plateau) and 19 (Pantijan) and are also provided

in PDF format in the Electronic Appendices to this report

as Mitchell Plateau Vegetation Map_simpliﬁed version and

Pantijan Vegetation Map_simpliﬁed.

The simpliﬁed Mitchell Plateau vegetation map groups

Communities A, C, E, F and H into Kimberley Sandy

Woodlands; Communities M, N, O, P and R into Kimberley

Clayey Woodlands and Communities I, S, U, T and V into

Kimberley Sandy Shrublands. All the azonal vegetation

types were retained on the simpliﬁed map.

The simpliﬁed Pantijan vegetation map groups

Communities B, C, D, E and L into Kimberley Sandy

Woodlands. Communities I, J and K are grouped into

Kimberley Clayey Woodlands and Communities G and H

into Kimberley Sandy Shrublands. All the azonal vegetation

types were retained on the simpliﬁed map.

65CHAPTER 2 FIELD SURVEY AND PROTOCOLS

scale that will remain accurate when the map is printed at

dierent scales.

The ﬁnal maps are provided electronically in printable

formats (PDF, TIFF and BMP) and as shape ﬁles in the

accompanying Electronic Appendices (CD).

REFERENCES

Archibald S & Scholes RJ. 2007. Leaf green-up in a

semi-arid African savannah—separating tree and grass

responses to environmental cues. Journal of Vegetation

Science 18: 583–594.

Arndt W & Norman MJT. 1959. Characteristics of native

pasture on Tippera clay loam at Katherine, N.T. CSIRO

Division of Land Research and Regional Survey, Technical

Paper 3: 1–20.

Bayes AJ & Mackey BG. 1991. Algorithms for monotonic

functions and their application to ecological studies in

vegetation science. Ecological Modeling 56: 135–159.

Beard JS. 1978. Vegetation survey of Western Australia.

1:1,000,000 Vegetation Series Sheet 1. The University of

Western Australia Press, Nedlands, WA.

Beard JS. 1979. Vegetation survey of Western Australia:

Kimberley. 1:1 000 000 Vegetation Series. Explanatory

Notes to Sheet 1. The University of Western Australia

Press, Nedlands, WA.

Beard JS. 1988. Vegetation map of the Mitchell Plateau

area. (Map with explanatory notes.) Department of

Conservation and Land Management, Perth, WA.

Benz UC, Hofmann P, Willhauck G, Lingenfelder I & Heynen

M. 2004. Multi-resolution, object-oriented fuzzy analysis of

remote sensing data for GIS-ready information. ISPRS Journal

of Photogrammetry and Remote Sensing 58: 239–258.

Bock MP, Xoﬁs P, Mitchley J, Rossner G & Wissen M. 2005.

Object-oriented methods for habitat mapping at multiple

scales—Case studies from Northern Germany and Wye

Downs, UK. Journal for Nature Conservation 13: 75–89.

Breiman L, Friedman JH, Olshen RA & Stone CJ. 1984.

Classiﬁcation and regression trees. Wadsworth and Brooks,

Monterey, CA.

Brocklehurst P, Lewis D, Owen G, Lynch D & Hill J. 2004.

Northern Territory Guidelines for Survey and Mapping

Vegetation. Technical Report 3/2004. Department of

Infrastructure Planning and Environment, Natural Systems

Division, Palmerston, NT.

Brooker MIH & Kleinig DA. 1994. Field guide to Eucalypts.

Volume 3. 2nd ed. Bloomings Books, Melbourne, VIC.

Buckley A & Vandegra D. 2008. Making better maps:

Cartographic design. Principles and practices for any map.

ESRI, Redlands, CA.Campbell JB. 2006. Introduction to

remote sensing. 4th ed. Taylor & Francis, London, UK.

schemes. A map reader visually relates objects to one

another and assigns meaning according to the map’s

visual structure. Research has shown that map users

perceive dark colours as important and bright colours as

dominant (Murad-al-shaikh 2004). Such a perception of

hierarchy should be avoided in a vegetation map, as no

vegetation unit is more important than another.

On our maps, we used ‘warmer’ colours (yellow to brown)

to depict vegetation units on basalt and ‘cooler’ colours

(greens) for those on sandstone. Sandstone shrublands

were emphasised by light colouring to make them easily

recognisable and reﬂect their ecological and evolutionary

speciﬁcity (see Chapter 3). Special intrazonal and relict units

(basalt rocky outcrops and rainforests) were stressed by the

use of deep bright colours (maroon and red). This facilitates

the recognition of these units on the map, despite their

occurrence as small patches. The azonal units associated

with water were presented in shades of blue, except for the

azonal saline habitats, which were allocated a colour of

bright yellow for the vegetation of beaches and coastal clis,

and a pale sandy yellow for the Kimberley samphire ﬂats to

make them more distinctive in relation to the neighbouring

mangroves presented by an aqua green colour.

We added major geographic features to the maps, such as

roads, rivers, wetlands, settlements and airﬁelds. Rivers

were labelled in blue italic serif font and splined to conform

to the curvature of the watercourse. Locality names were

labelled in a black sans-serif font, with the preferred label

position being top right, followed by bottom right, top le

and bottom le (Murad-al-shaikh 2004). Large geographic

features were labelled in large, uppercase type letters

with sucient character spacing to convey their expanse.

Character spacing is equidistant for visual balance and

ease of reading.

Map elements contain ancillary information such as the

title, explanatory text, north arrow, scale, legend and

logos. These aid the map reader to interpret, comprehend

and extract information from the map. The presentation

of these elements sets the tone of the map and should

support and promote the desired mood. All map elements

should be aligned with attention to visual balance and

visual ﬂow. Simplicity is a good design principle for maps

and the use of any unnecessary labels, objects or elements

should be avoided (Buckley & Vandegra 2008).

Our maps include a legend box containing the name of

the map, the authors and their aliation, sources of the

mapping data and the map datum. The most signiﬁcant

feature of the legend box, however, is the map symbology.

This identiﬁes the colour that symbolises each mapping

unit, including any associated subordinate units and

synonyms. This is followed by the symbol coding of

geographic features. The legend box also contains a rule

Centre for Plant Biodiversity Research 2006. EUCLID:

eucalypts of Australia. CSIRO Publishing, Melbourne, VIC.

[CD ROM].

Fielding AH. 2007. Cluster and classiﬁcation techniques for

biosciences. Cambridge University Press, Cambridge, UK.

Fox ID, Nelder VJ, Wilson GW, Bannink PJ, Wilson BA,

Brocklehurst PS, Clark MJ, Dickinson KJM, Beard JS,

Hopkins AJM, Beeston GR, Harvey JM, Thompson EJ,

Ryan TS, Thompson SL, Butler DW, Cartan H, Addicott

EP, Bailey LP, Cumming RJ, Johnson DC, Schmeider M,

Stephens KM & Bean AR. 2001. The vegetation of the

Australian tropical savannas. Queensland Government,

Environmental Protection Agency, Brisbane, QLD.

Gao B-C. 1996. NDWI—A normalized dierence water index

for remote sensing of vegetation liquid water from space.

Remote Sensing of Environment 58: 257–266.

Gardner CA. 1942. The vegetation of Western Australia

with special reference to climate and soils. Journal of the

Royal Society of Western Australia 28: 11–87.

George AS. 1978. Notes on the vegetation and ﬂora of

the Cape Londonderry peninsula. Western Australian

Herbarium Research Notes 1: 1–15.

Google Corporation 2012. Google Earth. Available at

www.googleearth.com.

Hajdu LJ. 1981. Graphical comparison of resemblance

measures in phytosociology. Vegetatio 48: 47–59.

Hansen M, Dubayah R & DeFries R. 1996. Classiﬁcation

trees: an alternative to traditional land cover classiﬁers.

International Journal of Remote Sensing 17: 1075–1081.

Hay GJ, Castilla G, Wulder MA & Ruiz JR. 2005. An

automated object-based approach for the multiscale image

segmentation of forest scenes. International Journal of

Applied Earth Observation and Geoinformation 7: 339–359.

Hijmans RJ, Cameron SE, Parra JL, Jones PG & Jarvis A.

2005. Very high resolution interpolated climate surfaces

for global land areas. International Journal of Climatology

25: 1965–1978.

Hnatiuk RJ & Kenneally KF. 1981. A survey of the

vegetation and ﬂora of Mitchell Plateau, Kimberley,

Western Australia. In: Biological survey of Mitchell Plateau

and Admiralty Gulf. pp. 13–94. Western Australian

Museum, Perth, WA.

Hubálek Z. 1982. Coecients of association and

similarity, based on binary (presence-absence) data: an

evaluation. Biological Reviews 57: 669–689.

Hüttich C. 2011. Mapping vegetation types in a savannah

ecosystem in Namibia: Concepts for integrated land cover

assessment. PhD thesis, Friedrich-Schiller-Universität,

Jena, DE.

Hüttich C, Herold M, Wegmann M, Cord A, Strohbach

B, Schmullius C & Dech S. 2011. Assessing eects of

temporal and varying observation periods for large-area

land-cover mapping in semi-arid ecosystems: Implications

for global monitoring. Remote Sensing of Environment

115: 2445–2459.

Kabay ED & Burbridge AA (eds). 1977. A biological survey

of the Drysdale River National Park, north Kimberley,

Western Australia in August, 1975. Wildlife Research

Bulletin Western Australia 6: 1–133.

Keith D. 2002. A compilation map of native vegetation for

New South Wales. Version 1.1. NSW National Parks and

Wildlife, Sydney.

Kenneally K, Willing T & McCann K. 2008. Anjo Peninsula

ﬂora and vegetation assessment. ENV Australia, Perth, WA.

Lance GN & Williams WT. 1967. A general theory of

classiﬁcatory sorting strategies. I. Hierarchical systems.

Computer Journal 9: 373–380.

Lawrence LR & Wright A. 2001. Rule-based classiﬁcation

systems using classiﬁcation and regression tree (CART)

analysis. Photogrammetric Engineering and Remote

Sensing 67: 1137–1142.

Legendre P & Legendre L. 1998. Numerical ecology.

2nd ed. Elsevier, Amsterdam, NL.

Liang S. 2004. Qualitative remote sensing of land surfaces.

Wiley-Interscience, Hoboken, NJ.

Maslin, B.R. (coordinator) 2001. WATTLE: acacias

of Australia. Australian Biological Resources Study,

Canberra and the Department of Conservation and Land

Management, Perth, WA. [CD ROM].

Mather PM. 2011. Computer processing of remotely-

sensed images: An introduction. 4th ed. J. Wiley & Sons,

Chichester, UK.

McCune B, Grace JB & Urban DL. 2002. Analysis of

ecological communities. MjM Soware Design, Gleneden

Beach, OR.

McCune B & Meord MJ. 2011. PC-ORD. Multivariate

analysis of ecological data. Version 6. MjM Soware,

Gleneden Beach, OR.

McKenzie N & Belbin L. 1991. Kimberley rainforest

communities: Reserve recommendations and management

implications. In: McKenzie N, Johnston R & Kendrick P.

(eds), Kimberley rainforests of Australia. pp. 453–468.

Surrey Beatty and Sons, Chipping Norton, NSW.

Michaelsen J, Schimel DS, Friedl MA, Davis FW &

Dubayah RC. 1994. Regression tree analysis of satellite

and terrain data to guide vegetation sampling and

surveys. Journal of Vegetation Science 5: 673–686.

Miles JM & Burbridge AA (eds). 1975. A biological survey

of the Prince Regent River Reserve, north-west Kimberley,

Western Australia in August 1974. Wildlife Research

Bulletin Western Australia 3: 1–116.

67CHAPTER 2 FIELD SURVEY AND PROTOCOLS

Miller J & Franklin J. 2002. Modeling the distribution of

four vegetation alliances using generalized linear models

and classiﬁcation trees with spatial dependence. Ecological

Modelling 157: 227–247.

Mucina L. 2010. Floristic-phytosociological approach,

potential natural vegetation, and survival of prejudice.

Lazaroa 31: 7–16.

Mucina L, Rutherford MC & Powrie LW. 2006. Logic of

the map: Approaches and Procedures. In: Mucina L &

Rutherford MC. (eds), The vegetation of South Africa,

Lesotho and Swaziland. pp. 12–29. South African National

Biodiversity Institute, Pretoria, ZA.

Mucina L, Schaminée JHJ & Rodwell JS. 2000. Common

data standards for recording relevés in ﬁeld survey for

vegetation classiﬁcation. Journal of Vegetation Science

11: 769–772.

Mucina L & van der Maarel E. 1989. Twenty years of

numerical syntaxonomy. Vegetatio 81: 1–15.

Murad-al-shaikh M. 2004. Cartography with ArcGIS course

manuals. Course version 1.2, revised October 2004. ESRI,

Redlands, CA.

Neldner VJ, Wilson BA, Thompson EJ & Dillewaard HA.

2012. Methodology for survey and mapping of regional

ecosystems and vegetation communities in Queensland.

Version 3.2. State of Queensland, Department of Science,

Information Technology, Innovation and the Arts,

Brisbane, QLD.

Nežerková-Hejcmanová P, Hejcman M, Camara AA,

Antonínová M, Pavlů V, Černý T & Bâ AT. 2005. Analysis of

the herbaceous undergrowth of the woody savannah in the

Fathala Reserve, Delta du Salom National Park (Senegal).

Belgian Journal of Botany 138: 119–128.

Nix HA. 1986. BIOCLIM—a Bioclimatic Analysis and

Prediction System. Research Report. CSIRO Division of

Water and Land Resources 1983–1985. CSIRO,

Canberra, ACT.

Orlóci L. 1978. Multivariate analysis in vegetation research.

2nd ed. Dr W. Junk Publishers, The Hague, NL.

Paczkowska G & Chapman AR. 2000. The Western

Australian ﬂora—a descriptive catalogue. Wildﬂower

Society of Western Australia, Nedlands, WA.

Petheram RJ, Kok B & Bartlett-Tor E. 2003. Plants of the

Kimberley Region of Western Australia. Rev. ed. University

of Western Australia Press, Perth, WA.

Podani J. 2000. Introduction to the exploration of

multivariate biological data. Backhuys, Leiden, NL.

Podani J. 2001. SYN-TAX 2000. Computer programs for

data analysis in ecology and systematics. User’s manual.

Scientia, Budapest, HU.

Richter R. 2004. ATCOR: Atmospheric and topographic

correction [online]. DLR-German Aerospace Centre, Remote

Sensing Data Centre, Wessling, DE. Available at:

www.dlr.de/caf/Portaldata/36/Resources/dokumente/

technologie/atcor_ﬂyer_march2004.pdf.

Sivertsen D. 2009. Native Vegetation Interim Type

Standard. NSW Government, Department of Environment,

Climate Change and Water, Sydney, NSW.

Sneath PHA & Sokal RR. 1973. Numerical taxonomy. The

principles and practice of numerical classiﬁcation. W.H.

Freeman & Co., San Francisco, CA.

Steinberg D & Golovnya M. 2006. CART 6.0 User’s Manual.

Salford Systems, San Diego, CA.

Tamás J, Podani J & Csontos P. 2001. An extension

of presence/absence coecients to abundance data:

a new look at absence. Journal of Vegetation Science

12: 401–410.

ter Braak CJF. 1986. Canonical correspondence analysis: a

new eigenvector technique for multivariate direct gradient

analysis. Ecology 67 1167–1179.

ter Braak CJF & Šmilauer P. 2002. CANOCO reference

manual and CanoDraw for Windows user’s guide. Soware

for canonical community ordination (version 4.5).

Biometris, Wageningen, NL.Tichý L & Chytrý M. 2006.

Statistical determination of diagnostic species for site

group of unequal size. Journal of Vegetation Science 17:

809–818.

Tichý L, Chytrý M, Hájek M, Talbot SS & Botta-Dukát

Z. 2010. OptimClass: Using species-to-cluster ﬁdelity

to determine the optimal partition in classiﬁcation of

ecological communities. Journal of Vegetation Science

21: 287–290.

Turner W, Spector S, Gardiner N, Fladeland M, Sterling

E & Steininger M. 2003. Remote sensing for biodiversity

science and conservation. Trends in Ecology and Evolution

18: 306–314.

Tüxen R. 1956. Die heutige potentielle natürliche

Vegetation als Gegenstand der Vegetationskartierung.

Angewandte Pﬂanzensoziologie (Stolzenau) 13: 4–42.

Walter H. 1964. Die Vegetation der Erde in öko-

physiologischer Betrachtung. Band I: Die tropischen und

subtropischen Zonen. G. Fischer Verlag, Jena, DE.

Walter H. 1973. Vegetation of the Earth in relation to

climate and the eco- physiological conditions. English Univ.

Press, London, UK.

Wheeler JR. (Ed.) 1992. Flora of the Kimberley region.

Western Australia Herbarium. Department of Conservation

and Land Management, Perth, WA.

Kimberley Vegetation Mapping book 2013 WEB

Data

February 2014

Matthew D. Barrett · Russell L Barrett · Milan Valachovič · Garth Stephenson · Adriaan Van Niekerk

Mucina&KimberleyVegMapCh2 2013b

Data

September 2013

Ladislav Mucina · Glen Daniel · Garth Stephenson · Ilze Boonzaaier · Milan Valachovič

Download

Release, deposition and elimination of radiocesium (Cs-137) in the terrestrial environment

Article

Full-text available

May 2014
ENVIRON GEOCHEM HLTH

Radionuclide contamination in terrestrial ecosystems has reached a dangerous level. The major artificial radionuclide present in the environment is (137)Cs, which is released as a result of weapon production related to atomic projects, accidental explosions of nuclear power plants and other sources, such as reactors, evaporation ponds, liquid storage tanks, and burial grounds. The release of potentially hazardous radionuclides (radiocesium) in recent years has provided the opportunity to conduct multidisciplinary studies on their fate and transport. Radiocesium's high fission yield and ease of detection made it a prime candidate for early radio-ecological investigations. The facility setting provides a diverse background for the improved understanding of various factors that contribute toward the fate and transfer of radionuclides in the terrestrial ecosystem. In this review, we summarize the significant environmental radiocesium transfer factors to determine the damaging effects of radiocesium on terrestrial ecosystem. It has been found that (137)Cs can trace the transport of other radionuclides that have a high affinity for binding to soil particles (silts and clays). Possible remedial methods are also discussed for contaminated terrestrial systems. This review will serve as a guideline for future studies of the fate and transport of (137)Cs in terrestrial environments in the wake of the Fukushima Nuclear Power Plant disaster in 2011.

Statistical determination of diagnostic species for site groups of unequal size

Article

Full-text available

Dec 2006
J VEG SCI

Aim: Concentration of species occurrences in groups of classified sites can be quantified with statistical measures of fidelity, which can be used for the determination of diagnostic species. However, for most available measures fidelity depends on the number of sites within individual groups. As the classified data sets typically contain site groups of unequal size, such measures do not enable a comparison of numerical fidelity values of species between different site groups. We therefore propose a new method of measuring fidelity with presence/absence data after equalization of the size of the site groups. We compare the properties of this new method with other measures of statistical fidelity, in particular with the Dufrêne-Legendre Indicator Value (IndVal) index. Methods: The size of site groups in the data set is equalized, while relative frequencies of species occurrence within and outside of these groups are kept constant. Then fidelity is calculated using the phi coefficient of association. Results: Fidelity values after equalization are independent of site group size, but their numerical values vary independently of the statistical significance of fidelity. By changing the size of the target site group relative to the size of the entire data set, the fidelity measure can be made more sensitive to either common or rare species. We show that there are two modifications of the IndVal index for presence/absence data, one of which is also independent of the size of site groups. Conclusion: The phi coefficient applied to site groups of equalized size has advantages over other statistical measures of fidelity based on presence/absence data. Its properties are close to an intuitive understanding of fidelity and diagnostic species in vegetation science. Statistical significance can be checked by calculation of another fidelity measure that is a function of statistical significance, or by direct calculation of the probability of observed species concentrations by Fisher's exact test. An advantage of the new method over IndVal is its ability to distinguish between positive and negative fidelity. One can also weight the relative importance of common and rare species by changing the equalized size of the site groups. Nomenclature: Ehrendorfer (1973).

Twenty years of numerical syntaxonomy

Article

Full-text available

Jul 1989
Vegetatio

The development of numerical syntaxonomy during its first 20 yr is reviewed. The use of methods of numerical classification and ordination is the dominating feature of the development. National and local phytosociological data banks were established, large data sets handled and many important vegetation monographs were methodically based on multivariate data analysis. Particularly the development in Italy, the Netherlands, Czechoslovakia, and Sweden contributed to new theoretical elements of numerical syntaxonomy. Ordination became a common tool of searching for reticulate synsystematic relations between community types. The most popular ordination techniques have been Principal Components Analysis and Detrended Correspondence Analysis. Hierarchical agglomerative techniques of clustering still prevail in classification, although the divisive strategy of TWINSPAN has also become an effective tool for phytosociological clustering and table sorting. Extensive program packages, also for personal computers have now become standard equipment for many vegetation scientists.

Kimberley rainforest communities: Reserve recommendations and management considerations.

Chapter

Full-text available

Jan 1991

An assessment procedure translated contour maps of species richness into compositional gradients. In terms of their perennial plants, birds and land snails, 28 assemblages of species are distinguished in these communities. The existing and proposed reserves encompass the compositional diversity of only 17 assemblages. From the compositional gradients the authors identify optimum positions for the 12 additional reserves necessary to encompass the diversity. Fire influences the area of Kimberley rainforest patches, and its impact is exacerbated by cattle. In the drier areas, rainforest patches only persist in perennially moist refugia that are fire and/or cattle-proof. -from Authors

SYN-TAX IV. Computer Programs for Data Analysis in Ecology and Systematics

Chapter

Jan 1991

Janos Podani

This new release of the package contains 27 FORTRAN and 4 BASIC programs written for IBM-PC and compatible machines and Apple Macintosh II, Plus and SE personal computers. Widely used standard multivariate methods and specific data analytical techniques, some of them suggested by the author, are represented in the package. The procedures programmed include hierarchical and non-hierarchical clustering, fuzzy classifications, block clustering, ordination, character ranking, comparison of classifications and ordinations, consensus methods, Monte Carlo simulation of distributions of partition agreement, simulated sampling based on digitized point patterns, and information theory functions for evaluating species assemblages. This paper gives general information on the programs; technical details are presented in the user’s manual.

Atmospheric Correction of Optical Imagery

Chapter

Jan 2005

Shunlin Liang

IntroductionMethods for Correcting Single-Viewing-Angle ImageryMethods for Correcting Multiangular ObservationsMethods for Estimating Total Column Water Vapor ContentSummary

Review: Plants of the Kimberley Region of Western Australia

Article

Jan 1986

Anita F. Cholewa

Rule-based classification systems using classification and regression tree (CART) analysis

Article

Oct 2001
PHOTOGRAMM ENG REM S

Incorporating ancillary data into image classification can increase classification accuracy and precision. Rule-based classification systems using expert systems or machine learning are a particularly useful means of incorporating ancillary data, but have been difficult to implement. We developed a means for creating a rule-based classification using classification and regression tree analysis (CART), a commonly available statistical method. The CART classification does not require expert knowledge, automatically selects useful spectral and ancillary data from data supplied by the analyst, and can be used with continuous and categorical ancillary data. We demonstrated the use of the CART classification at three increasingly detailed classification levels for a portion of the Greater Yellowstone Ecosystem. Overall accuracies ranged from 96 percent at level 1, to 79 percent at level 2, and 65 percent at level 3.

Canonical corespondence analysis: A new eigenvector technique for multivariate direct gradient analysis.

Article

Oct 1986
ECOLOGY

CJF TERBRAAK

Analysis of ecological communities. MJM Software Design, Gleneden Beach, OR

Technical Report

Jan 2002

Die Vegetation der Erde in Okologischer Betrachtung: I. Die Tropischen und Subtropischen Zonen

Article

Dec 1962

Floristic-ecological mapping in the Northern Kimberley: field survey methods and mapping protocols

Abstract and Figures

Supplementary resources (2)

Recommendations

Functional-trait approach to plant community assembly of kwongan shrublands in post-mining rehabilitation context (ARC Linkage Grant LP150100339)

Systematics & Taxonomy of Ptilotus (Amaranthaceae)

Vegetation of Europe

Vegetation of Africa

The Effect Of Transformational Leadership, Intrinsic Motivation On Employee Performance With Work Fa...

Floristic-ecological mapping in the Northern Kimberley: Field survey methods and mapping protocols

The Rapid Decay property and centroids in groups

Perceptions of Women Entrepreneurs and Their Performance: A Study in Sylhet City, Bangladesh