Journal of Hydrology (2006) 329, 75 97

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jhydrol

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge on a semi-arid oodplain, Australia
Rebecca Doble
a b

a,b,*

, Craig Simmons a, Ian Jolly b, Glen Walker

Flinders University of South Australia, Australia CSIRO Land and Water, PMB 2, Glen Osmond, South Australia 5064, Australia

Received 16 June 2004; received in revised form 31 January 2006; accepted 3 February 2006

KEYWORDS
Evapotranspiration; Spatial distribution; Floodplain geomorphology; Vegetation; Numerical modelling; MODFLOW 2000

Summary Insight into spatial patterns of groundwater discharge through evapotranspiration is critical for understanding responses of groundwater dependent ecosystems. In the semi-arid oodplain environment of the lower River Murray, South Australia, patterns of steady state groundwater ux may indicate potential areas of anthropogenic salinisation due to adjacent irrigation and aid in management decisions for the protection of indigenous oodplain vegetation. This paper outlines the use of relationships for evapotranspiration response to groundwater depth and salt concentration, within the MODFLOW 2000 framework, to model spatial patterns of groundwater ux. This modelling approach is used to develop relationships between the steady state net rates evapotranspiration, seepage and baseow, and oodplain properties including elevation, geometry, soil and vegetation distribution relevant to the eld site. The model was applied to a key research site on the River Murray, South Australia, to facilitate understanding of the factors contributing to the development of spatial patterns of irrigation induced groundwater discharge. The inclusion of elevation and vegetation data in the methodology revealed the effects of microtopography and plant water use on groundwater discharge. Spatial distribution of vegetation had the highest correlation with patterns of groundwater evapotranspiration. Floodplain elevation, oodplain width and geometry, which related to vegetation cover, also affected both groundwater discharge and baseow to the river. c 2006 Elsevier B.V. All rights reserved.

Introduction
* Corresponding author. Tel.: +61 8 8303 8705; fax: +61 8 8303 8750. E-mail address: rebecca.doble@csiro.au (R. Doble).

In arid and semi-arid regions, oodplains are ecologically and hydrologically important parts of a catchment. Due to their proximity to rivers, oodplains and their associated

0022-1694/$ - see front matter c 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2006.02.007

76

R. Doble et al.

Nomenclature
Parameter Denition (dimensions) a parameter in the K(S) function of (Gardner, 1958) (Ln+1T1) A aAn (Warrick, 1988) (Ln+1T1) An [(p/n)csc(p/n)]n (Warrick, 1988) () Cg groundwater solute concentration (ML3) Cth threshold soil water solute conc. at which plants can no longer extract water (ML3) d, Zg depth to groundwater from the soil surface (L) dext extinction depth, below which there is no further discharge from the groundwater (L) D cumulative discharge through plant water uptake (L) Dh conned piezometric head elevation above ground surface (L) K hydraulic conductivity (LT1) Ksat saturated hydraulic conductivity (LT1) L aquifer thickness (L) n exponent in the K(S) function of (Gardner, 1958), dependent on soil characteristics () qlim maximum steady state moisture ux through the soil (LT3) q qmax R t td tr U Zg, d Zf c h0 h1 hFC hsat groundwater ux, positive upward (LT3) maximum discharge rate when groundwater is at the soil surface (LT3) cumulative recharge to groundwater (L) the point in time at which groundwater ux is calculated (T) the length of time between ood inundation (T) duration for which the site is inundated (T) plant groundwater uptake ux (LT1) depth of groundwater below the soil surface (L) depth of salt front (plane of evaporation) below the soil surface (L) Zf/Zg () initial water content (L3L3) water content within capillary zone, approximately eld capacity (L3L3) water content at eld capacity (L3L3) water content under saturated conditions (L3L3)

wetlands have higher biodiversity value than surrounding areas, providing habitat for many aquatic and riparian species, including vegetation, birds, mammals and sh species. In lowland gaining streams groundwater owing from the catchment moves through the oodplain before either owing to the river or being attenuated by evapotranspiration.1 The shallow watertables found within oodplains mean that evapotranspiration rates are signicant, and understanding the nature of groundwater discharge processes through evapotranspiration and seepage in this environment is vital to produce an accurate water balance for the system. In semi-arid areas where saline regional groundwater discharges to streams, such as the lower River Murray in South Australia, evapotranspiration from rising oodplain watertables and altered ow regimes have led to the dieback of environmentally signicant riparian vegetation such as red gum (Eucalytpus camaldulensis) and black box (Eucalytpus largiorens) (Jolly et al., 1993). The naturally saline regional groundwater tables have risen as a result of higher river levels and a three orders of magnitude increase in recharge from adjacent irrigation developments, leading to the accumulation of salt in the soil and alluvial aquifer through evapoconcentration (Jolly, 1996). The hydrology, geomorphology and vegetation coverage of these oodplains are complex, leading to high spatial variability in evapotranspiration, and variation in vegetation health impacts. Management of vegetation health is undertaken through pumping regional groundwater to lower riparian watertables, targeted spear point pumping under key eucalypt communities, injection of fresh water under communities, and

For the remainder of this paper, any reference to evapotranspiration refers to capillary upow from shallow groundwater, rather than direct evaporation of rainfall.

localised articial ooding or environmental irrigation. These management techniques require an understanding of spatial patterns of salinisation risk on a ne scale, and the true ecological impact of land management and irrigation decisions requires knowledge of the proximity of salinisation risk to vegetation communities. The ability to quantify spatial patterns of groundwater discharge through modelling and site characterisation is required in order to understand the potential impacts from salinisation on vegetation. To do this, a conceptual model of discharge that accounts for the variability in elevation, soils and vegetation cover on oodplains is needed. Groundwater loss due to evapotranspiration, however, is known to be the most difcult component of the groundwater balance to estimate (Freeze and Cherry, 1979). Much soil science research of the 19501980s was focused on the understanding, quantication and representation of capillary upow from shallow watertables (Gardner, 1958; Warrick, 1988), and more recently transpiration within the context of environmental science (Thorburn et al., 1995; Mac Nish et al., 2000) and silviculture (Silberstein et al., 1999; Feikema, 2000). Despite this understanding, numerical modelling of groundwater loss has tended to simplify the evapotranspiration process to the point of representing it as a simple sink with no interface denition, or as linear approximations. Where patterns of groundwater evapotranspiration representing salinisation risk at a ne scale are of interest, however, simplied models of groundwateratmosphere ux can limit the understanding of spatial variation and heterogeneities. A model that incorporates the key factors inuencing evapotranspiration, including vegetation type, groundwater salinity, watertable depth and ooding frequency is required to investigate the origins of spatial patterns of groundwater discharge and salinisation.

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge The objectives of this study were therefore to model evapotranspiration at a scale and complexity appropriate for understanding salinity effects on vegetation health, and to use these results to understand the relative contribution of physical, ecological and hydrological parameters in governing groundwater ux distribution. Specically, the aims of the paper were to: Use relationships for evapotranspiration in semi-arid environments within a nite difference groundwater model in order to understand the long-term patterns of evapotranspiration and salinisation for the purpose of determining vegetation health in a oodplain environment; and Develop a conceptual understanding of the hydrological relationships between regional groundwater inows, vegetation water use and baseow to the river for a key research site to understand the factors that govern groundwater ow processes within a oodplain environment. The rst part of this study briey describes the modelling process undertaken, including the use of a spatially distributed evapotranspiration relationship as an input for MODFLOW 2000 in order to facilitate better representation of groundwater discharge uxes. This section provides a background to the modelling, and a more detailed description of this methodology is available in Doble et al. (2005). The second section of this paper applies the model to a oodplain on the River Murray, South Australia in order to understand the oodplain hydrological processes. Model results are compared against spatial patterns of discharge from remote sensing images and ne scale baseow data from geophysical measurements. A conceptual understanding of the relative importance of factors such as oodplain elevation, geometry, soil and vegetation type on the distribution of discharge is then developed for the eld site.

77

The hydrogeology of Clarks Floodplain is typical of the eastern part of the lower River Murray (Jarwal et al., 1996). The Coonambidgal Clay ranges from 3 to 7 m thick, while the Monoman Formation is approximately 7 m thick in this area. The cliffs adjacent the oodplains consist of a layer of Woorinen Sands over Blanchtown Clay, each approximately 2 m thick (not represented in the model), overlying a layer of Loxton Sands up to 35 m in depth. The whole area is underlain by the Bookpurnong Beds, which act as an aquitard basement to the shallow aquifer that encompasses the Monoman Formation and Loxton Sands. Groundwater salinity in the Loxton Sands and Monoman Formation is in excess of 30,000 mg L1, while irrigation recharge salinity is typically 5000 mg L1. The Bookpurnong Irrigation District comprises more than 1500 ha of irrigated land, with major irrigation having commenced in the mid-1960s, and has been increasing since that time (Telfer and Overton, 1999). Excess recharge from the irrigation area has led to the formation of a groundwater mound, which displaces saline groundwater towards the oodplain and has led to increased waterlogging and salinisation, and groundwater seepage at the break of slope adjacent to the cliffs. Aerial photographs show that extensive degradation has taken place between 1972 and 1998, with most of the vegetation health decline occurring at the back and centre of the oodplain. This is associated with the seepage and presence of a saline backwater that is thought to intercept groundwater and concentrate salt further (Fig. 3). Black box (E. largiorens) and red gum (E. camaldulensis) tree communities have been most affected by the salinisation of the oodplain. Clarks Floodplain is situated below a weir (Lock 4; Fig. 1), and therefore the occurrence of vegetation health decline due to weir-induced water logging is minimal. This paper therefore focuses specically on irrigation-induced salinity.

Simulation of groundwater ux patterns Site description

Clarks Floodplain, adjacent to the Bookpurnong Irrigation District (Fig. 1), is located on the lower reaches of the River Murray in South Australia (3421 0 S, 14037 0 E). The eld site is located within the semi-arid inland of Australia, with rainfall varying between 200 and 300 mm a1 and potential evaporation of approximately 1800 mm a1. The lower River Murray oodplain is characterised by a at, wide, meandering river within a deep river valley, excised during the early to middle, and then again in the late Pleistocene, and redeposited with coarse-grained quartz sands of the Monoman Formation during higher, post-glacial sea levels between 17,000 and 6000 BP (Twidale et al., 1978). In the late Holocene (approximately 4000 BP), the Coonambidgal Clay, typically consisting of clays and silts, was deposited in the river valley overlying and partly conning the Monoman Formation (Fig. 2). The modern oodplain surface tends to have rapid variation in soil type due to geomorphological processes of deposition and erosion that are continually occurring. Palaeochannels and backwaters formed from previous river courses are present in many parts of the oodplain. The intent of the modelling in this section of the paper was to use a better representation of evapotranspiration processes within a nite difference groundwater model to develop a conceptual understanding of the processes leading to spatial patterns of evapotranspiration. A more detailed discussion of the model development is given in Doble et al. (2005).

Conceptualisation
The methodology used in this paper involves representing spatial distributions of net groundwater ux by developing a relationship to use within a nite difference groundwater ow model, in this case MODFLOW 2000. This relationship uses spatial information for vegetation, soil and groundwater salinity distributions, as well as ooding duration and frequency to develop a spatially variable, depth-dependent ET function that can be used within MODFLOW. The function, which provides a surface boundary condition to the groundwater model, is similar to the segmented evapotranspiration function (ETS1, Banta, 2000) but also incorporates a recharge component, with the equilibrium depth, rather than

78

R. Doble et al.

Figure 1 Clarks Floodplain on the lower River Murray, South Australia (3421 0 S, 14037 0 E) showing irrigation areas of the Bookpurnong Irrigation District and ow divisions (numbered 111). Local grid renement graduates from 200 m by 50 m on the highland region, to 50 m by 50 m on the oodplain, to 6.25 m 6.25 m at the break of slope in the region indicated by the dashed line.

extinction depth describing a net ux of zero. The net rate of groundwater ux (recharge subtracted from evapotranspiration) is important for understanding the long-term or ultimate rate of salt accumulation, and is a common measure of salinisation potential (Petheram et al., 2003; Slavich et al., 1999). Similarly, the equilibrium depth, where the net ux is zero is critical for determining whether salinisation will occur. The model is run at steady state, in order to understand the ultimate salt accumulation potential for future riverine and vegetation management planning. The intent of the study was not produce a model to accurately represent eld conditions for management purposes, and to calibrate the model to produce a close t with historical data. Instead, the aim was to conceptualise relationships between oodplain groundwater processes, leading

to patterns of evapotranspiration. The modelling therefore required, rather than full calibration, only that the model parameters were within a reasonable range, likely to be found in the eld situation being modelled. Spatial and numerical validation of the model was performed using a series of geophysical, remote sensing and eld measurement techniques. A brief description of the spatial validation is given later in the paper in order to justify the model results.

Previous solutions for groundwater discharge

To determine a threshold for salinisation due to steady state evaporation of saline groundwater, Warrick (1988) proposed that the limiting rate of upward ow in a soil was inversely

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

79

Figure 2 Geological cross section of Clarks Floodplain, indicating the excised river valley, Monoman Formation alluvial aquifer, and Coonambidgal Clay which overlies much of the oodplain. Increased recharge from irrigation taking place on the highland causes a groundwater mound to form adjacent the oodplain, leading to shallow groundwater and seepage at the break of slope.

Figure 3 Vegetation health on the oodplain is shown in greyscale with black representing dead vegetation, dark grey signifying vegetation in poor health, light grey representing healthy vegetation and white designating no vegetation present. For Clarks Floodplain, vegetation health decline was observed at the back of the oodplain and the saline backwater, and extends out towards the centre of the oodplain. Most of the oodplain peninsula areas support healthy vegetation.

proportional to the depth to the watertable but varied according to soil properties, as described by: qlim An ad n . 1

From Eq. (1), the maximum moisture ux or diffuse discharge from a soil can be determined for varying groundwater depths. Shallower groundwater will lead to higher

evaporation rates, and therefore greater concentration of salts in the surface soil. The science of predicting both dryland and irrigation induced soil salinisation using groundwater discharge rates has already been established in soil physics, but is less frequently described by groundwater hydrologists. Using relationships described by Gardner (1958), Wind (1955) and

80 others, Talsma (1963) and Peck (1978) dened thresholds for salinisation in terms of groundwater uxes of 103 m d1 for irrigated areas and 104 m d1 for dryland conditions, respectively. This denition relates salinisation to rates of groundwater discharge and accounts for soil type and groundwater depth. In numerical models, a linear function, or a constant ux is often used to represent evapotranspiration. MODFLOW 96 (McDonald and Harbaugh, 1996) assumes a linear evapotranspiration function where evapotranspiration varies from a maximum rate at the ground surface, to zero when the watertable is at some depth below the surface (extinction depth), such that: q qmax ;   d q qmax 1 ; d ext q 0; d 6 0; 0 < d < d ext ; d P d ext . 2

R. Doble et al. and the salinity threshold at which vegetation can no longer extract groundwater derived soil water. This relationship for instantaneous ux is described by the equation:   n   n 1ACg td n1 Cg h1 h0 n1 U A Zg 1 c . 1 h1 Cth h1 Cth 3 The cumulative groundwater discharge (D) through plant water uptake is found by integrating Eq. (3) over the time period 0 to t: !  1 n 1ACg td n1 n1 D Zg 1 c Zg 1 c h1 Cth     Cth 4 1 h0 . h1 Cg For a oodplain environment, this represents the cumulative discharge occurring for the continuous period between ood cycles, where td represents the length of the dry period between the end of one period of inundation (t = 0) to the start of the next (t = td). The length of this dry period is critical for vegetation dependent on groundwater, as evapotranspiration processes will concentrate salt, which in turn limits plant water uptake. When an overbank ow event occurs, recharge from ooding will freshen soil and ground water, allowing plant water uptake to increase. More frequent ooding (shorter td) will therefore support higher volumes of discharge in the long term than would be possible with longer dry periods. Recharge into the oodplain will be dependent on both the vertical conductivity of the soil and the period for which that part of the oodplain is inundated. This volume of inltration, less the water volume that lls the unsaturated zone to eld capacity, gives the net volume of water recharging the groundwater. Recharge may be represented by: R K sat tr hFC h0 Zf hFC h1 Zg Zf if hsat hFC Zg > 0. 5

The linear nature of this relationship may cause discharge to be under-predicted where the watertable is deeper, and over-predicted where it is shallow, a problem when knowledge of groundwater discharge is required on a ne scale. MODFLOW 2000 includes the module ETS1, which represents evapotranspiration as a piecewise linear function (Banta, 2000), which is used to prevent this problem. Setting up an accurate curve, however, requires knowledge of the spatially varying oodplain conditions. Eq. (1) describes unsaturated zone groundwater movement at a point scale. Difculties arise in applying this model at a regional, or even local scale to consider spatial distribution of groundwater discharge. The linkage of point-scale unsaturated models with large-scale saturated groundwater ow models has been trialled extensively, but these models tend to have stability issues due to the sensitivity of the Richards Equation to lower boundary uxes (Boone and Wetzel, 1996). Scaling up data from point scale unsaturated models to local scale groundwater models requires some means of extrapolation (Hatton, 1998) to match spatial and temporal scales, and the processes that are relevant to the system can change between small scales of the unsaturated model and large scales of the saturated model. The methodology used in this paper employs an equation for groundwater discharge from plants developed for saline, semi-arid conditions, plus information on recharge rates and ood recurrence patterns, and applies it spatially to the MODFLOW 2000 ETS module using a geographic information system (GIS). Recharge and discharge An analytical model for the uptake of saline groundwater by plants in semi-arid regions was developed by Thorburn et al. (1995), based on unsaturated zone water and solute balances. In this model, discharge of groundwater by vegetation depends on soil water content, groundwater salinity

For simplication, the moisture content within the capillary fringe (h1) may be approximately equal to the moisture content at eld capacity (hFC), therefore the second term will cancel. Groundwater recharge in conductive soils will also be limited by the total unsaturated pore space that may be lled during oods. The equation for recharge is therefore the minimum of the following equations: Minimum: R K sat tr h1 h0 Zf ; R hsat h1 Zg ; 6

where (hsat h1) represents the pore space available to receive recharge water. Combining recharge and discharge for a cycle of one ood and one discharge period, an approximation of the net groundwater ux from plant water use (q) is:

 min q

K sat tr h1 h0 Zf hsat h1 Zg



Zg 1 cn1 tr td

n1ACg td h1 Cth

1 n 1

    th Zg 1 c h1 C 1 h0 Cg .

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

Table 1 Plant water use and soil parameters from Thorburn et al. (1995) and Jolly et al. (1993) Values for each soil, vegetation, groundwater category Silty sand 0.35 0.2 0.15 0.02 5 0.028 0.05 Red gum 8 Zone 1, extreme EC 0.5 0.1 0.2 20 Clayey sand 0.4 0.28 0.2 0.01 3 0.018 0.01 Black box 16 Zone 2, high EC 3 0.4 0.13 15 Sandy clay 0.45 0.34 0.25 0.005 2.9 0.009 0.005 Lignum 6 Zone 3, moderate EC 3 0.5 0.17 8 Clay 0.6 0.4 0.35 0.001 2.5 0.002 0.001 Grassland 7 Zone 4, low EC 3 0.6 0.2 2 Palaeochannel 0.6 0.44 0.35 0.001 2.5 0.002 0.001

81

Parameter Soil: hsat (v/v) h1 (v/v) h0 (v/v) a (mn d1) n A (mn d1) Ksat (m d1) Vegetation: Cth (g L ) Groundwater: Zg (m) Zf (m) c () Cg (g L1)
1

Eq. (7) reects the long term balance between groundwater discharge during dry periods, and recharge during oods. The relationship contains information on evapotranspiration from groundwater dependent ecosystems, and ood frequency and duration patterns, evaluated as a long term trend to determine whether areas are dominated by recharge or discharge, dependent on depth to groundwater. In areas with saline regional groundwater, discharge dominance will lead to secondary salinisation. This equation describes a relationship between groundwater ux and watertable depth that can be used within a nite difference groundwater ow model as a surface boundary condition. The function was used to calculate input parameters for the ETS1 module using GIS coverages of elevation, soil and vegetation conditions for each part of the oodplain and each model cell. Data used for this spatial calculation is shown in Table 1, and durations of inundation and dry periods (td and tr) provided from ood recurrence data. Seepage At many points along the river valley, the high incoming regional groundwater inows cause the watertable to intersect the ground surface at the break of slope, causing direct seepage of groundwater. Seepage at the break of slope was modelled with an additional segment of the evapotranspiration function, such that the maximum ET rate at the soil surface was equal to the potential evapotranspiration for the region (Table 2). At a point 0.2 m below the surface, the evapotranspiration function reverted to the groundwater ux relationship appropriate for the soil and vegetation conditions. Conning layers For Clarks Floodplain, and most areas of the lower River Murray, the Coonambidgal Clay forms a conning layer over the Monoman Sands aquifer. The effect of this conning layer is to increase the pressure head required to force groundwater to the surface, or high enough that the capil-

lary fringe reaches the surface for evaporation to occur. This process may be modelled within MODFLOW using a false evaporation surface above the actual ground surface to represent this additional head requirement. This stretches the rechargedischarge curve both upwards above the equilibrium point, and downward within the recharge section below the equilibrium point. The degree of upward stretching is given by the Darcy equation, which species the head required for maximum groundwater ux at the surface. That is: qK Dh . L 8

Rearranging and substituting into Eq. (4) for the condition of the watertable at the soil surface, and therefore K = Ksat gives: Dh L ET max ; K sat 9

where recharge is zero, or:  1    L n 1ACg td n1 Cth h1 1 h0 . Dh h1 Cth Cg K sat

10

For the groundwater ux curves, the maximum rates of evapotranspiration and recharge remain the same.

Coupling with MODFLOW

The relationships described in the previous sections are applied to MODFLOW 2000 (Harbaugh et al., 2000) through a modied version of the ETS module (Banta, 2000), which combines both recharge (RCH) and discharge (ETS) together as a single groundwater ux dependent on groundwater depth (Fig. 4). The resulting combined rechargedischarge (CRD) module is a simple additive combination of these two functions. The function applies a net recharge or discharge to the model, depending on whether the groundwater head is above the ground (max ET), between the surface and

82
Table 2 Model parameters Source Range 133243 0.1 40440 Black box 360720 Red gum 17 17001800 mm a1 0.10.5 13.2 m 9.810.5 m 12 m15 m 3540 m

R. Doble et al.

Parameter

Adopted Value 175 mm a1 0.1 mm a1 Ranges spatially mm a1 Ranges spatially 1.55 m 1800 mm a1 0.2 m 13.2 m 10.5 m 12 m15 m 40 m

Recharge and discharge parameters Irrigation recharge AWE (1999) Dryland recharge AWE (1999) Maximum evapotranspiration Thorburn (1996); Thorburn et al. (1993) rate on oodplains Extinction depth Thorburn (1996), Talsma (1963) Open water evaporation rate Bureau of Meteorology Seepage extinction depth River and elevation parameters River stage upstream of Lock 4 River stage downstream of Lock 4 Floodplain elevation Highland elevation Aquifer parameters Upper Loxton Sands Kh Kv Coonambidgal Clay Kh Kv Monoman Sands Kh Kv Lower Loxton Sands Kh Kv SA Water SA Water Overton (2005) Overton (2005), Armstrong et al. (1999) Armstrong et al. (1999), AWE (1999), Jolly et al. (1998), Evans and Kellett (1989), Barnett and SA Department of Mines and Energy (1991)

110 0.0351 0.050.1 0.0050.01 1035 0.351 0.054 0.0050.4

5 m d1 1 m d1 0.1 m d1 Ranges spatially m d1 10 m d1 3 m d1 0.2 m d1 0.2 m d1

Discharge

Recharge

Groundwater Flux

(a) (b)

rived from soil logs and geomorphology, and groundwater chloride concentration data. Each of the three coverages was divided into categories shown in Table 1 and assigned values for parameters used in Eq. (7). A GIS framework was used to calculate four points on the CRD curve, dening maximum ET, maximum recharge, and equilibrium depth. Spatial distributions of maximum ET, maximum recharge and equilibrium depth were used as inputs for MODFLOW, which provided the hydrological component and groundwater depth required to calculate long term groundwater ux.

Depth to groundwater
Figure 4 Combined recharge and discharge in MODFLOW showing: (a) the equilibrium depth where ET is equal to recharge, and (b) the extinction depth or maximum recharge, below which the ET component of ux is zero.

MODFLOW model development

Model parameters and their sources are listed in Table 2. Fig. 5 shows an example cross-section of the model, including the geological layers used. On the highland, the upper, unconned aquifer consisted of Upper Loxton Sands (Kh = 5 m d1), which was underlain by the Lower Loxton Sands (Kh = 0.2). On the oodplain, the alluvial aquifer within the Monoman Formation (Kh = 10) is again underlain by Lower Loxton Sands. Overlying the alluvial aquifer is a spatially variable semi-conning layer, consisting of heavy Coonambidgal Clay (K variable) at the break of slope, to a silty sand deposit near the river. Thicker Coonambidgal Clay was found in some low lying palaeochannels. Vertical hydraulic conductivity and other soil properties used for the evapotranspiration function (Eq. (7)) are shown in Table 1. The MODFLOW model was discretised into a regular, square grid with cell dimensions ranging from 200 50 m

equilibrium depth (a) (net ET), between the equilibrium depth and extinction depth (b) (net recharge) or below the extinction depth (max recharge). The point at which groundwater discharge is equal to zero is termed the equilibrium depth, that is where recharge and discharge uxes are equal in the long term. With this method of dening groundwater ux it is possible to include varying recharge functions such as zero recharge when groundwater is at the surface. Vegetation input parameters were applied spatially using a GIS vegetation survey (Telfer and Overton, 1999), and soil and groundwater parameters using a map of soil types de-

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

83

Figure 5 Cross-section of the conceptual model showing the partially conned Monoman Sands alluvial aquifer and the highland Loxton Sands Formation that form the main groundwater ow path toward the river (not to scale). Note the intersection of the ground surface by the piezometric surface at the break of slope, resulting in seepage.

to 6.25 6.25 m, rened about the break of slope, as indicated in Fig. 1. The model used constant head cells on the outer boundaries with elevations of 12.5 m, 13.9 m, 15.2 m and 12.5 mAHD, starting from the north-western corner and progressing clockwise around the corners of the model domain. Floodplain elevation data was available in spatial form at vertical intervals of 100 mm on a 30 30 m grid cell scale (Fig. 6, Overton, 2005). The river boundary was simulated using the MODFLOW river module. River stage was set at an elevation of 10.5 m below the weir and 13.2 m above it. The river is regulated into a series or weir pools, and river stage variation is minimal for the length of the oodplain. The model was run in steady state for current irrigation conditions, both for conned and unconned alluvial aquifer states.

Vegetation and soil parameters from Table 1 were adopted from Thorburn, 1996, and extrapolated to sandier soils using Gardner n parameters provided by Jolly et al. (1993) from eld experiments. Groundwater parameters were taken from eld measurements in March 2003. The spatial distribution of soil, vegetation and groundwater parameters are shown in Fig. 7ac. Historical ood data was used to determine the average return period between oods, and the duration of inundation for various river stages. These values were then used in Eq. (7). Data for River Murray ows into South Australia was obtained for the years 19572003 (post regulation) from gauging station GS426200SA (Porter, 2002). A rating curve measured at a point downstream of Lock 4 corresponding to Clarks Floodplain was used to link ow

Figure 6 Floodplain elevation distribution, showing surface elevation varying between 11 and 15 m. Low lying areas are found at the ends of the oodplain peninsulas and in the centre of the oodplain associated with the saline backwater.

84

R. Doble et al.

Figure 7 Distribution of: (a) groundwater chloride concentrations (g L1) for Clarks Floodplain obtained from eld data, (b) soil type distribution, and (c) surveyed vegetation type, used as spatial inputs for the combined rechargedischarge function. Parameter values associated with these distributions are shown in Table 1.

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

Consecutive Inundated and Dry Periods vs Floodplain Elevation td tr
3000 2500 2000 1500 1000 500 0 12 12.5 13 13.5 14 14.5 Floodplain Elevation (mAHD) 15
100 80

85

GW Flux vs Depth
RG Z4 GL Z1
tr (d)

BB Z3 RG, FO Z3

L Z2

GW Flux (mm/d)
-4 -3 -2 -1 0 1 2 3 0 1

td (d)

60 40 20 0

Figure 8 Average consecutive days between oods and during inundation as a function of oodplain elevation. Once assigned spatially, linked to the oodplain elevation of each polygon, this information provided the data for td and tr to be used in Eq. (7).

frequency data to the spatial coverage of oodplain elevation (Fig. 6). Fig. 8 shows the resulting smoothed relationship for average period of inundation and duration between ooding as a function of oodplain elevation. This means of including ood information allows the long term trends of the rechargedischarge balance to be investigated. Spatial distributions of the MODFLOW CRD module inputs: maximum ET, equilibrium depth and maximum recharge, were produced using Eq. (7), and the spatial data described above. Fig. 9 shows examples of various rechargedischarge relationships that cover several different vegetation, soil and groundwater scenarios on oodplain. The locations at which data for these curves were selected are indicated in Fig. 7a. The MODFLOW model was run twice more using uniform evapotranspiration and elevation parameters, the results of which were used to determine how much additional information was gained by considering variation of the oodplain system at a ner scale. The rst run consisted of a constant oodplain elevation of 14 m (herein referred to as the uniform model), and the second with a spatially variable surface elevation (variable elevation model, Fig. 6), but with evapotranspiration not dened by vegetation parameters. Both models used a constant maximum ET rate of 0.3 mm d1 and an equilibrium depth of 3.5 m, zero oodplain recharge and a constant vertical hydraulic conductivity (0.01 m d1).

GW Depth (m)

4 5 6 7 8 9 10

Figure 9 Groundwater ux versus depth relationships for various oodplain vegetation, soil and groundwater scenarios, including: red gums found on clayey sand with low salinity groundwater (RG Z4), black box community on sandy clay with moderate salinity groundwater (BB Z3), lignum on heavy clay, with high salinity groundwater (L Z2), grassland on the heavy clay ood out at the break of slope, with extremely saline, shallow groundwater (GL Z1), and red gum communities found at the river end of the heavy clay ood out, with moderate salinity groundwater (RG FO Z3).

MODFLOW model results

A spatial distribution of net groundwater ux was produced using the methodology described above (Fig. 10). This was compared with the ux outputs from the uniform (Fig. 11) and variable elevation (Fig. 12) models. Inclusion of elevation, vegetation, soil type and ooding information within the MODFLOW model provided ner-scale predictions for discharge patterns and revealed a variation in discharge rates across the oodplain. Not only could areas at risk of salinisation be identied, but variations in the rate of salinisation were also quantiable. The MODFLOW groundwater ow model predicted high net discharge of groundwater at the back of the oodplain, near the break of hill slope where the depth to groundwater

was at a minimum (Fig. 10). High groundwater discharge was also associated with red gum and black box forests at the ends of oodplain peninsulas and adjacent the river in divisions 1, 7 and 8. Flow divisions used to describe discharge locations are numbered in Fig. 1. Localised elevation, or microtopography, was revealed by high evapotranspiration rates in the spatial patterns of groundwater discharge, both at the break of slope, and associated with the low-lying palaeochannel that extended along the centreline of Clarks oodplain in division 7. Patterns of elevation were evident within vegetation classes, for example comparing the elevation prole at the river end of division 3 (Fig. 6) with the groundwater discharge patterns in Fig. 10. The effect of elevation between vegetation communities was not evident.

86

R. Doble et al.

Figure 10 Predicted groundwater discharge on Clarks Floodplain, conned model. Note the high rates of discharge at the break of slope, associated with high plant water use at the ends of the peninsulas, and through the centre of the oodplain associated with a low-lying palaeochannel.

Figure 11 Predicted groundwater discharge on Clarks Floodplain from the uniformly parameterised MODFLOW model, incorporating a at oodplain, constant hydraulic conductivity and vegetation distribution.

Validation In light of the objectives of this paper, the validation process described below provides a means of establishing that both the modelled spatial patterns of evapotranspiration and the rates of groundwater ux lie within a range that is appropriate for the eld conditions under investigation. Since the model was not used for detailed oodplain management, a through calibration with historical data was neither necessary nor productive. This validation is therefore intentionally concise and indicative in order to give condence that the processes representing evapotranspiration are adequate, rather than providing detailed calibration results. Higher rates of salinisation from evapotranspiration at the break of slope and through the centre of the oodplain

are indicated by dead and unhealthy vegetation in Fig. 3, as well as observations of seepage and signicant accumulation of salt in the soil prole during eld investigations. Point scale transpiration measurements from the eld were compared with modelled results. Sap ow measurements from Thorburn and Walker (1994), Jolly et al. (1991) and Akeroyd et al. (1998) indicate transpiration rates within the lower River Murray region from 70 to 700 mm a1, and 35 to 150 mm a1 for red gum and black box eucalypt species, respectively. Evapotranspiration from bare soils on Chowilla Floodplain in the same region of the River Murray was approximated using 2H isotope proles to be 10 mm a1 (Jolly et al., 1993) with a groundwater depth of 45 m. Accounting for recharge, modelled ET rates were approximately 180360 mm a1 for red gum areas,

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

87

Figure 12 Predicted groundwater discharge on Clarks Floodplain from the MODFLOW model with variable oodplain elevation, incorporating constant hydraulic conductivity and vegetation distribution.

40130 mm a1 for black box and 040 mm a1 for the elevated grassland, lying within the ranges described. A shallower watertable (23.5 m) may explain why the grassland evaporation rates were slightly higher than that described by Jolly et al. (1993). Satellite imagery was used as an indicator of vegetation water use (Goodrich et al., 2000). Spectral indices such as normalised difference vegetation index (NDVI) that are derived from satellite images can indicate vegetation distribution and health or density changes due to salt accumulation. In groundwater dependent ecosystems, they are also able to provide a direct indication of groundwater discharge

through transpiration, which may be correlated with groundwater discharge model outputs. NDVI and other vegetation water use indices indicate plant water use only, however, and do not account for direct soil evaporation. Landsat-5 TM images with a 25-m resolution were obtained for the 18th February 2002, and the NDVI was calculated using the normalised difference of the red and infra red spectral bands (bands 3 and 4). This data was not used to create or calibrate the model. Fig. 13 shows the spatial distribution of NDVI across the oodplain, with darker areas indicating high water use. The image indicates high water use at the ends of the oodplain peninsulas and adjacent

Figure 13

NDVI at Clarks Floodplain on the 18th February 2002, inferring evapotranspiration rates.

88 the river. High water use is also evident along the break of slope, indicating areas where fresh irrigation recharge has broken through to the oodplain seepage face and is providing a permanent water supply to waterlogging tolerant vegetation such as grasses and reeds. There were strong correlations in high water use at the ends of the peninsulas and adjacent the river in both model output and NDVI. This indicates that the modelled discharge function is providing an adequate basis for further analysis of spatial patterns of discharge. NDVI values and long term discharge uxes from the model output were compared within different eucalypt vegetation categories on Clarks Floodplain (Fig. 14, Table 3). Areas where evaporation from annual species or bare soil dominated, such as the break of slope and oodplain centre, were excluded from this analysis. Within the tree categories, a trend of increasing modelled discharge rates with higher NDVI values was observed, suggesting a positive relationship between modelled and indicated discharge. To test the hydraulics of the model, baseow to the river was tested against eld measured and indicated baseow.

R. Doble et al. Monitoring of river salinity has estimated that Clarks Floodplain contributes salt loads of between 30 and 50 t d1 (Porter, 2002). This also correlated with modelled salt loads from Clarks Floodplain, which were equivalent to 45 t d1. Transient electromagnetics at a nanometre scale (NanoTEM) has been used to measure resistivity in river sediments along the stretch of river between Lock 3 and 4 of the River Murray in South Australia (Berens et al., 2004) in order to identify the conductive anomalies associated with concentrated saline groundwater inows. NanoTEM analysis does not provide actual values of salt loads or baseow, but provides a qualitative means of comparing the relative groundwater inputs to a watercourse. Modelled baseow showed higher groundwater discharge in areas where the river is close to the highland, and negative baseow at the ends of the peninsulas, correlating strongly with the nanoTEM resistivity patterns. Multiplying baseow rates by a regional groundwater salinity of 30,000 mg L1 resulted in values of modelled salt inputs. Modelled salt loads at approximately 250 m intervals were compared quantitatively with nanoTEM results using a threshold resistivity of 27 X m to distinguish positive and negative inows (Fig. 15). Patterns of high conductivity correlated well with the modelled baseow. A map of modelled depth to groundwater is provided in Fig. 16, compared with groundwater depth measured from alluvial groundwater wells. Root mean square error of modelled vs eld groundwater elevation was calculated to be 0.59 m, with the data biased toward the lagoon and break of slope where the majority of boreholes were located. The greatest variation between eld and modelled heads was found between the lagoon and break of slope where the change in hydraulic gradient was greatest. Whilst the depth to groundwater provides an opportunity for a standard model calibration and indicates the underlying groundwater conditions, it does not provide information on the distribution of vertical groundwater ux, as there is no provision for spatially variable soil, vegetation or groundwater salinity conditions. Measuring the potential for salinity as depth to watertable will be of little benet if the spatial distribution of salinisation potential is of interest.

Comparison of Modelled Discharge and NDVI

Discharge (mm/yr) 0 0 -0.05 1r(f) -0.1 NDVI -0.15 -0.2 -0.25 2r -0.3 1b 1r 1r(r) 2r1bc 2rl 1bl R = 0.82
2

20

40

60

80

100

Mallee

1br

Figure 14 Comparison between modelled long term groundwater discharge and NDVI for Clarks Floodplain for areas with eucalypt vegetation. Vegetation classications are found in Table 3.

Salt Loads: nanoTEM and Modelled

Table 3 Classications for the vegetation survey of Clarks Floodplain, used in the NDVI analysis (Fig. 14) for model validation Classication 1r(f) 1br 1r 1r(r) 1b 1bl 2r 2r1bc 2rl Vegetation type Healthy red gum forest Healthy mixed black box red gum woodland, black box dominant Healthy red gum woodland Red gum regeneration Healthy black box woodland Black box with lignum understory Unhealthy red gum woodland Unhealthy red gum, healthy black box and cooba Unhealthy red gum with lignum understory
Modelled Load 7 Salt Load (t/d) 250 metre interval 6 5 4 3 2 1 0 518 516 514 512 510 508 River Km 506 504 nanoTEM 7 5 4 3 2 1 0 502 Linearly Scaled Conductivity 6

Figure 15 Salt load patterns from Clarks Floodplain showing modelled data and resistivity from the nanoTEM information using a threshold value of 27 X m to distinguish positive and negative baseow, and linearly scaling conductivity to salt loads.

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

89

Figure 16

Depth to groundwater from modelled data compared with measured groundwater depth from boreholes indicated.

Spatial patterns of high groundwater discharge While Fig. 10 shows a distribution of groundwater ux similar to that found from vegetation indices, the simpler uniform and variable elevation MODFLOW models indicated high discharge associated with the break of slope (Figs. 11 and 12), but none of the groundwater ux associated with vegetation water use. Some discharge was predicted on the low-lying edges of the oodplain for the variable elevation model run, however, and higher evapotranspiration rates are evident around the palaeochannel in the centre of the oodplain. While these simplications may be sufcient for broad regional scale modelling, the lack of distributed vegetation data meant that representation of groundwater discharge was less accurate on a sub-oodplain scale. Table 4 shows a budget of groundwater ow for each oodplain division. As a general trend, groundwater inow from the highland increased for each division from the northwest to the southeast depending on the proximity of the break of slope to the highest point of the groundwater mound. Seepage only occurred within divisions 510, while

evapotranspiration was present on all divisions, and was highest for divisions 1, 7 and 8. Baseow was negative for four of the divisions (1, 2, 7 and 8), indicating ow from the river into the alluvial aquifer. The areas in which the net groundwater ux is recharging the aquifer are predicted by the model to be minimal, and occur only in very low lying areas on the ends of oodplain peninsulas where sandy deposits occur and vegetation is yet to establish. The dominance of ET within the system is expected to be due to the signicantly decreased ood recurrence interval from river regulation in the lower River Murray. Net groundwater recharge from ooding is predicted to occur predominantly in oodplain divisions 1, 3, 7 and 8, characteristically the low-lying and sandy inside of meanders (Table 4). Table 5 shows the width, amplication factor, average elevation, average hydraulic conductivity and area of trees for each oodplain division. The area of trees indicates the area of oodplain supporting black box and red gum eucalypts, which have high water requirements compared with bare ground, lignum or grassland vegetation. The

Table 4 Division 1 2 3 4 5 6 7 8 9 10 11 Total

Groundwater balance for each oodplain division Highland inow (m3 d1) 9.60 7.22 50.78 79.57 131.84 147.26 164.53 140.01 126.09 239.85 389.72 1486.48 Net recharge (m3 d1) 7.16 0.03 4.56 0.71 0.00 0.00 2.41 3.47 0.00 0.00 0.54 18.87 Net ET (m3 d1) 94.05 26.32 54.92 13.03 14.89 35.18 155.09 116.35 15.65 16.22 6.77 548.47 Seepage (m3 d1) 0.00 0.00 0.00 0.00 0.85 9.39 20.60 13.15 15.19 23.02 0.00 82.20 Baseow (m3 d1) 73.93 3.90 6.35 100.31 87.80 141.81 45.97 34.25 136.62 204.64 376.54 896.02

90
Table 5 Division 1 2 3 4 5 6 7 8 9 10 11 Floodplain characteristics used in the analysis, varying with division Floodplain width (m) 1492 496 830 461 494 626 2059 1895 544 261 119 Amplication factor () 12.36 2.00 3.90 2.00 1.11 0.79 11.17 9.33 1.27 0.93 1.00 Area trees (ha) 39.22 14.64 16.44 4.78 3.60 8.44 52.44 36.68 2.04 2.59 2.54 Mean height above river at break of slope (m) 3.91 4.13 3.99 3.96 3.42 3.13 2.90 3.41 3.35 3.56 3.70

R. Doble et al.

Kv average (m d1) 0.0085 0.0034 0.0042 0.0048 0.0033 0.0040 0.0045 0.0028 0.0018 0.0028 0.0017

description of area of trees in the following sections dene low tree area as 010 ha of trees, medium tree area as 10 30 ha and high tree area as more than 30 ha of trees. The spatial patterns of groundwater discharge tended to be quite robust, and did not change signicantly when the groundwater discharge relationship was altered. The quantitative rate of groundwater discharge was far more sensitive to changes in the discharge function than the spatial patterns themselves. Conned and unconned aquifer responses The evapotranspiration output for which surface conning effects were not modelled (Fig. 18) retains the overall spatial effects of the conned case (Fig. 10). Groundwater discharge is still predominant at the break of slope, ends of peninsulas and in the low-lying palaeochannel, but is higher than for the conned case. Seepage is especially more pronounced at the break of slope, as this area tends to be represented by a heavier clay with lower vertical conductivity, and is therefore more conning of vertical groundwater movement. The conning effects of the Coonambidgal Clay had little impact on the spatial patterns of groundwater discharge other than reducing the rates of groundwater seepage with respect to overall discharge rates. Groundwater discharge was lowered overall, but patterns associated with vegetation types and microtopography were retained.

The distribution of groundwater discharge with distance from the break of slope (Fig. 17) conrms that while the unconned scenario gives higher rates of groundwater discharge across the entire oodplain, the difference is more pronounced in the rst 200 m from the break of slope where seepage occurs. In addition to the smaller vertical conductivity in this location, the conning effects from the Coonambidgal Clay are most signicant where the watertable is close to or above the ground surface, such as at the break of slope, as the distance of upward uid ow through clay before evaporation occurs is greater. Modelling the surface conning layers may not be necessary in situations with deeper groundwater or coarser soils, but should be considered in situations where seepage or high discharge rates from very shallow groundwater are of interest.

Interpretations of model results

The remaining sections of this paper are directed toward the development of relationships between evapotranspiration and factors such as oodplain geometry, elevation and distribution of vegetation, soil parameters. Surface elevation is a key parameter in determining where salt accumulation will occur, as the greatest rates of ground-

Distribution of groundwater discharge with distance from break of slope

Confined Unconfined 250 200 150 100 50 0

Discharge (m 3/d)

40 0

60 0

80 0

20 0

Distance from break of slope (m)

Figure 17 Distribution of groundwater discharge with distance from break of slope for conned and unconned models. The greatest difference occurs in the seepage area close to the break of slope.

14 00

10 00

12 00

16 00

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

91

Figure 18

Predicted groundwater discharge on Clarks Floodplain, unconned model.

water discharge by evapotranspiration will be located in low lying areas where the depth to watertable is a minimum. Microtopography, or small-scale variation in elevation, denes both the areas of groundwater discharge and the frequency of ood inundation. As ecosystem communities are linked with microtopography, it also denes the impact on biodiversity in the region (Cramer and Hobbs, 2002). Soil type (Talsma, 1963; Warrick, 1988), proximity to the river and vegetation characteristics (Holland, 2002) will also affect the extent of salt accumulation. The episodic leaching of salt from the soil prole by ood-induced recharge further complicates the establishment of salt stores. Floodplain geometry will affect distribution of groundwater ow to a river or water body, as observed in the embayments of Green Bay and Lake Michigan (Cherkauer and McKereghan, 1991), and meanders of the Swan River, Perth (Linderfelt and Turner, 2001). The sinuosity of the river, and associated geometry of the oodplain is also likely to determine the distribution of groundwater discharge from a oodplain environment. It was hypothesised that each of these parameters would affect the distribution of long term groundwater ux and therefore salt accumulation. The order of importance of these parameters for dening discharge patterns, however, was unknown.

The average rate of evapotranspiration for a division, calculated at steady state, was compared with average elevation above the river, oodplain width and vertical hydraulic conductivity. Amplication or divergence of the division, that is, the length of river boundary divided by the width of the division at the break of slope was also compared with evapotranspiration and baseow, as was the proportion of healthy and total trees found in each division. These results were compared with those from the uniform model. Results from the elevation only example were almost identical to the uniform model, and are therefore not shown.

Dependence of seepage on local elevation

Constriction of groundwater ow as it moves from the unconned highland into the partially conned oodplain aquifer led to groundwater heads at the break of slope exceeding the oodplain elevation and caused seepage to occur. Seepage was greatest in the area of the palaeochannel, which corresponded to the points of lowest elevation along the cliff. These predictions match with observed seepage at the site. There was very little correlation between the fraction of inow from the highland expressed as seepage and the mean oodplain elevation for each of the ow divisions (R2 = 0.12). Comparison with the mean elevation of a 200 m wide strip of oodplain at the break of slope where groundwater heads are highest however, gave an inverse correlation (R2 = 0.61; Fig. 19). No grouping of divisions with high tree areas or wide oodplain (points 1, 7 and 8) were found, suggesting that, as expected, vegetation type had far less effect on seepage than elevation or geomorphology. Seepage is an elevation controlled process, and is most affected by low depths to watertable, whether by low surface elevation, or high groundwater table. While the volume of seepage is dependent on inow rates, seepage only appears in divisions with break of slope elevations lower than 3.7 m above the river stage. Apart from changes in hydraulic head, and therefore depth to

Analysis
The oodplain was divided into a series of ow divisions (Fig. 1) in order to compare characteristics of the oodplain within each division. The ow lines were dened as the ow path that groundwater took from a set of equally spaced points along the edge of the oodplain to the river, and were determined by generating a set of equipotential contours with groundwater ow being perpendicular to these and toward the river. This minimised groundwater ow between ow divisions. Zonebudget (Harbaugh, 1990) was used to perform water balance calculations for each of these divisions.

92
Seepage/Highland Inflow - Mean Elevation Above River at Break of Slope
0.2 Seepage/Infow
7
2

R. Doble et al.
ET - Floodplain Width
200.0 R = 0.96 ET (m /d)
R = 0.61 0.1
8
2

150.0
3

100.0 50.0
Increasing tree area High tree area

0.0 2.7

2.9 3.1 3.3 3.5 3.7 3.9 4.1 Mean elevation above river at break of slope (m)

4.3

0.0 0 500 1000 1500 2000 2500 Floodplain Width (m)

Figure 19 Seepage/highland inow vs mean elevation above river at break of slope, showing highly treed divisions in each subset. Vegetation coverage has little signicance for groundwater seepage. No relationship can be derived between local elevation and seepage from the uniform model results, as no elevation variation was included within the model.

groundwater, no other geometric factor signicantly affected the seepage rate from the oodplain.

Figure 21 Evapotranspiration vs oodplain width. There is a strong correlation for the modied model results (), despite the highly variable inows for each of the divisions. The relationship between ET and oodplain width for the uniform MODFLOW model ( ) is much less dened, and suggests that the ET is more dependent on vegetation than oodplain width alone.

Effects of vegetation coverage and oodplain width

The spatial distribution of vegetation across the oodplain had the greatest overarching effect on patterns of groundwater discharge. The area of high water using trees within each oodplain division correlated very closely with evapotranspiration (R2 = 0.96; Fig. 20). While the maximum rate of groundwater discharge was found at the break of slope, associated with shallow groundwater and seepage, the total volume of discharge from the eucalypt communities was also high due to the large areas of vegetation with high water requirements and deep extinction depths. Wider sections of oodplain result in a longer groundwater travel time within the oodplain. As the hydraulic head is constrained at both the break of slope and the river, longer ow paths reduce the hydraulic gradient and groundwater velocity. This consequently subjects groundwater to evapotranspiration for longer periods of time. Larger oodplain widths also provide a greater area over which evapotranspiration may take place. Fig. 21 shows a strong relationship between the rate of evapotranspiration from

a division and the oodplain width (R2 = 0.96) despite varying groundwater inows from the highland. However, the area of trees within a division was also found to increase with oodplain width, indicating some interdependence. Results from the uniform MODFLOW model showed no relationship with either tree area, which was expected due to the lack of vegetation parameters, nor oodplain width, despite the inclusion of oodplain spatial geometry within the uniform model. The lack of relationship between oodplain width and evapotranspiration in the uniform model conrms that ET is more dependent on vegetation distribution than oodplain width alone, and that inclusion of the spatial patterns of vegetation is fundamental for studies of groundwater discharge on a sub-oodplain scale.

Dependence of groundwater discharge on oodplain elevation and soil type

Warrick (1988), Gardner and Fireman (1958) and Thorburn et al. (1992) show that the relationship between watertable depth and evaporation in soils is an inverse power relationship. It was hypothesised that given a relatively level piezometric surface, oodplain elevation, which varies more rapidly in a spatial context than groundwater elevation, is likely to be the primary variable for depth to watertable. Relative elevation of the oodplain is therefore thought to govern distribution of groundwater discharge on a spatial scale. Fig. 22 shows evapotranspiration rates plotted against oodplain height relative to the river level. While there are no strong overall trends, there are sub-groups within the data set dened by the area of trees present within the division. The divisions with a high tree area exclusively represent the group with the highest ET rates. The second group, below this, is represented by divisions with a low to medium area of trees, and shows a trend of increasing ET with decreasing elevation (R2 = 0.64). The location of low-lying regions within a division will also dictate the relative effect of surface elevation. Lowlying areas that coincide with the break of slope (divisions 59) or eucalypt communities (ends of divisions 1, 3, 7

ET vs Area of trees
200 R = 0.96 150 ET (m /d)
3 2

100 50 0 0 10 20 30 Tree Area (ha) 40 50 60

Figure 20 Evapotranspiration vs area of trees. There is a strong positive trend for the modied modelling results (), even for divisions represented by points 1 and 2, which lie below the trend line, and have extremely low inow. There is no relationship between ET and tree area for the uniform MODFLOW model ( ).

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

ET - Floodplain Height Above River
200.0 150.0 100.0 50.0 0.0 3 3.2 3.4 3.6 3.8 Mean Height Above River (m) 4
Med-low tree area High tree area

93

this overall trend is overwhelmed by the effects of vegetation type on ET.

ET (m /d)

Effects of oodplain amplication on baseow

The ratio between river frontage of a oodplain division to its width at the break of slope, or amplication factor, strongly inuenced where baseow to the river occurred. High amplication factors were associated with negative baseow rates (Fig. 24), and within the divisions with low areas of trees, baseow increased with both higher groundwater inows and lower amplication rates. Divisions with long river frontages coincided with a wider oodplain and greater area of trees, facilitating high evapotranspiration rates adjacent the river. To achieve this evaporation rate, river water inltrates into the alluvial aquifer, facilitated by the lower groundwater heads under the eucalypt communities. This behaviour has been recorded in riparian environments in the San Pedro River, described by Mac Nish et al. (2000) and Stromberg et al. (1996). While there was a notable correlation between baseow and division amplication or river frontage, the area of trees present on a oodplain division was also represented in data clusters. The interdependent nature of river frontage and vegetation coverage made these parameters difcult to distinguish, and suggested that baseow was dened by a combination of river geomorphology and ecology, rather than any single parameter. Fig. 24 shows that the uniform model results were similar to the more detailed modelling for oodplain divisions with low amplication factors and low tree cover. Baseow followed the same trend, increasing with rising highland inows. For divisions with high amplication factors, however, the lack of spatial variation in the ET function meant the uniform model still predicted positive baseow. Baseow was shown to be inuenced rstly by groundwater inows to the oodplain, then by amplication factor and therefore also by proximity to the highland. Spatially variable ET rates were only important for areas further from

Figure 22 Evapotranspiration vs oodplain elevation above the river. Note the cluster of points within tree area classes of high, and low and medium. No relationship can be derived between elevation and evapotranspiration from the uniform model results, as no elevation variation was included within the model.

and 8) are likely to produce higher discharge rates than if they coincide with grassland or low water using lignum (Muehlenbeckia cunninghamii) covered areas further from the break of slope (centre of division 8). Within each vegetation class, elevation is also a factoring determining patterns of groundwater discharge. The effects of localised elevation patterns (Fig. 6) are visible within different vegetation areas, rather than between them, but as elevation is an indicator of ooding frequency, vegetation distribution and oodplain elevation are not completely independent variables. The weaker dependence on elevation than expected is potentially as a result of the higher recharge rates associated with lower lying areas. Although lower elevation leads to higher rates of evapotranspiration, it is partly offset by increased recharge through more frequent inundation, so the sensitivity of net ux to elevation is lessened. Groundwater discharge and average vertical hydraulic conductivity (Kv) of a division appear to have a limited relationship for Clarks Floodplain (R2 = 0.12 for increasing discharge with increasing Kv) as shown in Fig. 23. Separating two groups of data points, high and low tree area, reveals a slight positive trend in discharge with increasing Kv, but

Baseflow - Amplification Factor

400.0 300.0 Baseflow (m /d) 200.0 100.0 0.0 0.0 -100.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0
Increasing highland inflow Med tree area
3

ET and Seepage - Average Kv

200.0 ET+Seepage (m /d)
3

150.0 100.0 50.0 0.0 0 0.002

High tree area Low-med tree area

High tree area

Amplification Factor

0.004 0.006 Average Kv (m)

0.008

0.01

Figure 23 Evapotranspiration and seepage vs averaged Kv for each oodplain division. There is a slight positive trend, but the high tree area class divisions overwhelm the results. No relationship can be derived between vertical conductivity and evapotranspiration from the uniform model results, as no conductivity variation was included within the model.

Figure 24 Baseow vs division amplication factor. Note again the separate tree area classes, and the effect of inow within the low tree area class within the modied model result results (). Uniform model results ( ) show a similar trend for baseow in divisions with low amplication factors, but baseow in the divisions with high amplication factors and high tree areas is positive in contrast to the negative baseow of the modied model.

94 the highland. This indicates that while patterns of ET are sensitive to spatially variable model inputs, baseow patterns are relatively robust.

R. Doble et al.

Linking discharge with salt accumulation

Net ux output from the model can also be used to understand the spatial patterns of salt accumulation. Salt accumulation rates were predicted by multiplying the coverage of groundwater discharge rates (Fig. 10) with a spatial map of local groundwater salinity determined from eld studies. High rates of salt accumulation are predicted at the break of slope, and also on the outsides of the bend in divisions 37 (Fig. 25). Although Fig. 10 shows a high discharge rate from red gum and black box vegetation at the ends of the peninsulas, long-term salt accumulation is minimal (Fig. 25). Local groundwater salinity at the ends of peninsulas is low, as most saline regional groundwater has been discharged as baseow at much narrower parts of the oodplain. High water use from vegetation has lowered groundwater tables and led to an inux of fresh water into the alluvial aquifer from the river. Vegetation health mapping (Telfer and Overton, 1999) shows a concentration of dead trees or trees in poor health around the backwater on Clarks Floodplain (Fig. 3), corresponding with areas of high salt accumulation (Fig. 25). Poor vegetation health is also evident in the combined red gum and black box stands around the outside of the bend correlating with higher salt accumulation within divisions 47. High rates of salt accumulation indicate areas where vegetation health or water bodies are under threat from salinisation, and focus where remediation and management efforts should be directed. Methods such as direct irrigation during drought, targeted spear-point pumping of groundwater below affected vegetation and ood augmentation may be used to reduce the salt impact on oodplain ecosystems.

Regression
A linear multiple regression was undertaken, comparing the modelled evapotranspiration with highland inow, oodplain width, area of trees, amplication factor, average Kv and mean height of oodplain section above the river level. This indicated the relative contributing factors to evapotranspiration, and allowed a series of variables to be compared with evapotranspiration simultaneously. A relationship for evapotranspiration was derived with an R2 value of 0.998. In order of signicance, the parameters contributing to evapotranspiration were the area of trees, highland inow, oodplain width and mean height above the oodplain, with P < 0.05 indicating signicance (Table 6). Amplication factor and average Kv had little effect on patterns of evapotranspiration. Baseow, oodplain width and tree area were highly interdependent, with wider oodplain divisions also having high tree coverage and negative baseow. Conversely, narrow divisions tended to have low tree coverage and higher, positive baseow. Results from the regression supported the conclusions drawn from the gures above.

Table 6

Regression results for evapotranspiration P-value

3 1

Parameter Highland inow (m d ) Floodplain width (m) Area trees (ha) Amplication factor () Mean height above river (m) Kv average (m d1)

0.015 0.020 0.003 0.325 0.041 0.904

Interrelationships between vegetation and groundwater hydrology

The analysis section of this paper has treated vegetation distribution as an independent variable in order to ascertain

Figure 25 Modelled salt accumulation on Clarks Floodplain. High salinity at the break of slope from regional groundwater inows causes salt accumulation to be highest in this area, while low salinity at the ends of the peninsulas due to fresh inows from the river lowers the rate of salinisation.

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge causal factors for groundwater ux distribution, when in reality it has complex relationships with geomorphology, groundwater hydrology and salinity. Vegetation with high water requirements such as the red gum eucalypt is more likely to be found in areas with access to fresh groundwater, such as the inside of meanders or other sandy deposits with good connection to the river. High rates of evapotranspiration are able to occur as a result of the available fresh water. Floodplain width, tree area and amplication factor were found to be highly correlated, with correlation coefcients ranging between 0.93 and 0.96, with wider oodplain divisions having high tree coverage and high amplication due to the presence of meanders. Correlation was less pronounced for oodplain elevation, soil type and highland inow. The value of these results lies in determining the most critical parameters to include in groundwater models to adequately represent groundwater discharge processes on a ne spatial scale. In nite difference numerical models, varying the evapotranspiration function spatially according to vegetation distribution will give a more appropriate representation of groundwater ow processes than uniform parameters, or variation of evapotranspiration with soil or other factors. For full ecological analyses, however, the lack of feedback on groundwater salinity from salt accumulation will be a disadvantage for temporal investigations of vegetation health responses using this methodology.

95

ter ow divisions. Floodplain elevation and width had a lesser effect, but the distribution of vegetation related to both of these parameters. Localised elevation is the most critical parameter in determining the proportion of inow to a oodplain expressed as seepage and is signicant in determining spatial patterns of evapotranspiration. Geomorphic features such as backwaters, lagoons and palaeochannel, which are commonly found in meandering river systems, can produce elevation changes that signicantly effect the spatial distribution of discharge. The complex meandering patterns of the river dened where baseow occurs on a ne scale. Higher proportions of baseow are associated with areas of oodplain that are close to the break of slope, and where river frontage, or division amplication is low. Negative baseow was found on the inside of river bends where eucalypt communities are present, and where division amplication is high. The uniformly parameterised model was able to represent baseow with some degree of accuracy, but could not predict the negative baseow associated with high vegetation water use. The uniformly parameterised and more complex distributed models produced similar baseow distribution despite very different spatial patterns of evapotranspiration. While many groundwater models are calibrated or validated against baseow calculations, further comparison with spatial indicators of groundwater discharge, such as vegetation water use or salt accumulation maps are required when vertical groundwater ux is of interest. Although long-term patterns of net groundwater discharge were found to depend on vegetation distribution, elevation, soil type and river geometry, the relationship between these interdependent parameters is complex, and based on the geomorphology and ecology of the riverine system. The value of these results, however, lies in determining the most critical parameters to include in groundwater models to adequately represent spatial groundwater discharge processes.

Conclusions
This paper has provided a discussion of the importance of geometry, geomorphology and elevation in determining the spatial distribution of discharge from a oodplain. It used a quasi-three-dimensional MODFLOW model to conrm the critical parameters required to represent groundwater discharge patterns affecting vegetation health, and compared these results with a uniform MODFLOW model using no spatial distribution of vegetation or elevation parameters. The key ndings from this study are that: To assess the magnitude of the impact on oodplain biodiversity and ecological value, groundwater modelling needs to take into account spatial patterns of groundwater discharge and rates of salinisation on a ne scale. A uniform model with no spatial variation or recharge information was not able to accurately represent spatial distributions of evapotranspiration. Progressively more information on the spatial patterns of discharge was gained when moving from a uniform parameter oodplain model to variation of elevation, to variation of vegetation and soil parameters. Better correlations between model results and eld measurements were achieved when vegetation parameters were included in the modelling. Due to the large range of vegetation present on the oodplain studied, the distribution of vegetation type had the most signicant effect on the spatial distribution and magnitude of groundwater discharge between groundwa-

Acknowledgements
The rst author acknowledges the support of Land and Water Australia and the Centre for Groundwater Studies. The work was also supported through the project Assessing Current and Future Impacts of Land Management Induced Groundwater Discharge on Floodplain Health, which is a partnership between the River Murray Catchment Water Management Board (RMCWMB), CSIRO Land and Water, the South Australian Department of Environment and Heritage, and South Australian Department for Water, Land and Biodiversity Conservation (DWLBC). The project was funded through the Natural Heritage Trust and by the project partners, and managed by RMCWMB.

References
Akeroyd, M.D., Tyerman, S.D., Walker, G.R., Jolly, I.D., 1998. Impact of ooding on the water use of semi-arid riparian eucalypts. Journal of Hydrology 206, 104117.

96
Armstrong, D., Yan, W., Barnett, S.R., 1999. Loxton irrigation area groundwater modelling of groundwater/river interaction. Report for the Department of Primary Industries and Resources, Adelaide, South Australia. AWE, 1999. Bookpurnong and lock 4 preliminary land and water management plan. Background report and appendices. Report for the Bookpurnong and Lock 4, Environmental Group Inc. Banta, E.R., 2000. MODFLOW-2000, the US geological survey modular ground-water model-documentation of packages for simulating evapotranspiration with a segmented function (ETS1) and drains with return ow (DRT1). US Geological Survey OpenFile Report 00-466, 127p. Barnett, S.R., SA Department of Mines and Energy, 1991. Renmark Hydrogeological Map (1:250,000 Scale). Bureau of Mineral Resources, Geology and Geophysics, Canberra, Australia. Berens, V., Hatch, M., Wilson, T., Hill, T., 2004. River-borne NanoTEM survey for location of salt accession to the River Murray at Loxton. In: Proceedings of the ASEG 17th Geophysical Conference and Exhibition, Sydney, Australia. Boone, A., Wetzel, P.J., 1996. Issues related to low resolution modeling of soil moisture: experience with the PLACE model. Global and Planetary Change 13, 161181. Cherkauer, D.S., McKereghan, P.F., 1991. Ground-water discharge to lakes: focusing in embayments. Ground Water 29 (1), 7280. Cramer, V.A., Hobbs, R.J., 2002. Ecological consequences of altered hydrological regimes in fragmented ecosystems in southern Australia: Impacts and possible management responses. Austral Ecology 27, 546564. Doble, R.C., Walker, G.R., Simmons, C.T., 2005. Understanding spatial patterns of discharge in semi-arid regions using a rechargedischarge balance to determine vegetation health. CSIRO Land and Water Technical Report 13/05. Evans, W.R., Kellett, J.R., 1989. The hydrogeology of the Murray Basin, southeastern Australia. BMR Journal of Australian Geology and Geophysics 11, 147166. Feikema, P.M., 2000. Tree growth and water relations in eucalypt plantations over shallow saline groundwater. PhD thesis for the Department of Civil Engineering, Monash University, Melbourne Australia. Freeze, R.A., Cherry, J.A., 1979. Groundwater. Prentice-Hall, Inc., New Jersey, USA. Gardner, W.R., Fireman, M., 1958. Laboratory studies of evaporation from soil columns in the presence of a water table. Soil Science 85, 244249. Gardner, W.R., 1958. Some steady state solutions of the unsaturated moisture ow equation with application to evaporation from a water table. Soil Science 85, 228232. Goodrich, D.C., Scott, R., Qi, J., Goff, B., Unkrich, C.L., Moran, M.S., Williams, D., Schaeffer, S., Snyder, K., MacNish, R., Maddock, T., Pool, D., Chehbouni, A., Cooper, D.I., Eichinger, W.e., Shuttleworth, W.J., Kerr, Y., Ni, W., 2000. Seasonal estimates of riparian evapotranspiration using remote and in situ measurements. Agricultural and Forest Meteorology 105, 281 309. Harbaugh, A.W., 1990. A computer program for calculating subregional water budgets using results from the US geological survey modular three-dimensional ground-water ow model. US Geological Survey Open-File Report 90-392. Harbaugh, A.W., Banta, E.R., Hill, M.C., McDonald, M.G., 2000. MODFLOW-2000, The US Geological Survey Modular Ground-water Model User Guide to Modularization Concepts and the Ground-water Flow Process. US Geological Society, Reston, VA. Hatton, T., 1998. Catchment scale recharge modelling. In: Zhang, L., Walker, G.R. (Eds.), Part 1 of Studies in Catchment Hydrology: The Basics of Recharge and Discharge. CSIRO, Vic., Australia. Available from: <http://www.publish.csiro.au/nid/ 18/pid/3286.htm>.

R. Doble et al.
Holland, K.L., 2002. Blackbox (Eucalyptus largiorens) tree health and groundwater discharge on the lower river Murray oodplain, South Australia: linking spatial and temporal scales. PhD thesis, Flinders University of South Australia. Jarwal, S.D., Walker, G.R., Jolly, I.D., 1996. General site description. In: Walker, G.R., Jolly, I.D., Jarwal, S.D. (Eds.), Salt and Water Movement in the Chowilla Floodplain. CSIRO Division of Water Resources, pp. 16309 (Chapter 2). Jolly, I.D., 1996. The effects of river management on the hydrology and hydroecology of arid and semi-arid oodplains. In: Anderson, M.G., Walling, D.E., Bates, P.D. (Eds.), Floodplain Processes. Wiley, Chichester, New York, pp. 577609. Jolly, I.D., Narayan, K.A., Armstrong, D., Walker, G.R., 1998. The impact of ooding on modelling salt transport processes to streams. Environmental Modelling and Software 13, 87104. Jolly, I.D., Walker, G.R., Thorburn, P.J., 1993. Salt accumulation in semi-arid oodplain soils with implications for forest health. Journal of Hydrology 150, 589614. Jolly, I.D., Walker, G.R., McEwan, K.L., Thorburn, P.J., Hatton, T.J., Bennett, J., Hacker, J.M., 1991. The role of oodplain vegetation in salt accession to the River Murray in South Australia. In: Proceedings of the International Hydrology and Water Resources Symposium, Perth. The Institution of Engineers, Australia. Linderfelt, W.R., Turner, J.V., 2001. Interaction between shallow groundwater, saline surface water and nutrient discharge in a seasonal estuary: the Swan-Canning system. Hydrological Processes 15, 26312653. Mac Nish, R.D., Unkrich, C.L., Smythe, E., Goodrich, D.C., Maddock, T., 2000. Comparison of riparian evapotranspiration estimates based on a water balance approach and sap ow measurements. Agricultural and Forest Meteorology 105, 271 279. McDonald, M.G., Harbaugh, A.G., 1996. Users documentation for MODFLOW-96, An update to the US Geological Survey modular nite-difference ground-water ow model. Scientic Software group, Washington, DC, USA. Overton, I.C., 2005. Modelling oodplain inundation on a regulated river: integrating GIS, remote sensing and hydrological models. River Research and Applications 21, 9911001. Peck, A.J., 1978. Note on the role of a shallow aquifer in dryland salinity. Australian Journal of Soil Research 16 (2), 237240. Petheram, C., Dawes, W., Grayson, R., Bradford, A., Walker, G., 2003. A sub-grid representation of groundwater discharge using a one-dimensional groundwater model. Hydrological Processes 17, 22792295. Porter, B., 2002. Daily river ows into South Australia. Report for the Department of Water, Land and Biodiversiy Conservation. Silberstein, R., Vertessy, R.A., Morris, J.D., Feikema, P., 1999. Modelling the effects of soil moisture and solute conditions on tree growth and water use. Agricultural Water Management 39, 283315. Slavich, P.G., Walker, G.R., Jolly, I.D., 1999. A ood history weighted index of average root-zone salinity for assessing ood impacts on health of vegetation on a saline oodplain. Agricultural Water Management 39, 135151. Stromberg, J.C., Tiller, R., Richter, B., 1996. Effects of groundwater decline on riparian vegetation of semiarid regions: the San Pedro, Arizona. Ecological Applications 6, 113131. Talsma, T., 1963. The control of saline groundwater. Mededelingen Van De Landbouwhogeschool Te Wageningen Nederland 63, 1 68. Telfer, A., Overton, I.C., 1999. The impact of irrigation on oodplain vegetation health adjacent the River Murray. In: Rutherford, I.D., Bartley, R. (Eds.), Proceedings of the Second Australian Stream Management Conference, Adelaide. Thorburn, P.J., Hatton, T.J., Walker, G.R., 1993. Combining measurements of transpiration and stable isotopes of water to

Spatial relationships between vegetation cover and irrigation-induced groundwater discharge

determine groundwater discharge from forests. Journal of Hydrology 150, 563587. Thorburn, P.J., Walker, G.R., Woods, P.H., 1992. Comparison of diffuse discharge from shallow water tables in soils and salt ats. Journal of Hydrology 136, 253274. Thorburn, P.J., 1996. Can shallow water tables be controlled by the revegetation of saline lands? Australian Journal of Soil and Water Conservation 9, 4550. Thorburn, P.J., Walker, G.R., Jolly, I.D., 1995. Uptake of saline groundwater by plants: an analytical model for semi-arid and arid areas. Plant and Soil 175, 111.

97

Thorburn, P.J., Walker, G.R., 1994. Variations in stream water uptake by Eucalyptus camaldulensis with differing access to stream water. Oecologia 100, 293301. Twidale, C.R., Lindsay, J.M., Bourne, J.A., 1978. Age and origin of the Murray River and gorge in South Australia. Proceedings of the Royal Society of Victoria 90, 2742. Warrick, A.W., 1988. Additional solutions for steady-state evaporation from a shallow water table. Soil Science 146 (2), 6366. Wind, G.P., 1955. A eld experiment concerning capillary rise of moisture in a heavy clay soil. Netherlands Journal of Agricultural Science 3, 6069.