Interim Marine Bioregionalisation for Australia
Towards a National System of Marine Protected Areas
Contents
The data acquisition and assembly components of this project are
time consuming and a 3-5 year time frame rather than the 18 months
was needed to incorporate all data sets. Consequently our methodology
has been fast tracked to meet this extremely tight deadline.
10.1.
Bathymetry Refinement/Generation
In order to generate the species polygons, the bounds for the
species were clipped with bathymetric contours generated from
a number of sources detailed below.
Dataset 1 (DS1), provided by Dr Neil Hamilton of CSIRO Division
of Wildlife and Ecology, was a 1:100000 series of contour lines
at depths of 0, 20, 50, 100, 150, 200, 250 and 300 metres. The
sizes of the contour maps range from 12 megabytes (Mb) for the
coast down to 2Mb for the 300 metres. Unfortunately this dataset
contained gaps in which there were no data and numerous contours
which were not closed ("dangling nodes").
Dataset 2 (DS2) was a commercially available set of contour lines
(GEBCO) at depths of 0, 50, 100, 200, 300, 400 and 500 metres
and thereafter in increments of 500 metres. In the area of interest,
except for the coastline, those contours below 200m were deemed
to be unusable due to the lack of data.
Dataset 3 (DS3) was a point dataset obtained, from the Royal Australian
Navy Hydrographic Office, to supplement those areas, mainly the
Cape York region, which were missing in the other two datasets.
10.1.1.
Accuracy
DS1 was considered to be highly accurate, however a spheroid of
projection was not given and as such could be considered to be
accurate at best to within +/- 200m. DS2 had a spheroid of WGS84,
however, when overlaid with DS1 and a position of known location
this dataset was inaccurate by approximately 500m. DS3 could not
be checked as no exact position was known near its bounds.
10.1.2.
Dataset Merging
Due to the gaps in the two contour line based datasets, it was
decided to combine all three datasets into one coherent set of
contour lines. Initially it appeared the best way to do this was
to generate a TIN (Triangular Irregular Network) which was then
used to extract the contour lines. Various subsets of the three
datasets were used in order to generate tins. These produced varied
results. Unfortunately due to the algorithmic difficulty involved
in generating a TIN from point and line data sources even the
best of the resultant TINs were less than satisfactory for the
majority of areas.
It was therefore decided to use the best from each dataset and
patch a segment of the TIN into those areas which were missing
and considered too large to be simply connected by a line. Boxes
around the areas which required patching were drawn. A contour
line at the required depth was extracted from the TIN. The contour
line was then clipped with the boxes and cleaned up using ArcEdit.
The resultant coverage was then added to the original coverage.
This allowed for a consistent/accurate contour line except for
the areas previously missing which were filled with the interpolated
TIN data. The contour lines chosen for the final bathymetry set
were 0, 20, 50, 100, 200, 500, 1000, 2000 and 3000 metres as reasonable
bathymetric information was already available for these depths.
10.1.3.
Reduction of bathymetry resolution and resultant errors.
Due to problems of storing and generating the species polygons
it was decided to decrease the resolution of the contour lines.
This was done by iteratively 'generalising and dissolving' the
contour lines until their size became less than a predetermined
level. For our purposes this was chosen to be a total coverage
size of 200 kilobytes. Since generalising a coverage removed only
vertices and not nodes of an arc some of the resultant coverages
were found to be inconsistent with adjacent coverages. i.e.
in some cases the 20m and 50m contour lines cross. It should be
noted however that these problems were also apparent in the original
datasets.
10.1.4.
Addition of Slope Flag.
Since the generation of the species polygons involved clipping
and erasing with the bathymetric contour lines some indication
of the slope was required to indicate regions. A coverage containing
regions still maintains polygon topology except that meaning is
given to those polygons which are considered part of the region.
That is, a region is typically a subset of the polygons. An example
of this is the 0m and 20m contour lines for Tasmania where only
the 0m to 20m region is considered a valid area. i.e. inside
of the 0m is invalid. In order to generate regions from polygons,
some indication of which polygons to include are needed and this
was done by adding an attribute (flag) to the bathymetry coverages.
This was one for those polygons which contained depths less than
its stated depth and zero for all others.
The bathymetry coverages were then used as clip-and-erase polygons
with the bounding box. All polygons in the resultant polygon coverage
with a flag value not equal to zero were chosen as being part
of the region.
The limitation of this technique is that it is only workable for
areas near to the coast. This is because the assumption is that
the maximum depth is used as the clipping polygon and the minimum
depth is used as the erasing polygon. As such all data outside
the maximum depth polygon are removed.
10.1.5.
String Generation
For the purposes of the string analysis (described below) a sample
coastline from the mainland of Australia and Tasmania was extracted
from the weeded bathymetry data. A route system for each area
was then created using ArcEdit. Although the documentation
suggested that the creation of a circular route system was possible
further investigation revealed that a problem existed in determining
a point of reference. This made it necessary to 'cut' the arc
in order to define the start of the route system. ArcEdit
was used to select the arcs and create a route system. The point
chosen for the mainland was near the tip of Queensland, just to
the west of Cape York. For Tasmania a cut in the Tamar River was
chosen as the origin.
The output from the string generation consisted of lines containing
the species number, start position and end position. There were
multiple lines for some species with disjoint distributions, or
where the bathymetry contour line cut the box in more then two
places. Species which spanned the 'cut' were shown as two lines
with the ending point of the first line being the length of the
coastline and the starting point of the second line being zero.
The initial algorithm used an aml script to determine the
intersections of the coastline with the bounding box defined in
the Oracle database. These intersection points which were in units
of degrees along the coast were then written out to a file. Any
polygons which did not intersect with the coast were logged. The
species from the BioTax '96 Workshop were handled
in a similar manner except that the coverages used for the intersection
were entered interactively and as such were not restricted to
boxes.
10.2.
Fish distribution data
Compilation:
The GIS currently comprises data gathered on Australian fish distributions
represented by ranges and points (which can be displayed on map
templates) which were captured in the form of latitude and longitude
co-ordinates, depths, other information associated with the records,
and record sources (see below). The initial data were entered
into an Excel spreadsheet. The information was then periodically
written out to a delimited text file and then imported into Oracle.
Species are tied to relevant information by a numerical coding
system (CAAB) used broadly by Australian fisheries agencies. There
is potential for expanding the GIS to incorporate other types
of information relating to species or for using captured information
for other purposes.
Currently the database contains 7540 polygons and 3229 point records,
representing a total of 3007 distinct species. These data was
assembled through examination of the information contained in
scientific literature selected by Peter Last and Martin Gomon.
A reference list for this literature appears in section 19 of
this report.
For each literature record of a species' distribution, information
was entered into the database in the following fields:
The information which has been used to create the distribution,
and the fields are as follows:
id record identification number in the Oracle database
name species name
species species number from the CAAB list
ref identification number and author(s) of the source for this
information (e.g. journal article, book) (please find enclosed
list of references)
lata latitude of point A in degrees.minutes (see diagram below)
longa longitude of point A in degrees.minutes (see diagram below)
desca description of point A (e.g. Kangaroo Island, NETas)
gradea reliability of point A (1=lat/long given in source, 2=lat/long
description given in source and lat/long derived from gazetteer,
3=no lat/long or description given, so coordinates chosen
arbitrarily)
latb latitude of point B (see diagram at right)
longb longitude of point B (see diagram at right)
descb description of point B
gradeb reliability of point B (as for gradea)
depthmin minimum of depth interval, in metres
depthmax maximum of depth interval, in metres
ecol ecological classification (e=estuarine, cm=coastal marine,
sd=shelf demersal, sp=shelf pelagic, sld=slope demersal,
ep=epi-pelagic, mb=meso-bathypelagic, a=abyssal)
disttype distribution type (0=normal, 1=larval/juvenile, 2=reproductive,
3=extralimital)
remarks any queries or problems with source or information (for example,
depth? in this field indicates that the depth range is considered
suspect)
lastmod date of most recent modification of this record
mod what the most recent modification consisted of
quality not used
In most cases we have more than one reference or source of information
for each species.
Points A and B are used by the GIS software to draw a rectangle
(see above diagram). All areas within the rectangle within the
specified depth interval form a polygon which represents all or
part of the species' described distribution. Sometimes the definition
of several rectangles was necessary to describe a species' distribution.
Distribution maps represent the geographical range of a species
at one moment in time (Miller, 1994). Mapped distribution is the
most complete possible, incorporating larval, reproductive and
migratory ranges in the rare cases in which such information is
available. Consequently, individuals of the species may be absent
from parts of the mapped distribution at certain seasons/years/life
stages. In addition, biodiversity hotspots may be temporally and
spatially dynamic, following upwelling events or seasonal changes
in species distributions (Yatsu, 1995).
Reliability scores were added either by the taxonomy group as
a whole during the BioTax '96 Workshop or later
as part of a cooperative effort on the part of Peter Last, Martin
Gomon and Patricia Kailola.
The information stored for the point distributions are as follows
:-
rec Oracle record number
srcfname Name of source
species Species number from the CAAB list
id Reference identifier used by museum/source
latitude
longitude
mindepth Minimum depth
maxdepth Maximum depth (=mindepth for most cases)
remarks
type Type of record (0=Unknown,1=Museum,2=Survey,3=Literature)
A list of potential Workshop participants was compiled to cover
Australian-based scientists with significant expertise in fish
taxonomy and distribution or with strategic involvement in the
Bioregionalisation Project. Due to financial constraints, the
provisional list of 45 had to be reduced to 16, the foremost selection
criteria being the ability to provide maximum input into the refinement
of species distributions and the possession of skills required
for an efficient operation of the Workshop. The list of Workshop
attendees is inside the cover of this report.
Participants were responsible for assisting in determining the
primary and secondary geographical distributions for species occurring
within the four coastal ecosystems inshore of the continental
shelf break. Some vetting of distribution maps was also required
before the Workshop.
The number of species distributions vetted and refined from the
BioTax '96 Workshop totals 1036 species with 175
of these species having breeding range information entered. The
bulk of the species distributions are for the coastal and shelf
bands (4925 records for coastal marine band, 1978 for shelf demersal,
499 for shelf pelagic). Reliability scores were also assigned
to each species, rating the probable accuracy of the described
distribution.
Data reduction process (A,B,C,D lists)
The method by which the Key List of species was selected for the
workshop is diagrammatically illustrated in Figure 10-1.
From the data set of ca. 3200 species, a rapid assessment
technique using an information index based on genera (rather
than species) provided relative scores of "potential importance".
These lists were divided into ecosystem lists where only the genera
represented in each of the 4 ecosystems (estuarine, coastal, shelf
demersal, pelagic) were ranked. From these lists the "top"
300 species were collated into a "priority list" of
about 900 species. From this list the "well-defined species"
list (list D) was extracted, as those with high reliability scores
were felt to not need examination at the workshop due to time
restraints. From the "poorly-defined species" list those
groups which probably could not get more information added to
them at the workshop (such as the sharks and rays) were extracted
(into an "excess key species list" (list B)). The remaining
"poorly-defined species" were combined with species
from the "String List" (list C) which was about 350
species which had been selected out from the original data set
due to their narrow ranges. The result was the final "Key
List" (list A) which was covered in the workshop.
Figure 10-1 Diagrammatic representation
of the reduction for the data list of polygons used to preselect
species for consideration by the BioTax '96 participants.
These reliability scores (c.f. 8.1) were subsequently used
as a further selection criterion in grouping species for the purposes
of bioregional analysis.
Genera were ranked from lowest score to highest within each of
the four ecosystems, and the resulting priority list was used
in concert with a listing of all polygons in the database to determine
(i) for which species and genera more distributional information
was required, and (ii) important species for scrutiny by experts
at the BioTax '96 Workshop.
The polygon refinement as part of the workshop involved manual
modification on the paper maps with the modifications being entered
into the GIS as part of the post-workshop task.
10.2.2.
Interactive Polygon Editing
Refinement of the polygons was done with an aml script
and menus. Although it can be thought of as editing, at no stage
were the original polygon coverages modified. Two methods were
evaluated to handle the refinement. Both involved the user interactively
creating a bounding polygon which was then clipped with the appropriate
bathymetry contour lines in order to generate the final polygon.
The first was to create complete new polygons whilst the editing
was being done. The advantage to this is that the user gained
some feedback as to possible errors in the bounds of the data.
The main disadvantage was that this was time consuming.
The second method was to store the bounding polygon which was
entered and re-create the polygon coverage at a later date. The
obvious advantage was the removal of the clipping and erasing
operations which allowed significant speed improvements. Also
since the majority of the polygon entry involved crossing the
coastline the generation of data for the string analysis was significantly
less complicated.
Polygons which were entered were named with a serial number so
that multiple entries could be manually generated if necessary.
This removed the possibility of an error removing a previously
valid polygon. As such, a requirement of the second approach was
to maintain links with the coverage being 'edited' and its attributes.
This was accomplished with an ascii logfile which detailed those
operations which were entered and the polygons to which they applied.
The logfile contains the following fields :-
B Species Number
C Initial Coverage Filename
ZD Date Of Entry
I Oracle ID Of Selected Record
R Remark/Comment Field
G Name of Coverage To Use In Generation Of Polygon
N Minimum Depth
X Maximum Depth
D Name of Coverage To Use In Deletion Of Polygon
ZB Breeding Range Polygon (True Or False)
F Finished Species Number
T Terminated Species Coverage
A preprocessing program was used to filter out those entries which
had been terminated. Three scripts were written to convert from
the logfiles into coverages.
The first was a simple union and rename. This script unioned the
bounding coverages contained within each record and gave it an
appropriate name. None of the original Oracle generated coverages
were used in the process. This version was ideal for string analysis
as it gave a bounding polygon for the species which crossed the
coastline at appropriate locations.
The second script was similar to the first except that the coverages
were clipped and erased with bathymetric contours before the union.
This generated the polygons which could be thought of as 'final'
and were ideal for viewing and mapping purposes.
The third script was envisaged to be the same as the second except
that the original Oracle generated coverages were to be used as
a starting point. In this scenario the user chose those records
which most closely matched the final distribution and proceeded
to add and delete polygons to this set. This script was not fully
developed as the entered polygons were based on zero records being
selected. That is, the polygons entered were to be considered
independent of the original Oracle records. Time consumption in
the execution of this script were also considered inappropriate
due to the large amount of merging involved.
The procedure for modifications at the workshop was essentially
identical for each session:
- the distribution map for the species to be discussed was called
up in ArcPlot and displayed on a projection screen via a computer
screen projection device
- any additional information such as the references used to
produce the polygon and whether depth ranges were given, or guessed,
was provided to the delegates
- discussion between delegates would then result in a consensus
decision on the most appropriate or accurate distribution and
depth ranges for the species and breeding range (if known)
- these were then recorded on the printed distribution maps
for each species for editing later.
The method of vetting polygons was as follows:
Polygon ranking Future action
Reasonably reliable or none
better
Differences of opinion none, S, GL
exist
Some uncertainty GL, S, SE
No idea SE
(S - specialist at workshop; SE - specialist elsewhere; GL - Gomon/Last)
The modifications noted on the printed distribution maps were
used for editing the polygons. Modifications were also provided
by Martin Gomon, Peter Last and Patricia Kailola during the weeks
subsequent to the workshop. Delegates also provided feedback after
the workshop by checking their records for information that could
contribute further to what had been discussed.
10.2.4.
Reliability assessment
It is widely appreciated that information on the breeding range
of a fish species is more useful than that of its total range.
More mobile species can have greater extralimital occurrences
than less active ones making a determination on the naturally
occurring range of the species that much more difficult. Unfortunately,
the breeding ranges of most marine species are ill defined or
undocumented. Apart from a few commercial species, impressions
are at best based on the understanding of the primary range of
adult individuals.
As the reliability of known distributional and breeding range
data varies dramatically between species, an attempt was made
to quantify differences. Two classifications that consider the
degree of reliability of distributional polygons were devised:
a total range reliability index (R) to define the basis of the
known distribution of species, and a breeding range index (B).
All shelf species were scored according to both B and R criteria
below.
Reliability Index (R) -
The broad distributional polygon of a species is based on:
- specific studies that examine the range limits of the species
- good survey/museum data and observations
- adequate information - but which may suffer from taxonomic
problems or patchy sampling
- basic information only - anecdotal sources, unreliable survey/sampling
data, etc.
- very little or no information
Where the understanding of the distribution of a species was considered
to be basic or worse (i.e. criteria 4 or 5), but where
one limit of the polygon (occasionally more where disjunct distributions
exist) was considered to be adequately known (or better), then
the reliability score was suffixed by compass direction of the
reliable limit (i.e. 'n', 's', 'e', 'w'). These boundaries
provided additional information for the delineation of zootones
and boundaries.
Breeding range index (B) -
The breeding range of a species was defined as either:
- well defined - based on reproductive studies of the species
- not based on specific reproductive studies but is known from
other research
- likely to mirror defined adult range of the species
- similar to extended species range but with outliers removed
- currently unknown but could be postulated
- unknown
Disjunct distributions were indicated by suffixing breeding range
scores with a 'd'.
Next Chapter: 11. Biological Regionalisation: Analytical Strategies