You are on page 1of 13


River Res. Applic. 21: 991–1001 (2005)

Published online in Wiley InterScience
( DOI: 10.1002/rra.867



CSIRO Division of Land and Water, Urrbrae, SA, 5064, Australia

The lower River Murray in South Australia is highly regulated through weirs and water extraction for irrigation. Management of
the river for environmental purposes requires an understanding of the extent of floodplain inundation from various flows and
weir manipulations. This study aimed to produce a floodplain inundation model for the 600 km long and 1–5 km wide portion of
the River Murray in South Australia from the New South Wales border to Lake Alexandrina. The model was developed using a
Geographical Information System (GIS), remote sensing and hydrological modelling. Flood inundation extents were monitored
from Landsat satellite imagery for a range of flows, interpolated to model flood growth patterns and linked to a hydrological
model of the river. The resulting model can be analysed for flows ranging from minimum flow to a 1-in-13-year flood event for
any month and weir configuration and has been independently tested using aerial photography to an accuracy of approximately
15% underestimate. The results have proven the approach for determining flood inundation over a large area at approximately
one-tenth of the cost of detailed elevation and hydrodynamic modelling. The GIS model allows prediction of impacts on infra-
structure, wetlands and floodplain vegetation, allowing quantitative analysis of flood extent to be used as an input into the
management decision process. Copyright # 2005 John Wiley & Sons, Ltd.
key words: flood inundation; modelling; hydrology; GIS; remote sensing; River Murray

The River Murray is Australia’s largest river and plays a vital role in Australia’s economic and environmental
resources. The Murray–Darling Basin catchment occupies one-seventh of Australia and contributes 70% of
Australia’s irrigated crops and pastures. The River Murray provides 100% of Adelaide’s water supply in dry
months and is the major water supply source to the towns in the basin. The study region receives almost all of
the catchment run-off, minus the water extracted due to irrigation and evaporation, as all tributaries including
the Darling River enter the River Murray further upstream.
Management of the River Murray in southeastern Australia has been implemented to mitigate large floods and to
protect infrastructure, while maintaining storages for regular water supply to irrigators. Concerns over river health
have increased the attention on environmental flow strategies over the past decade. The focus is on releasing and
managing flows to provide environmental benefits to the floodplain, wetlands and in-stream water quality. No
quantitative tool for measuring the extent and volume of flood inundation existed for the lower part of the River
Murray prior to the original work (Overton et al., 1999). This project was commissioned to create such a model to
provide input into the management decision process. Previous studies on modelling river flow and inundation have
been undertaken for wetlands on the Darling River in New South Wales (Shaikh et al., 2001) and floodplains on
Roanoke River in North Carolina (Townsend and Walsh, 1998) and the River Murray in Australia (Overton et al.,
1999; Frazier et al., 2003).
Spatial decision support is one of the main roles of Geographical Information Systems (GIS) and provides an
excellent framework for the integration of multi-criterion evaluation results (Taylor et al., 1999; Jankowski et al.,
2001). The flood inundation model was developed using a spatial information system framework that allowed the
integration of non-spatial river flow models with the extent of flood inundation. Riverine ecosystems benefit from
spatial analysis studies because they encompass three important temporally dynamic spatial dimensions (along the
*Correspondence to: I. C. Overton, CSIRO Land and Water, Urrbrae, SA, 5064, Australia; School of Earth and Environmental Sciences,
University of Adelaide, SA, 5005, Australia. E-mail:

Received 12 October 2004

Revised 13 December 2004
Copyright # 2005 John Wiley & Sons, Ltd. Accepted 19 January 2005

channel, river height and floodplain width) and the physical and ecological processes taking place are complex.
Much work has been done on the longitudinal (upstream and downstream) dimension examining the ecological
impacts of river regulation on native flora, fauna and the physical changes occurring in the littoral zone (Stanford
et al., 1996). However, long-term modification of flow rates, changes to the frequency of flooding events and altera-
tion to the timing of flows have now been identified as causing degradation beyond the littoral zone into both the
lateral and vertical dimensions (Young, 2001). This is particularly prevalent in the River Murray ecosystem where
rising saline groundwater has contributed to land degradation throughout the region and increased accession of
saline groundwater to the river (Jolly, 1996). Furthermore the semi-arid nature of the Murray floodplain means
that the variability in the size and timing of flows is extreme and unpredictable (Walker and Thoms, 1993). Con-
sequently, river regulation in the Murray has severely impacted the balance of physical and ecological processes
that maintain this unique riverine environment. Managing the river system to balance social, economic and envir-
onmental requirements, when resources are scarce, requires quantitative analytical tools (Young et al., 2000). This
paper describes the development of a predictive model to assess the effect of flow management options on flood-
plain inundation.

Mapping flood inundation

The degradation of the River Murray has increased interest and the need for information and predictive model-
ling of river health and the impacts from management and land use options. An Environmental Flows Decision
Support System (Young et al., 2000) has been developed to look at a number of river environmental factors such as
area of wetland and aquatic and terrestrial fauna. This model has been advanced into the Murray Flow Assessment
Tool (Young et al., 2003). Neither model, however, looks at the extent of varying flood sizes and considers flood-
plains as one unit with a commence-to-fill flow and a rate of fill.
Remote sensing is particularly useful for monitoring flood extents because it provides basic data more cheaply
and efficiently than ground-based methods (Whitehouse, 1989). Previous studies to determine the aerial extent
of flooding have commonly involved optical satellite image analysis (Usachev, 1985; Walker et al., 1986;
Townsend and Walsh, 1998; Overton et al., 1999; Shaikh et al., 2001; Sheng et al., 2001; Frazier et al., 2003),
radar remote sensing (Townsend and Walsh, 1998) or an integration of remote sensing and GIS (Brivio et al.,
2002). These methods have proved to be very useful and economical for large-area flood analysis. More detailed
studies have used digital elevation models to create a floodplain surface that can be inundated at certain river
heights (Townsend and Walsh, 1998). Elevation methods are particularly useful for predictive studies of changing
flow paths through the floodplain by manipulating flow barriers. However, surface modelling may not give the best
representation of flood inundation, as there are numerous impediments and small channels across a predominantly
flat floodplain. The modelling of flood inundation from surface elevation also requires detailed information on
stage heights, backwater curves, flow impedances and roughness coefficients as simple height levels are insuffi-
cient in this dynamic environment.
Due to the high level of control over flows in this region, the river exists as a series of pools under most con-
ditions. Floods rise and fall in days or weeks and last for several weeks to months, meaning that the extent of flood
inundation for a given flow has usually extended to its full potential extent before the flood peak passes. This type
of regulated river is therefore modelled sufficiently from images of various events. Upstream in the unregulated
reaches, stream flooding is very rapid and sporadic and two flood peaks of equal magnitude are unlikely to create
the same extent of inundation (Frazier et al., 2003).

Study area
The study area was chosen to capture the floodplain environments of the River Murray in South Australia from
the New South Wales border to Wellington (Figure 1). This is generally referred to as the lower River Murray and is
highly regulated by six weirs (which are situated alongside locks for river navigation—the total structure is usually
referred to as a lock) and the barrages at the mouth of the river. It has large extents of floodplain and is a discharge
area for the regional saline groundwater system of the Murray–Darling Basin. The ecological significance of the
Chowilla floodplain, the largest floodplain in this region, is highlighted by its listing under the UNESCO Ramsar
Convention as a Wetland of International Importance (National Environmental Consultancy, 1988) and its

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)

Figure 1. Study area and detail showing locks and localities

inclusion in the list of Significant Ecological Assets of the River Murray under the Murray–Darling Basin
Ministerial Council’s ‘The Living Murray Initiative’ (Murray Darling Basin Commission, 2003). The regulation
of this section of the river allows flows to be manipulated for environmental management to mitigate increasing
salinization on the floodplain and enhance wetland environments.

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)

Figure 2. Cross-section of the local geomorphology

The floodplains along this stretch of the river range from gorge sections below Overland Corner (Figure 1),
2–3 km wide and 30–40 m deep, to valleys 5–10 km wide flanked by broad floodplains (Walker and Thoms,
1993). Figure 2 shows a schematic cross-section of the floodplain. GIS data are available for the study area includ-
ing the floodplain boundary and the permanent water in the river channel and wetlands; however, no data were
available for the flood extent under different flows other than the extent of the 1956 flood (the largest on record)
at approximately 290 000 Ml/day which covered what we now define as the floodplain boundary.

Satellite imagery was obtained for this region at a range of magnitudes of flood events and therefore correct inter-
pretation of these data provides information on the spatial extents of flood inundation. The process of developing
the flood inundation model is described in detail below. Figure 3 shows a schematic flow diagram of the process

Figure 3. Flow diagram of the process of building the Flood Inundation Model

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)

Production of flood inundation maps from satellite imagery

Most studies on identifying water bodies have identified the landscape through image classification procedures
on multi-band imagery (Munyati, 2000). To reduce costs, most of the information required for flood mapping can
be obtained from satellite image bands that detect reflected light from the Earth’s surface in the mid-infrared wave-
length. Radiation in this region is almost completely absorbed by water and hence images show sharp contrast
between the high reflectance in soil and vegetation on dry areas and the very low reflectance of water (Whitehouse,
1989). This study used Landsat Thematic Mapper (TM) satellite data to provide a complete coverage of the study
site at a cell (pixel) resolution of 30 m by 30 m. Landsat TM Band 5 (which detects light waves in the mid-infrared
region, 1.55–1.75 mm) was therefore chosen as the single band to determine the presence of surface water.
Water is detected by isolating the pixels within the image that have a very low reflectance value, indicating that
the light was absorbed and not reflected, for this band. However, other features, especially shadow, also have very
low reflectance in this region of the light spectrum. Detecting water from dark shadow is not possible using a sin-
gle-band image. This problem is compounded by the slightly higher reflectance of turbid or shallow water leading
to these features being detected as non-water. The assessment on the cut-off value of percentage reflectance for
determining surface water in a single-band image (referred to as a density slice) is a judgement made by the analyst
based on the histogram (Figure 4) and image characteristics. Despite the problems involved in individual pixel
misclassification, the method of density slicing a single mid-infrared band to detect surface water has been suc-
cessfully used in previous studies and represents an economical method for determining flooding over large areas
(Sheng et al., 2001). Frazier and Page (2000) found that a simple density slice on Band 5 achieved an overall accu-
racy of 96.9% when compared to aerial photographic interpretation, and proved as successful in delineating water
as a six-band maximum likelihood classification (a multi-band statistical analysis commonly used in remote sen-
sing for image classification). Figure 4 shows the cut-off made for one of the images in the study area.
Twenty-one satellite images (Table I) were chosen to correspond to a range of flow values as recorded at the
gauging station near the New South Wales border (Gauging Station 426 200). The flows range from 3460 Ml/day to
101 900 Ml/day with dates ranging from October 1986 to November 1996. The images were chosen to occur in
periods of rising hydrographs (increasing flow in the river); this reduced the potential for remnant water from
higher flows to still be present on the floodplain, and flood edges are harder to delineate with a receding flood
boundary. Rainfall events were not taken into consideration in choosing the image dates as water not connected
to the river was able in most cases to be filtered out at a later stage.

Figure 4. Histogram of a 102 Gl/day flow with the central line indicating the point at which the reflectance value is cut (density slice) to
indicate water or not water. Two low-reflectance-value peaks represent Lake Victoria (with deep water) and the River Murray (with shallow
turbid water)

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)

Table I. Image path and dates for the 21 satellite Landsat TM images used in the study. All images were from Row 84

Flow Path 97/Row 84 þ 85 Path 96/Row 84 Path 95/Row 84

(Gl/day) Lakes to Overland Corner Overland Corner to Lake Victoria Lake Victoria to Wentworth

Date (Gl/day) Date (Gl/day) Date (Gl/day)

<20 9-Nov-94 6.45 30-Sep-94 3.5 9-Nov-94 6.1

20–30 20-May-89 26.7 20-Nov-89 25.8 8-May-90 27.6
30–40 10-Aug-96 37.8 11-Oct-92 38.2 12-Aug-96 38
40–45 25-Jul-90 40.6 27-Oct-86 43.8
45–50 3-Aug-90 46.7 21-Sep-93 47.4
50–60 20-Sep-96 57 2-Sep-95 55.1
60–70 23-Nov-96 68.3 23-Nov-96 68.3 7-Oct-93 66.4
70–75 31-Jul-89 70.5
75–80 16-Aug-89 78.1 16-Aug-89 77.4
80–100 20-Sep-90 93.4 3-Oct-89 82.4 26-Sep-89 80.2
>100 22-Nov-93 109.9 26-Oct-90 101.6 29-Sep-90 101.9

Registration and spatial accuracy of the imagery

The successful comparison of images to model flood growth with increasing flow relied on accurate image regis-
tration. The satellite images had to be registered to map grid co-ordinates so that flood masks could be used within
the GIS. Registration was performed on the first image using both map co-ordinates and digital data. All subse-
quent images were registered to this image.
The accuracy of the registration was visually assessed as being plus or minus one pixel. Problems with over-
laying registered images must be considered as errors get compounded with additional overlays. This problem is of
special concern in this study as individual pixels were being monitored over time for flooding or not flooding.
Errors in the registration caused a shadowing effect on the edges of the flood extent in some cases.

Flood map editing and coding of the flood masks

The flood inundation images were used to predict the flood extent for a given flow once any anomalies had been
removed. Flood extent anomalies can occur from the identification of water that is not hydrologically connected
to the river, such as remnants from previously larger flows, rainfall events, irrigation practices or misclassified
pixels. These areas were removed in order to determine the exact extent of flooding that would occur at a particular
flow. The process involved removing any identified water outside the 1956 flood boundary; this was the largest
flood event in recorded history. The process also involved investigating the relationship between images of increas-
ing flow rate. It was necessary to remove water that was identified to be present at lower flows but not at higher
flows. The choice of imagery on rising hydrographs assisted with the reduction in such anomalies. This will be an
ongoing process of editing the flood masks as new information becomes available from satellite imagery, aerial
photography and ground recordings. Satellite imagery does not detect every pixel of water within the scene as
some areas are covered by vegetation or have high turbidity or shallow depth. For this reason, the areas that lie
within the river channel itself and areas classified as permanent wetlands (Pressey, 1986) were assigned a unique

Interpolation of flood masks

Satellite images were obtained to provide a range of flows, but it was necessary to consider flows between these
events to provide a more continuous predictive model. Interpolation between the discrete flow intervals was per-
formed to produce finer intervals of flood extent.
The extents of the flood masks of each flow from the satellite imagery provided a boundary line of equal flood
extent. These pixels were interpolated to obtain the flow at all other pixels in the image. There are many ways to

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)

perform this kind of interpolation. The true situation is defined by the local topography of the area which is
unknown in this case. Therefore interpolation of the flow level is similar to the problem of interpolating the land-
scape height at each point. Kriging is the most common surface interpolation method (Burrough, 1986) but is influ-
enced by areas outside the adjacent known boundaries being the two closest satellite masks. It was decided to use
an image morphological process called a ‘marker-based watershed segmentation algorithm’ rather than traditional
interpolation methods to ensure that the information from each satellite image was preserved. The ‘watershed algo-
rithm’ is commonly used in mathematical morphological problems and is often used in relation to topographic
analysis of digital elevation models (Vincent and Soille, 1991). The advantage of this method over other
approaches to contour interpolation is that it can be applied to very noisy data with broken contours. This was
the case here since the data are quantized so coarsely.
Flood interpolation was undertaken by dividing the combined satellite mask image into regions of constant
minimum flow. Using this method, interpolated values were ensured of lying within the minimum and maximum
flow bounds by interpolating each region independently. In each region, contours of equal flow are interpolated
from the boundary points, at which the minimum flow is known. This contour interpolation is repeated iteratively,
each time based on the contours that have already been estimated. Each contour is interpolated using a flooding
simulation extending from the next higher and lower boundaries. Regions having the minimum flow value or repre-
senting land that did not flood at the largest flow were not included and were used as the lower and upper limit of
interpolation accordingly.
The difficulty in applying the watershed algorithm to flood interpolation is in choosing the source and sink
points. The sink points are points adjacent to the region that have the next highest quantized flow level. The source
points are those adjacent points that have the highest flow that is lower than the sink points for this region. The
locations of these sink and source points were chosen to model the behaviour of flood growth across the floodplain
but also to represent the filling of wetlands from a single inflow channel. A set of colour aerial photographs of a
70 000 Ml/day flood were used to validate the growth behaviour of floods, and a small number of edits were made
to the final flood inundation grid.

GIS development
The result of the interpolation stage was a raster grid of cell values that represent the commence-to-flow based
on the flows on the day of the satellite images. This grid was then filtered to remove noise and converted to a
polygon coverage.
Filtering is a technique used to enhance the quality of digital imagery by changing the values of cells in raster
images, and can be used to ease the computation burden of raster to vector conversion by removing isolated pixels
(Trotter, 1991). Filtering uses neighbouring cells to determine the value of the cell in question and can sharpen or
smooth images to emphasize features or reduce the effects of erroneous cells (Wilkinson, 1996). To remove
anomalies in the interpolated flood masks, such as higher or lower value pixels in the middle of lakes, a majority
filter was used. This filter replaced the value of each pixel with the value of the eight nearest neighbours if there is a
majority of neighbouring cells with a different value from the cell.
At this stage of the processing the imagery was still in raster format (pixel-based data). The raster data were
converted to vector data (area- or shape-based data) to reduce the data redundancy and allow for easy retrieval,
updating and generalization of graphics and attributes that are especially important if equations are to be used to
generate models.

Hydrological modelling
Output from the hydrological models was provided as input data to the GIS. The GIS model then used table
look-ups to query the required hydrological parameters that then link to the spatial layers in the GIS. This type
of loose coupling allows existing hydrological models to be run without reprogramming within the GIS (Karami
and Houston, 1996; Sui and Maggio, 1999). As the hydrological parameters do not change, this proved an effective
and economical approach.
River flow and height model. To create a model that would predict flood inundation from a regulated river, the
river height at every kilometre is required. Actual recorded data of river height at certain kilometres from the

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)

River height at kilometre markers from state border to lock 2



River height (m)


Flow rate at border

26 10.00
102 5.00


























River km markers

Figure 5. River levels for kilometres of the river from weir 2 to the SA–NSW border for different flows

mouth of the river, and flow at the gauging station during the flooding events, were obtained from the South
Australian Water Authority, the organization which manages the river in this region. The limited number of
gauging stations meant that a modelled series of backwater curves were also used. These were a series of water
surface elevation curves between weirs for a steady uniform flow computed by the Murray River Level Module
(MURLEV), part of the River Murray Flow and Salt Transport (RMFST) computer model (Water Studies, 1992).
Figure 5 shows the flow curves derived from these methods for a section of the river from the border to weir 2. The
flow rate of 3000 Ml/day equates to a river height referred to as weir pool level.
The river is held higher than pre-regulation by the weirs, which causes the water levels to have a backwater curve
rather than a straight flow down hill. River heights near the mouth of the river are much lower than near the border
due to downfall flow of the river and attenuation of the flow. Once flow levels have reached 60 000 Ml/day the river
is above the maximum height of the weirs and backwater curves become straight.
Flood units. As river heights can be manipulated by the six weirs, predominantly independent of each other, the
floodplain needed to be divided up into regions that could be related to the local conditions in the river. Areas where
the floodplain inundation responded to the same point in the river were identified using flood behaviour and river
morphology as a guide. Regions identified that respond to a particular point in the river were termed Flood Units.
Flood Units were assigned the closest kilometre marker from the hydrological model (trigger) or the trigger that
was assumed would most likely cause that area to flood. As a height difference of several metres can occur between
the upper and lower pool levels of the weirs, behaviour of the flow path around the weir is complex. A weir manip-
ulation trial at Lock 5 demonstrated that the area around the weir responded to changes in the river height below the
weir. There is very little difference between the river height within 2-km intervals in the modelled backwater
curves, which implies that there is probably an error margin of approximately two triggers up- or downstream
in the assignment of a Flood Unit to a trigger. Each polygon for each particular reach now contained attributes
for Reach, Flood Unit, Flow and Trigger.
Once the data had been converted to polygons in the GIS it was clear that each Flood Unit contained small areas
of floodplain that were inundated at the same particular flow. These areas were termed unique Ecological Units as
they will have similar ecology due to their similar elevation and the fact that distribution of floodplain vegetation
communities is strongly related to flooding frequency (Van Der Sommen, 1987).
Hydrological inputs to the model. The same flow rate at the border gauging station can produce different river
elevations along the river depending on the weir elevations, the antecedent conditions and the time of year. Local
river height predominantly determines whether the river will break the banks and flood an adjacent Flood Unit.

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)
Plate 1. Example of seasonal attenuation curves

Backwater Profile for Lock Pool 3,

at 5cm above normal pool level
Flow (ML/day)
13.4 56295
12.8 50000
12.4 40000
12.0 30000
AHD (m)

11.6 20000
11.2 15000
10.8 10000
10.2 5000
9.8 pool level
430 440 450 460 470 480 490 500 510 520
AMTD (km)

Plate 2. Example of a backwater profile for Lock 3

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: (2005)
Plate 3. Example output of the GIS Flood Inundation Model showing an area around Lock 6 at Chowilla with two flood predictions of
60 000 Ml/day (dark) and 100 000 Ml/day (light)

Plate 4. Area of flood inundation versus flow size for the South Australian Border (Chowilla floodplain) to Wellington. The graph shows the
area of inundation compared to the area of permanent water and the extent of the 1956 flood (  258 000 Ml/day)

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: (2005)

Other factors such as antecedent conditions were not included in the model. In the hot dry summer period a flow
rate at the border will be further reduced downstream than in wetter months as water is evaporated and extracted
from the river and less is replaced by rainfall. Curves were generated which show the attenuation of flow at each
weir for a given month. Plate 1 shows an example of the change in flow rate during the year in each lock reach for a
flow of 60 000 Ml/day at the border.
The RMFST model was used to generate backwater curves for a range of weir manipulations and flow values.
The model estimates the stage–discharge matrices for a nominated water level and discharge at the weir, which
defines the upstream water levels to the next weir. In this way relationships can be derived between the water levels
at the downstream and upstream ends of the river segment and discharge. The river level and discharge relation-
ships for each weir reach were established using a linear relationship. Plate 2 shows the backwater curve for the
weir pool of Lock 3 at a level of 5 cm above normal pool level. The precision of the model is 5 cm in the river
height predictions.

Decision support system development

The height of the river at triggers had to be related to the management areas, the Ecological Units derived from
the original flow values and the interpolation process. With area inundated extremely sensitive to slight alterations
in river height, it was decided that a relationship between river height and area of flood inundation should be used
in each Flood Unit. This relationship has the potential to be improved with further image acquisition and statistical
analysis. The model will always be limited to the spatial resolution at which the images are acquired, which for
Landsat TM is 30 m, regardless of the number of images used. If a change in the height of the river results in less
than a 30 m change at the flood boundary, the area of inundation will not increase. Other satellite systems may be
useful in reducing the pixel size and therefore identifying finer changes in flood inundation.
In the region downstream of Lock 3, the height of the river continues to increase without breaking its banks until
it reaches approximately 35 000 Ml/day. As the water flows onto the floodplain there is a gradual increase in area
with increasing flow until the river level rises to reach the edges of the trench (Figure 2). In the region upstream of
Lock 3 the river height increases until it breaks its banks and continues to increase as it fills up wetlands and then
gradually tapers out as the water spreads over the floodplain. The relationships at each floodplain unit were used to
code the flood masks with a river height which would cause the Ecological Unit to be inundated. The height of the
river was assigned to the Ecological Unit for all Flood Units using the Flood Unit trigger as the point that would
cause this Ecological Unit to flood. A dialogue screen was then designed with input fields to ask the user the values
for the flow at the border, the month of the year and the height of each lock. These values are then used to perform a
query which selects all the river height codings in the map layer where the local river height is less than that pre-
dicted for the given flow (attenuated for the time of the year).
The flood inundation model is a steady-state model that predicts the extent of flooding from a given flow on the
first day of the flood. It does not consider the effect of antecedent conditions or the effect of flood duration. Further
research on the wetting and drying behaviour of the floodplain and its wetlands needs to be incorporated into the
model to be able to predict time sequences for management scenarios. The model is limited by its ability to predict
areas of inundation given the same floodplain topography. Changes in elevations due to construction of levee banks
and regulators will cause changes in the area of inundation not modelled by the Flood Inundation Model. The
model can provide wetting and drying cycles for wetlands, days since last flooding, average return period, volume
of water, extent of inundation and other hydrological indicators that could be linked to floodplain ecological pro-
cesses. Better understanding is needed to determine, or at least demonstrate in most cases, these linkages between
the hydrological indicators and the benefits to the environment.

The Flood Inundation Model produced has been used as the basis for an environmental flow strategy in South
Australia (SKM and Mapping and Beyond, 2000) and a number of other studies on the hydrology and ecology
of the floodplain. The Flood Inundation Model was used to predict the area of inundation expected to occur as
a result of the raising of Lock 5 near Renmark in 2000 (DWLBC, 2002). Aerial photography was obtained during

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)
1000 I. C. OVERTON

the raised weir and was interpreted for area of inundation. The spatial accuracy of the model in predicting the area
inundated was estimated to be 15% underestimate of the actual inundated area from the photography. A database of
the surveyed heights of 126 bores on the floodplain was used to independently test the accuracy of the height data
derived by the model. The mean difference between the surveyed height and modelled height was 0.07 m with a
standard deviation of 0.88 m. Plate 3 shows an example of the GIS Flood Inundation Model.
The initial stages of the project provided us with Ecological Units and the height of the river at a particular
trigger that would cause these areas to flood. The hydrological modelling allows the monthly simulation of a flow
from the border under different weir configurations, with the GIS providing the spatial view of the area inundated.
The inputs to the decision support system are the flow at the border, the weir configurations of all six weirs and the
month. The outputs of the model will be the river heights at each trigger kilometre and the area of inundation. The
flood inundation model created is in a GIS framework and has incorporated layers such as riparian vegetation,
transport and water infrastructure and major wetlands. The science of incorporating spatial patterns in decision
support systems through the integration of GIS and process models is suggested as a critical tool for the future
of information integration in the environmental sciences (Taylor et al., 1999; Hendriks and Dirk, 2000).
The total area of inundation of the floodplain in South Australia, including the whole of the Chowilla floodplain,
which is partly in New South Wales, down to Wellington, is 117 827 hectares (downstream of the gauging station
on Figure 1). The floodplain is defined as the extent of the 1956 flood which was approximately 258 000 Ml/day.
Plate 4 shows the increasing area of inundation of the River Murray floodplain from 5000 Ml/day to 100 000 Ml/
day. The area of permanent water is approximately 24 000 hectares and is made up of the river channel, wetlands
and anabranch creeks.

Satellite imagery of known flood events has been interpolated to create a model of flood growth. This flood growth
has been linked to a hydrological model of the river, making predictions of the extent of flood inundation from
given river flows and manipulation possible. The GIS framework could be utilized to determine flow patterns and
losses across the floodplain, incorporating wetlands as sources and sinks to create a more dynamic model of flood-
ing (Costelloe et al., 2003).
The integration of the hydrological model with the GIS has enabled simulation for both scientific research and
policy management. The visualization, quantitative analysis and spatial correlation of environmental and infra-
structure data of the GIS has improved the usefulness of the hydrological modelling.
The model has been used successfully to predict flood inundation in a weir manipulation trial (DWLBC, 2002)
and has been used as the basis for the development of a flow management strategy for the River Murray in South
Australia (SKM and Mapping and Beyond, 2002). The model has also provided a useful surface elevation model in
further floodplain modelling work. Current work on modelling floodplain vegetation health affected by soil sali-
nization processes will be linked to the flood inundation model to improve its usefulness as a decision support
system. Further research on the behaviour of floods on the floodplain and in wetlands is required before it can
be used as a temporal model.


The author wishes to acknowledge funding through the Murray Darling Basin Commission, the River Murray
Catchment Management Board and the South Australian Water Corporation (MDBC Natural Resource Manage-
ment Strategy R6045 project). Assistance on the initial project design was provided by Brenton Erdmann and Bob
Newman of the South Australian Water Corporation. Jamie Sherrah performed the mathematical image interpola-
tion, while Suzanne Slegers and Jane Lawley, of Mapping and Beyond Pty Ltd, assisted with the data manipulation.


Brivio PA, Colombo R, Maggi M, Tomasoni R. 2002. Integration of remote sensing data and GIS for accurate mapping of flooded areas. Inter-
national Journal of Remote Sensing 23(3): 429–441.
Burrough AP. 1986. Principles of Geographical Information Systems for Land Resources Assessment. Oxford University Press.

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)

Costelloe JF, Grayson RB, Argent RM, McMahon TA. 2003. Modelling the flow regime of an arid zone floodplain river, Diamantina River,
Australia. Environmental Modelling and Software 18(8–9): 693–703.
DWLBC. 2002. Lock 5 Flow Enhancement Trial. Murray-Darling Division, Department of Water, Land and Biodiversity Conservation, South
Australia. Report No. 1.
Frazier PS, Page KJ. 2000. Water body detection and delineation with Landsat TM data. Photogrammetric Engineering and Remote Sensing
66(12): 1461–1467.
Frazier PS, Page KJ, Louis J, Briggs S, Robertson A. 2003. Relating wetland inundation to river flow using Landsat TM data. International
Journal of Remote Sensing 24(19): 3755–3770.
Hendriks P, Dirk V. 2000. From geographical information systems to spatial group decision support systems: a complex itinerary. Geographical
and Environmental Modelling 4: 1.
Jankowski P, Andrienko N, Andrienko G. 2001. Map-centred exploratory approach to multiple criteria spatial decision making. International
Journal of Geographical Information Science 15(2): 101–127.
Jolly ID. 1996. The effects of river management on the hydrology and hydroecology of arid and semi-arid floodplains. In Floodplain Processes,
Anderson MG, Walling DE, Bates PD (eds). John Wiley and Sons: New York; 577–609.
Karami HA, Houston BH. 1996. Evaluating strategies for integrating environmental models with GIS: current trends and future needs. Com-
puters, Environment and Urban Systems 20(6): 413–425.
Munyati C. 2000. Wetland change detection on the Kafue Flats, Zambia, by classification of a multitemporal remote sensing image dataset.
International Journal of Remote Sensing 21(9): 1787–1806.
Murray Darling Basin Commission. 2003. The Murray Darling Basin Commission website. [November 2003].
National Environmental Consultancy. 1988. Chowilla Salinity Mitigation Scheme—Draft Environmental Impact Statement. Report prepared by
National Environmental Consultancy for the Engineering and Water Supply Department of South Australia: Adelaide.
Overton IC, Newman B, Erdmann B, Sykora N, Slegers S. 1999. Modelling floodplain inundation under natural and regulated flows in the Lower
River Murray. Second Australia Stream Management Conference Proceedings, 8–11 February 1999, Adelaide, Australia: 493–498.
Pressey RL. 1986. Wetlands of the River Murray below Lake Hume. Environmental Report 86/1. Murray-Darling Basin Commission: Canberra.
Shaikh M, Green D, Cross H. 2001. A remote sensing approach to determine the environmental flows for wetlands of the Lower Darling River,
New South Wales, Australia. International Journal of Remote Sensing 22(9): 1737–1751.
Sheng Y, Gong P, Xiao Q. 2001. Quantitative dynamic flood monitoring with NOAA AVHRR. International Journal of Remote Sensing 22(9):
SKM and Mapping and Beyond. 2002. Decision Support System for Managing Flows in the River Murray, South Australia. Report prepared for
the River Murray Catchment Management Board, the South Australian Department of Water, Land and Biodiversity Conservation, and the
Natural Heritage Trust.
Stanford JA, Ward JV, Liss WJ, Frissell CA, Williams RN, Lichatowich JA, Countant CC. 1996. A general protocol for restoration of regulated
rivers. Regulated Rivers: Research and Management 12: 391–413.
Sui DZ, Maggio RC. 1999. Integrating GIS with hydrological modelling: practices, problems and prospects. Computers, Environment and
Urban Systems 23: 33–51.
Taylor K, Walker G, Abel D. 1999. A framework for model integration in spatial decision support. International Journal of Geographical
Information Science 13(6): 533–555.
Townsend PA, Walsh SJ. 1998. Modelling floodplain inundation using an integrated GIS with radar and optical remote sensing. Geomorphology
21: 295–312.
Trotter C. 1991. Remotely-sensed data as an information source for geographical information systems in natural resource management: a
review. International Journal of Geographical Information Systems 5(2): 225–239.
Usachev VF. 1985. Evaluation of flood plain inundations by remote sensing methods. Hydrological applications of remote sensing and remote
data transmission. Proceedings of the Hamburg Symposium, August 1983, Goodison BE (ed.). Publication No. 145. International Association
Hydrological Sciences: Washington; 235–243.
Van Der Sommen FJ. 1987. A Review of Components and Processes Influencing the Ecological Status of the Vegetation Component of the
Riparian Ecosystems of the River Murray Floodplain in South Australia. South Australian Department of Environment and Planning Report.
Vincent L, Soille P. 1991. Watersheds in digital spaces: an efficient algorithm based on immersion simulations. IEEE Transactions on Pattern
Analysis and Machine Intelligence 13(6): 583–598.
Walker KF, Thoms MC. 1993. Environmental effects of Flow Regulation on the Lower River Murray, Australia. Regulated Rivers: Research and
Management 8: 103–119.
Walker RD, Aubrey MC, Fraser SJ. 1986. Remote Sensing in Australia. Canbec Press: Melbourne.
Water Studies. 1992. River Murray Flow Model, Vol. 3. User Manual. Department of Engineering and Water Supply.
Whitehouse G. 1989. Flood monitoring and floodplain studies. In Remote Sensing in Hydrological and Agrometeorological Applications, de
Vries D (ed.). COSSA Publications: Canberra.
Wilkinson GG. 1996. A review of the current issues in the integration of GIS and Remote Sensing data. International Journal of Geographic
Information Systems 10(1): 85–101.
Young WJ (ed.). 2001. Rivers as Ecological Systems: The Murray-Darling Basin. Murray Darling Basin Commission: Canberra, Australia.
Young WJ, Lam DC, Ressel V, Wong IW. 2000. Development of an Environmental Flows Decision Support System. Environmental Modelling
and Software 15: 257–265.
Young WJ, Scott AC, Cuddy SM, Rennie BA. 2003. Murray Flow Assessment Tool: A Technical Description. CSIRO Client Report for the CRC
for Freshwater Ecology.

Copyright # 2005 John Wiley & Sons, Ltd. River Res. Applic. 21: 991–1001 (2005)