australian journal · 1. a review of recharge studies in australia cuan adpetheram abcf, abglen...

P u b l i s h i n g

Australian Journal of Soil ResearchCSIRO PublishingPO Box 1139 (150 Oxford St)Collingwood, Vic. 3066, Australia

Telephone: +61 3 9662 7628Fax: +61 3 9662 7611Email: [email protected]

Published by CSIRO Publishing for CSIRO and the Australian Academy of Science

w w w . p u b l i s h . c s i r o . a u / j o u r n a l s / a j s r

All enquiries and manuscripts should be directed to:

Volume 40, 2002© CSIRO 2002

Australian Journalof Soil Research

An international journal for the publication oforiginal research into all aspects of soil science

Aust. J. Soil Res., 2002, 40, 397–417

© CSIRO 2002 0004-9573/02/03039710.1071/SR00057

Towards a framework for predicting impacts of land-use on recharge: 1. A review of recharge studies in Australia

Cuan PetheramABCF, Glen WalkerAD, Rodger GraysonAB, Tomas ThierfelderE,and Lu ZhangAC

ACooperative Research Centre for Catchment Hydrology.BDepartment of Civil and Environmental Engineering, University of Melbourne,

Victoria 3010, Australia.CCSIRO Land and Water, Canberra, ACT 2601, Australia.DCSIRO Land and Water, Adelaide, SA 5064, Australia.EDepartment of Biometry and Informatics, Swedish University of Agricultural

Sciences, PO Box 7013, 750 07 Uppsala, Sweden. FCorresponding author; email: [email protected]

Abstract

This work investigated the potential for developing generic relationships from measurements of rechargemade in previous studies that would allow the assessment of the impact of land-use change on recharge.Forty-one studies that measured recharge from across Australia were reviewed to generate a database.Studies were characterised on the basis of broad soil type (sand or non-sand), land-use/vegetation (annual,perennial, or trees), and annual rainfall. Attempts to develop quantitative recharge relationships met withlimited success because of the limited geographical coverage of the studies, lack of details on the studysites, and high variability in the data. Nevertheless, the following relationships for annual vegetation werestatistically valid:

Ln(recharge) = –19.03 + 3.63 ln(rainfall) [for sandy soils]; F(1, 96) = 149.03; R2 = 0.60

Ln(recharge) = –12.65 + 2.41 ln(rainfall) [for non-sandy soils]; F(1,151) = 46.87; R2 = 0.23

The low degree of explanation of rainfall for the annual × non-sand data suggests that it is likely thatsoil structure becomes more important for higher clay content soils. Recharge under trees was negligiblecompared with that under annuals. These relationships should not be used in areas such as those where:preferential pathway flow is the dominant recharge mechanism, rainfall is summer dominant, lateralhydraulic gradients are high, water holding capacities are very low, or there are fresh, high-yieldingaquifers.

Collectively, the results show that: (1) rainfall explains a significant proportion of the observed rechargevariation; (2) there is a significant difference between mean recharge under trees and annual vegetation; (3)there is a significant difference between mean recharge under annual vegetation on sand soils and non-sandsoils; (4) the land-use groups had a greater influence on recharge than the broad soil groups used in thisstudy; (5) there is a lack of annual recharge measurements under perennial pastures/crops, under trees inhigh rainfall zones (i.e. >600 mm/year) and in areas of summer dominant rainfall; (6) across a broad rangeof locations, recharge is higher under shallow-rooted annual vegetation than deep-rooted vegetation; and(7) the estimator of Zhang et al. (1999) for ‘excess water’ may provide a useful indication of the upper limitto the long-term average recharge measurements. Large variation in the data resulted from disparity in therecharge techniques used, the coarse soil categories used, failure to account for land management factors,and complications due to macropores and shallow water tables. It is proposed that the value of theinformation presented here may be enhanced in future studies by incorporating information from qualitativestudies, particularly paired-site studies, and by drawing information from unsaturated zone andgroundwater modelling studies, particularly comparisons of different land-uses at similar locations.Furthermore, the results of this study can be used to identify gaps in knowledge and, hence, target areas forfuture research such as annual recharge measurements beneath perennial vegetation.

Additional keywords: dryland salinity, groundwater management, deep drainage.

398 C. Petheram et al.

SR00057I mpact of landuse on r echar geC. Petheramet al .

Introduction

Secondary salinisation in Australia is generally accepted to have been caused by the aquiferdischarge capacity of groundwater systems being exceeded, due to an increase in recharge(Walker et al. 1999). Quantitative estimation of recharge is important in assessingalternative land management options, as well as for providing input into groundwatermodels that assess impacts on groundwater systems. While a number of studies haveestimated recharge for specific areas or groundwater systems, there has never been anintegrated review of Australian recharge studies in a way that would allow generalpredictions of recharge for sites elsewhere. This paper summarises the results of manymeasured recharge studies from across Australia and examines them collectively in aneffort to determine the extent to which the derivation of quantitative relationships can bejustified, based purely on the data. This review is a first step towards the ultimate goal ofdeveloping simple empirical relationships that can be applied in a generic way across arange of landscapes.

There are different ways of estimating recharge; it can be modelled using water balancetechniques (Gee and Hillel 1988), inferred from groundwater flow modelling (Hatton1998), and measured using a wide range of physical and chemical methods. This reviewfocuses on only those studies in which recharge was actually measured. The data wereobtained from journal articles, technical reports, conference papers, and unpublishedsources. If they can be established, generic relationships should enable rapid preliminaryestimates of recharge or deep drainage without the need for detailed fieldwork.

The utility of such a concept depends on the existence of simple relationships betweendeep drainage and factors such as soil type, rainfall, and land-use, as well as on our abilityto map these factors, and on the collated data being representative of these relationships.Previous work has demonstrated considerable variability in measured deep drainage, evenwithin a paddock (Cook et al. 1989). Hence, the effects of many factors such as landmanagement, plant disease, and micro-topography are likely to be very complex.Nevertheless, the intention here was to establish what predictions could reasonably bemade, using the measured data on recharge alone.

For this study, we considered the 3 primary factors controlling recharge in semi-aridAustralia to be land-use, soil type, and climate (see e.g. Kennett-Smith et al. 1994; Smettem1998). In a study done in central Kansas (USA) by Sophocleous (1992), depth to water tableduring the spring months was found to be one of the most influential variables affectingrecharge. However, at many sites in Australia where recharge measurements have beentaken, the water table has been very deep (e.g. Allison et al. 1985; Johnston 1987a; Cooket al. 1989; O’Connell et al. 1997). Hence, depth to water table is unlikely to be a key factoraffecting recharge over large parts of southern Australia.

Data were reviewed and collated from previous studies across Australia, with noadditional measurements taken for this study. This approach necessitated an assessment ofthe techniques and associated estimates as part of the analysis for this study. While efforthas been directed into developing relationships for the larger continental scale, there isemphasis here on the application of the findings to salinity management.

This is the first in a series of studies that investigate methods for providing first-cutestimates of recharge under different land-uses. This study focuses on analysis of existingdata while future papers will assess the potential of water balance and groundwatermodelling (i.e. Walker et al. 2002). Here we bring together the results of many measuredrecharge studies and aim to determine across a broad range of locations, whether:

Impact of landuse on recharge 399

1. collectively the results confirm commonly held tenets in salinity management, suchas the notion that removal of native vegetation leads to an increase in recharge;

2. there is potential to develop generic relationships for recharge;3. we understand the causes of the variability in measured recharge estimates.

In the next section, we outline the methods by which data were categorised (including thereasons for the exclusion of some data) and present the data graphically. We then describethe hypotheses and statistical procedures used in the quantitative analysis. The Resultssection provides summary tables of the statistical analysis while the Discussion puts theresults into perspective and addresses reasons for the variability in the data. We concludeby examining the extent to which the aims were achieved.

Methods and data presentation

Only those studies that measured annual recharge or inferred annual recharge through measurements werereviewed. Several studies reporting ‘measured’ annual recharge values were not analysed because: (1) theywere in rainfall zones very different to most other studies (e.g. 4239 mm/year in the case of Bonell et al.1983); (2) site runoff characteristics had been altered (e.g. Ridley et al. 1997); or (3) rainfall and rechargewere measured/presented as a seasonal rather than an annual total (e.g. Nulsen and Baxter 1982; Lane 1996;Smith et al. 1998).

A database was developed to summarise information from the resulting 41 papers in which informationwas available on the amount of recharge, method used, land-use, soil type, and annual rainfall.

The terms ‘potential recharge’ and ‘deep drainage’ refer to water fluxes (measured or inferred) at somedepth within the unsaturated zone, and are not necessarily equivalent to the ‘actual’ recharge (i.e. theamount of water entering the aquifer at that time). Most techniques ‘infer’ recharge from measurementsmade of other parameters, which can be related to recharge (e.g. water table rise or chloride distribution inthe profile). Thus, in this paper, as in many previous studies, the term ‘measured’ covers both directlymeasured and inferred recharge values.

Site description

The 41 recharge studies came broadly from 18 sites across Australia. As illustrated in Fig. 1, distributionof studies is heavily biased towards southern Australia and other low- to medium-rainfall zones. This islargely related to the relatively high importance of salinity as a land degradation issue in those areas.

In the studies reviewed, recharge has been estimated at sites across a wide range of environments.Rainfall ranged from 100 mm/year to 1150 mm/year, while soil types varied from very coarse sands toheavy clays (i.e. clay contents >65%). Land-use also varied widely, from deep-rooted vegetation such asMallee scrub, banksia, jarrah, and pine plantations, to herbaceous perennial vegetation such as buffel grasspasture, to annual vegetation like wheat, oats, clover, and ryegrass (with many combinations of rotationsand lengths of fallow).

Data handling and analysis

Prior to analysis it was necessary to discard misleading and inappropriate data from the 41 selected studies.Data were discarded for the following reasons: (1) data had already been reported in another study (e.g.Kennett-Smith et al. 1994); (2) measurements were deemed unrepresentative by the authors of the citedpaper (e.g. Pakrou and Dillon 2000, observed that their repacked lysimeters produced considerably higherestimates of drainage than their monolith lysimeters); (3) recharge was estimated under trees <5 years old(e.g. Barker et al. 1995)*—3 measurements were discarded for this reason; (4) preferential pathway flowwas the dominant recharge mechanism (by mention of the authors of the cited paper)†—5 measurementswere discarded for this reason; (5) one of the 3 key factors controlling recharge (as discussed in theIntroduction) was absent, i.e. measurements taken on skeletal soils (Allison and Hughes 1978) or where no

* Vertessy et al. (1996) showed that it took about 5 years of growth before the transpiration rate from astand of re-growth mountain ash reached the pre-clearing rate.

† The scale and frequency of distribution of preferential flow paths is such that they are often missed bymeasurements. It is difficult to contemplate any relationship that could be applied to such an areabecause of the difficulty of mapping such features.


vegetation had been established over the time period of measurement (Thorburn et al. 1991) werediscarded—8 measurements were discarded for this reason; (6) cited study failed to specify soil type (e.g.O’Connell et al. 1997)—10 measurements were discarded for this reason.

The number of recharge measurements, rainfall, land-use, soil group, and recharge estimation techniqueused for each study are summarised in Table 1.

Because of the limited number of recharge qualifying studies, it was necessary to keep the number ofkey land-use categories small, to ensure a reasonable number of data values for each category. Also, lackof clear documentation made it difficult to further subdivide the categories.

Land-use was divided into 3 broad categories: annuals (shallow-rooted annual crops or pasture),perennials (perennial crops, pastures and native herbaceous vegetation), and trees (very deep-rootedvegetation).

Data from the 41 measured studies (i.e. Table 1) are presented in Fig. 2, in which all rainfall and rechargevalues are annual totals. Where site-specific rainfall data were not provided for a long-term averagerecharge measurement, we allocated a rainfall value using information from the Bureau of Meteorology orfrom other studies conducted in the same area. Where necessary, long-term annual rainfall was interpolatedfrom nearby sites of known rainfall.

Figure 2 suggests that at Gnangara Mound in Western Australia there are different factors limitingtranspiration/recharge to those applied generally in this study. This region is covered by very deep, coarsesands, and recharge estimates were considerably higher (i.e. sometimes an order of magnitude) than any ofthose found in the literature from other parts of Australia. That this region has very fresh water and theaquifer is high yielding is indirect evidence that this system is atypical of many Australian groundwatersystems, particularly those that are salinised. The reasons for the very high recharge rates are not clear. Itis apparent that this region and other regions with similar soils will need to be examined separately. Thesedata highlight that relationships will not be applicable across all areas, and that it may be possible to identifysuch outlier areas.

Excess water curve

Excess water is the term given to the non-transpired component of rainfall (i.e. P – Et) and so includesrecharge and runoff. To place the recharge values into context with other components of the water balance,recharge values were compared with estimates of ‘excess water’ by Zhang et al. (1999), who developed arational function approach for estimating actual evapotranspiration based on mean annual rainfall andvegetation cover. By studying over 250 catchments from many parts of the world, Zhang et al. (1999)

Fig. 1. Median annual rainfall (mm/year) and seasonality map of Australia. Numbers indicate recharge study site location and correspond with the third column in Table 1. Information derived from The Ausmap Atlas of Australia, AUSLIG.


showed that, for a given vegetation type, there is a good relationship between long-term averageevapotranspiration and rainfall. The relationship between the annual non-transpired (or excess) water, andannual rainfall is shown in Fig. 3 and this will be referred to as the rational function approach to excesswater — which represents the upper limit to recharge (i.e. if runoff equals 0). It should be noted that these‘average’ curves take no account of soil type. Hence, recharge estimates at sites with sandy-soils (i.e. withhigh infiltration and low water holding capacity) and few surface drainage features may lie above the excesswater curve because the vegetation may not be able to transpire as much water as on clay soils.

In their present form, the curves in Fig. 3 provide a hypothetical upper limit of water that can rechargeunderlying aquifers. The excess water curves should only be used as an approximate upper limit of rechargefor long-term average recharge estimates. It is inappropriate to compare single year recharge values with acurve generated from mean long-term data.

Figure 4 is a subset of Fig. 2 where data from the Gnangara Mound (3 of the 41 studies) have beenexcluded. In Fig. 4, a distinction has been made between single year and long-term average annual rechargeand rainfall values.

Recharge estimation techniques

The collated data represent estimates of recharge at different temporal and spatial scales as well as atdifferent depths in the profile (Table 2). To enable useful comparisons between different studies, therecharge data were divided into 3 groups, based on the measurement technique used, particularly withregard to depth of measurement.

Group 1 techniques are those that infer recharge at depths many metres below the root-zone.Measurements are generally made at a point scale over many metres of depth and recharge is averaged overdecades. Group 2 techniques are those that infer recharge from changes in the water table. Rechargeestimates are at the paddock to catchment scale and the time scale of measurement may vary from an eventbasis to an annual basis. Group 3 techniques are those that estimate potential recharge immediately belowthe root-zone. These techniques are generally at the point scale and measurements are at short time intervals(i.e. days or even hours). For Group 1 and 2 measurements the annual totals are long-term/historicalaverages, while the yearly totals for Group 3 techniques are for a single year.

The influence of different estimation technique groupings on recharge was not investigated for the entiredata set because few recharge estimation techniques are suitable for estimating recharge under trees. Figure5 is a subset of Fig. 4 and shows recharge estimates made using different techniques, but only in respect ofthe annual vegetation subclass.

Soil groups

It is proposed that much of the variation in the data in Fig. 4 was caused by different soil types and thataccounting for this factor may lead to more robust relationships. However, difficulties with categorising soiltypes meant that soils were ultimately divided into only 2 very broad textural groups: sand and non-sand(Fig. 6). Where the soil description was ambiguous or unclear (e.g. Allison and Hughes 1978; Harringtonet al. 1999), the authors drew upon personal experience and that of others in the area or areas with similarsoils, to help categorise the data. Where only a quantitative description was provided (i.e. percentage claycontent), non-sand sites were comprised of soils with a clay content >10%. In this case allocation was partlydependent upon vegetation type (i.e. rooting depth). For a soil planted with annual vegetation, the maximumclay content in the top 2 m of the soil profile was considered in categorising the soil; for perennials themaximum clay content of the top 3 m was considered, and for trees the maximum clay content of the top 5m was considered.

The matrix in Table 3 illustrates the number of soil–land-use combinations encountered in thequalifying studies from the literature.

Quantitative analysis of the data and development of generic relationships

Because data on land-use and soil type could only be defined on an ordinal scale, the simplest means ofdeveloping generic recharge relationships is to relate recharge to rainfall for each land-use/soil typecombination. The small number of measurements under perennial vegetation precluded this land-usegrouping from being analysed statistically.

Before developing recharge–rainfall relationships for different land-use/soil type combinations, it needsto be shown that recharge differs significantly across the land-use and soil groupings. In developing genericrelationships the following hypotheses were tested:


Table 1. Summary of site description and data used in analysisWA, Western Australia; SA, South Australia; Vic., Victoria; NSW, New South Wales; Qld., Queensland; NT, Northern Territory. Site numbers correspond with those in Fig. 1; the numbers illustrate the broad location in which the study was conducted. Rainfall is presented as the long-term annual average. Soils

were divided into 2 broad soil groupings, i.e. sand and non-sand. No. of estimates is the number of recharge estimates that were plotted from each study in Fig. 2; reasons for data being discarded are indicated under Data handling and analysis; parentheses indicate that the recharge estimates were

reported as an average value; the numbers within parenthesis correspond with number of measurements taken to produce the average value; where N/A appears, the authors did not report this information. The

numerals in the ‘Recharge Technique’ column correspond with recharge techniques listed in Table 2

Author Location Siteno.

Rainfall Soil group Land-use No. of estimates

Recharge technique

Allison (1987) New Kalimurina, Central Australia

06 100 Sand Tree 1 1

Allison and Hughes (1972) Gambier Plain (SA) 10 700 Non-sand Tree 1 (11) 4Allison and Hughes (1978) Gambier Plain (SA) 10 710–760 Sand,

non-sandAnnual 21 1,4

Allison and Hughes (1983) Walpeup (Vic) 11 335 Sand Annual, tree

2 (6) 1,2

Allison et al. (1985) Mallee (SA) 08 300 Sand,non-sand

Annual,tree

4 1,3,4

Allison et al. (1990) Maggea (SA)

Kulkami (SA)

08 270

370

Sand

Sand

Annual,tree

Annual,tree

11

23

1,2

1,2

P.J. Baker, pers. comm. (2000) Mitchell (Qld) 17 567 Non-sand Annual,tree

2 1,2

Barker et al. (1995) Albury (NSW) 15 300B Non-sand Tree 1 6Brinkley et al. (1997) Goulburn catchment (Vic) 14 621

629660976

1038

Non-sand Annual,tree

1 (33)1 (20)1 (23)1 (4)1 (14)

6

Carbon et al. (1982) Gnangara Mound (WA)A 01 669B

668B

803B

Sand Annual,perennial,

tree

11 11

Colville and Holmes (1972) Mt Gambier (SA) 10 483B

805B

617B

Sand Tree 3 7

Cook et al. (1989) Borrika (SA) 08 340 Sand Annual,tree

12 1,3

Cook et al. (1990) Woolpunda Groundwater Mound (SA)

08 270 Sand Annual 1 (7) 1

T. Ellis, pers. comm. (1999) Dergholm (Vic) 12 701 Sand Tree 1 1R. George, pers. comm. (1999) East Belka (WA)

Eastern Wheatbelt (WA)02 330

330Non-sandNon-sand

AnnualAnnual,

tree

2 (N/A)2 (N/A)

61,6

George and Frantom (1990) Merredin Catchment (WA) 03 328 Sand,non-sand

Tree 2 1

Gordon, I. et al. pers. comm. (1999) Liverpool Plains (NSW) 16 550750

Non-sand Annual 7 10

Harrington et al. (1999) Ti-Tree (NT) 05 300 Sand Tree 2 (63) 1,8Holmes and Colville (1970a) Mt Gambier (SA) 10 600 Sand Tree 6 11Holmes and Colville (1970b) Mt Gambier (SA) 10 566B

438B

566B

463B

805B

617B

Sand Annual 12 9

Johnston (1987c) Don, Ernie, Lemon (WA) Salmons, Wights (WA)

02 8001150

Non-sandNon-sand

TreeTree

1 (20)1 (12)

11

Johnston (1987a) Salmons (WA) 02 1150 Non-sand Tree 7 1Kennett-Smith et al. (1990) Buronga (NSW)

Balranald (NSW)Euston (NSW)

12 310322312

Non-sandNon-sand

Sand,non-sand

AnnualAnnualAnnual

246

10

222

(continued next page)


Table 1. (Continued)

Author Location Site no.

Rainfall Soil group

Land-use No. of estimates

Recharge technique

Kennett-Smith et al. (1992a) Mallee (NSW) 12 255 Sand,non-sand

Annual,tree

1721

1,2,3

Kennett-Smith et al. (1992b) Glendook (Vic)

Eureka (Vic)

11 430

430

Non-sand,sandSand,

non-sand

Annual,tree

Annual,tree

14

10

1,2,3

1,2,3

Kennett-Smith et al. (1993) Goroke (Vic) 13 526 Non-sand Annual,tree

26 1,2,3

Kennett-Smith et al. (1994) Balranald (NSW)Borrika (SA)Maggea (SA)Cooke Plain (SA)Walpeup (Vic)Buronga (NSW)Euston

12080808111212

322340270380340295312

Non-sandSand

Non-sandSand

Non-sandSand

Non-sand

AnnualAnnualAnnualAnnualAnnualAnnualAnnual

42852331

2222222

Leaney and Allison (1986) Mallee near Morgan (SA) 08 300 Sand Tree 2 (33) 8,1Leaney et al. (1995) South East (SA) 09 490

530

Non-sand

Sand

Annual,tree

Annual,tree

2

2

1,2

1

Leaney and Herczeg (1999) SW Murray Basin 09 390–500 Sand,non-sand

Annual 9 1

Loh and Stokes (1981) Wights (WA)Lemon (WA)Bingham (WA)Batalling (WA)Bakers Hill (WA)Lake Toolabin (WA)Salmon Gums (WA)

02 1150750725650490410390

Non-sandNon-sandNon-sandNon-sandNon-sandNon-sandNon-sand

AnnualAnnualAnnualAnnualAnnualAnnualAnnual

1 (N/A) 6666666

O’Connell et al. (1997) Walpeup (Vic)

Timberro South (Vic)

Wimmera (Vic.)

11

13

13

338

338

420

Sand

Sand

Non-sand

Annual,tree

Annual,tree

Annual,tree

4

10

9

2,3

2,3

2,3

Pakrou and Dillon (2000) Mt Gambier (SA) 10 714B

985B

642B

625B

Non-sand Annual 4 9

Salama et al. (1993) Cuballing (WA) 02 462 Non-sand Annual,tree

4 1,2,6

Sharma et al. (1991) Gnangara Mound (WA)A 01 770–800756B

798B

1227B

Sand Annual,tree

26 1,11,12

Thorburn et al. (1991) NE Australia (Qld) 18 630 Non-sand Annual,perennial,

tree

12 10

Thorpe (1987) Gnangara Mound (WA)A 01 830 Sand Tree 8 1,4Turner et al. (1994) Kalgoorlie (WA) 04 250 N/A Tree 1 8Walker et al. (1987) Tatiara (SA)

Binnum (SA)Joanna (SA)

09 489524582

Non-sandNon-sand

Sand,non-sand

AnnualAnnualAnnual

113

555

Walker et al. (1992a) Upper SE (SA) 09 530 Sand Annual,tree

4 1,2,3

Walker et al. (1992b) Cooke Plains (SA) 07 380 Sand Annual, perennial,

tree

4 1

AData from the Gnangara Mound (WA) are not included in Fig. 4.BA single year value.


Hypothesis 1 — recharge differs across the categories of land-use and soil type. Hypothesis 2 — rainfall explains the greatest degree of variation in the data.Hypothesis 3 — within each land-use category (i.e. annual vegetation and trees), recharge differs across

the categories of soil type.Hypothesis 4 — within the land-use category of annual vegetation, recharge differs due to estimation

technique.

In addition to testing each of the hypotheses, log-transformed rainfall is linearly regressed againstlog-transformed recharge within the categories of nominal data. As regression slopes principally expressthe response of recharge to rainfall, the possibility of different regression slopes across the nominalcategories is assessed.

Statistical analyses

The statistical analysis is first concerned with determining whether recharge could be considered differentacross the categories of nominal data. Since rainfall generally influences recharge, effects of the nominaldata cannot be assessed without compensating for different rainfall across nominal categories. This isusually done with standard analysis of covariance (ANCOVA), which assumes equal regression slopesacross categories (parallelism). However, since the possibility of different slopes is assessed in the linearregression analysis, standard ANCOVA does not apply. Instead, as discussed below, a generalised ANCOVAis used that does not assume parallelism.

General analysis of covariance

The ANCOVA designs used in this study are essentially 2- and 4-factor ANOVAs (analysis of variance),with one factor (the covariate) continuously distributed and all effects considered as random (e.g. Sokal andRohlf 1995). The method assumes a linear relationship between the dependent variable and the covariate,and compensates for different covariate interactions across the categories of the nominal predictor. In otherwords, the analysis attempts to simulate the results that would have been obtained had a constant covariate(rainfall) been used across the categories of the nominal predictor (e.g. land-use). Hence, the effect of thecovariate is eliminated, allowing an independent analysis across the categories of the nominal data.

Assuming a linear relationship between the dependent variable and the covariate implies a model with2 parameters, the intercept and the slope. In order not to over-parameterise the system of equations usedfor statistical inference in standard ANCOVA, the method requires equal slopes across the categories of the

0

100

200

300

400

500

600

0 200 400 600 800 1000 1200

Rainfall (mm/year)

Rec

harg

e (

mm

/yea

r)

AnnualsPerennialsTreesAnnuals (GM)Perennials (GM)Trees (GM)1:1 line

Fig. 2. Annual recharge versus annual rainfall for different vegetation types. The hollow data symbolsindicate measurements made at Gnangara Mound (GM), Western Australia. The 1:1 line indicates thepoint at which all rainfall becomes recharge.


nominal predictor. Since the analysis performed here wishes to assess the possibility of different slopes(different recharge responses across the categories of nominal data), the standard methodology isinappropriate. The appropriate design for modelling the influence of the predictors in this situation is theGLM (general linear model) separate slope design (e.g. Finn 1974; Morrison 1990). It copes with

0

500

1000

1500

2000

0 500 1000 1500 2000 2500 3000

Rainfall (mm)

Exc

ess

wat

er (

mm

)

Forest

Grassland

0

50

100

150

200

250

300

0 200 400 600 800 1000 1200

Rainfall (mm/year)

Rec

harg

e (

mm

/yea

r)

Annuals - Long-term

Perennials - Long-term

Trees - Long-term

Annuals - Single year

Trees - Single year

Excess water curve

Grassland

Trees

Fig. 3. Relationship between excess water (i.e. recharge and runoff)and rainfall for different vegetation types. (Source: Zhang et al. 1999.)

Fig. 4. Annual recharge versus annual rainfall for different vegetation types and time scales ofmeasurement. Each symbol represents either an individual measurement or an area-averaged estimate ofrecharge. Solid symbols represent long-term average annual data and hollow symbols represent singleyear data. Symbols enclosed by a circle were measurements made under conditions of preferentialpathway flow (by mention in cited paper) and were not included in the statistical analysis. Data from theGnangara Mound have not been included in this figure.


Tab

le 2

.R

ech

arge

tec

hniq

ues

and

th

eir

dep

th a

nd s

cale

s of

mea

sure

men

t.

Num

bers

in p

aren

thes

es n

ext t

o th

e re

char

ge te

chni

que

indi

cate

the

num

ber

of s

tudi

es u

sing

this

tech

niqu

e to

est

imat

e re

char

ge

Rec

harg

e te

chni

que

Sca

le o

f m

easu

rem

ent

Dep

th o

f m

easu

rem

ent

Wha

t doe

s it

mea

sure

?F

urth

er r

efer

ence

Spa

tial

Tem

pora

l

Gro

up 1

Env

iron

men

tal t

race

rs1.

CM

BA

A (

25)

2. C

DF

MB (

12)

Poin

tH

isto

rica

l ave

rage

Bel

ow r

oot-

zone

Infe

rs p

oten

tial

rec

harg

eW

alke

r (1

998)

3. S

oil w

ater

mas

s ba

lanc

e (7

)Po

int

His

tori

cal a

vera

geB

elow

roo

t-zo

neIn

fers

pot

enti

al r

echa

rge

Ken

nett

-Sm

ith

et a

l. (1

992a

)H

isto

rica

l tra

cers

4. T

riti

um (

4)5.

Cl (

1)Po

int

His

tori

cal a

vera

geB

elow

roo

t-zo

neIn

fers

pot

enti

al r

echa

rge

Wal

ker

(199

8)

Gro

up 2

6. P

iezo

met

eric

ris

e (5

)C

atch

men

tE

vent

/his

tori

cal a

vera

geW

ater

tabl

eIn

fers

rec

harg

eA

rmst

rong

and

Nar

ayan

(1

998)

7. R

atio

of

hydr

ogra

phPa

ddoc

kN

/AW

ater

tabl

eIn

fers

rec

harg

eC

olvi

lle

and

Hol

mes

(19

72)

am

plit

udes

(1)

8. 14

C (

3)C

atch

men

tH

isto

rica

l ave

rage

Wat

er ta

ble

Infe

rs r

echa

rge

Coo

k an

d H

ercz

eg (

1998

)

Gro

up 3

9. L

ysim

eter

s (2

)Po

int

Sea

sona

l/an

nual

R

oot-

zone

Mea

sure

s po

tent

ial r

echa

rge

Bon

d (1

998)

10. S

OD

ICS

/ M

odel

of

Ros

e et

al.

1979

(2)

Poin

tA

nnua

lR

oot-

zone

Infe

rs p

oten

tial

rec

harg

eW

alke

r (1

998)

11. W

ater

bal

ance

(3)

Poin

tS

easo

nal/

annu

alR

oot-

zone

Infe

rs p

oten

tial

rec

harg

eZ

egel

in e

t al.

(199

2)12

. Bro

mid

e (1

)Po

int

Sea

sona

l/an

nual

Roo

t-zo

neIn

fers

pot

enti

al r

echa

rge

Wal

ker

(199

8)

AC

hlor

ide

mas

s ba

lanc

e ap

proa

ch.

BC

hlor

ide

disp

lace

men

t fro

nt m

etho

d.


over-parameterisation via asymptotic estimation (Finn 1974), and resembles standard nested ANOVAdesigns since it omits the main effect of the covariate.

With ANCOVA relying on the central limit theorem in cases of unknown probability density functionsof the continuously distributed variables, normal distribution cannot be guaranteed in categories containingsmall sample sizes. In such cases (i.e. under hypothesis 4, in Group ‘2’ of ‘sample technique’), theANCOVA results should be treated with caution. In the linear regression analysis, all continuouslydistributed variables were transformed to their natural logarithms to approximate normal distributions.

0

50

100

150

200

250

300

0 200 400 600 800 1000 1200

Rainfall (mm/year)

Rec

harg

e (

mm

/yea

r)

Group 1

Group 2

Group 3

Excess water curve

Grassland

0

50

100

150

200

250

300

0 200 400 600 800 1000 1200

Rainfall (mm/year)

Rec

harg

e (

mm

/yea

r)

Annuals and sand

Trees and sand

Annuals and non-sand

Trees and non-sand

1:1 line

Fig. 5. Annual recharge versus annual rainfall for annual vegetation using different rechargeestimation techniques. This figure is a subset of Fig. 4.

Fig. 6. Annual recharge versus annual rainfall for sand and non-sand soils. This figure is a subset ofFig. 4. The 1:1 line indicates the point at which all rainfall becomes recharge. The excess water curves arenot illustrated because both long-term and single year data are presented together.


Since all statistics involved are invariant with respect to linear (in a broad sense) transformation (Box andCox 1964), regression results based on such transformations are just as valid as if performed on the primaryvariables themselves.

Additional methodological ANCOVA requirements

With the methodological requirement of parallelism overcome, the application of ANCOVA requires:(1) homogeneity of variance (i.e. the variance of the generic population is constant across categories); and(2) that the generic population is homoscedastic (i.e. the variance should not depend on the mean).Homogeneity of variance was tested using the Hartley F-max statistic, the Cochran C statistic and theBartlett chi-square test (Winer 1962; Sokal and Rohlf 1995), whereas homoscedasticity was visuallyexamined in plotted regression residuals.

Linear regression analysis

Within the categories of nominal data, linear regression analysis was used to explore the existence oflinear relationships between recharge and rainfall, with significant relationships used for two mainpurposes: (1) the slopes of the relationships were used to infer the responses of recharge to rainfall withinthe different categories of nominal data; and (2) in categories with strong enough relationship, the modelequation may be used for predictive purposes.

The hypotheses of whether regression slopes differ across two categories of nominal data is tested witha standard 2-sample t-test in accordance with Sokal and Rohlf (1995).

Under (2) above, to use the regression equations for predictive purposes, a general model precisionrecommended for safe prediction within earth sciences is 70% degree of explanation (R2) (Håkanson andPeters 1995), although recommendations vary widely with application and discipline.

Results

Unless otherwise stated, the methodological requirements mentioned above were allsatisfied, and all the linear regression coefficients were highly significant.

Hypothesis 1: recharge differs across the categories of land-use and soil type

These hypotheses were tested univariate, with effects of rainfall accounted for. Rechargewas shown to be significantly different across the land-use categories (i.e. annualvegetation and trees, Table 4) and across the soil type categories (i.e. sand and non-sand,Table 5).

Hypothesis 2: rainfall explains the greatest degree of variation in the data

In order to simultaneously rank all effects, a 4-factor ANCOVA was required (Table 6). Itwas found that rainfall had the greatest influence upon recharge, i.e. explained the greatestproportion of observed recharge variance as expressed by the F-value, followed bygroupings of land-use, soils, and estimation technique respectively.

Table 3. Number of recharge estimates for different soil–land-use combinations shown in Fig. 4

Numbers in brackets indicate number of studies that estimated recharge for each particular soil–land-use combination

Vegetation Sand Non-sand

Annual 98 (17) 158 (19)Perennial 01 (1) 03 (1)Tree 51 (17) 071 (17)


Hypothesis 3 — within the land-use categories (i.e. annual vegetation and trees), recharge differs across the categories of soil-type

Again, this hypothesis was tested univariate with effects of rainfall accounted for. For bothannual vegetation (Table 7) and trees (Table 8), recharge was shown to be significantlydifferent across the sand and non-sand categories.

Hypothesis 4 – within the land-use category of annual vegetation, recharge differs due to estimation technique

This hypothesis was also tested univariate, with effects of rainfall accounted for. It wasshown that an overall significant difference existed across the three categories of estimationtechnique (Table 9), a significance that, according to the principles of post-hoc analysis(Sokal and Rohlf 1995), allows further inference regarding the distribution ofsignificances. The post-hoc analysis reveals (Table 10) that the significance is distributedbetween the technique Groups ‘1’ and ‘3’ and between Groups ‘2’ and ‘3’, but not betweenGroups ‘1’ and ‘2’.

Table 5. ANCOVA of soil type categories

General effect Recharge F Recharge P

Intercept F(1,376) = 99.04 0.00Land-use × rainfall F(2,375) = 136.63 0.00Land-use F(1,376) = 66.94 0.00

Table 6. ANCOVA results for key factors and key factor combinations

Factor F P

Rainfall F(376) = 66.80 <0.01Soil {1} 2.51 0.32Land-use {2} 9.24 0.21Technique {3} 0.72 0.591 × 2 7.06 0.011 × 3 1.84 0.162 × 3 -- --1 × 2 × 3 -- --

Table 4. ANCOVA of land-use categories


Intercept F(1,376) = 65.64 <0.01Land-use × rainfall F(2,375) = 164.86 <0.01Land-use F(1,376) = 56.48 <0.01

Table 7. ANCOVA of soil type categories given land-use is Annual


Intercept F(1,254) = 132.73 <0.01Soil × rainfall F(2,253) = 182.90 <0.01Soil F(1,254) = 23.09 <0.01


Despite homogeneity of variance being satisfied across all categories, there were too fewobservations within the technique Group 2 to guarantee methodological compliance(although deviations from the results presented probably are minor).

ANCOVA analyses were not performed for the tree data because of insufficientsample-sizes in categories ‘2’ and ‘3’ of the nested technique factor. The results of theregression analysis for all factor combinations analysed are provided in Table 11.

Discussion

The natural logarithm of recharge regressed against the natural logarithm of rainfall for allthe collated recharge data (i.e. Fig. 4) explained 21% of the variation in the data (Table 11).It was found that there was a significant difference between the tree and annual vegetationcategories, and between the sand and non-sand categories.

Rainfall was found to be the factor that explained the greatest amount of variance,followed by the land-use, soil type, and technique groups used. Accounting for differencesin broad land-use improved the degree of explanation to 43% and 36%, respectively, fortrees and annual vegetation, while accounting only for differences in broad soil-typeexplained 33% and 16% of the observed variation for sand and non-sand soils, respectively.

Combinations of different land-use and soil groupings were analysed in an attempt tofurther improve the degree of explanation of rainfall provided by these factors alone.Regressing the natural logarithm of recharge on the natural logarithm of rainfall improvedthe degree of explanation of rainfall to 60%, 45%, and 51% for the annual × sand, tree ×sand, and tree × non-sand combinations, respectively. However, the degree of explanationdecreased to 23% for annual × non-sand combination. That the annual × non-sandcombination comprised 42% of all data explains why the overall amount of varianceexpressed by land-use × soil type was slightly less than that expressed by land-use alone(Table 6). The low degree of explanation of rainfall for the annual × non-sand data suggeststhat it is likely that soil structure becomes more important for higher clay content soils, butthis information was not available from these studies.

Table 8. ANCOVA of soil type categories given land-use is Trees


Intercept F(1,120) = 24.39 <0.01Soil × rainfall F(2,119) = 21.06 <0.01Soil F(1,120) = 18.52 <0.01

Table 9. ANCOVA of technique groups provided land-use is Annual


Intercept F(1,254) = 35.01 <0.01Technique × rainfall F(3,252) = 83.61 <0.01Technique F(2,253) = 5.02 <0.01

Table 10. Post-hoc analysis of technique group provided land-use is Annual

Technique 1 Technique 2 Technique 3

Technique 1 — 0.78 <0.01Technique 2 0.78 — 0.02Technique 3 <0.01 0.02 —


For all the data, and just that data within the annual vegetation category, it was found thatthere was a significant difference between the technique Group ‘1’ and ‘3’ data and Group‘2’ and ‘3’ data, but not between the Group ‘1’ and ‘2’ data. The reasons for theseobservations are not clear, but it is thought that the scales and depth of measurement aremore similar for between the technique Groups ‘1’ and ‘2’ than between Groups ‘1’ and ‘3’or Groups ‘2’ and ‘3’.

Not surprisingly, the relationship between the natural logarithm of recharge and thenatural logarithm of rainfall was stronger for all 3 recharge estimation technique groupingsfor annual vegetation than for all vegetation (Table 11). For annual vegetation, regressingthe natural logarithm of recharge on the natural logarithm of rainfall improved the degreeof explanation to 30%, 63%, and 16% for Groups 1, 2, and 3, respectively. That therelationship for the Group 2 data was so much stronger than the other groups isencouraging. It is worth noting that a number of studies have shown a strong relationshipbetween rise in groundwater level and rainfall (e.g. Armstrong and Smith 1974;Sophocleous 1992; Wu et al. 1996). This is thought to be because piezometric responsetechniques are integrators of the processes occurring in the unsaturated zone at the largerscale and hence are not subject to the same spatial variability as point scale measurements.However, the limited number of Group 2 and 3 measurements (18 and 25, respectively)means that these results should be treated cautiously.

As can be seen in Table 11, the combination of statistical significance and relatively lowexplanatory power of the regression relationships means that the factors analysedsignificantly explain a relatively small proportion of observed recharge variation, and thatmore specific land-use and soil type categories are required. The poor predictive ability ofthe regression relationships is due in part to: (1) the application of linear models in anon-linear reality*; and (2) the errors associated with estimating recharge. The followingsection describes some possible reasons for error in recharge estimates.

Table 11. Summary of regression analysisColumns titled Intercept and Gradient refer to the coefficients of the equation: Ln(recharge) = gradient ×

ln(rainfall) + Intercept. In all regressions the P level was <0.01

Land-use Soil type Technique group No.

F-value Intercept Gradient Adjusted R2

All All All F(1,361) = 94.75 –15.35 2.75 0.21Annual All All F(1,249) = 141.86 –16.25 3.07 0.36Tree All All F(1,110) = 85.30 –15.57 2.42 0.43All Sand All F(1,141) = 70.35 –26.52 4.67 0.33All Non-sand All F(1,218) = 41.51 –10.66 1.94 0.16All All 1 F(1,320) = 52.94 –12.82 2.30 0.14All All 2 F(1,16) = 35.43 –9.26 1.98 0.67All All 3 F(1,21) = 5.04 –13.81 2.72 0.16Annual Sand All F(1,96) = 149.03 –19.03 3.63 0.60Annual Non-sand All F(1,151) = 46.87 –12.65 2.41 0.23Tree Sand All F(1,43) = 37.0651 –30.42 4.87 0.45Tree Non-sand All F(1,65) = 69.31 –13.08 2.03 0.51Annual All 1 F(1,214) = 165.44 –16.25 3.07 0.30Annual All 2 F(1,13) = 49.84 –8.79 1.90 0.63Annual All 3 F(1,23) = 13.16 –13.81 2.72 0.16

* Linear models can only hope to explain the linear component of the variance observed.


Reasons for variation in estimates

Lerner et al. (1990) suggest that the accuracy of recharge estimates is prone to 4 types oferror: an incorrect conceptual model, neglect of spatial and temporal variability,measurement error, and calculation error. In this study, these errors were not separatedbecause insufficient detail was presented in the various studies. Hence all errors werelumped together, although it is likely that the majority of variability in results was due tothe variation in the data arising from natural spatial and temporal variability.

Apart from the different recharge estimation techniques and the coarseness of soilgrouping already discussed, the broad land-use categories used also contributed to thevariation in the data — a result of a wide range of vegetation types, management practices,and other factors (e.g. soil). For example, the annual vegetation category included bothpasture and cropping, as well as differing management practices such as fallow,fertilisation, and grazing. Such differences are known to cause variation in recharge (e.g.Kennett-Smith 1992a; O’Connell et al. 1995). Even within the tree category there can behigh variation. Sharma et al. (1991) compared the groundwater recharge to an unconfinedaquifer under different land-uses on the Swan Coastal Plain (Fig. 2). Potential recharge wasgreater under the native Banksia vegetation (15% of annual rainfall) than under mature pinetrees (<1% of annual rainfall). It was also noted that potential recharge was higher under a‘sparse’ pine plantation (16% of annual rainfall) than under a dense pine plantation (10%annual rainfall), and higher under young pines (>32% of annual rainfall) than mature pines.The variation caused by the broad land-use groupings used in this paper was unavoidablebecause more detailed land-use information was seldom available for the individual studies.

Variation in recharge estimates can be introduced by ignoring temporal rainfalldistribution. Data reported in large time-steps, such as annual rainfall data, introducescatter by failing to distinguish the differences in temporal variation within years. It is welldocumented that the duration and intensity of rainfall events are at least as important torecharge, as the total annual rainfall (Leaney et al. 1995). This is particularly the case insemi-arid regions where rainfall can be very episodic in nature and much of a year’s raincan fall within a short space of time, causing greater recharge than would occur if the samevolume of rainfall were spread over an entire year. Because of this high degree ofepisodicity, it is often misleading to talk of a mean annual recharge. Mean annual rechargedata are more useful if the mean has been derived over a long period that contains astatistically significant number of extreme events (Barnes et al. 1994). The variabilitycaused by ignoring the temporal distribution of rainfall events is likely to be greater forsingle year estimates of recharge (i.e. Group 3) than for long-term averages (i.e. Group 1).This is because single year values will be affected by the temporal variability in rainfallbetween years as well as the temporal variability between sites.

In the preceding discussion, a number of secondary factors were introduced, associatedwith the 3 primary factors. These were land management, episodicity, and seasonality ofrainfall (Thorburn et al. 1991), and soil structure. Other secondary factors include shallowwater tables (Allison and Hughes 1972; Colville and Holmes 1972; Sophocleous 1992) andpreferred pathways, both of which can vary spatially and temporally.

For some areas, preferential flow pathways can be a major form of recharge (Johnston1987a, 1987b, 1987c). These pathways are caused by macropores like cracks and old rootchannels as well as larger scale sink holes and geological discontinuities, which enablewater to flow more rapidly than through the soil matrix (Allison and Hughes 1983). Severalmeasurements made under conditions of preferential pathway flow can be seen in Fig. 4(those symbols enclosed by circles). In conditions of such preferential flow, the influence


of land-use, rainfall and soil texture is reduced, e.g. Allison et al. (1985), Johnston (1987a).The ability of such channels to transmit recharge depends on the depth of vegetation rooting(Johnston 1987c; Allison et al. 1990) and the likelihood of continuous macropores to thewater table (Walker 1998). It would be expected that recharge in areas of preferential flowrecharge will be much more dependent on the temporal variability of rainfall than in areaswhere matrix flow dominates. It is likely that such preferred pathways will be important inareas of fractured hard rock, gilgai areas, dissolution features in limestone, and in areas ofshallow water tables.

The excess water curves appear to indicate an approximate upper limit to the long-termaverage recharge. Where recharge values lie above the excess water curve (Fig. 3), rechargehas either been estimated on sandy-soils with few drainage features (e.g. Allison andHughes 1978; Kennett-Smith et al. 1992b) or the site has been influenced by secondaryfactors (e.g. Allison et al. 1985). The general trend observed in the data was a movementaway from the rational function approach to excess water curve, as the soil texture becomesfiner, as the vegetation changes from shallow to deeper rooted and as the rainfall increases,because all of these increase the runoff to recharge ratio.

Application of the generic relationships

In all of the land-use × soil type combinations, strong relationships were established withall regression coefficients being highly significant. None of the relationships had R2 greaterthan the 70% recommended for predictive use. The predictive ability of the mainrelationships (bold in Table 11) varied from R2 of 60% for annual vegetation on sandy soilsto 23% for annual vegetation on non-sand soils, with the equivalent relationships for treeshaving R2 of 45% and 50% respectively. The most appropriate situation in which to use theresults in Table 11 for estimating recharge, is where long-term, ‘first cut’ estimates ofrecharge are required over a regional scale, although even then, the high level of uncertaintyin predictions must be accepted. It is also important to consider other limitations andassumptions underlying the analysis presented here. The main assumptions were thatpotential recharge/deep drainage will become recharge and the collated measurements area representative sample. The main limitations are: (1) studies reporting high numbers ofrecharge estimates have a stronger influence than those reporting few; (2) greater errors inestimates of recharge are likely to occur in higher rainfall regions because of thepredominance of studies encountered for low rainfall regions; (3) the generic relationshipsdeveloped from this review can be used only where matrix flow conditions are dominant,or they may underestimate recharge (because estimates of recharge where preferential flowwas reported were not included in the analysis); (4) the lack of data from summer-dominantrainfall regions means the relationships may not be applicable in areas whereevapotranspiration exceeds rainfall over most of the year; (5) the data should not beconsidered representative of areas in winter-dominant rainfall zones of <800 mm/year thathave very fresh water and are high yielding (e.g. Gnangara Mound); and (6) the applicationof these results to a regional scale (i.e. use in a geographical information system) caninvolve difficulties since most of the data were effectively point scale measurements.Application of such relationships may be appropriate in some catchments, but generally thewater balance of a particular section of a landscape is not independent of the upslope data(Hatton 1998). In landscapes where lateral hydraulic gradients are high and subsurface flowis significant, the results in Table 11 may be unreliable. Even in landscapes where theresults are applicable, it is important to recognise the scatter apparent in the data, use it to


gain an indication of uncertainty and assess this against the level of detail required for thetask and the potential risk associated with an erroneous value.

Potential avenues for improvement

The low scatter obtained in the recharge estimates made using piezometric techniques(Fig. 4) suggests an alternative approach may be to explore the prospect of developinggeneric recharge relationships using recharge data from groundwater flow modellingstudies. These data may provide a range within which catchment scale recharge could beexpected to lie. However, from these data it is unlikely that it will be possible to distinguishbetween the impacts on recharge made by different land-uses within the catchment. Henceit may be necessary to further enhance the relationships by drawing on quantitative andqualitative information from (1) this study of measured recharge; (2) existing measuredpair-site studies; and (3) water balance modelling studies, particularly with regard tocomparing different land-uses at similar locations.

Conclusion

This paper provides the first collation of measured recharge studies across all of Australiaand the first in a series of studies investigating the potential for developing generic rechargerelationships. The analysis of collated recharge studies showed that, across a wide range oflandscapes, recharge is significantly greater under shallow-rooted annual vegetation thanunder deep-rooted vegetation. It also showed that recharge under sandy soils is significantlyhigher than under non-sandy soils.

Rainfall was able to explain 21% of the variation in the collated data. The degree ofexplanation was improved by accounting for variations in land-use and soil type. Theland-use × soil type combination with the strongest relationship between the naturallogarithm of recharge and the natural logarithm of rainfall, was the annual vegetation × sandsoil combination. For this combination, rainfall was able to explain 60% of the variation inthe data. However, this value is still too low for safe prediction. For Group 2 data underannual vegetation, rainfall was able to explain 63% of the variation, but this figure shouldbe treated with caution because of the small sample size. Nevertheless, the results suggestthat measurements from piezometric methods may offer greatest promise for thedevelopment of generic relationships. To improve upon these relationships, more soil typeand land-use categories need to be included in the analysis to reduce some of the variation.However, to do this more data is required. In particular, this work has highlighted a lack ofannual recharge data in summer-dominant rainfall zones, under perennial vegetation, andunder trees in high rainfall zones.

This study broadly showed that we could identify factors that can cause variability inrecharge estimates. Because of the differences in behaviour of different landscapes (e.g.Gnangara Mound), rules regarding land-use and climate may need to be developedindependently for particular landscape types (i.e. classes in a recharge classification). Areasthat may require separate analysis include; soils with low water holding capacity (i.e. deepsands and skeletal soils), storage limited systems (e.g. fractured rock environments, karsticlandscapes, and regions with shallow water tables), areas with duplex soils and crackingclays, and areas with fresh, high yielding aquifers.

Acknowledgments

The authors would like to thank Dr Warren Bond and Dr Peter Cook for their comments ona draft of this report. Dr Richard George is acknowledged for providing a spreadsheet of


results for many Western Australian recharge studies and Dr Jeff Wood for statistical adviceon an earlier draft. The senior author was supported by an Australian Postgraduate Awardand scholarship funded by the CRC for Catchment Hydrology, which also supported theresearch.

References

Allison GB (1987) Estimation of groundwater discharge and recharge with special reference to arid areas.In ‘Proceedings of the International Conference on Groundwater Systems Under Stress’. Brisbane,1986. pp. 231–238. (Australian Government Publishing Service: Canberra)

Allison GB, Cook PG, Barnett SR, Walker GR, Jolly ID, Hughes MW (1990) Land clearance and riversalinisation in the western Murray Basin, Australia. Journal of Hydrology 119, 1–20.

Allison GB, Hughes MW (1972) Comparison of recharge to groundwater under pasture and forest usingenvironmental tritium. Journal of Hydrology 17, 81–95.

Allison GB, Hughes MW (1978) The use of environmental chloride and tritium to estimate total rechargeto an unconfined aquifer. Australian Journal of Soil Research 16, 181–195.

Allison GB, Hughes MW (1983) The use of natural tracers as indicators of soil-water movement in atemperate semi-arid region. Journal of Hydrology 60, 157–173.

Allison GB, Stone WJ, Hughes MW (1985) Recharge in karst and dune elements of a semi-arid landscapeas indicated by natural isotopes and chloride. Journal of Hydrology 76, 1–25.

Armstrong D, Narayan K (1998) Using groundwater responses to infer recharge. In ‘Studies in catchmenthydrology. The basics of recharge and discharge’. (Eds L Zhang, G Walker) Part 5. (CSIRO Publishing:Melbourne)

Armstrong D, Smith PE (1974) Recharge and the groundwater regime of the Toowoomba city catchment.Darling Downs Institute of Advanced Education, Unpublished Report, Toowoomba, Queensland.

Barker PJ, Gates G, Moore JC (1995) Agroforestry studies for groundwater recharge control near Albury.In ‘Proceedings of Murray Darling Basin Commission Workshop, 1995’. Wagga Wagga. pp. 19–23.(MDBC Groundwater Working Group)

Barnes C, Jacobson G, Smith G (1994) The distributed recharge mechanism in the Australian arid zone.Soil Science of America Journal 58, 31–40.

Bond W (1998) Soil physical methods for estimating recharge. In ‘Studies in catchment hydrology. Thebasics of recharge and discharge’. (Eds L Zhang, G Walker.) Part 3. (CSIRO Publishing: Melbourne)

Bonell M, Cassells DS, Gilmour DA (1983) Vertical soil water movement in a tropical rainforest catchmentin north-east Queensland. Earth Surface Processes and Landforms 8, 253–272.

Box GEP, Cox DR (1964) An analysis of transformations. Journal of the Royal Statistical Society 26,211–253.

Brinkley A, Linke G, Potter M (1997) Salinity Strategy Assessment — The salt and water balanceperspective. In ‘Proceedings of ‘Murray Darling Basin Commission Workshop, 1997’. Toowoomba. pp.72–76. (MDBC Groundwater Working Group)

Carbon BA, Roberts FJ, Farrington P, Beresford JD (1982) Deep drainage and water use of forests andpastures grown on deep sands in a Mediterranean environment. Journal of Hydrology 55, 53–64.

Colville JS, Holmes JW (1972) Water table fluctuations under forest and pasture in a karstic region ofsouthern Australia. Journal of Hydrology 17, 61–80.

Cook PG, Herczeg AL (1998) Groundwater chemical methods for recharge studies. In ‘Studies incatchment hydrology. The basics of recharge and discharge’. (Eds L Zhang, G Walker) Part 2. (CSIROPublishing: Melbourne)

Cook PG, Walker GR, Jolly ID (1989) Spatial variability of groundwater recharge in a semi-arid region.Journal of Hydrology 111, 195–212.

Cook PG, Walker GR, Jolly ID, Allison GB, Leaney FW (1990) Localised recharge in the vicinity of theWoolpunda Groundwater Mound. The Centre for Groundwater Studies, Report No. 30.

Finn JD (1974) ‘A General Model for Multivariate Analysis.’ (Holt, Rinehart and Winston: New York)Gee GW, Hillel D (1988) Groundwater recharge in arid regions: review and critique of estimation methods.

Hydrological Processes 2, 255–266.George R, Frantom P (1990) Preliminary groundwater and salinity investigations in the eastern wheatbelt

2. Merredin catchment. Division of Resource Management Department of Agriculture WA, TechnicalReport 89.


Harrington GA, Herczeg AL, Cook PG (1999) Groundwater sustainability and water quality in the Ti-TreeBasin, central Australia. CSIRO Technical Report 53/99.

Hatton T (1998) Catchment scale recharge modelling. In ‘Studies in catchment hydrology. The basics ofrecharge and discharge’. (Eds L Zhang, G Walker.) Part 4. (CSIRO Publishing: Melbourne)

Holmes JW, Colville JS (1970a) Forest hydrology in a karstic region of southern Australia. Journal ofHydrology 11, 59–74.

Holmes JW, Colville JS (1970b) Grassland hydrology in a karstic region of southern Australia. Journal ofHydrology 10, 38–58.

Håkanson L, Peters RH (1995) ‘Predictive limnology — Methods for predictive modelling.’ (SPBAcademic Publishing: Amsterdam)

Johnston C (1987a) Mechanisms of water movement and salt mobilization in profiles of south-westWestern Australia. In ‘International Conference on Groundwater Systems Under Stress’. Canberra. pp.389–398. (Australian Government Publishing Service: Canberra)

Johnston C (1987b) Distribution of environmental chloride in relation to subsurface hydrology. Journal ofHydrology 94, 67–88.

Johnston C (1987c) Preferred water flow and localised recharge in a variable regolith. Journal of Hydrology94, 129–142.

Kennett-Smith AK, Budd GR, Cook PG, Walker GR (1990) The effect of lucerne on the recharge to clearedmallee lands. The Centre for Groundwater Studies. Report No. 27.

Kennett-Smith AK, Budd GR, Walker GR (1992a) Groundwater recharge beneath woodlands cleared forgrazing, south western New South Wales. CSIRO, Division of Water Resources, Divisional Report No.92/1 and Centre for Groundwater Studies Report No. 35.

Kennett-Smith AK, Cook PG, Thorne R (1992b) Comparison of recharge under native vegetation anddryland agriculture in the Big Desert region of Victoria. Centre for Groundwater Studies Report No. 46.

Kennett-Smith A, Cook P, Walker G (1994) Factors affecting groundwater recharge following clearing inthe south-western Murray Basin. Journal of Hydrology 154, 85–105.

Kennett-Smith AK, Thorne R, Walker GR (1993) Comparison of recharge under native vegetation anddryland agriculture near Goroke, Victoria. CSIRO, Division of Water Resources, Glen Osmond, S. Aust.

Lane PNJ (1996) The spatial variability of groundwater recharge at the local scale. PhD thesis, TheAustralian National University, Australia.

Leaney FW, Allison GB (1986) Carbon–14 and stable isotope data for an area in the Murray Basin: Its usein estimating recharge. Journal of Hydrology 88, 129–145.

Leaney FWJ, Herczeg AL (1999) The origin of fresh groundwater in the SW Murray Basin and its potentialfor salinisation. CSIRO Land and Water, Technical Report 7/99.

Leaney FW, Herczeg AL, Love AJ (1995) Fresh groundwater in the S-W Murray Basin resulting fromrecharge during wet climatic periods. In ‘Proceedings from Murray Darling Basin CommissionWorkshop’. Wagga Wagga. pp. 165–169. (MDBC Groundwater Working Group)

Lerner DN, Issar AS, Simmers I. (Eds) (1990) ‘Groundwater recharge. A guide to understanding andestimating natural recharge.’ (Heise: Hannover)

Loh IC, Stokes RA (1981) Predicting stream salinity changes in south-western Australia. AgriculturalWater Management 4, 227–254.

Morrison DF (1990) ‘Multivariate statistical methods.’ 3rd edn (McGraw Hill: New York)Nulsen RA, Baxter IN (1982) The potential of agronomic manipulation for controlling salinity in Western

Australia. The Journal of the Australian Institute of Agricultural Science 48, 222–226.O’Connell MG, O’Leary GJ, Incerti M (1995) Potential groundwater recharge from fallowing in north-west

Victoria, Australia. Agricultural Water Management 29, 37–52.O’Connell M, Thorne R, O’Leary G, Mason G, Hoxley G (1997) Water movement beneath cleared and

native vegetation in the Victorian Mallee and Wimmera region. In ‘Proceedings of Murray DarlingBasin Commission Workshop’. Toowoomba. pp. 36–41.

Pakrou N, Dillon PJ (2000) Comparison of type and depth of lysimeter for measuring the leaching lossesof nitrogen under urine patches. Soil Use and Management 16, 108–116.

Ridley AM, White RE, Simpson RJ, Callinan L (1997) Water use and drainage under phalaris, cocksfoot,and annual ryegrass pastures. Australian Journal of Agricultural Research 48, 1011–1023.

Rose CW, Dayanda PWA, Nielson DR, Biggar JW (1979) Long-term solute dynamics and hydrology inirrigated lowly permeable soils. Irrigation Science 1, 77–87.

Salama R, Farrington P, Bartle G, Watson G (1993) Salinity trends in the wheatbelt of Western Australia:results of water and salt balance studies from Cuballing catchment. Journal of Hydrology 145, 41–63.


http://www.publish.csiro.au/journals/ajsr

Sharma ML, Barron RJW, Craig AB (1991) Land use effects on groundwater recharge to an unconfinedaquifer. CSIRO, Division of Water Resources, Divisional Report 91/1.

Smettem KRJ (1998) Deep drainage and nitrate losses under native vegetation and agricultural systems inthe Mediterranean climate region of Australia. Occasional Paper RAPPS02/98.

Smith CJ, Dunin FX, Zegelin SJ, Poss R (1998) Nitrate leaching from a Riverine clay soil under cerealrotation. Australian Journal of Agricultural Research 49, 379–389.

Sokal and Rohlf (1995) ‘Biometry. Its principles and applications.’ (Wiley: New York)Sophocleous M (1992) Groundwater recharge estimation and regionalization: the Great Bend Prairie of

central Kansas and its recharge statistics. Journal of Hydrology 137, 113–140.Thorburn PJ, Cowie BA, Lawrence PA (1991) Effect of land development on groundwater recharge

determined from non-steady chloride profiles. Journal of Hydrology 124, 43–58.Thorpe PM (1987) Tritium as an indicator of groundwater recharge to the Gnangara Groundwater Mound

on the Swan Coastal Plain, Perth, Western Australia. In ‘Groundwater recharge’. (Ed. ML Sharma) pp.41–56. (Balkema Publishers: Brookfield)

Turner J, Townley L, Rosen M, Milligan N (1994) Groundwater recharge to paleochannel aquifers in theeastern goldfields of western Australia. In ‘Water Down Under 94’. Adelaide. pp. 511–516. (TheInstitute of Engineers: Canberra)

Vertessy RA, Hatton TJ, Benyon RG, Dawes WR (1996) Long-term growth and water balance predictionsfor a mountain ash (Eucalyptus regnans) forest catchment subject to clear-felling and regeneration. TreePhysiology 16, 221–232.

Walker GR (1998) Using soil water tracers to estimate recharge. In ‘Studies in catchment hydrology. Thebasics of recharge and discharge.’ (Eds L Zhang, G Walker.) Part 7. (CSIRO Publishing: Melbourne)

Walker GR, Blom RM, Kennett-Smith, AK (1992a) Preliminary results of recharge investigations in theupper south-east region of South Australia. CSIRO, Division of Water Resources — Centre forGroundwater Studies Report No. 47.

Walker GR, Dillon PJ, Pavelic P, Kennett-Smith AK (1992b) Preliminary results of recharge and dischargestudies at Cooke Plains, South Australia. CSIRO, Division of Water Resources — Centre forGroundwater Studies Report No. 48.

Walker G, Gilfedder M, Williams J (1999) Effectiveness of current farming systems in the control ofdryland salinity. CSIRO Land and Water general publication.

Walker GR, Jolly ID, Stadter MH, Leaney FW, Stone WJ, Cook PG, Davie RF, Fifield LK (1987) Estimationof diffuse recharge in the Naracoorte Ranges region South Australia: An evaluation of chlorine-36 forrecharge studies. Department of Primary Industries and Energy, Canberra. Research project final report87/10.

Walker GR, Zhang L, Ellis TW, Hatton TJ, Petheram C (2002) Estimating impact of changed land-use onrecharge. Review of modelling and other approaches as appropriate for dryland salinity management.Hydrogeology Journal 10, 68–90.

Winer BJ (1962) ‘Statistical principles in experimental design.’ (McGraw-Hill: New York)Wu J, Zhang R, Yang J (1996) Analysis of rainfall-recharge relationships. Journal of Hydrology 177,

143–160.Zegelin SJ, White L, Russell GF (1992) A critique of the time domain reflectometry technique for

determining soil-water content. In ‘Advances in measurement of soil physical properties.’ (Eds GCTopp, WD Reynolds, RT Green) Soil Science Society of America Special Publication 30, 187–208.

Zhang L, Dawes W, Walker G (1999) Predicting the effect of vegetation changes on catchment averagewater balance. CRC for Catchment Hydrology Technical Report 99/12.

Manuscript received 20 July 2000, accepted 4 October 2001

australian journal · 1. a review of recharge studies in australia cuan adpetheram abcf, abglen...

Documents