Sie sind auf Seite 1von 30

Ecological Modelling, 46 (1989) 165-194 Elsevier Science Publishers B.V.

, Amsterdam - Printed in The Netherlands





Alligator Rivers Region Research Institute, Office of the Supervising Scientist, P.O. Box 387, Bondi Junction, N.S.W. 2022 (Australia)
(Accepted 9 January 1989)

ABSTRACT Vardavas, I.M., 1989. A water budget model for the tropical Magela flood plain. Ecol. Modelling, 46: 165-194. A water budget model is presented for the tropical Magela flood plain, which is located downstream from the Ranger uranium mine in the Northern Territory, Australia. Using input data on rainfall and water discharge from Magela Creek, the model provides daily estimates of the volume of surface water on the flood plain and the rate of discharge at the outlet of the flood plain. Annual variations in the relationship between gauge height and rate of discharge at the outlet are interpreted in terms of a correlation between early wet-season rainfall and seed germination On the flood plain. The daily volume of water on the flood plain can be used to estimate the dilution and subsequent evapoconcentration of any waterborne contaminants that may be transported by Magela Creek from the mine site downstream to the flood plain. The model was validated by testing its ability to predict the measured daily water depth at the outlet channel; comparison of prediction with 12 years' data produced an average error of about 16%.

INTRODUCTION T h e R a n g e r u r a n i u m m i n e in A u s t r a l i a ' s N o r t h e r n T e r r i t o r y is l o c a t e d close to M a g e l a C r e e k (Fig. 1), a w a t e r c o u r s e in w h i c h w a t e r flows o n l y in the wet season. S o m e 12 k m d o w n s t r e a m f r o m the m i n e site the creek enters its associated f l o o d p l a i n s y s t e m which, in a typical year, has a n i n u n d a t e d area of a b o u t 200 k m 2. T h e w e t l a n d s o f the f l o o d p l a i n region are rich in n u m b e r s o f species of a q u a t i c a n i m a l s a n d p l a n t s a n d f o r m p a r t of K a k a d u N a t i o n a l Park. W a s t e w a t e r f r o m the m i n e site c a n enter the surface w a t e r s y s t e m either b y discharge or b y seepage f r o m r e t e n t i o n p o n d s or the tailings i m p o u n d m e n t structure. A s s e s s m e n t s of the p o t e n t i a l i m p a c t of such d i s c h a r g e s o n


Mudginbenl Conidor

H n / v y

Gauging station Pltiograph Catchment boundary Escarpment

Fig. 1. Magela Creek catchment and flood plain.

the ecosystem and on Aboriginal food gathering activities are being undertaken. As part of these assessments a suitable hydrological model of Magela Creek and its associated flood plain is needed to enable estimates to be made of the effects of dilution and evapoconcentration on the constituents of any water originating from the mine. Such a model has been published for Magela Creek (Vardavas, 1988); in the present paper a hydrological model for the Magela flood plain is described.


The model has been validated by examining its ability to predict the measured daily water depth above the flood plain's outlet levee for the twelve years for which full data sets are available; it does so successfully, with an average error of 16%, using a single set of nine parameters selected by optimising the agreement between measured and model water depth for only four of those years. The model provides daily estimates of the volume of surface water on the flood plain and the discharge rate at the outlet channel. The results indicate that there is a correlation between early wet season rainfall and channel roughness, i.e. density of vegetation cover (or standing crop) lining the bed and banks of the outlet channel.

Magela Creek flood plain, shown in Fig. 1, covers an area of about 200 km 2 and receives water from the surrounding lowlands ( ~ 780 km 2) and from the Magela Creek catchment ( ~ 600 km 2) upstream of gauging station GS821009. The region's climate is monsoonal with a distinct wet season commencing, on average, in November and ending in April. The dry season, extending from May to October, experiences less than 5% of the annual average rainfall. The average annual rainfall is about 1550 mm while the average annual evaporation rate is about 2000 mm (Vardavas, 1987). Magela Creek begins to flow about a month after the commencement of the wet season and flow ceases by the end of April. The wet-season flow on the flood plain is high during floods but outside flood periods water moves very slowly over the flood plain because of its low gradient, wide expanse and dense aquatic vegetation. The flood plain discharges into the East Alligator River usually over the months February to July, but water can persist in some swamps, and depressions at the edge of the plain and several billabongs (deep pools) are permanent. These surviving waterbodies play a vital role in the ecology of the area by providing water through the dry season for land animals, large numbers of water birds, fish and other aquatic organisms. The amount of water and number of waterbodies remaining on the flood plain during the dry season varies markedly from year-to-year, depending on annual rainfall and the length of the wet season. The flow out of the flood plain is impeded by Oryza grasslands and Melaleuca open forest and woodland at the outlet, while upstream towards the Mudginberri Corridor the flow is impeded by Hymenachne and Pseudoraphis grasslands. The western central arms of the plain are covered by Eleocharis sedgelands and swamps (e.g. Finlayson, 1986). In a recent publication, Hart et al. (1986) reported the nutrient and trace metal fluxes carried by Magela Creek to the flood plain and concluded that


Interception Store MogelOcotchmentCreek Qr Lowlonds Catchment IQi [ JU Surfoce Store D , Vd J


Wedge Store T , V.

Prism Store H , Vp

- Qo

Soil Stores

t p Groundwcoter Store 7 G

Fig. 2. Flood plain water flow model. most of the heavy metal transport occurs during high flow periods and that it is largely associated with particulate matter. The volume of water that is transported to the flood plain determines the amount of heavy metals on waterborne material, their initial dilution and subsequent evapoconcentration. Rainfall onto the flood plain is monitored by three pluviographs (R821019A, -019B and -017), the water depth at the outlet is monitored by gauging station GS821019, while the discharge of water from the Magela Creek catchment is monitored by gauging station GS821009, at the major 'inlet' to the flood plain (see Fig. 1). MATHEMATICAL MODEL A flow diagram of the model is given in Fig. 2. The flood plain system is considered to comprise two feeder catchments, that of Magela Creek and of the lowlands surrounding the plain. These provide the input runoff Qr = Q m + Q] to the flood plain which itself provides a catchment of about 200 km 2 for direct rainfall P. Part of the rainfall is intercepted by the vegetation cover on the plain while any excess rainfall, U, beyond the capacity of this interception store, L, and the input runoff contribute to the water filling up the storage D at the surface. This surface store consists of the many depressions, stores created by levees and a high porosity uppermost alluvium layer on the flood plain. Once the store below this alluvium layer has been filled to capacity, the infiltration, F, becomes negligible and the surface store is rapidly filled up by Qr and U. Water in excess of the capacity of the



~ '.
: ...

I oter Surface 2 3

Wedge Prism Surfoce

Store Store Store



~, ~

_1 C3 "1.< 6


iiii!iii !i!i!iiii ii
i i i
" " ', ..... 4 8

'~ ' 12 "'i 16 ',' 2~ ' ~ 24from ""; 28 " i 32 .... f 36 (kin) " downstreom GS821009


ii.i)i;ii!i iiil, i
', ' i 4-0 4-4


Fig. 3. Longitudinal cross-section along the centre of the flood plain. The water surface corresponds to a flood which occurred in February 1980.

surface store plus any lateral subsurface water movement or interflow, flS, and groundwater or base flow, 7G, flows downstream to the plain's outlet. The uppermost alluvial layer, which is part of the surface store in the present model, is about 0.5-1 m thick and has a high infiltration capacity. Below the alluvium layer is an = 10-m layer of blue-grey estuarine muds. This layer lies above a sandy alluvium layer of about 20 m thickness (East, 1987). Hence the soil and groundwater stores are saturated and are not expected to significantly affect the water budget of the flood plain and hence the contribution of interflow (flS) and base flow ('fG) to surface water storage is ignored as is infiltration ( F ) from the surface store down to the estuarine muds. Because of the travel time to the outlet, water entering the flood plain can be sufficiently delayed for the water surface to develop a significant average slope in excess of that arising from the slope of the flood plain surface. This backwater, Vw, is referred to as wedge-storage in Fig. 3. In the present model this travel time is effectively incorporated as a time delay by allowing only a fraction, , of this backwater store to reach the plain's outlet each day. The depth, H, of the water at the outlet defines the amount of water on the flood plain held in prism-storage, Vp, that is, water in excess of the surface store capacity, Vdm, but with zero backwater component. Discharge, Qo, at the outlet is constrained b y a channel which is about 5 km wide and b y the levee which forms the base of the channel. In Fig. 3 these three flood plain stores are depicted on a longitudinal section


taken along the centre of the flood plain (Uranium Province Hydrology, 1980). The example of the water surface shown corresponds to a storm event which started on 4 February 1980 (Uranium Province Hydrology, 1982).
Water conservation equations

The total volume of above surface water on the flood plain, VT (m3), is given by:

VT: Vd+ Vp+ Vw


where Vd, Vp and Vw are the volumes in the surface-store, prism-store and wedge-store, respectively. The surface store capacity, Vdm, defines the volume of water held on the flood plain before there is discharge above the outlet levee. The inundated surface area when the surface-store is filled to capacity, i.e. when Vd = Vdm, is taken to be the base area A 0 of the flood plain, where A 0 ~ 200 km 2. If Qin ( m3 d - l ) represents the daily net water input to the flood plain and Qo the discharge out of the flood plain then the conservation equation for the total water volume can be written as:

dVT/dt =

Qin - Qo


While the surface store is filling up there is no discharge out of the flood plain, i.e. Qo = 0, Vp = Vw = 0 and d V p / d t = d V w / d t = 0. Thus, from equation (2) the conservation equation for the surface-store volume, Vd, is then given by:
d V d / d t = Q,n = Qr + A o( U - ET)


where E T is the daily evapotranspiration rate (m d - l ) , and U is the rainfall not intercepted by the vegetation canopy. Once the surface-store is full and there is discharge, Qo, out of the flood plain, then Vd = Vdm, d V d / d t = 0, and V = Vp + Vw is given by:
d V / d t = Qr + A ( U -

ET) -- Qo


where A (m 3) is the inundated area and A > A 0. Since the wedge-store loses water to prism-storage at a rate V w (m 3 d -1) then the conservation equation for the wedge-store volume, Vw, can be written as:
d V w / d t = Qr + A ( U E I ) - ~Vw


and the conservation equation for the prism-store volume, Vp, is:
d V p / d t = KVw - Qo





Fig. 4. Average vertical cross-section of the flood plain adopted in the model.

I n u n d a t e d surface area

The surface area A inundated by water can be computed by adopting an average vertical cross-section for the flood plain as shown in Fig. 4. If a c represents the cross-sectional area delimited by the water surface of width b and the surface of the plain then the total volume, Vv, on the flood plain is given by VT = a c L o, where L 0 represents the effective length of the flood plain, assumed to be constant. The inundated surface area when the surface store is filled to capacity, i.e. when D = Din, is given by A 0 = b o L o. The vertical cross-section of the base of the flood plain is taken to be u-shaped, and the banks have an average slope tan ~. The volume of water held in surface storage can be written in terms of the average water depth D as:
Vd = A o D

(7) the surface store , the inundated with Vd so that A = ~ A 0 with prism-storage depth, H, and the be written as:

and because of levees and depressions in area, A, is assumed to simply increase = D I D m. The inundated area when the wedge-storage depth, T, are non-zero, can A = A 0 + 2 L 0 ( H + T ) / t a n q, and so now:
=A/A o = 1 + 2(H+ T ) / ( b o tan ~)


More generally:
= D/D m + 2(H+ T ) / ( b o tan ~)


with D < Dm, and the average width b of the inundated flood plain is given by b = ~b0. For the Magela Creek flood plain tan q,---0.01 ( U r a n i u m Province Hydrology, 1980) while the mean width bo ~ 4 km and since A 0 --- 200 km 2 then L 0 ~ 50 km.


Surface storage

The input to the surface store consists of catchment runoff Q r ( m3 d - a ) and a fraction of the direct rainfall, P (m d - a ) , onto the flood plain catchment area A 0, in excess of the interception store L provided b y the vegetation canopy. The interception store is computed as in Vardavas (1988) based on a quick wetting phase followed by a slower evaporation phase. After the wetting phase the interception store on day i is given by:
Li = Li-a + Pi


Any rainfall U in excess of the interception storage capacity L m flOWS down to the surface store, where:
Ui = n i - L m


If U > 0 then the interception storage becomes L~ = Era, whereas if U < 0 then U is set to zero. After the drying phase L i = L, - E T and if Li < 0 it is set to zero, where E T is the daily evapotranspiration rate. The conservation equation for the surface-store volume, Vd, is then given by:
d V d / d t = Qr + A o ( U -



Equation (13) can be re-written in terms of the characteristic depth, or volume per unit area A0, vd = V d / A o , as
dvd/dt = R + UEv = u


where R = Q i n / A o with vd = D. Equation (14) can be easily solved for day i if written in the form: Udi = Udi_ 1 q- A t ( u i q- U , _ a ) / 2 (15)

with the conditions that Vd0 = 0 at t = 0 and va~ _< Udm with At = 1 day. If Vd~> Odin then the excess water Ave = Vd~- Udm is transferred to wedge-storage and Vd, is set equal to Yam.
W e d g e a n d prism storage

Once the surface store is full, the rate at which water is transferred to wedge-storage is given by:
dve/dt= R + ~( U ET)

where now ~ is given by equation (9). The conservation equation for the wedge-storage can be written in terms of the characteristic wedge-store depth vw = V w / A o as:
d v w / d t = d v e / d t - xv w = f


173 where x represents the fraction of backwater or wedge-storage lost to prism-storage per day and is a function of runoff R. The function x ( R ) was chosen so that when R = 0 then x = x0, representing decay of the wedgestorage in the absence of any runoff input to the flood plain. Theoretically, when R >> 1 then x ~ 1, representing the rapid transfer of all the backwater to prism-storage by a massive flood. The variation of x with R was thus of the form: x ( R ) = ~0 + (1 - %)[1 - exp(-K1R~2)] (17)

where x0, xl and x 2 are model parameters which need to be selected through model optimisation. The conservation equation for the prism-storage can be written in terms
o f Up = V p / A 0 a s :

d v p / d t = x ( R ) v w - q - i~(Vw)~ET = g


where 8(Vw) is the Dirac delta function defined by: 8(Vw )( = 1 =0 for for Vw= 0 VwV~0

and is introduced to ensure that once the wedge-storage, which overlays the prism-storage, is depleted then evapotranspiration removes water from the prism-storage. In the next section we shall see that the modified flood plain discharge q = Q o ( H , T ) / A o (m d - 1 ) is a function of the prism and wedgestorage water depths or heads H and T which in turn are related to vp and Vw. From Fig. 4:

Vw= r [ 1 + (T + 2H)/(bo tan , ) ]

and % = H[1 + H / ( b o tan q~)]



The coupled equations equations (16) and (18) can be re-written in the Euler-trapezoidal form for day i:

Vw,= Vw,-, + At ( f + f , _ l ) / 2
%, = %,-1 + At (gi + g i - a ) / 2 (21) and these together with equations (19) and (20) can be easily solved using N e w t o n - R a p h s o n iteration at each time step to obtain vw and Vw~ and h e n c e / / s and Ti, given the initial conditions Vd0 = Vw0= 0 at t = 0.
Outlet channel discharge

In order to obtain a relationship between gauge height and discharge at the floodplain outlet channel, the theory of discharge from a simple channel

174 is explored and a more general form for the discharge is obtained suitable for the flood plain's outlet. The discharge at the outlet, Qo, can be written in terms of the vertical water cross-section a of the outlet channel as: Qo = aU (22)

where U is the discharge velocity which can be expressed as a function of the total energy slope Sf via the semi-empirical Chezy Equation (e.g. see Henderson, 1966, chapter 4):

u = c ( p s , ) 1/2


where p = a l p ; p = 2 H +/30 is the wetted perimeter of the channel of average width /30. An idealised sketch of the Magela floodplain outlet channel is shown in Fig. 5. The channel is wide and essentially rectangular with a = H/30 and /3o = 4.8 km. Since the depth of water H at the outlet channel is of the order of 1 m then p = H and the Chezy factor C is given by Manning's equation: C = pl/6/n = H 1 / 6 / n (24)

where n (m -1/3 s) is Manning's channel roughness factor (Henderson, 1966). The total energy slope, Sf, consists of the floodplain bed slope S o and the wedge-storage slope Sw. For the Magela Creek flood plain S o -- 2 10-5 (see Fig. 3) while the wedge-storage slope is parameterised by S w -- a T / L o, so that Sf = S o + a T / L o, and using equations (22) to (24) the discharge Qo (m 3 d -1) can then be written as: Qo = c0 H/30 ( H 1 / 6 / n )( H S f ) I / 2 = coHS/3(/3oS~/2/n )(1 + a T / S o L o ) 1/z which has the more general form: Qo = CoqoHm( 1 + q l T ) 1/2 where c o = 8.64 104 s d -1, m = 5 / 3 , qo =/3o8~/2/rt and qa = a l S o L. Manning's n (m ~/3 s) in terms of q0 is given by: n =/3oS~/2/qo ,! (27) (26)

and so for the Magela Creek flood plain /30=4-8 k m = 4 8 0 0 m and S~/2 = 4.47 10 -3 hence n = 21.5/q o (m -1/3 s). For a channel with thickly vegetated banks and bed, Henderson (1966) gives n - - 0 . 3 for slow flow through very tall grasses ( > 30 cm) and hence we expect qo to have a value of about 70:



vE 3.6
_o '

.. . . . . . . . ; . . . . . . . . ~ >
':~ :::~' :: :. " ~ ~ : i ! . \ i~

3.4 7.<


i i i


iiii,i i! i,


6 (kin)



8 E



iiiil!iiil!i!i i i!!iiiiiiiiiiiiiiii!i!iii!i i i ! ili!!!iil i

0 I




5 (kin)


Outlet levee
East Alligator river


-= o

--./ !
: :
i t i i

:-:"~i~".. :.i i?
:: 1i

'.."". -. : .... -











Fig. 5. Longitudinal section and cross-section of the outlet channel of the flood plain and typical water levels at the channel and East Alligator River.

The daily water depth Hg at the outlet channel has been recorded by the Water Division of the Northern Territory Department of Mines and Energy since 1974 using a gauging board whose zero is at 2.34 m A H D (Australia Height Datum). The outlet channel's bed is a levee at 3.64 m A H D (see Fig.


2.8 2.6

E 2.4 ~= 2.2 ~c 2

o o 8

1.6 1.4 1.2

i I I I I I I



600 Discharge

800 (rn3s -1)




Fig. 6. Measured gauge height (m) versus discharge (m3 s -a) at outlet channel together with regression curve. The measurements shown correspond to data obtained over a number of years.

5) so that the prism-storage head H = Hg - 1.3 m. Below Hg = 1.3 m any outflow from the flood plain is confined to a few narrow channels ( = 10 m 2) leading to the East Alligator River and the flow is subject to tidal influence. The water levels shown in Fig. 5 correspond to the measured gauge height at gauging stations GS821019 and GS821041 located at the levee and b e y o n d the levee near the river, respectively (see Fig. 1). The daily average levels shown correspond to an above average Wet season, with day 1 corresponding to 1 N o v e m b e r 1984, and it can be seen that the river does not impede significantly the flood plain's outflow above the levee. Estimates of the outflow, Qo ( = 50% error) versus gauge height have been made by the N.T. Water Division over the 10 years 1975-85, usually a few per wet season, and these are shown in Fig. 6. A two variable regression curve (Q = 2.29 H 6 = 2.29 ( H + 1.3) 6) is also shown; however, because of backwater storage on the flood plain, the discharge-gauge height rating curve will be in reality a complex loop-rating curve determined b y the total energy slope Sf. While there is backwater on the plain, the discharge will be higher for a given value of Hg than when the plain is gently draining out at the outlet. Further, any seasonal variations in the channel roughness due to variations in the density of vegetation will also modify the discharge-gauge height relationship and produce the type of scatter shown in the measurements on Fig. 6.


Input data The input data for the model are the average daily rainfall P recorded by the four pluviographs (R821009A, -017, -019B and -019A) whose location is shown in Fig. 1. The measured daily discharge Qm at the flood plain inlet, representing the runoff from the Magela Creek catchment upstream of GS821009, is also required and so is the runoff Q1 from the surrounding lowlands. The daily discharge Qm is measured by the Water Division at GS821009 while the discharge Q1 must be estimated. One method is to assume that the average daily runoff coefficient (runoff/rainfall) of the lowlands is the same as that of the Magela Creek catchment and so: Ql = ( A , / A m ) Q m (28)

where A a and A m are the surface areas of the lowland and Magela Creek catchments. However, the runoff coefficient will be determined by the temporal and spatial variations in the rainfall, by evapotranspiration and by the physical features of the catchment. For low cumulative rainfall (over a time period) most of the rain will go into infiltration and evaporation while for very high rainfall the evaporation process is the dominant loss mechanism which determines runoff (e.g. Pilgrim, 1983). The following simple expression for the cumulative runoff is thus used to obtain an estimate of the daily runoff from the lowlands: Q,c = AIPc e x p [ - E c / P c ] (29)

where the parameter E c is a measure of evaporation and Pc is the cumulative rainfall. As Pc increases the runoff coefficient tends to unity while as E c increases the runoff coefficient increases more slowly. For the Magela Creek catchment Ec = 1.5 m gave good agreement between the measured and estimated monthly runoff and reasonable agreement in the daily runoff. Thus using the rainfall on the lowlands the runoff Q~ on day i was calculated from: Qli = Qlci- Qlci-1 (30)

with Ec a model parameter to be selected by optimisation of an objective function. The other input data to the model are the daily gauge height Hg = H + 1.3 m recorded at the flood plain outlet channel by GS82019 and monthly averaged daily pan evaporation Ep. The true free-water surface evaporation, E, was evaluated from E = p c E p with the monthly lake-to-pan coefficients

178 for the region taken from Vardavas (1987). The evapotranspiration rate E T over the flood plain was evaluated from: ET =( 1--rf \l_rw)E (31)

(see Vardavas, 1988), where r w = 0.07 is the free-water surface albedo to solar radiation and re is the flood plain albedo to solar radiation. Henderson-Sellers and Wilson (1983) give r = 0.12 for rice lands or areas where paddies dominate. In the present model re is selected as a model parameter between 0.07 and 0.16 but it is expected that re for the flood plain will have a value close to 0.12. As will be shown later re = 0.10 for the flood plain.

Objective function
The model involves ten parameters which determine the volume of surface water on the flood plain. These parameters are selected so that the differences between the measured gauge height H 0 ( = Hg - 1.3 m), measured above the tidal level (ATL) or levee level which forms the outlet channels bed at Hg = 1.3 m, and the model prism-storage head H are minimised. The objective function, ~, which is the quantity that needs to be minimised, was taken to be the average relative error between the two profiles Ho(t ) and H(t) over the wet season:

where A = I H 0 - H I / H o

over the N days of measurements, i.e. when

Ho(t ) > O. Parameter selection

The model parameters are as follows: Dm Lm
/0, K1 a n d I2

q0, m and ql Ec rf

surface storage capacity interception storage capacity constants which determine x (equation 17) constants which determine Q0 (equation 26) constant which determines Q1 (equation 29) floodplain albedo to solar radiation (equation 31)

The parameters were selected b y using an iterative Fibonacci search technique (Vardavas, 1989) which minimised the objective function c (equation 32), for the parameter ranges shown in Table 1.

179 TABLE 1 M o d e l p a r a m e t e r values a n d sensitivity Parameter Constraining range 10--200 0-2 1-2 0-50 0-1 1-3 0-0.2 1-10 0.07-0.16 0.005-0.030 Mean value 68.8 1.65 1.77 33.4 0.522 1.47 0.147 4.60 0.094 0.021 Mean sensitivity g 4.57 3.21 2.14 0.88 0.77 0.69 0.49 0.22 0.12 0.05 SE N/

qo ( m4/3 s - l )
m E c (m) ql ( m - l ) D m (m) x2 Ko x1 rf L m (m)

12.41 0.13 0.093 5.91 0.058 0.24 0.011 1.35 0.003 0.003

0.83 0.25 0.11 0.16 0.09 0.11 0.04 0.07 0.01 0.01

Each parameter was constrained within a range of values based, where possible, on field measurements. The parameter q0 = 2 1 . 5 / n is inversely proportional to Manning's channel roughness factor n which takes a value of about 0 . 3 ( m - 1 / 3 s) for channels lined by tall grasses. For the Magela Creek floodplain cutlet which is lined by tall grasses, in particular b y Oryza, q0 should be about 70 and so the constraining range was taken to be 10-200 (m 4/3 s-1 if r n = 5 / 3 in equation 26). The parameter m was expected to be near to the theoretical value of 5 / 3 and so the range was taken to be 0 - 2 . The range for parameter Eo was taken to be 1.0-2.0 m based on the 1.5-m value for the Magela Creek catchment. Parameter ql for an idealised triangular shaped wedge-storage from inlet to outlet would be about 2 b u t from Fig. 3 we expect the slope to be steeper near the plain's inlet so the range was chosen to be 0 - 5 0 m -1. The surface storage capacity parameter D m w a s taken to be between 0 - 1 m based on the longitudinal cross-section given in Fig. 3. The parameter x2, which is a power index in equation (17), was taken to be between 0 - 3 while x0 was taken to lie between 0-0.2 based on the 0.1 value for the Magela Creek catchment (Vardavas, 1988). The parameter re, which represents the albedo of the flood plain to solar radiation, varies from about 0.07 for a free-water surface to 0.16 for tropical woodlands (Henderson-Sellers and Wilson, 1983) so the range was taken to be 0.07-0.16. Finally, the interception storage capacity L m w a s taken to be within the range 0.005-0.030 m based on the 0.007-m value for the Magela Creek catchment (Vardavas, 1988). The model parameters were first optimised b y matching the model and measured gauge height (m ATL) for the 1978-79 wet season and the optimal

180 parameter vector was denoted by Pl- Optimal parameter vectors were then obtained for the wet seasons 1979-80, 1980-81 and 1981-82, denoted by P2, P3 and P4, respectively. The values of c corresponding to p~ through to P4 where 0.07, 0.10, 0.06 and 0.07, respectively, thus establishing ' m o d e l verification', i.e. the model is capable of fitting the measurements of a given season with reasonable accuracy (e.g. Stra~kraba and Gnauck, 1985). The mean parameter vector ~ for the four years and the mean sensitivity vector where then evaluated from:
4 4





Here each element sij of the vector s i was obtained from: s,j= (AqJqj)/(ApiJp~j) (34)

where Aeij is the error in the objective function q j evaluated with ApiJp,j = 0.10. The elements of fi and ~, are given in Table 1 in decreasing order of mean sensitivity g together with the standard error (SE) in each parameter. The largest sensitivity corresponds to the parameter q0, which determines discharge from the flood plain, and which varies inversely with Manning's channel roughness factor n. This is not unexpected since n will vary from season to season depending on the density of the vegetation which lines the banks and bed of the flood plain's Outlet channel. A high sensitivity of the objective function to the parameter m, which also determines the discharge, is also shown in Table 1. However, an error analysis reveals that the error in the value of q0 due to averaging over the four Wet seasons results in a large error in the objective function. If ej is the standard error in the m e a n value of fij, corresponding to parameter j, then an estimate of the resultant mean fractional error Acj in the objective function, e, can be obtained from:

AEj= ejsJpj.


F r o m Table 1 we see that Aq = 0.83 which supports the hypothesis of a strong seasonal variation in Manning's n or q0- This will be discussed further in the next section.

Model validation
In Table 2 are given the values of the objective function e which corresponds to the relative error between model and measured gauge height (m ATL). In column one are the years which correspond to the beginning (November) of each wet season, hence 1974 corresponds to the wet season November 1974 to July 1975.

181 TABLE 2 Model objective function values


Year 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 Mean

0.10 0.06 0.05 0.10 0.07 0.10 0.06 0.07 0.09 0.08 0.09 0.08 0.08

(qo) 0.16 0.14 0.19 0.18 0.19 0.18 0.09 0.28 0.37 0.11 0.32 0.17 0.20

(qo) 0.13 0.12 0.18 0.15 0.18 0.11 0.07 0.10 0.15 0.10 0.10 0.16 0.13

0.18 0.15 0.21 0.12 0.07 0.18 0.18 0.15 0.10 0.21 0.16

In column two are given the values of e with all parameters optimised for that particular year's gauge height, with the 12-year mean value given in the last row. In column 3 the error e is evaluated using the 4-year mean parameter values of Table 1, in particular the value of q0 is fixed to q0 = 68.8. As can be seen the mean value of e(q0) is about 0.20 with individual year values reaching -- 0.40. Based on the expectation that n, and hence q0, varies from one year to the next and that its magnitude is determined by the growth of vegetation in the outlet channel, a correlation was sought between some meteorological quantity and vegetation density or standing crop. Since seed germination will determine the density of grasses growing in the bed and banks of the outlet channel and because this must occur early in the Wet season before the flood plain is inundated then we expect that the early wet season rainfall Pe (i.e. rainfall from the beginning of the Wet season up to the end of December say) to show some correlation with Manning's n or q0- The expectation is that Manning's channel roughness (or flow impedance) will increase with Pe, i.e. q0 should be inversely proportional to PeThe early wet season rainfall, Pe (cm), recorded at the outlet channel by pluviograph R821019A, is shown in Fig. 7 for each year. Also shown is the variation in q0 obtained by keeping all other model parameters fixed to the 4-year mean values, given in Table 1, and optimising the value of q0 for each year. Clearly there is an inverse relationship between q0 and Pe" It is interesting to note that the two recent E1 Nino years, indicated by an asterisk, correspond to unusually dry years with higher than usual values of q0

70 ......... pe(cm) 60 50 40 30 20 10 0

* El Nino Years

- -


74 75 76


78 79 80 81 82 83 Year

84 85 86 87

120 100



40 20 0 0
i i

1 O0

200 300 400 500 600 700 Early Wet S e a s o n Rainfall ( r a m )


Fig. 7. Model parameter q0 (inversely proportional to Manning's channel roughness factor n) correlation with early wet season rainfall Pe-

preceded by years of unusually low values of qo- In Fig. 7 is also shown a regression line drawn through the optimised qo values, for each year, corresponding to different early wet season rainfall values. There is a statistically significant (correlation coefficient 0.94) decreasing variation in q0 versus P with the regression line given by:

qo(Pe) =

128.1 -- 0.171P e


In column 4 of Table 2 are given the error values e(q0) with q0 optimised for each year keeping all other parameters fixed to the 4-year mean values given in Table 1. It should be noted that the c(q0) values obtained with all other parameters fixed to the 12-year mean values were found not to differ significantly from those using the 4-year averages.


The error values c(q0(Pe) ) obtained using the above linear relationship, equation (36), are given in column 5 of Table 2. There were no Pe measurements for the years 1974 and 1975. The quantity q0 can now be removed from the model parameters as it can be reasonably estimated from the early wet season rainfall. The vegetation density (or standing crop) at the flood plain's outlet channel will be

2 1.6 1.2

- Model Measured



.4 0 2 1.6 1.2

, ~)~




I , i i . , i , i , ~,


2 1.6 1.2




-~ ~

.4 13 2 1.6 1.2


' 19,8a~



















Fig. 8. Gauge height above the outlet levee (above tidal level, ATL) as a function of time m days from the 1 November. The model values were generated using the average parameter set for the 4 years 1977-1981.


proportional to Manning's n which from equation (36) can be estimated from: n = 21.5/(128.1 0.171Pe) (37)


E .x


c _c







180 210 Day






Fig. 9. Observed flood plain inundation between 12 March and 14 March 1973 and model inundated flood plain area as a function of days after the 1 N o v e m b e r 1973.

185 Manning's n varies from about 0.2 during unusually dry wet seasons to about 0.5, in agreement with the values given in Henderson (1966) for channels covered by tall grasses (height > 30 cm). The comparison between model and measured daily gauge height (m ATL) measured as water depth above the outlet channel's bed is given in Fig. 8 for twelve wet seasons from 1974-75 to 1985-86 with day 1 corresponding to 1 November. The model is thus validated by its ability to adequately predict the measured daily gauge height for the four years 1978, 1979, 1980 and 1981, using the one set of nine model parameters and by its ability to predict the daily gauge height for the following four years 1982-85 and the preceding two years 1976-77. There were no Pe records for the two years 1974 and 1975, although using the mean value of q0, given in Table 1, averaged over the four years 1978-81, the model was able to adequately predict the daily discharge for these two years. The model was further validated by its ability to predict the variation in the inundated area, A, before and after a storm event in March 1973. Between 12 March and 14 March 1973 the inundated area on the flood plain increased by about 60%, as shown in Fig. 9. The flood inundation maps were produced largely from helicopter sightings by the Water Division of the Northern Territory (Christian and Aldrick, 1977). Below the two maps is the model's calculated inundated area, A, for the wet season 1972-73. It can be clearly seen that in mid-March (about day 130) there is a rapid rise in A from about 130 km 2 to 200 km 2. This corresponds to a rapid filling up of the flood plain's surface store D. Once the surface store is filled to capacity, the inundated area increases by little as the water rises up relatively steeper banks (see Fig. 4). Note that A decreases smoothly due to evaporation loss during the dry season. FLOOD PLAIN HYDROLOGICAL CHARACTERISTICS Some of the typical hydrological characteristics of the Magela Creek flood plain are given in Fig. 10a-d for the four wet seasons 1974-75, 1977-78, 1980-81 and 1983-84. It can be seen that the direct rainfall onto the flood plain ( ~ 200 km 2) is small compared with the runoff from its catchment areas (-- 1380 km2). The rainfall period begins around November while the runoff onto the flood plain begins about January and on average both last until May. The model generates the daily prism-storage water depth, H, the backwater or wedge-storage depth, T, and D, the depth associated with the surface storage, shown for the above four wet seasons in Fig. 10. The prism-storage depth, H, which represents the depth of water above the levee


varies from an average of about 0.5 m during a drier than usual Wet season to about 1.5 m for an unusually wet season. The prism and surface storages are the dominant storages on the flood plain, with the surface storage



Rainfall input

R u n o f f input

.300 E E 200 100 0 2 ~.E 1.6


- ...... Model H Measured H

...... r
- H



'6 .4 ' 6'0 120 180 240 300 360 Day 0 60 120 180 24-0 300 360 Day


Total s t o r a g e


plus wedge


2 vE e 1.5




~n 250 200

& 15o

100 F, 50 0 60 120 180 24-0 ..300 360 Day 1.2 1.5



2 t (m)



Fig. 10a. Flood plain hydrological characteristics for the wet season beginning 1 N o v e m b e r 1974.


capacity determining the time water begins to discharge at the outlet channel. These three storages determine the total volume of water and the discharge from the flood plain. Once the rainfall and runoff onto the flood
1977--78 Rainfall input




E E 100

- .... Model H H

A F:

...... - - ' .... T H D




~ ~

.6 .4
, i
. . . .


120 180 240 300 360



180 240 300 360

Total storage

wedge storage



"~ 1.5


o 2oo _~
. 150



60 120 180 Day 24-0 300 ,.360
1.2 1.4


1.8 (m)

Fig. 10b. F l o o d p l a i n h y d r o l o g i c a l characteristics for the wet s e a s o n b e g i n n i n g 1 N o v e m b e r 1977.


plain cease the backwater storage becomes negligible and the volume of water begins to decrease smoothly as water gently discharges out of the channel. This can be seen in Fig. 10 where the total volume per unit base








o 2 Z1.6

... J J,J,
- Model H H Measured

, Id

- -



0 60



120 180 24E1 300" 360




180 240 300



3 2.5


P r i s m plus wedge



.5 0 625 500
. . .


250 125 0 60 120 180 2 ; 0 " 3 0 0 ' 3 ; 0 1.2 1.6 2 2.4" 2.8 3.2


Gauge height (m)

Fig. 10c. Flood plain hydrological characteristics for the wet season beginning 1 November 1980.










E 200

1 O0

_,..L, L ~ , &

,, ,

ii i i ii i i i - .... Model H H ......





-: 1.2




180 240 Day





180 240 DQy










~ E

w 1.5
2 1

( r

300 fi




~ 100
0 0 60

120 180 240 300 360 Day

1.6 2 2.4 Gauge height ( m )


Fig. 10d. Flood plain hydrological characteristics for the wet season beginning 1 N o v e m b e r 1983.

area (v v = VT/A o) is given as storage (m). Eventually the discharge stops and the flood plain water is lost by evapotranspiration, as can be seen b y the change in the gradient of the storage versus time at about t = 240 days,

190 which corresponds to about the end of June. Shallow' water can persist on the flood plain until the time of the next monsoons thus supplementing the small, but ecologically vital, permanent stores of water in deep pools or billabongs. The model also generates the discharge (m 3 s -a) from the flood plain and this is also shown in Fig. 10 as a function of time and as a function of gauge height Hg (m, with zero at 2.34 m AHD). The true discharge-gauge height rating curve is a complex loop curve determined by the influx of runoff onto the flood plain. The higher the influx of water the larger the discharge out of the flood plain for a given gauge height. The theoretical loop rating curves are in good agreement with the field measurements of the Water Division. VOLUME ESTIMATION The volume of water on the flood plain on any given day during the wet season is determined by three hydrological phases. During the early part of the Wet season there is the filling up of the surface store until the water level reaches the height of the levee. Once the water rises above the levee (3.64 m AHD or Hg = 1.3 m) there is discharge into the East Alligator River (see Fig. 5). While there is rainfall and runoff onto the flood plain the volume of surface water is determined by surface-storage, prism-storage and backwater or wedge-storage. During this time the volume versus time curve is highly irregular depending on the temporal pattern of water influx onto the plain. When runoff ceases at the end of the Wet season, the volume is determined by the prism-storage and surface storage as can be seen in Fig. 10. The water volume then decreases smoothly as the flood plain empties into the East Alligator River which is sufficiently below the levee at the flood plain outlet channel so as not to interfere with the flow out of the flood plain; this was discussed earlier with reference to Fig. 5. The volume generated by the model is shown in Fig. 11 for the twelve wet seasons beginning 1 November 1974 to July 1985. The peak volume can vary between about 100 1 0 6 m 3 during unusually dry years to 500 1 0 6 m 3 during unusually wet years with an average peak volume of about 300 106 m 3" While there is discharge out of the flood plain the total volume VT -AoD m + V, where V is the volume in excess of the surface store and since = 1 then: VT= Ao( H + T + Dm) (38)

where from Table 1 the surface storage maximum depth D m = 0.522 __ 0.058. On average the model and measured prism-storage depth, H, differ by about 16% (see Table 2) and the backwater or wedge-storage depth, T, is always

600 500 400 300 200 1O0 0 600 500 1977-78 1978--79 1979-80 1974.--75 1975-76 1976--77

300 200 100 0 ~-" E 600 1980-81 1961-62



300 20O 100 0 6O0 500 4.0O 30O 2 O O 100 0





120 180 240 300 360


120 180 240 300 360


120 160 24.0 300 360


Fig. 11. Model daily variation in the flood plain's surface water volume beginning with 1 November in each wet season.

much smaller than H (< 11/3). An estimate of the error in the model's volume can then be obtained from:

'v/-/= + 32(1-7/-/g2 - 4.3Hg + 3.0) 1/2


where Cv/_ is in units of 1 0 6 m 3 and Hg is the gauge height measured by / GS821019 at the flood plain outlet. For example, during the 1 9 8 0 - 8 1 wet season the peak volume occurred at about t = 120 days after i N o v e m b e r

192 1980. F r o m Fig. 10 we have H ~ 1.7 m and so H e = 3.0 m and Vv = 520 _+ 70 1 0 6 m 3. A quick estimate of the volume of water, V, in excess of the surface store can be obtained from the following simple expression: V = 2 5 7 H g - 334 (40)

The above linear relationship for V is based on the 12-year average line of best fit through the storage-gauge height loop rating curves as shown in Fig. 10. The mean error was calculated b y taking the line of best fit of all points which lie above the line defined b y equation (40) and the line of best fit of all points below and averaging the two to obtain the mean dispersion of points above and below equation (40). The mean error in V is given by: EvL = 23He -- 18 The total volume Vv can be estimated from: Vv = 257Hg - 282 with an error: Err = _+32(2.2H 2 - 5.1Hg + 3.3) 1/2 (43) (42) (41)

where CVT ----(C2L + e2n)l/2. For example, for Hg = 3 m then Vv = 490 _+ 90 106 m 3, in agreement with the total volume given b y the model for t = 120 days from 1 N o v e m b e r 1980. Once water ceases to flow out of the flood plain the volume of water remaining on the flood plain in surface store depth, D, decreases b y evapotranspiration according to:

= --Ev(t )


An estimate of the rate at which D decreases and hence the time, after flow ceases at the outlet, it takes for the flood plain to dry up, can be obtained by using the long-term average variation in ET over the months July to October. E v can be approximately expressed as a linear function of time: E T ( t ) = a + bt with a = 4.5 1 0 - 3 ( m ) and b = 2.0 10 -5 (m d -a) with t = 0 now corresponding to the day flow ceases, usually around early July. Then surface storage depth D decreases according to:
D / D m = 1 - 8.6 10 -3 t -

1.9 10 -5 t 2


In Fig. 11 a change of slope in the volume versus t curves usually occurs at about 240 days from 1 N o v e m b e r which now corresponds to t = 0 in equation (45). Thus the flood plain dries up about 100 days after flow ceases at the outlet leaving only isolated swamps areas and billabongs.


When water input to the flood plain ceases and backwater or wedge-storage depth, T, is zero, i.e. at the end of the wet season, the discharge Qo can be estimated from:
Qo ~ qo H5/3


with a relative error A Q o / Q o ~ 5 / 3 A H / H and since the model can predict the value of H with an average relative error of about 0.16 then A Q o / Q o 0.27. A rough estimate for T during the Wet season can be obtained from equations (38) and (40): T = 0.29Hg - 0.38 hence T = 0.29H and using equation (26) with ql = 33.4 we have:
Qo = qoHS/3(1 + 10H) '/a



with an average relative error:

A Q o / Q o = 5/18 + 5 / / / ( 6 + 6 0 H )


For the peak flow during the 1980-81 wet season H ~ 1.7 m (Hg ~ 3.0 m) and using q0 -- 70 (Fig. 8) then Q0 = 720 _+ 250 m 3 s-1 which compares well with the model value of = 700 m 3 s -1 given in Fig. 10.

The present water budget model for the tropical Magela flood plain provides an estimate of the daily surface water volume on the flood plain and the rate of discharge from the flood plain outlet in terms of the daily rainfall and daily discharge into the flood plain from Magela Creek. The flood plain model together with the simple rainfall-runoff model for the Magela Creek system (Vardavas, 1988) can be used to estimate the initial dilution available in the creek for any contaminated waters that m a y be released into the creek from the Ranger uranium mine site, and the subsequent dilution and evapoconcentration on the flood plain. The model indicates that there is a correlation between early wet season rainfall and water flow impedance provided by vegetation at the outlet. This suggests that there is a correlation between early wet season rainfall and seed germination on the flood plain.

It is a pleasure to thank m y colleagues Drs. John East and Max Finlayson for helpful discussions regarding the Magela Creek flood plain. I would like to thank Drs. Arthur Johnston and Glen Riley for their comments on the


manuscript, Lisa Cannon for computing assistance and Dr. Liz Gyurasits for assisting in the preparation of the manuscript.
REFERENCES Christian, C.S. and Aldrick, J.M., 1977. Alligator Rivers Study: A review report of the Alligator Rivers Region environmental fact-finding study. Australian Government Publishing Service, Canberra, A.C.T., 174 pp. East, J., 1987. Late Quarternary evolution of the Magela backwater plain. In: Annual Research Summary 1985-86, Alligator Rivers Region Research Institute. Australian Government Publishing Service, Canberra, A.C.T., 142 pp. Finlayson, C.M., 1986. Recycling of nutrients of heavy metals by flood plain vegetation. In: Annual Research Summary 1984-85, Alligator Rivers Region Research Institute. Australian Government Publishing Service, Canberra, A.C.T., 155 pp. Hart, B.T., Ottaway, E.M. and Noller, B.N., 1986. Nutrient and trace metal fluxes in the Magela Creek System, Northern Australia. Ecol. Modelling, 31: 249-265. Henderson, F.M., 1966. Open Channel Flow. Series in Engineering, 4. Macmillan, New York, 522 pp. Henderson-Sellers, A. and Wilson, M.F., 1983. Surface albedo data for climate modelling. Rev. Geophys. Space Phys., 21: 1743-1778. Pilgrim, D.H., 1983. Some problems in transferring hydrological relationships between small and large drainage basins and between regions. J. Hydrol., 65: 49-72. Stragkraba, M. and Gnauck, A.H., 1985. Freshwater Ecosystems: Modelling and Simulation. Developments in Environmental Modelling, 8. Elsevier, Amsterdam, 309 pp. Uranium Province Hydrology, 1980. Data Summary 1978-79. Volume 11, Water Division of the Northern Territory Department of Transport and Works, Darwin, N.T., 300 pp. Uranium Province Hydrology, 1982. Data Summary 1979-81. Volume 12, Water Division of the Northern Territory Department of Transport and Works, Darwin, N.T., 252 pp. Vardavas, I.M., 1987. Modelling the seasonal variation of net all-wave radiation flux and evaporation in a tropical wet-dry region. Ecol. Modelling, 39: 247-268. Vardavas, I.M., 1988. A simple water balance daily rainfall-runoff model with application to the tropical Magela Creek catchment. Ecol. Modelling, 42: 245-264. Vardavas, I.M., 1989. A Fibonacci search technique for model parameter selection. Ecol. Modelling, 47 (in press).