Effect of Climate Change on Groundwater Recharge

Transcription

1 Effect of Climate Change on Groundwater Recharge By Thaddeus Chew Supervisor Professor Murugesu Sivapalan Final Honours Thesis Centre for Water Research November 23

2 Abstract Abstract Groundwater recharge is of extreme importance with regard to both water quantity and water quality studies, especially in Western Australia. It has been very difficult in the past to estimate and model recharge, especially when subjected to climatic and/or land uses changes. The objective of this thesis is to quantify the relationship between climate changes, simulated as decline in rainfall, on groundwater recharge. A multiple wetting front recharge model was utilized to simulate recharge. The model, due to its parsimonious nature and its ability to simulate recharge for a hundred year period over a relatively short period of time on the computer, was selected over other Richards Equation solvers which do not have the same advantages. Simulated rainfall was produced from a rainfall generator, incorporating multi-annual variability, including an overall decline in rainfall variations within and between storm events. The recharge model was calibrated to rainfall and soil characteristic data for five regions. Four simulations were then run through the calibrated recharge model for each location, outputting the corresponding recharge data for the hundred-year period. Analysis of the data involved generating recharge/rainfall ratio curves and a comparison between linear approximations for rainfall and recharge decline. The analysis concluded that recharge is highly episodic and in areas with high soil retention and low average rainfall, threshold dominated. The analysis also demonstrated that the relationship between recharge and rainfall is highly non-linear. The relationship between recharge decline and rainfall decline was also as such that an assessed mean decline of 14% in annual rainfall produced a decline of 75% in recharge over the hundred year period. I

3 Acknowledgements Acknowledgements Before this thesis proceeds any further, I would like to take the opportunity to thank a number of people who have made the accomplishment of this research feasible. One of the first people I would like to express my gratitude to is Professor Sivapalan, or just Siva, who offered this project in the field that I had exhibited interest in and who was the guiding influence throughout the course of my research. I would also like to acknowledge my fellow thesis colleagues: Cameron, Christina and Dina, for their contribution throughout the year and their valuable input and assistance. I am also extremely grateful to Iain Struthers and Jos Samuel, PhD students at the Centre for Water Research, who have assisted me with modelling difficulties. Professor Keith Smettem, Dr Richard Silberstein and Doctor Richard George also deserve mention. I have with consulted them on numerous occasions during the year and they have aided me in providing data, insights and different approaches towards resolving the project objectives. And last but not least I would like to thank my family and friends for sharing my highs and lows. This endeavour is a testament to their support, thank you. II

4 Table of Contents TABLE OF CONTENTS 1 INTRODUCTION BACKGROUND SITE LOCATION Regional Climate Geology and Vegetation LITERATURE REVIEW DEFINING GROUNDWATER Unconfined Aquifers Defining Recharge RECHARGE MODELS Analytical vs Numerical Modelling Scales of Modelling ASSESSING RECHARGE Water Balance Chloride Tracers Hydrogaphs and Specific yield ONE-DIMENSIONAL FLOW: BUCKET MODELS Complex models CLIMATE CHANGE Defining Climate Change Downscaling of GCM s El Nino Southern Oscillation METHODOLOGY SUMMARY OF MODELS: Recharge model Rainfall Model CALIBRATION OF MODELS Calibration of Recharge Model CALIBRATION PROCESS PROJECTED SIMULATIONS Scaling Data III

5 Table of Contents Generating a regional estimate Generating Rainfall & Recharge Declines RESULTS AND DISCUSSION CALIBRATION Accuracy of Models Difficulties with Calibration SIMULATED RESULTS: Simulation Simulation Simulation Simulation Trends in Rainfall and Recharge Comparison of Recharge and Rainfall Declines FURTHER DISCUSSION Limitations on Rainfall Input Recharge CONCLUSION RECOMMENDATIONS FOR FURTHER RESEARCH REFERENCES IV

6 Other Tables TABLE OF FIGURES FIGURE 2-1: SITE LOCATION (SOURCED WALIS 23)... 3 FIGURE 2-2: AVERAGE ANNUAL RAINFALL FOR WESTERN AUSTRALIA (SOURCED BOM 23)... 4 FIGURE 2-3 OBSERVED RAINFALL DURATION IN SOUTHWESTERN WESTERN AUSTRALIA (SOURCED HIPSEY ET AL 23)... 5 FIGURE 2-4: ANNUAL AVERAGE PAN EVAPORATION FOR AUSTRALIA (SOURCED BOM 23)... 5 FIGURE 2-5: GEOLOGICAL MAP OF THE SOUTHWEST REGION WITH KEY BELOW (SOURCED DNMA 1978)... 6 FIGURE 2-6: VEGETATION MAP IN SOUTHWEST REGION FOR THE MONTHS JANUARY AND SEPTEMBER... 7 FIGURE 3-1: DIAGRAM OF RECHARGE MECHANISMS FIGURE 4-1: FLOW DIAGRAM OF THESIS METHODOLOGY FIGURE 4-2: DIAGRAM DEMONSTRATING FORMATION AND MERGING OF WETTING FRONTS (SOURCED STRUTHERS 23) FIGURE 4-3: LONG TERM RECHARGE VS RAINFALL RELATIONSHIP FOR CLEARED AND UNCLEARED LAND FIGURE 4-4: FLOW DIAGRAM FOR STOCHASTIC RAINFALL GENERATOR (SAMUEL 23) FIGURE 4-5: MAP OF LOCATIONS (SOURCED WALIS 23) FIGURE 4-6: MEAN ANNUAL EVAPORATION FOR THE SOUTHWEST REGION FIGURE 4-7: ONE HUNDRED YEAR RAINFALL SIMULATION FOR SERPENTINE CATCHMENT AND LINEAR APPROXIMATION FIGURE 5-1: SITE RECHARGE OUTPUT CALIBRATED AGAINST NATIVE VEGETATION RECHARGE RAINFALL CURVE FIGURE 5-2: GENERATED RAINFALL AND RECHARGE PLOTS FOR BRIDGETOWN OVER 2 YEARS FIGURE 5-3: GENERATED RAINFALL AND RECHARGE PLOTS FOR BRUNSWICK OVER 2 YEARS FIGURE 5-4: GENERATED RAINFALL AND RECHARGE PLOTS FOR CAPE NATURALISTE OVER 2 YEARS FIGURE 5-5: GENERATED RAINFALL AND RECHARGE PLOTS FOR DENMARK OVER 2 YEARS FIGURE 5-6: GENERATED RAINFALL AND RECHARGE PLOTS FOR DUMBLEYUNG OVER 2 YEARS FIGURE 5-7: GENERATED RAINFALL AND RECHARGE PLOTS FOR KURANDA OVER 2 YEARS FIGURE 5-8: GENERATED RAINFALL AND RECHARGE PLOTS FOR MANJIMUP OVER 2 YEARS FIGURE 5-9: GENERATED RAINFALL AND RECHARGE PLOTS FOR NARROGIN OVER 2 YEARS... 5 V

7 Other Tables TABLE OF EQUATIONS EQUATION 3-1: ESTIMATING RECHARGE FROM TRACERS EQUATION 4-1 : WATER BALANCE EQUATION 4-2: WATER TRANSPORT- INSTANTANEOUS CONDUCTIVITY EQUATION 4-3: BARE SOIL EVAPORATION EQUATION 4-4: EVAPOTRANSPIRATION EQUATION 4-5: EVAPOTRANSPIRATION COEFFICIENT EQUATION 4-6: INTERSTORM DURATION: EQUATION 4-7: STORM DURATION:...31 EQUATION 4-8: AVERAGE RAINFALL INTENSITY EQUATION 4-9: PERCENTAGE DECLINE IN RAINFALL OVER 1 YEARS... 4 EQUATION 4-1: PERCENTAGE DECLINE IN RECHARGE OVER 1 YEARS... 4 EQUATION 5-1: CALCULATING AVERAGE RECHARGE TABLE OF TABLE TABLE 4-1: CHARACTERISTICS OF FOUR DIFFERENT SOILS (SOURCED SMETTEM & GREGORY 1996) TABLE 4-2: CALIBRATION SITE DATA TABLE 5-1: TABLE OF SOIL & VEGETATION CHARACTERISTICS, RECHARGE AND PERCENTAGE ERROR TABLE 5-2: TABLE OF AVERAGE RAINFALL, MODEL RECHARGE AND PROJECTED RECHARGE FOR EACH REALISATION TABLE 5-3 PERCENTAGE DECLINE RAINFALL AND RECHARGE FOR REGIONAL AVERAGES OVER THREE SIMULATIONS AND OVERALL AVERAGE VI

8 Introduction 1 Introduction The climate of Western Australia since the beginning of the century has altered considerably, with the changes in global circulation affecting a decline in the volume of rainfall and significantly modifying the rainfall patterns. Due to the semi-arid climatic nature of most of the southwest region of Western Australia, a considerable amount of water resource is supplied through groundwater as opposed to surface water. The significance of extensive groundwater demand has accentuated the need for assessing the effect of climate change on groundwater recharge and recharge patterns. The severity of the water crisis in the south west of Western Australia is evident on a daily basis, with reference specifically to water restrictions in reticulation systems. A study conducted by final year students for the Centre of Water Research has produced a projection of the impact of climate change on the water resources in Western Australia (CWR 22). The increasing population size and the assessed decrease in rainfall will increase the pressure on water resources within the state. The surface water catchments have been projected to be unable to meet the elevated demands on water resources, inducing an increase in the dependence on extraction from groundwater (CWR 22). The implementation of desalination and water reuse policies will reduce the amount of extraction required of groundwater bores, however this will not alleviate the increasing dependence on groundwater extraction significantly (CWR 22). Water resources, including groundwater resources, are only of finite quantity and can be depleted if extracted at an uncontrolled rate. It should be ensured that the extraction of groundwater does not exceed the rate of regeneration. Extended extraction of the resource will produce detrimental effects on the surrounding environment, including salinisation of land through rising water tables, reducing available water for resident vegetation and alterations to soil characteristics which have been known to hamper anthropogenic development. Previous research conducted by the Centre for Water Research final year students accounted for alterations in the level of groundwater resource resulting from variation in climate and the increasing population in Western Australia. However the assumption made by that study that if 1

9 Introduction rainfall declines by a certain percent, the same percentage will be reflected in the decline in recharge, i.e., the ratio of recharge on rainfall will remain constant, may not hold true. This assumption may result in the under- or over-estimation of recharge as a function of rainfall, and may seriously affect estimation of the sustainable level of removal of groundwater from an aquifer. The objectives of this thesis are to verify the assumption of linearity in the relationship between rainfall and recharge for changes in climate and if this assumption does not hold valid, to specify a more appropriate relationship between rainfall and groundwater recharge. The computations however will assume that the change in climate is a result of natural variability as opposed to the result of anthropogenic activities. The relationship between rainfall and groundwater will be quantified through implementing a rainfall model to simulate climate change, and through combining synthetic traces of rainfall sequences produced by it with the recharge model, to generate a set of recharge time series to compare to the rainfall data. Analysis of the data should yield sufficient information to obtain a preliminary statement on the relationship between rainfall and recharge, under altered climate conditions 2

10 Background 2 Background 2.1 Site Location The site location for the thesis is taken to be the area bounded by Perth and Albany. Previous definitions of South West for Western Australia have varied, including assigning the region in accordance with aquifer regions, rainfall or geological similarity (BOM 23). But as the objective of the thesis is to assess the rainfall-recharge relationship over different locations in Western Australia, sufficient variations of atmosphere and soil conditions are required so as to provide a more universal assessment of the situation within Western Australia. The boundaries of the site area are between the latitudes S and 35 2 S and between the longitudes E and E (refer to highlighted area below in Figure 2-1). Figure 2-1: Site Location (sourced WALIS 23) The site area spans 83 km 2 as demonstrated by the scale in the above figure, but will during the course of this thesis, not address the Gnangara mound system near Perth due to the different aquifer dynamics (Bekele et al 2). The site area includes both the coastal regions of Western 3

11 Background Australia and the inland wheatbelt region. The area contained within the designated boundaries is highly diverse in geology, vegetation and rainfall patterns, with the latter most influential for recharge estimates Regional Climate The rainfall patterns as outlined graphically by the Bureau of Meteorology are only described in terms of average annual rainfall as evident in Figure 2-2 rather than the other parameters including duration or intensity. Figure 2-2: Average Annual Rainfall for Western Australia (sourced BOM 23) However research at the Centre for Water Research has produced modelled data on the average observed rainfall durations in January and July as demonstrated below: 4

12 Background Figure 2-3 Observed Rainfall duration in Southwestern Western Australia (sourced Hipsey et al 23) An average regional duration can hence be derived from an average of the estimated mean durations for each of the above figures (Hipsey et al 23). The variation of pan evaporation levels geographically is not as significant as the average regional duration as demonstrated in the figure below: Figure 2-4: Annual Average Pan Evaporation for Australia (sourced BOM 23) From other available data, it has been found that the monthly pan evaporation during summer varied at an increased level compared to monthly pan evaporation in winter geographically. The southwest region of Western Australia geographically has the greatest deviation (BOM 23). 5

13 Background Geology and Vegetation The southwest region of Western Australia also demonstrates high rates of spatial variability of vegetation and geology. The soil composition of the southwest regions can be outlined in the following table corresponding to the figure below: Relevant Soils: Figure 2-5: Geological Map of the Southwest Region with key below (sourced DNMA 1978) A 2 Duplex soils Bb 4 Massive Earths Cd 1 Bleached sands 2 Yellow duplex soils Cf 1 Shallow Sand soils 6

14 Background As can be observed the variability within the southwest region of Western Australia geologically is significant, with soil types ranging from duplex soils (loams) to sandy soils, although the dominance of sandy soils in the south of the region is apparent, while loamy soils become increasingly common further inland. The vegetation for the southwest region is difficult to estimate, not just due to spatial variability but also because of temporal variability. The temporal variability is less significant along the coastline as represented in Figure 2-6, but in the wheatbelt regions the vegetation fluctuates from sparse vegetation in the summertime to lush in the winter as specified by the NDVI index (BOM 23). The following figures illustrate both the temporal and spatial variability over the southwest region as outlined previously: Sparse Lush Figure 2-6: Vegetation Map in Southwest Region for the months January and September (sourced BOM 23) As verified in the above figure, the coastal fringe of vegetation is composed primarily of what is termed lush vegetation and is less affected by temporal variations, while the inland regions appear to be sensitive to monthly fluctuations in climate with temporal variation in sparse and lush vegetation. The geographical variability in the spring months corresponding to a period of regrowth in the southwest region is significantly less than variability in the summer months. Although this variability in vegetation is expected, the fluctuations will affect the accuracy of evaporation and evapotranspiration rates during modelling of recharge. 7

15 Literature Review 3 Literature Review 3.1 Defining Groundwater The term groundwater is ambiguous in hydrology literature, while groundwater recharge is even less clearly defined. However the final destination of groundwater recharge is ultimately an aquifer. This thesis will use the definition of an aquifer as a mass of rock or soil capable of storing and transmitting water (Armstrong et al 1998 (1)). Aquifer types consist of confined, unconfined, fractured rock and semi-confined aquifers; however the primary aquifers are the confined and the unconfined, with the latter being more common. It is however important to first determine what aquifer will be applied to the model due to the extremes in system dynamics Unconfined Aquifers An unconfined aquifer is defined as an aquifer with an upper boundary of groundwater flow as a water surface (Dingman 1994), with a base consisting of clay or bedrock, usually occurring close to the soil surface. The water table within the porous medium also defines the water pressure within an aquifer (Armstrong et al 1998 (1)), or that the total head at the water table is equal to the elevation above a defined datum (Dingman 1994). As the water table is only confined along one boundary, the water table has the freedom to fluctuate vertically, with the effect of a nonconstant hydraulic conductivity which may influence calculations accordingly. The unconfined aquifer consists of an impermeable base as stated above, saturated zone and unsaturated zone. The transition between the saturated zone and the unsaturated zone is defined as the water table depth (Fetter 21). The amount of water retained in an aquifer is known as the Volumetric Water Content and is related to Specific Yield. Recharge for unconfined aquifers is commonly sourced via infiltration, with vertical percolation through the unsaturated zone towards the saturated zone. As unconfined aquifers are commonly found on hillslopes or near rivers, horizontal recharge is not uncommon due to elevation gradient (Dingman 1994), although a pressure gradient may also induce horizontal recharge (Fetter 21). Discharge from the aquifer may occur horizontally as with horizontal recharge or through direct evaporation and evapotranspiration processes (Armstrong et al 1998 (1)). 8

16 Literature Review This thesis will assume that an unconfined aquifer will be the destination for recharge, as such aquifers are present throughout the Western Australian wheatbelt in general, although it is recognised that confined aquifers are existent in other areas of the southwest. This is the reason the model does not hold true to recharge response to climate change in Gnangara mound due to the different recharge system dynamics with a mixture of unconfined and confined aquifers (Bekele et al 2). A confined aquifer is classified as a saturated aquifer, bounded above and below by aquicludes or confining layers (Dingman 1994). The confining layers may consist of bedrock or clay, which are relatively impermeable. Confined aquifers have been known to be found kilometres beneath the surface. It is not uncommon for unconfined aquifers to be located above confined aquifers. As confined aquifers are confined and saturated, breaks in the confining layer will induce the water level to rise up to a certain depth known as the piezometric head (Armstrong et al 1998 (2)). Confined aquifers are always assumed to be saturated, making mathematical calculations relatively simple. Recharge and conversely discharge to and from confined aquifers are primarily sourced from an upstream unconfined aquifer that leads into the confined aquifer (Fetter 21). Infiltration to confined aquifers is virtually negligible due to the impermeability of the upper boundary. However semi-confined aquifers or leaky aquifers involve a semi-permeable upper boundary allowing some small amount of infiltration to occur across the boundary in both directions along the vertical axis Defining Recharge There are also many different ways to interpret recharge. One method is to define recharge as addition of water to the body of water already stored in the ground or water-table recharge (Armstrong et al 1998 (2)). Another definition for recharge is drainage in which drainage can be considered as potential recharge but may not actually reach the groundwater system. (Walker & Zhang 22). The above definitions only apply to unconfined aquifers as confined aquifers derive their form of recharge from their connection to unconfined aquifers (Fetter 21) 9

17 Literature Review Water-table Recharge vs Deep Drainage Water-table recharge in this case is dependent on the amount of water that infiltrates toward a groundwater system (Armstrong et al 1998 (1)) or as part of the rainfall that reaches groundwater via the unsaturated zone (Bekesi & McConchie 1999). The rate at which recharge occurs is primarily dependent on the thickness of the unsaturated zone and the hydraulic conductivity of the unsaturated layer (Fetter 21). Permeable soils with a high water table will obviously receive more recharge than impermeable layers with a deep water table. However water-table recharge can be further divided into gross recharge and net recharge. Gross recharge pertains only to the recharge processes, as compared to net recharge which includes discharge processes. The discharge processes include evapotranspiration (plant uptake) or evaporation. When tree roots breach the saturated zones or during periods of high sunlight intensity, the net recharge can be significantly less than gross recharge (Ragab et al 1997). Scientific papers still often confuse the terms net and gross recharge when relating to water-table recharge, although many relate net recharge as the change in the piezometric head, or the change of the water table pressure heads, which is the easiest component to measure rather than using tracers (Leduc et al 1997). Deep drainage is different in that it is actually a component of water-table recharge (Armstrong et al 1998 (1)). Deep drainage, which is sometimes known as vadose seepage (Avner et al 1998) is similar to net recharge in that it occurs when recharge is more than discharge components such as evaporation or evapotranspiration. The difference is that drainage is not calculated as recharge directly to an aquifer, but rather as the amount of water infiltrating past a selected height in the soil (with respect to one-dimensionality). Previously in simple soil models drainage/recharge has been assumed to be the amount of water percolating beyond the root zone (Ragab et al 1997). Deep drainage generally comprises the smallest fraction of the water balance process (Armstrong et al 1998 (1)). 3.2 Recharge Models Models have been developed over the years to estimate the amount of recharge in a system. As described in section 3.1, recharge has a variety of definitions. Models have been developed to 1

18 Literature Review accommodate the different notions of recharge. Models provide solutions through analytical methods, governed by equations, or numerical models, which are based on subdividing an area into a grid and calculating the interactions between each cell (Hatton 1998). The models also differ in the scale, dimensionality and complexity in estimation of recharge processes. Predictive modelling in particular is utilised to model changes in recharge due to land use and climate change (Hatton 1998), the latter the subject of this thesis Analytical vs Numerical Modelling Analytical expressions for recharge do exist, including one defined by Morel-Seytoux, who derived an expression that calculated recharge through unit infiltration (Wu et al 1997). The foundation of an analytical model is the analytical equation governing the groundwater system dynamics (Fetter 21). The model is generally solved within a short space of time and is a relatively inexpensive procedure. The model assumes homogeneity of parameters and in the case of groundwater recharge, the analytical model assumes uniform hydraulic conductivity of an aquifer. The necessary assumptions for boundary conditions incorporate a specified pressure head and a constant value for recharge and discharge (Hatton 1998) The model is also assumed temporally and spatially continuous, meaning that the model is applicable for any location at any point in time. Numerical models divide an area into a grid (spatial discretisation) or in the case of onedimensional models, layers (Hatton 1998). The dynamics of each grid cell/layer is determined over an allotted period of time, with the period known as a time step. The dynamics are in turn determined in terms of algebraic equations or, when relating to groundwater recharge, flow equations. Numerical models can be used regardless of the groundwater system s boundary conditions, non-uniform root uptake or soil property heterogeneity (Van Dam & Feddes 2). Comparison of the two model types highlights advantages and disadvantages to both models. On the one hand analytical models are quickly solved without extensive effort. However analytical models are unable to include temporal and spatial alterations, for example seasonality of recharge (Hatton 1998) or non-uniform soil and vegetation properties (Van Dam & Feddes 2). The approximations made for aquifer characteristics may not be applicable to large groundwater 11

19 Literature Review systems, rendering the analytical model invalid (Hatton 1998). Numerical models in contrast are able to solve complex boundary conditions and account for heterogeneity in the input parameters (Fetter 21). Numerical models are also better equipped to handle three dimensional flow, as with MODFLOW rather than one dimensional vertical flows (Batelaan et al 23). However, numerical models are also more expensive to develop and take longer to complete when the increments are extremely small. Conversely the smaller the increments are, the more accurate the model. Despite assumptions of one dimensionality (in the vertical) for the movement of water, due to climate variability existing as a major influence, the implementation of numerical models cannot be avoided Scales of Modelling There are different scales at which models are produced at including scales of temporal nature and spatial variability. The non-linear relationship between the two scales has previously been acknowledged, although the relationship is further complicated by the meteorologicalhydrological process non-linearity (Dionysia & Dimou 1997), especially at catchment scale. However, previous attempts have been made to further determine the accuracy of models at a particular time and/or over a defined space. Constructing a model at a predefined scale involves: upscaling, downscaling or the establishment of new laws at the given scale and validating the laws through measured parameters thereafter (Dionysia & Dimou 1997). This thesis will utilise the concept of upscaling, where it is assumed that the calibration of a model over a number of smaller areas will validate the model at the desired scale. Overcoming Spatial Variability The range of spatial variability is more extreme when the area of land governed by the analysis increases. In the more extreme cases it has been recognised that upscaling from a single location may not generate sufficiently accurate results (Dionysia & Dimou 1997). Analysis of individual locations within the region is also prohibitive due to the large amount of data required (Bekesi & McConchie 1999). However some simplification of interacting soil and hydrological processes is still necessary to achieve a manageable level of sophistication of estimates of recharge values (Bekesi & McConchie 1999). The Monte Carlo technique is useful to test the order of uncertainty 12

20 Literature Review within the model. The analysis involves creating random variables for each of the input parameters, running the model for a defined number of times governed by the equation: s 2 f ( a) < u n where f(a) is the reverse confidence interval, s is the standard deviation, n is the number of runs required and u is the range of values. The number of runs required is generated by substituting the standard deviation and range values from an initial analysis, which in some cases involves up to 5 runs (Bekesi & McConchie 1999). After the number of runs required has been gauged, the simulation should run n number of times, where via statistical analysis (linear regression) on the range of outputs variability can be assessed through correlation of data. An average figure for recharge can then be generated. Previous Monte Carlo analysis has been carried out on the effect of rainfall variability on recharge over a regional area involving fixed inputs (rainfall, evaporation) and randomised parameters (soil properties) (Bekesi & McConchie 1999). This technique is especially useful when the necessary data is unavailable as accurate data is not required for regionalisation. 3.3 Assessing Recharge Assessing actual recharge data is difficult and often imprecise, however reasonably accurate recharge data is necessary for model calibration. The most common techniques for assessing recharge are: water balance, chloride tracers and the specific yield; though the first and last are often related. The variation within each type for assessment of recharge has also been assessed Water Balance The term water balance can refer to the measurement of the volumetric water content, which is often difficult to discern. However one method for the measurement of water balance is the change in piezometric heads. The change in piezometric heads reflects the change in water table height, which represents the change in volumetric water content given a specified area or region (Armstrong et al 2 (1)). 13

21 Literature Review Chloride Tracers The use of chloride tracers for estimation of recharge rates has already been proven historically to be both popular and accurate (Bekele et al 2). As chloride is a conservative tracer, the amount and distribution of chloride reflect evaporation and pan evaporation and in arid conditions under native vegetation, it is not uncommon for a high concentration of chloride to exist just below the soil (Bekele et al 2). Measuring recharge via chloride tracers assumes steady state conditions, therefore the reading is most accurate in locations that have already reached steady state conditions (Bekele et al 2). During the estimation of recharge, advection was assumed dominant over diffusion hence the equation as assumed by Bekele et al (2): Equation 3-1: Estimating Recharge from Tracers R = Pc p / c g R is the mean annual recharge [mm/yr] P is the mean annual precipitation [mm/yr] c p is the chloride concentration in precipitation [mg/l] c g is the groundwater chloride concentration at the water table [mg/l] The groundwater recharge is thus proportional to the precipitation and the ratio of chloride in precipitation to that in the water table. This also assumes that recharge is understood as addition to a water table rather than drainage past a certain depth Hydrogaphs and Specific yield The recharge can also be estimated from a combination of hydrograph output for a region and the specific yield (Bekele et al 2). However the specific yield is extremely difficult to estimate and is often simulated under laboratory conditions, where the estimate is then applied to the field. 14

22 Literature Review 3.4 One-Dimensional Flow: Bucket Models Many different models exist for the purpose of generating recharge estimates on a regional scale. It is not uncommon for models to incorporate horizontal flow as well as vertical infiltration requiring a numerical model grid (Armstrong et al 1998 (2)). Numerical models may also incorporate other factors of heterogeneity including geography, inconsistent horizontal geology and other miscellaneous subsurface bodies or point inputs influencing recharge. Programs, namely Modflow, are effective at incorporating such factors during simulations (Batelaan et al 23). However modelling recharge to such an extent can be counterproductive due to the lack of availability of data and the sheer amount of time necessary to complete the complex calculations (Bekesi & McConchie 1999). It is therefore reasonable to assume homogeneity to a certain extent over a regional scale leading to the development of simple models to assess recharge (Walker & Zhang 22). By assuming homogeneity over a regional scale and assuming recharge as deep drainage, the simple single layer bucket model provides a suitable estimate for recharge estimation. The simplest form of the bucket model consists of two primary components: evapotranspiration and rainfall which factor as output and input respectively (Walker & Zhang 22). Drainage is assumed to occur when rainfall exceeds evapotranspiration leading to groundwater recharge. The single layer bucket model is governed by a simple equation as follows: RECHARGE = RAINFALL Infiltrated EVAPORATION TRANSPIRATION Input data into the simple bucket model includes daily rainfall, daily potential evapotranspiration, storage capacity and soil water deficit at the initial stages of the simulation (Walker & Zhang 22). The single layer bucket model can be represented thusly: (refer to figure on next page) 15

23 Literature Review Figure 3-1: Diagram of Recharge Mechanisms Complex models The single layer bucket model is efficient in that it does not require a large amount of data for simulating recharge for a given rainfall event. However the single layer bucket model does not account for heterogeneity in the soil layer or for plant growth which includes changes in the depth of the root zone and the leaf area index, both factors for influencing deep drainage (Walker & Zhang 22). However it is also possible to add complexities to the single layer bucket model to accommodate for the above non-linear variability. One method is to utilise multiple layering or cascading buckets representing different layers in the soil as ascribed by Richie and Otter-Nacke 16

24 Literature Review in 1985 (Walker & Zhang 22). Inter-layer water transport occurs once the bucket has reached maximum capacity, tipping into the next bucket (Walker & Zhang 22). The other method is to use Richards Equation which models water transfer dependent on water potential gradients and the initial and boundary conditions (Walker & Zhang 22). Models utilising Richards equation can incorporate plant growth parameters and multiple layers, but the primary difference between this form of modelling and the cascading bucket is that the different non-linear processes such as infiltration, redistribution, plant growth, etc do not need to be treated separately (Zhang & Walker 22). 3.5 Climate Change Climate change and climate variability are issues that have been previously researched by meteorologists and hydrologists through history and are associated with changes in patterns of rainfall and temperature. However the driving force behind climate change has been subject to much debate and is still poorly understood (Hunt et al 1999). One of the main driving factors behind climate change that has previously been proposed is the El Nino-Southern Oscillation or ENSO (Houghton 1996). Debate also exists over the viability of assessing climate change over a regional scale through simulating Global Climate Models (Houghton 1996) and about the category of model that is most feasible for simulating rainfall over a regional scale Defining Climate Change Climate variability has previously been defined numerous ways, however common argument proposes that climate is either more variable (temporally variable) or extreme (event variable) (Houghton 1996). Variability in results reported by others is possibly due to the calculation process, for example taking variation in extreme events as compared to taking variation over an averaged number, as well as the way climate change/variability is defined (Houghton 1996). The variability, especially when related to recharge, is primarily related to the change in rainfall or the change in evapotranspiration, with the former proportionally dominant in high rainfall areas and the latter in regions of low rainfall (Houghton 1996). Again, to simplify the assumptions, many texts have previously supported the factor for climate change as the change in rainfall (Houghton 17

25 Literature Review 1996; Bekele et al 2; Nicholls et al 1999;). Climate change also previously has been defined as the result of the effect of anthropological activities on atmospheric circulation or as the result of natural variability. This thesis will utilise the definition of climate change as the result of natural variability. However the underlying fact remains that rainfall has declined in the Southwest as seen in Manjimup, a reasonable indicator for the rest of the Southwest Region, which has experienced a decline in May-October rainfall of about 25% since the middle of the century (Nicholls et al 1999). This fact defines climate change over a regional scale as a decline in rainfall. Variable vs Extreme Previous research by Yu and Neil (Nicholls et al 1999) has established that decadal scale decrease in annual rainfall is not accompanied by decrease in high intensity rainfall. However further research on climate variability and predictability has indicated that both the changes in the number of rain days and the intensity during extreme events have produced the current trend in rainfall patterns (Nicholls et al 1999). A correlation of.8 for total rainfall to the number of rain days exists, where the number of rain days is increasing for a decrease in rainfall; similarly a correlation for total rainfall of.75 is associated with the decrease in the intensity and amount of rainfall on extreme rainfall events (Nicholls et al 1999). The high correlation factor for the first statement infers that the total rainfall is inversely related to the number of rain days. From the high coefficient for the second statement it is apparent that intensity during high rainfall events may also drive a decrease in rainfall Downscaling of GCM s Physically based models designed to simulate weather patterns are known as general circulation models and are effective at simulating climate at sub-continental scales with a minimum time scale of seasonal averages (Houghton 1996). However the accuracy of these GCM s depreciates markedly when utilised at a local or regional scale (Houghton 1996). The downscaling of general circulation models has led to the development of statistical and stochastic techniques, to reduce the error between the simulation and historical data. 18

26 Literature Review Statistical Downscaling Statistical scaling is a widely used technique and has been used by a variety of researchers. Statistical downscaling was utilised during investigations into climate variability in the southwest of Western Australia (Nicholls et al 1999) and for assessing multi-seasonal predictability and climate trends (Hunt et al 1999). However statistical models have been known to be ineffective at regional scales. A form of statistical modelling known as Limited Area Models subdivides the computational grid to produce a smaller boundary. The daily precipitation processes were then modelled through multivariate probability distributions conditional on the derived patterns (Bates et al 23). However LAM s still overestimate the frequency and underestimate the intensity of daily precipitation, substantially deviating from historical data (Bates et al 23), producing weather patterns independent of precipitation and experiencing only limited success in simulating wet and dry spell statistics. Stochastic Modelling: Markov As deterministic models have not been used since the turn of the mid-century due to their lack of success in simulating random rainfall events, a type of probabilistic model known as the stochastic model is frequently used instead. A stochastic model has been developed by Bates, Charles and Campbell (23) to simulate precipitation amounts in the southwest of Western Australia. The downscaling technique in particular incorporates the hidden Markov model to simulate precipitation occurrence and multiple linear regressions. (Bates et al 23). The NHMM determines definite patterns of precipitation over a spatial domain rather than patterns in the atmospheric circulation (Bates et al 23). However most importantly, the Markov chain model is not based on a generalised set of parameters for defined weather classifications, but instead on a set of atmospheric predictor variables (Bates et al 23). This means that the next projected weather state, in many cases precipitation, is dependent on a set of atmospheric parameters present in the current weather state instead of the average of previous weather data. The stochastic model has previously incorporated mean sea level pressure as a predictor (Bates et al 23). The application of the NHMM to CSIRO general circulation models and limited area 19

27 Literature Review models, altering a purely statistical model into a physical stochastic model, has overall allowed the models to produce feasible simulations of historical atmospheric patterns in the South West of Western Australia El Nino Southern Oscillation History of ENSO The phenomenon of the El Nino Southern Oscillation is a mode of climate variability with a duration of about 2-5 years (Houghton 1996). ENSO is suspected to have triggered climate change through the release of latent heat, affecting circulation in the atmosphere (Houghton 1996). The phenomenon has only existed in instrumental recordings only as far back as the late 19 th century causing great difficulty in deriving a history of ENSO events. The Southern Oscillation Index or SOI is an index of ENSO based on surface atmospheric pressure, with negative indices indicating El Nino events and positive indices an indicator of La Nina events. Records have indicated that periodicity of ENSO has fluctuated after the 195s with a periodicity of about 5 years between 195 and 1965, since then shortening to approximately 4 years (Houghton 1996). However since 1976, the ENSO pattern has shifted so that there are more frequent intrusions into El Nino patterns and conversely less in La Nina (Houghton 1996). Analysis by Trenberth and Hoar (Houghton 1996) has also indicated that this pattern has never previously been found in the 12 years of recorded data and the probability of extended El Nino behaviour has evidenced from 199 to 1995 has a one in 2 chance of occurring, although research Allan and D Arrigo has demonstrated that such an event has occurred before the advent of recorded data. Association of ENSO with Rainfall Patterns The change in ENSO has been previously associated with the change in climate, especially in Western Australia (Nicholls et al 1999). ENSO, with the release of latent heat, is possibly responsible for changes in atmospheric circulation over a local scale. It has previously been recognised that at least some of the recent variation in rainfall within the tropics and subtropics has been in part due to unusual ENSO behaviour (Houghton 1996). However any direct 2

28 Literature Review relationship between rainfall and ENSO behaviour is still unestablished. Nicholls et al 1999 concluded in a report for climate variability and predictability for Western Australia that half of any observed decline in rainfall is related to changes in regional atmospheric circulation, for example the Perth MLSP, part of which is due to changes in ENSO. However little of observed rainfall decline is attributable to long-term changes in ENSO. Although ENSO can be factored out as a long term, general parameter it can still be used effectively as an atmosphere predictor variable as listed previously. The NHMM does not require a parameter describing long term change, as the projections are based on present conditions or indicators rather than long term statistics, making ENSO an ideal indicator for projecting rainfall fluctuation. 21

29 Methodology 4 Methodology The study consisted of two phases: calibration and simulation. Two models were utilised in the course of this study, one recharge model and one rainfall model. The recharge model selected is known as the multiple wetting front model and was devised by Iain Struthers (Struthers 23) at the Centre for Water Research. The rainfall model was originally developed by Robinson and Sivapalan (1997), and further refined by Sivandran (1999) and Jos Samuel (23). The recharge model was then calibrated against data sourced from research by Hatton et al (2) estimating the average recharge per rainfall amount in Western Australia. Following completion of calibration, input of rainfall realisations generated from the stochastic rainfall model into the recharge model commenced to estimate the effect on recharge over a regional area for a projected change in rainfall. The diagram below summarises the processes: Gather Rainfall & Soil Data Calibrate Recharge Model Eight Locations Alter Parameters Fit to Curve Save parameters Simulation Scale Rainfall to Locations Perform Four Simulations for each location Analyse Output Figure 4-1: Flow Diagram of Thesis Methodology 22

30 Methodology This section will commence by providing a summary of the models and elaborate on the governing theory to provide a better understanding of how the analysed results were produced. The procedures for the calibration and simulation procedures are next outlined in more detail, in part to provide insight into the generation of recharge and rainfall trends that were essential towards the estimation of the relationship between the two variables, and highlight some of the limitations of the above procedures. 4.1 Summary of models: The two models utilized within the thesis were a multiple wetting front model, for the purpose of predicting recharge, and a stochastic rainfall generator to generate rainfall data. Both the models were developed by postgraduate students within the Centre for Water Research. The thesis ultimately involved the incorporation of the data generated through the stochastic rainfall generator into the multiple wetting front model to simulate changes in recharge due to a decline in rainfall Recharge model The recharge model was developed in Matlab, a language similar to C++, generating recharge in daily time step increments and incorporating a water balance equation into the frontal transport of water. The model was initially developed to the concept of the bucket model, estimating drainage through the subsurface exceeding a user-determined depth. The recharge model also incorporated the added complexity of penetration of multiple fronts, evapotranspiration via root uptake and interception of water through foliage cover, both of which are again subject to user discretion. Other user determined variables include the soil characteristics porosity, saturated conductivity, residual volumetric water content and the van Genuchten Parameter. The van Genuchten Parameter is a variable that defines the water retention capabilities of a particular soil type, with a higher value corresponding to less water retention (van Genuchten 198). Rainfall input is computed via a daily data set over a period in time, with the average intensity of rainfall controlled through the user defined parameter for average storm duration. Moisture is also withdrawn from the system through pan evaporation and evapotranspiration, which is again defined in daily increments by the user. 23

31 Methodology Basic Theory It is assumed that water from rainfall, upon contact with the ground infiltrates into the soil forming a wetting front, with characteristics determined by the local soil characteristics. The volume of water per unit area or the water balance within a soil column is governed by Equation 4-1. The individual wetting fronts are constructed from different rainfall events taken upon a daily basis. The rainfall characteristics such as the intensity of the rainfall are determined by an average rainfall event duration and rainfall data, both of which are predetermined by the user. The development of a wetting front is divided into two characteristics: depth and saturation. The following diagram introduces the wetting front concept: Figure 4-2: Diagram demonstrating formation and merging of wetting fronts (sourced Struthers 23) The calculated moisture content and depth of the front are dependent on the variables mentioned previously in the section concerning soil characteristics. Contact between wetting fronts may produce a merging of fronts into a single front, assuming the tolerance for merging has not been exceeded, which then resumes travel through the soil column as a single front towards the depth 24

32 Methodology of drainage as set by the user. The above figure displays the redistribution process when two fronts come in contact. The evapotranspiration and evaporation are also factored into the model. Evaporation induces a loss of moisture through a predefined depth in the soil column and has been known to vary throughout the year on a seasonal basis. Variable evapotranspiration is also evident in regrowth areas until maturity is reached by the vegetation, allowing the evapotranspiration to remain constant. Evapotranspiration also induces the loss of moisture throughout a soil column over the user defined root depth. Moisture is also retained throughout the depth of the soil column, depending on the soil parameters in the section relevant to the Water Transport Equations. The remaining water in the fronts that breaches the depth determined by the user is ultimately considered recharge or drainage. Water Transportation Equations The rate at which the wetting front progresses at and the amount of moisture stored in each front as the front moves through the soil is dependent on the soil characteristics. The main characteristics include retention coefficient, porosity (saturated volumetric water content), saturated conductivity and the van Genuchten Parameter. The amount of water transported out of the system is dependent on the evaporation and evapotranspiration parameters. The pertinent equations are: Equation 4-1 : Water Balance t θ i, t j where = x i, j t t t { min[ f i, K( θ i, j )] K( θ i, j )}[ θ i, j θ i, j 1 ] [ θ θ ] + min[ f, K( θ )] min[ f, K( θ )] i, j i, j 1 { }t i i, j t t θ ij and x ij are the moisture content and extension (respectively) of front j in layer i at time t within the model timestep i i, j 1 f i is the infiltration rate (constant during a given timestep) into layer i θ ij and x ij are the moisture content and extension (respectively) of front j in layer i at the start of the timestep 25

33 Methodology K is the hydraulic conductivity Equation 4-2: Water Transport- Instantaneous Conductivity K = K where sat S (1 (1 S ( VWC VWCr ) S = ( VWC VWC ) s r ) (1/ VGPm) VGPm K is the Conductivity K sat is the Saturated Conductivity VGPm is the van Genuchten Parameter m VWC is the volumetric water content VWC s is the saturated volumetric water content (porosity) VWC r is the residual volumetric water content ) 2 Equation 4-3: Bare Soil Evaporation E bs = e α 1 p bs ( C ) Ebs is the bare soil evaporation ep is the pan evaporation v α bs is the ratio of the storage of all fronts within the Bare Soil Evaporation Zone and the saturation storage within the Bare Soil Evaporation Zone C v is the percentage vegetation cover Equation 4-4: Evapotranspiration E t = e α C p t v Et is the evapotranspiration ep is the pan evaporation C v is the percentage vegetation cover α t is the coefficient for evaportranspiration worked out from the equation below: 26

34 Methodology Equation 4-5: Evapotranspiration Coefficient θ θ wp θ θ wp α t = if θ wp < θ < θ c θ c θ wp 1 θ θ c θ Saturation θ wp Saturation of Wilting Point = max(, min( θ, θ ) max( θ, θ j ( θ θ ) c c wp wp j 1 )) θ c Saturation Field Capacity The storage of water in a front as outlined in Equation 4-1 : Water Balance, is dependent on the level of saturation within the front, the conductivity for the time step, a constant for infiltration rate and the moisture content already present within the soil structure. The depth and moisture content for each front during a time step is dependent on the instantaneous conductivity, with the amount of water progression in the front bounded by the parameters for residual water content (minimum) and the porosity (maximum) and the van Genuchten Parameter. Residual water within the pores may then attach to the next front. The amount of water removed from the system per unit time step is dependent on the evaporation and evapotranspiration values, with removal sourced from the front with the highest moisture content. However the model also accounts for the evapotranspiration rate within the model itself using the pan evaporation values as a base, factored into Equation 4-4 without requiring additional vegetation data. All these equations drive the amount of recharge over a defined time period without long simulation times, while the simplicity of the model phases out the necessity for a high quantity of parameters. These qualities ensure the suitability of the model to project recharge over long periods of time, as required for long term climate change scenarios. Model Structure The model is structured to allow the main script to directly receive user inputs, utilizing subfunctions to perform the appropriate calculations as outlined previously in the Water Transportation Equations. Frontal distribution is calculated on a daily basis with separate 27

35 Methodology functions factoring: evaporation/evapotranspiration, gravitational drainage, infiltration, vegetation growth, front merging/redistribution; while water balance checks ensure accuracy of front distribution, contributing to the recharge estimate, is maintained. The main script initially included a plot command to illustrate the fluctuation of recharge but was then further modified to plot the annual rainfall, calculate an average rainfall value, the average recharge value and the recharge projected for the average rainfall value from the lower recharge-rainfall curve, representing recharge under native conditions, as illustrated below in Figure 4-3. Figure 4-3: Long term Recharge vs Rainfall Relationship for Cleared and Uncleared land (sourced Hatton et al 2) Please note that the equations as outlined on each curve should be reversed, but the R 2 values remain correct. The recharge curve represented in the figure above is the result of recharge data gathered over a period of time for a certain average rainfall as represented by the different points on the graph. The data points cover spatial variations in the southwest region of Western Australia. It does not represent recharge in deep sandy soils located in isolated areas in Western Australia, including the Gnangara mound system. The system dynamics in these deep soils are 28

36 Methodology different from that of soils in other locations of Western Australia (Petheram et al 2). As the model is calibrated to the data represented in the above figure, the model results are not indicative of recharge fluctuations in the aforementioned regions. However as this thesis represents a preliminary assessment of the relationship between recharge and rainfall in the southwest region of Western Australia and recovered long term statistics, the utilisation of the data for calibration purposes was deemed satisfactory. Model Limitations and Assumptions The recharge model is similar to the bucket model concept as it assumes only one dimensional flow, which may not be appropriate in structures with complicated groundwater systems that induce horizontal flow as with the Gnangara mound (Bekele et al 2). The model is still in development and has not yet incorporated a number of variables that need to be accounted for in the physical world. The main assumptions are associated with the soil characteristics. These characteristics are assumed stationary. One of the assumptions is the bucket model assumption of homogeneity in a single soil layer, required to utilize the Water Balance Equation (refer to Equation 4-1). Many regions in Western Australia comprise of duplex soils, a mixture of sand, clay and loam, having a significant effect on system dynamics. The model also assumes a static rainfall duration value, which is not realistic when assuming climate change is due to alterations in atmospheric circulation including the value for rainfall duration. The model has also not as of yet incorporated the effect of macropores, and thereby preferential water flow pathways through the soil, on recharge. The model has also ignored the effect of a rising water table on recharge or drainage which may induce variations in recharge. Finally the model has not included the effects of suction or capillary pressure, but instead suction was assumed negligible due to the low level of saturation of Australian soils. User Assumptions Assumptions were also made during the modelling process that were not as per se part of model limitations. One of the more important assumptions was that recharge was taken as drainage as opposed to direct input of water to the water table, assumed in the RIRDC report (Hatton et al. 29

37 Methodology 2) affecting the viability of calibrating recharge output against the data curve (refer to Figure 4-3). The variation of recharge figures between drainage values and direct recharge to the water table was explored in previous research, concluding that significant variation occurred (Petheram et al. 2). However, subject to the lack of available resources on available recharge data, due to the difficulty of obtaining accurate recharge estimates, and by reason of allotted time, there was no choice but to utilize the recharge/rainfall curve during calibration. The next assumption is that vegetation characteristics are stationary and at peak state, even when exposed to climate change over a 1 year period. The term peak state refers to the maximum interception of rainfall by vegetation, foliage cover and achievement of maximum root depth. This assumption was used as the added complexity of vegetation growth can be factored into the simulation in future research once an initial simulation was completed for stationary vegetation characteristics. The evaporation was also assumed to be constant and assumed as pan evaporation as opposed to evapotranspiration, as the evapotranspiration value had been accounted for within the model (refer to Equation 4-4). As seen previously in the Section monthly levels of fluctuation in evaporation within the wheatbelt region remain high, influencing the level of accuracy of calibration. The greatest influence of evaporation fluctuation is primarily isolated to low rainfall areas where the amount of recharge is determined by evaporation rather than amount of rainfall (Petheram et al. 2). A certain degree of freedom has also been allocated to the vegetation and the saturated conductivity values to allow for variation between locations. The other soil characteristics are based on values from previous research by Smettem & Gregory (1996) Rainfall Model The rainfall model this thesis utilised was developed by Robinson and Sivapalan (1997), and further refined and extended by Sivandran (22) and Jos Samuel, PhD student in the Centre for Water Research. The rainfall model is a stochastic model, and previously applied to rainfall data collected from within the Serpentine catchment, generating synthetic rainfall data for 1 year intervals. The rainfall model incorporates a variety of rainfall parameters including hourly or daily rainfall data, sea surface temperature from the IPO index and mean sea level temperature 3

38 Methodology from the SOI index. The following diagram demonstrates the model processes for the rainfall generator: INPUT -hourly rainfall data Rainfall Data -daily rainfall data -Sea surface temperature PROCESS Modeling Events Between Events Boundary Random Random Deterministic OUTPUT Rainfall Model CHECK Figure 4-4: Flow Diagram for Stochastic Rainfall Generator (Samuel 23) Key Storm Parameters The key storm parameters required for the model to work are: storm duration, interstorm duration and storm intensity and in for this thesis, an average decline in rainfall over the 1 year period. The parameters were randomly generated, and then distributed based on the following equations: Equation 4-6: Interstorm duration: Equation 4-7: Storm duration: ft b ft r 1 t = γ γ b ( t γ ) exp b 1 t = δ δ ( ) r t δ exp r Equation 4-8: Average Rainfall Intensity f I ( i t ) r = Γ λ κ ( ) ( i) κ λ 1 exp ( λ i) 31

39 Methodology The formulae are essential to generating rainfall distribution on an intra/interstorm basis during rainfall projections. The formulae are utilised during analysis of historical rainfall data, featuring the previous equations outlined (Equation 4-6 to Equation 4-8). Through applying the parameters in the rainfall projection viable variability in rainfall, consistent with historical data, over short time scales can be achieved while incorporating a decline in rainfall to incorporate climate change. Model Processes The model was initially calibrated to the Serpentine Catchment within Western Australia. Recalibrating the model to the other sites described would have required a significant amount of time due to the high variability of rainfall patterns. The relevant patterns include variation in volume rainfall, storm duration, interstorm duration and number of storms. Another method was required to scale the rainfall to other parts, outlined in section Scaling Data. The rainfall model was structured into two primary components: the main rainfall generator script and the ENSO script. Executing the ENSO script initially creates the necessary parameters for the distribution and magnitude of rainfall events to be included in rainfall projections for the next 1 years. The main script is then executed to produce the simulated rainfall as daily rainfall data for input into the recharge model to produce an estimate for the decline in recharge over the simulated rainfall period. 4.2 Calibration of Models Although the recharge model was established to produce an accurate estimate of recharge, calibration was required to ensure that the model produces a reasonably accurate depiction of a localized situation. The multiple wetting front model was previously calibrated to German soil and vegetation properties. Appropriate recalibration was required incorporating the addition of local regional soil and vegetation properties. As the rainfall model had already been calibrated to the Serpentine region, no modification was required at this stage. 32

40 Methodology Calibration of Recharge Model The diversity of rainfall patterns and soil properties is inherent in this region, in part due to the immense area of Western Australia. To incorporate the variability into recharge estimates, it was necessary to apply the model to specific locations. The model was calibrated over 8 different areas within the predefined region in the Southwest of Western Australia. The regions were selected on the basis of availability/regularity of rainfall data; and the calculated average rainfall to ensure sufficient variety of rainfall. All data was sourced from rainfall data collection by the Bureau of Meteorology (BOM 1998). A single pan evaporation constant of 19 mm/year was selected as an average for pan evaporation, for further details refer to section Figure 2-4 (BOM 23). The average pan evaporation values as stated in pan evaporation maps by the Bureau of Meteorology (23) can be used as the effect of vegetation cover is taken into account during the model simulation. Recalibration commenced methodically, with the alteration of a single soil or vegetation parameter for each trial run until the recharge estimated is similar to the results produced in the recharge/rainfall trend (Hatton et al 2). The initial parameter values were based on soils from four different locations as outlined in the table below: Location Wongan Hills Dandaragan Narrogin Beverley soil type yellow sand loam sand deep duplex shallow duplex N Theta s Theta r Table 4-1: Characteristics of Four different soils (sourced Smettem & Gregory 1996) The saturated hydraulic conductivity of soils was also assumed between appropriate parameters found in research by Clarke et al (2). The extreme values in saturated hydraulic conductivity for sands were assumed as 27 and 1 mm/day, while hydraulic conductivity for loam was assumed similar to conditions for tertiary mixed soils, assigning a value for saturated conductivity between 19 and 18 mm/day (Clarke et al 2). The soil parameters that were altered during calibration composed of the saturated conductivity, initial volumetric water content and included minute changes to the van Genuchten parameter during some calibrations. The other parameters were not altered. The regions selected for calibration are outlined on the next page in alphabetical order: 33

41 Methodology Figure 4-5: Map of Locations (sourced WALIS 23) Latitude Longitude Average Rainfall (mm) Soil Type Bridgetown (1) Duplex Soils Brunswick (2) Shallow Sands Cape Naturaliste (3) Massive Earths Denmark (4) Bleached Sands Dumbleyung (5) Duplex Soils Kuranda (6) Duplex Soils Manjimup (7) Massive Earths Narrogin (8) Duplex Soils Table 4-2: Calibration Site data Parameterisation of Rainfall Input A minimum rainfall period of 2 years was assumed and incorporated into the selection criteria, as this period of time is sufficiently long to assume a stable system. The range of average annual rainfall varied between 35 mm to 985 mm, providing sufficient variation in rainfall (refer to 34

42 Methodology Table 4-2). The rainfall data series was extracted directly from a publication of rainfall data by the Bureau of Meteorology (BOM 1998), evaporation data was taken from an annual evaporation figure from the southwest region on the average annual evaporation map in (BOM 23) and dividing the figure by the number of days in that year. The pan evaporation value was sourced from the following figure extracted from the map for Australian Pan Evaporation: Figure 4-6: Mean Annual Evaporation for the Southwest region The appropriate variables were stored in a temporary file which was then accessed and utilized by the main wetting front model script. The main wetting front script also required a parameter for average storm duration. This average duration drives the rainfall intensity, influencing the rate of infiltration. An average duration of 4 hours was taken off research pertaining to the average storm duration for Western Australia and was assumed as a uniform parameter over the Southwest region of Western Australia (Hipsey et al 23). Parameterisation of Soil and Vegetation Properties The soil types accounted for in the main wetting front recharge model were divided into three main categories: sand soil (Cf2), duplex soils (Cd2 + A2) and massive earths (Bb4). The soil types were derived from a soil map of Australia (DMNA 1978), with soil categories allocated to 35