2.1 Introduction
2.2 Vegetation Sampling
2.3 Data Storage
2.4 Vegetation Data Analysis
2.5 GIS Modelling of the Vegetation Map Units
2.6 Comparisons of map units to other vegetation classifications

2. Method

2.1 Introduction
The approach involved the completion of several clearly defined stages. These were to collate all available floristic site data;
* complete new survey work in unsampled environments;
* derive vegetation communities using statistical agglomeration techniques;
* extrapolate the derived ecosystems across all land tenures across the sub region;
* produce a map and report which identifies the variation and location of vegetation communities across the sub region;

2.2 Vegetation Sampling

2.2.1 Data Audit
An audit of all systematic vegetation survey data was undertaken. Systematic vegetation survey sites needed to conform to the following characteristics:
* a fixed plot size within which an inventory of all vascular plant species recorded;
* a measure of relative abundance for each species is recorded at each site; and,
* an accurate location reference (Australian Map Grid) to within 100 metres.
As highlighted by Keith & Bedward (1999) it is possible to use data without abundance records however this would require reducing all other data to this level. Cover abundance scores are important indicators when attempting to define patterns in vegetation (Keith & Bedward, 1999). For this reason they were considered a requirement for data used in the analysis.

A total of 780 existing systematic vegetation sites were available for use in this project. Most sites have been extracted from systematic surveys completed within conservation reserves and state forests over the last 10 years. More recent regional surveys have provided systematic data across a range of different tenures (NEFBS 1992, CRA, 1998). The source of all site data used is shown in Table 1.1. All site floristics and attribute data (where collected) are now stored in the NPWS Flora Survey database.

Modifications to the existing site data were limited to a taxonomic review and a conversion of relative abundance scores used by Thomas (1998), Clarke and Benson (1986) and Clarke and Benson (1988) to a six point Braun-Blanquet index. Table 2.1 provides the conversion table for these sites.

Relative abundance scores used by Clarke (1986) in the Popran NP data have been converted according to the following table:

Clarke and BensonÕs data from Dharug (1986) was recorded in sub quadrats nested within the 20 x 20m plots. The presence data available from the original field sheets was converted to a six point Braun-Blanquet index.

2.2.2 Use of External Data Sources
External data listed in Tables 2.2: Vegetation Data Sets and Table 2.6: Classifications Used for Identifying Extant Vegetation Cover have granted permission from the relevant data custodian. This data has kindly been provided for use in this project only.

2.2.3 Environmental Stratification for Site selection
The aim of the stratification in this process is to provide a structure to support the project goals by ensuring the efficient allocation of field survey effort. The focus of this study has been to gather information on private land to establish a solid data resource to support future data acquisition and analysis. To this end, the approach sought to sample the widest variation in plant communities present in the study area while having sufficient replication to explain them. A key component was to visit plant communities that were less likely to receive survey effort in the near future, so that a complete picture of all communities could emerge.

A field survey program was planned in order to sample major gaps in site coverage across the study area. Development of the stratification and sampling regime needed to consider existing data, recent CRA mapping, long-term data objectives and financial resources. Funds were allocated for approximately 300 new sites, which, in combination with available sites provided for a target of over 1100 sites in the study area.

To efficiently sample, the location of potential variations in vegetation communities is needed. Factors that are most likely to affect plant distribution and abundance are moisture, nutrients and solar energy and fire although this is not easily mapped (NPWS, 1997). Surrogates can be used to describe these factors using rainfall distribution maps, soil maps and aspect models. Features within these digital layers can be combined to form parcels of the landscape, which comprise similar environmental characteristics (Neldner, 1995; Margules & Redhead, 1995). These parcels or ÒstrataÓ are then assumed to represent features that are likely to support similar assemblages of plant species.

Recent CRA work completed surveys across a large area from the Hawkesbury River to the Queensland border. Surveys focused primarily on public land to provide information to address the forestry reform process. Within the Hunter Sub-region (Hawkesbury to the Hunter River) a stratification system had already been established to sample variation in environmental characteristics across a much larger area than that of the LHCCREMS study area. In that analysis five data layers were used in the ARC View GIS system to highlight potential environmental variations. These were rainfall, temperature, dominant lithology, aspect and broad forest structure (NPWS, 1999b). As the LHCCREMS study area represents a subset of this study area, portions of the landscape already had some survey effort within them.

Two environmental factors were selected for stratification in this project: soil fertility and aspect. As previous survey effort had captured the major variations in elevation and rainfall it was considered that soil fertility and aspect would capture the widest variation of communities, particularly if sites were well spaced across the region.

Soil landscapes were used as a surrogate for soil fertility. They were chosen for three reasons. Firstly they form the most recent and detailed mapped layer available, which describe variations in substrate. Secondly, they inherently account for variations in topography such as slope. Thirdly they are major recognisable land management resource that are widely available to the community and councils.

Aspect was used to review local variations in vegetation that may arise from the influences of shelter and exposure. Four aspect classes were used: 0: flat, 1: 301-30¡, 2: 2-211¡, 3: 31-120¡, 4: 121 - 210¡. ARCVIEW GIS was used to combine the above aspect classes with each soil landscape. Not all landscapes were merged with an aspect class. Landscapes that contained slopes less than three degrees were considered to be flat. Typically these areas contained mangrove and wetland environments. A total of 325 strata were derived.

A proportional sampling strategy was used to identify sampling priorities for each stratum. Sampling was initially directed toward soil landscapes greater than 50 hectares. The proportional sampling strategy (Sanders et al, 1988; Keith and Sanders, 1990; Keith & Bedward, 1999) allocated sampling effort using the extant vegetation remaining on each soil landscape in relation to the total vegetation cover present. As an example if the Beresfield soil landscape comprised 10% of the vegetated landscape then 10% of the total number of sites expected would be allocated to that strata.

Progress on the sampling performance was reviewed regularly. These reviews also considered the number of sites allocated in proportion to the uncleared area of each stratum to ensure that heavily cleared strata received adequate sampling wherever possible.

2.2.4 Site Selection and Access
All existing survey plots were compiled and mapped across the strata (soil landscapes and aspect) in order to identify sampling adequacy. The position of sites was cross-referenced against soil landscape and aspect and tabulated. Deficits in the sampling of each strata highlighted areas requiring new survey. The majority of existing data was obtained from public tenures or land in the reserve system. Therefore targeting of private land was priority in this survey to minimise sampling gaps across these areas and reduce potential tenure bias in the classification of vegetation.

Following the identification of these areas, maps indicating survey priorities within each local government area were produced. Each council was visited to identify land ownership details for priority areas. First survey priorities were those areas containing several target strata within single owner properties. This minimised travel time between survey sites and reduced time consuming access negotiations. Where owners were identified, each was assigned a code. The RBCS team arranged access to these properties. Letters were sent requesting permission to survey. In most instances this was followed up by a phone call. Where access was approved the area was marked for survey. Acknowledgment goes toward all private landholders that cooperated in the survey.

2.2.5 Field survey
The field survey collected data using standard NPWS field proformas. Field plots were fixed to 0.04 hectares based on 20m x 20m quadrats. Modifications were made occasionally to the plot size (usually 10 X 40 m) to match patterns in vegetation. This was applied to features such as riparian vegetation. This quadrat size is the standard used by NPWS and the Sydney Royal Botanic Gardens (NPWS, 1997). Over 80% of the existing plots in the region conform to this size and it was considered both consistent and time efficient to maintain this plot size in the region. Descriptions of field data collected are described in more detail elsewhere (NPWS, 1995 and NPWS, 1999b).

Three field teams were established to undertake five-day field trips to survey unsampled or insufficiently sampled strata. Each team consisted of one experienced botanist and an assistant. New plot data was collected between March and July 1999. Strata maps were prepared at a scale of 1: 25 000 with the matching topographic series. These maps highlighted the soil landscape and aspect that required sampling within the area marked for survey. Standard NPWS Survey techniques were applied. These involve:

• Avoiding boundaries of strata to prevent the sampling of ecotone communities
• Avoiding local disturbances such as roads, mines quarries and other areas of gross disturbance
• Aligning the quadrat with the contour of hillsides or elongated vegetation communities (eg. riparian) to avoid significant local environmental gradients.

At each site standard field survey proformas were completed (see Appendix A). Completed survey forms provided:

1. 1. lists of plant species with respective cover-abundance values. Cover-abundance values conformed to a six-point Braun-Blanquet scale (1-<5% and uncommon, 2- <5 and common, 3- 5-25%, 4- 25-50%, 5- 50-75%, 6- 75-100%;
2. estimates of the height, cover and dominant species of each vegetation stratum;
3. measurements of slope, aspect and horizon azimuths;
4. parent material; and
5. qualitative notes on soil moisture, texture and depth, and disturbance history.

Species that could not be identified in the field were recorded to the nearest possible family or genera. These were then collected, tagged and pressed with the appropriate site code. All major features of the specimen were collected where possible ie. fruit and/or flower. Subsequent to the fieldwork the botanist identified outstanding specimens. These were noted on the site proforma. Where positive identifications could not be made, specimens were sent to the National Herbarium for resolution.

In the event that any rare or threatened plants were recorded in the plot a threatened plant proforma was completed (Appendix B). Opportunistic records were also taken where a rare or threatened plant was located on route to the site. Interesting specimens and range extensions were also lodged at the herbarium at the botanistsÕ discretion.

2.3 Data Storage

2.3.1 Site Identification
For the purpose of managing existing and new field data each survey plot was given an 8-digit alphanumerical survey identification number, taken from a system used by Bell et al (1993). A separate survey identification code was given to all data to distinguish its source. Using this system enables the reader to understand basic geographical information about the survey site.

Example:

The first three letters ÒGSFÓ refer to the first three consonants in the 1:25 000 map sheet name

The fourth and fifth digits Ò07Ó refer to the site number, ie. The seventh site on this map sheet

The sixth character ÒHÓ refers to the geology using the following categories.

N = Narrabeen sandstone
H = Hawkesbury sandstone
W = Wianamatta shale
B = Basalt
A = Alluvium
Q = Quaternary sand
P = Permian sediments
C = Carboniferous sediments

The seventh character Ò3Ó refers to the aspect using the following categories.

1 = 67.6 ø112.5 degrees
2 = 112.6 ø 157.5 degrees
3 = 157.6 ø 202.5 degrees
4 = 202.6 ø 247.5 degrees
5 = 247.6 ø 292.5 degrees
6 = 292.6 ø 337.5 degrees
7 = 337.6 ø 22.5 degrees
8 = 22.6 ø 67.5 degrees

The eighth character ÒCÓ refers to the morphology codes as used in the field survey proforma (see appendix C).

Data from other sources was transferred electronically into the ACCESS database.

2.3.3 Data validation
On completion of the field survey program, all data was checked for errors. A number of steps were involved as described below.

1. The database has a series of checks to prevent gross errors. These include restrictions on the amount of numbers that can be entered into eastern and northing fields. The database was also designed so that vegetation structure descriptions could only be sourced from species identified from the site species list.
2. A manual check of each site held in the database was compared to the original field sheet. Grid references were checked, as were total number of species recorded at each site. Searches for missing cover abundance scores were also completed.
3. A number of queries using a crosscheck of field data with the GIS system were completed. These reviewed: (1) topographic sheet of the site code against the topographic sheet name in which the GIS records the site (2) comparison of field elevation and the digital elevation model in the GIS and (3) comparison of tenure marked in the field and that recorded by the GIS.

In all cases, where errors where detected, sites were reviewed and the correct information entered.

2.4 Vegetation Data Analysis

2.4.1 Taxonomic Review
Data sourced for this project has been collected over a 13-year period by numerous observers. Over this period the taxonomy of plants has undergone continual revision. It becomes necessary to establish a botanical nomenclature to equate species collected from different periods. For this project, all nomenclature was reviewed and standardised across data sets. Synonyms for the same taxon were updated to reflect currently accepted revisions. The treatment given in Harden (1990-93) was used as a standard with the revisions outlined in Table 2.3.

The principle outcomes of the taxonomic review follow.

1. all exotic species were identified and excluded from the analysis dataset;
2. highlighted species that were likely to have been incorrectly identified or incorrectly entered into the database. Original field sheets were reviewed to determine the status of these species and where data entry errors were detected changes were made to the database. Where data entry errors were not detected species were reviewed against existing literature. Where this indicated them to be outside their likely range, and no confirmation had been made, the record was deleted;
3. inconsistently collected records of species containing sub-species or varieties. In such cases, sub-species were either lumped to species level or were assigned to a single sub-species or variant if only one variety is present in the study area;
4. species hybrids which are not recognised formally in the literature were assigned to one or other of the species based on the predominance of either in proximate environments; and;
5. species identified to genus level only. These were deleted from the analysis dataset.

2.4.2 Classification Analyses
The use of semi quantitative field survey methods allows the application of numerical techniques to assist with the identification of vegetation communities. Numerical classification techniques are not an end themselves but a means to review and compare surveyor field experiences with raw data. In this way, biases and patterns easily overlooked in the field are avoided. Such techniques are aimed at present hypotheses to help describe patterns in the field. Whether these are accepted rejected or modified remains a subjective element of the chosen classification system. However, the benefits of such an approach are that an explicit and repeatable method is employed.

The PATN analysis package (Belbin 1994) was used to group sites based on the similarities of species occurrence and abundance between sites. PATN provides a range of modules and algorithms from which to undertake analyses of ecological data. ASO, FUSE, GDEF and DEND were used to establish the association matrix for hierarchical classification, clustering and the production of a dendrogram.

PATN analyses were performed on 2360 sites drawn from the wider Hunter Region to include the Upper Hunter region, Wollemi and Yengo National Parks. Of these, 1117 sites fell within the LHCCREMS study area. The pooling of all available site data in the region enabled site data occuring on the edge of the study area to be placed in a broader regional context.

Several different analyses were performed to identify sensitivities in the data. Bray-Curtis and Kulzcynski measures of association were both tested on data using raw cover abundance codes for each species and a simpler presence-absence data set to derive dissimilarity scores between sites. These scores are calculated by comparing all possible pairs of sites in the data set (eg. Site1 and Site 2, Site 1 and 3 and so on.). A large (association) matrix describing the dissimilarity of the pairs is produced. An unweighted pair-group arithmetic averaging (UPGMA) clustering strategy was applied to the matrix to derive a hierarchical classification. The default beta value of ø0.1 was used for all analyses.

Dendrograms were derived to show the degree of dissimilarity between individual sites and groups of sites. The review of groups commenced at the highest levels of the dissimilarity scale. The dendrograms provided a coarse overview of the way in which the sites were positioned relative to one another.

To commence the fine scale analyses of data, homogeneity analyses (Bedward et al., 1992) was applied to the association matrix. This technique provides a useful guide to the benefits of improvements in species homogeneity for increasing numbers of groups or clusters made from the data set. If every single site in the dataset were made a group or Community, its characteristics would be perfectly homogenous. This would mean 2360 groups. At the other end of the scale, using only two groups would result in large within group variation so homogeneity would be low. The aim is to select the number of groups appropriate to the dataset to maximise returns to group homogeneity while minimising the number of groups used to describe the patterns in the data

Seventy groups were initially used as a means of interrogating the data. Group definition files were extracted, and species characteristics of each group were produced. A check for misclassified sites within the group was undertaken using 'GDFChk'. This reviews the grouping and highlights misclassified sites by examining the characteristics of the neighbouring sites in the group. Where identified, sites were reviewed and either withheld from further analyses, reallocated to a new group or kept with the existing group.

Groups were successfully split, where such splits reinforced sub groups based on physical characteristics of sites and floristic composition. Most often splits were made where distinct changes in substrate characteristics were apparent. Further, groups were refined at lower levels in the dissimilarity index where such splits improved the definition and recognition of floristic composition. No further splits were made when patterns presented did not accord with field experience or where features were unlikely to be mappable.

A number of core groups were readily apparent using these methods and required no further analyses. Other groupings required individual review of sites to determine the reasons behind clustering.

A software program known as FIDEL was developed for use in the vegetation mapping of the South East Forests (Keith and Bedward; 1999). This program allows relevant data to be extracted from the sites that make up the defined Map Units to indicate which species are driving the distinction between one Map Unit and another. Keith and Bedward (1999) termed these diagnostic species.

In some Map Units a species is dominant both in terms of its frequency of detection at sites and its abundance while in other Map Units it occurs patchily, at low abundances or not at all. This can be quantified by setting user defined rules to determine diagnostic species.

A table mirroring that used by Keith and Bedward (1999) is presented here. It should be noted however that the criteria for defining positive species has been amended to allow for greater number of locally abundant species to be included as positive diagnostics.

* C/A = Cover abundance

Based on this table two categories have been defined. Positive diagnostic species are those that have a higher frequency recorded within the group and / or have been recorded at a higher median cover abundance than in all other Map Units. Also species that are only recorded within the target Map Unit and in no other are considered positive. Uninformative species are those species that do not display unique characteristics to the target Map Unit. They are likely to be species common to many communities, or are those recorded at low frequency and / or very low abundance.

Map Unit profiles contain all species defined as positive diagnostics. A list of uninformative species is given for all canopy species and species considered by the authors to assist with the presentation of a visual picture of the Map Unit.

Summary structural data is provided for each Map Unit. The mean upper height, the range in field heights, the mean cover abundance, the standard deviation of the cover abundance estimates and the number of sites used to describe each structural layer is given.

A summary species richness figure and the associated standard deviation amongst sites in the Map Units are also provided.

A canopy label is also provided to assist those familiar with tree species only to impart a rough guide to the type of environment expected.

It needs to be reinforced that the presence or absence of some of these species does not indicate that the mapping is incorrect, rather that other non canopy species are likely to be influencing the classification. Consultation with the full floristic profiles and on site presence is required.

2.5. GIS Modelling of the Vegetation Map Units

2.5.1 Development of Spatial Data Layers
Spatial data layers describing the abiotic characteristics of the study area were compiled at 25 metre square grid cell resolution for use in vegetation modelling. These data layers were derived by NPWS in the following way:

• Terrain variables including topographic position, roughness index, wetness index, stream order, slope, aspect, solar radiation indices were derived from a 25m grid digital elevation model supplied by the NSW Land Information Centre.
• Climatic surfaces (Table 2.5) were derived using ESOCLIM (Hutchinson 1989). This provided layers describing mean annual figures as well as coldest and hottest months and driest and wettest months. The Bureau of Meteorology provided rainfall and temperature data.
• Dominant lithological features were grouped from soil landscape mapping within the region (DLWC, 1992-7). Three different layers representing soil parent material were derived. The first represented a 14 scale class which amalgamated mapped landscapes according to their dominant lithology (tertiary alluvium, quartz sandstone, quaternary sand, quaternary sediments, quaternary alluvium, acid volcanics, basic igneous, granitic rocks, sedimentary (coarse grained) sedimentary (fine grained). Secondly, the 14 class lithology was grouped into five classes to provide a relative index of soil fertility. This ranged from basic igneous as the highest fertility through to quartz sandstones at the lowest. The unique soil landscape code provided the third substrate layer.
• Other variables were generated directly in the Arcview GIS system to provide a grid for easting and northing values and to generate layers indicating distance from features including streams, geological features and the coastline.

TABLE 2.5: SPATIAL DATA LAYERS USED IN MODELLING

• GIS COVERAGE	DESCRIPTION
  Altitude	Elevation above sea level (metres)
  Slope	Inclination from horizontal (degrees)
  Aspect	Deviation from grid north perpendicular to slope (degrees)
  Aspect Index	Categorical index of aspect (0: flat, 1: 301-30¡, 2: 211-300¡, 3: 31-120¡, 4: 121-210¡)
  Solar Radiation Index ø January (Summer)	Index representing topographic exposure to solar radiation calculated from slope, aspect, horizon azimuth and latitude. Varies below 100 for sheltered sites and above 100 for exposed sites
  Solar Radiation Index ø July (Winter)	Index representing topographic exposure to solar radiation calculated from slope, aspect, horizon azimuth and latitude. Varies below 100 for sheltered sites and above 100 for exposed sites
  Solar Radiation Index	Continuous index representing topographic exposure to solar radiation calculated from slope, aspect, horizon azimuth and latitude. Varies below 100 for sheltered sites and above 100 for exposed sites
  Wetness Index	Continuous index representing the volume of water draining to a given point in the landscape
  Local Topographic Position (S)	Continuous index (0-100) representing proportional distance between local ridge (100) and local gully (0)
  Neighbourhood Topographic Position (100)	Difference between altitude of a central cell and mean altitude of cells within a 1 x 1 neighbourhood
  Neighbourhood Topographic Position (300)	Difference between altitude of a central cell and mean altitude of cells within a 3 x 3 neighbourhood
  Neighbourhood Topographic Position (500)	Difference between the altitude of a central cell and mean altitude of cells within a 5 x 5 neighbourhood
  Neighbourhood Topographic Position (700)	Difference between the altitude of a central cell and mean altitude of cells within a 7 x 7 neighbourhood
  Neighbourhood Topographic Position (900)	Difference between the altitude of a central cell and mean altitude of cells within a 9 x 9 neighbourhood
  Neighbourhood Topographic Roughness (100)	Standard deviation of altitude within a neighbourhood of 1 x 1 cells
  Neighbourhood Topographic Roughness (300)	Standard deviation of altitude within a neighbourhood of 3 x 3 cells
  Neighbourhood Topographic Roughness (500)	Standard deviation of altitude within a neighbourhood of 5 x 5 cells
  Neighbourhood Topographic Roughness (700)	Standard deviation of altitude within a neighbourhood of 7 x 7 cells
  Neighbourhood Topographic Roughness (900)	Standard deviation of altitude within a neighbourhood of 9 x 9 cells
  Annual Rainfall	Mean total yearly rainfall (mm)
  Rainfall of Wettest Month	Maximum mean monthly rainfall (mm)
  Rainfall of Driest Month	Minimum mean monthly rainfall (mm)
  Minimum Temperature of Coldest Month	Mean minimum monthly temperature (¡C)
  Maximum Temperature of Hottest Month	Mean maximum monthly temperature (¡C)
  5-class Parent Material	Major Fertility Classes based on amalgamation of geologies (Low Fertility- Hawkesbury Sandstone, Quartz sands to High Fertility -Basic Igneous)
  9-class Parent Material	Dominant lithologyÕs
  393-class Parent Material	Raw Soil Landscape Codes
  Vegetation Structure	Vegetation Structure was derived from all available digital mapping in the study area. Classifications were converted to standardise with the 17-class structure used by DLWC.
  Forest Types	Types and mosaics interpreted from aerial photos according to Baur (1989). All vegetation mapping using canopy species descriptors was transcribed to this classification system.
  Distance from Coast	Shortest distance from coast (metres)
  Easting	Australian map grid
  Northing	Australian map grid
  Distance from Basalts	Shortest distance from Basalt soils (metres)
  Distance from Fine Sediments	Shortest distance from Fine sediment soils (metres)
  Distance from Narrabeen	Shortest distance from Narrabeen sandstone soils (metres)
  Distance from stream order ø stream order 1	Shortest distance from stream order 1
  Distance from stream order ø stream order 1-2	Shortest distance from stream orders 1 and 2
  Distance from stream order ø stream order 1-3	Shortest distance from stream orders 1 to 3
  Distance from stream order ø stream order 2-4	Shortest distance from stream orders 2 to 4
  Distance from stream order ø stream order 4-9	Shortest distance from stream orders 4 to 9
  Distance from stream order ø all orders	Shortest distance from all stream orders

2.5.2 Deriving the Extant Vegetation Layer
Mapping of vegetation cover was not available as a consistent data layer across the study area. Sixteen data layers described in Table 2.6 were merged together in ArcView to provide a single coverage to a scale of at least 1:25 000. Some mapping has been captured from larger scale aerial photography and these have been noted in the table where known. Considerable variation is apparent between each of the coverages in terms of geo-referencing accuracy and attributes collected. The most notable discrepancies will emerge between criteria used to map remnant vegetation. Some mapping may delineate remnants less than five hectares others will not map under this size and hence small patches of vegetation may be excluded.

Mapping completed by DLWC (1999) covers a wide range of structural types, native and exotic vegetation by using canopy cover estimates and understorey features. Extant vegetation was delineated by any forest or woodland structure that supported a canopy cover greater than 10%. All mapped areas of Rainforest, Littoral rainforest, Mangrove, Heath and Swamp were included. Areas described as No Mature Trees, Scattered Trees or Trees in Clumps were only included where canopy cover was greater than 10% and contained a visible understorey.

Scale and accuracy of attributes in each of the coverages was unknown. Attributes relating to either vegetation classification or vegetation structure were retained for use in deriving the map of vegetation Map Units (see 2.5.3).

No aerial photo interpretation is available over large areas of Yengo National Park west of the Wollombi Valley. Extant vegetation has been derived for this area using map interpretation of Landsat images taken in 1989 (NPWS, 1990).

No information on the condition of vegetation has been consistently mapped or described in the study area.

TABLE 2.6: DATA LAYERS USED FOR DERIVING EXTANT VEGETATION COVER

2.5.3 Deriving the Vegetation Map Unit Modelling Rules
The technique used to describe the distribution of each ecosystem in the region draws on that used by Keith and Bedward (1999) for the Eden CRA. This technique seeks to derive statistical relationships between sites used to define a map unit and their environmental characteristics. In essence, a series of equations is developed that combines selected values of the environmental data layers to describe the distribution of floristic assemblages. However, in most instances multiple rather then single equations or Ôdecision rulesÕ as it is known, are used to model a Map Unit using GIS.

A new piece of interactive modelling software (ALBERO) developed by (Keith and Bedward, 1998) was used to generate decision rules by statistical induction and expert input. The software is a tool to structure arguments that may describe relationships between the location of ecosystems and spatial variables. These potential relationships, supported by a pre-defined level of statistical significance, can be accepted or rejected by the user. ALBERO builds a decision tree or arms of equations to describe the distribution of all communities in the study area. Building a decision tree is a manual one with ALBERO presenting relationships between Map Units and data layers, and suggesting critical thresholds that continually discriminate one Map Unit from another by using its position in the landscape. ALBERO relies heavily on the user knowledge of the study area.

To use ALBERO a data matrix is needed to provide a raw value for each variable at each site. This is generated using a sites file and an output file known as ÔEnvarsÕ. Final vegetation Map Unit numbers are allocated to each site. At the commencement of decision rule construction ALBERO contains a list of each Map Unit and the number of sites used to describe them. ALBERO then searches for significant splits using the chi square statistic to discriminate between Map Units based on the data matrix values. Each split provides two new branches requiring resolution. At the top of the tree all sites appear. The aim is to reach an end point in the tree to account for a unique environmental niche where one or more sites describing the Map Unit fall. At the start of the tree spatial variables which depict regional variation are used most frequently eg rainfall and broad geology. To resolve variation between Map Units at the end of the tree, finer local scale variables are used such as shelter indices.

There are many combinations of rules that could be used to describe the distribution of Map Units. The process is one of iteration, where review and further review of derived maps is carried out. Fine scale adjustment of rule sets can be made at any time.

2.5.4 Mapping Pre-1750 Distribution of Map Units
Where all sites and Map Units are resolved, the derived rules are then exported and run in the GIS system to derive a map of the Map Units. Each 25 metre grid cell is allocated to a Map Unit using all spatial variables used in the decision rules. The derived map then represents the pre-European distribution of all vegetation assemblages. The map is then cleaned using a filter of communities less than a four by four neighbourhood. That is, the data is smoothed in outlying grids by evaluating the values in the 4 surrounding grid cells and is amalgamated with the dominant neighbouring Map Unit. This removes small slithers and artefacts of the intersects between the GIS layers which are less than 0.25 hectares.

1. Where mapped polygons of existing vegetation were enclosed within mapped units of the vegetation model these were directly attributed the model value unless overstorey species contradicted the diagnostic species of the modelled map unit. In such instances, polygons were reviewed and ascribed an analogous map unit based on proximate sites, field knowledge and community description. Most examples dealt with fine scale swamp communities that were not delineated on soil landscapes given their size, but conformed to characteristics of neighbouring larger mapped swamp patches.
2. Polygons delineating rainforest-dominated canopy were attached to the model based on overlap with modelled rainforest øwet sclerophyll Map Units and sites.
3. Polygons defining mangroves and saltmarshes were directly assigned to the analogous Map Unit.
4. Where API delineated variations in structure within a modelled map unit this was ascribed a structural sub unit. This was prevalent on sandstone and quaternary sand heath-woodland complexes. This recognises that local structural variations are present although floristic composition is relatively constant.
5. Polygons that were well sampled in existing mapping (Tomaree, Glenrock, Awabakal and Lake Macquarie NPWS reserves) were directly ascribed the regional map unit to which the sites conformed.
6. Polygons which mapped vegetation extant only (Biosis, 1999) or were ascribed a single structural value (DLWC, 1999) were cut directly to the modelled map units. The modelled Map Unit of similar structure, which comprised the greatest proportion of each polygon, was ascribed.
7. Where the modelled map unit contained a vegetation structure similar to the mapped vegetation community but different character species the polygon was reviewed using field knowledge and proximate sites. Where information supported the former, the polygon was ascribed to the model. Where the later, the polygon was given the analogous regional map unit. Where it remained uncertain the model map unit was accepted and attributed to the polygon.

All polygons were systematically reviewed in this way. As would be expected some joins between each of the different extant vegetation mapping sources are not smooth. These were resolved where ever possible using the model. Where resolution was not possible unmatched polygons were not resolved.

Slithers and artefacts of the GIS layers were filtered in ARC-INFO. All polygons less than 600 m2 were merged to the largest neighbouring polygon. To assist in removing long slithers larger than this size an index was generated to indicate an area to perimeter ratio (area (m2)/perimeter (m)) for each polygon. Polygons that contained index values less than 10 were merged to the largest adjoining polygon.

2.6 Comparisons of map units to other vegetation classifications

2.6.1 Within Rems Study Area
Comparisons were drawn between some existing classifications and assemblages defined in LHCC REMS as an example of how communities may relate. This was achieved by assessing floristic composition, structural form, geological substrate and distribution. No comprehensive review was made of all communities previously described in the study area.

A match was only drawn where the majority of dominant species were shared in the canopy, mid and lower stratum. This comparison is limited to Eucalypt dominated forests. Some further suggestions have been made where possible between sand and swamp communities based on floristic data presented by Myserscough and Carolin (1986) for adjacent the Myall Lakes area.

Profiles for each Map Unit provide an indication of the extent of similarity with assemblages outside the region

2.6.3 Map Accuracy
FEWG (1987) recommended that all vegetation maps should be evaluated with an explicit validation process. Keith and Bedward (1999) argue that it is important where large scale conservation planning exercises such the CRA use such information as a key information source. Mapping can never perfectly represent all features in the landscape and to this end all maps will be incorrect somewhere. The general test is whether a sample taken describes a Community that matches that identified on the map. Most mapping exercises do not explicitly test for accuracy. It is generally assumed that in the case of qualitatively derived maps the level of reliability will be proportional to the amount of field time available and the skill of the mapper.

Two approaches were used to refine and test map accuracy during this project. Firstly, several iterations of mapping rules were constructed in concert with all available digital vegetation mapping in the study area. Secondly, 54 sites used to describe the vegetation communities were extracted and set aside during the modelling exercise. These sites were selected from those communities that were likely to be most extensive and therefore most difficult to map. These sites were then placed back on the map at the completion of the modelling process. The accuracy of the Community described by the site to that presented on the map was tested using windows of a 100 metre and 250 metre radii.

An interim review period was implemented for the maps to be reviewed by the seven councils. Errors in identifying extant vegetation and model inaccuracies were highlighted. These areas were targeted in a two-week field validation period and the model was subsequently reviewed where problems were identified.

Regional Biodiversity Conservation Strategy
     - User Guide
     - Local Government area Plant Species Lists
     - Module 1 Fauna Surveys
     - Comprehensive technical report
         * Acknowledgements
         * Method
         * Results
         * Discussion
         * References
         * Appendix A
               ~ Survey Form
         * Appendix B
         * Appendix C
         * Appendix D
              ~ Map Unit Profiles
         * Appendix E

Copyright 2003, Hunter Councils Inc as legal agent for the
Lower Hunter & Central Coast Regional Environmental Management Strategy