The Effect of Climate Change and Variability on Streamflows in South-West Western Australia

Transcription

1 The Effect of Climate Change and Variability on Streamflows in South-West Western Australia Christina Young Honours Project School of Water Research Department of Environmental Engineering The University of Western Australia Supervised by Murugesu Sivapalan

2 November 2003 Abstract Over recent decades, the South-West of Western Australia has experienced exceptional and mostly unpredictable climatic shifts. In particular, drier than average winters for approximately the last 25 years have resulted in a decline in mean annual rainfall for the region. Currently, it is believed that this phenomenon is attributed primarily to the effect of natural climate variability and El Nino-Southern Oscillation (ENSO) related events. The effect of natural climate variability and a declining rainfall was captured using a stochastic model of rainfall time series which produced rainfall simulations over the next 100 years. Subsequently, the impact of the rainfall simulations on streamflow response was examined using the lumped conceptual hydrological model LASCAM. Analysis techniques were conducted to investigate the relationship between rainfall and runoff and determine the key controls governing runoff generation. The rainfall-runoff relationship is shown to be strongly non-linear and dominated by a threshold limit. Substantial decreases in runoff were observed in response to moderate decreases in simulated rainfall. The main controls governing runoff production were shown to include both meteorological and hydrological factors. Of particular importance are the extent of vegetation cover, antecedent soil moisture capacity, interception losses and evapotranspiration processes.

3 Acknowledgements I would like to convey my sincerest thanks and appreciation to my honours supervisor, Professor Murugesu Sivapalan for his continued support, encouragement and knowledge throughout the duration of the project. Without his guidance, this project could not have been possible. Deepest gratitude to Jos Samuel for his contribution to this study with his stochastic rainfall model, and for his time and effort in providing me with data and help. I would also like to express my gratitude to Dr Christian Zammit of the Water and Rivers Commission for his time and effort in supplying me with data and running the very complex LASCAM model, which was very much appreciated. A special thanks goes to my fellow final year environmental engineering class, particularly Dina Rahmah, Thaddeus Chew and Cameron Hanush, for making the experiences of this study enjoyable and for sharing their knowledge and providing support throughout the duration of this thesis. Finally, I would like to thank my family and friends for their understanding and patience which has been invaluable and for their inspiration which has sustained me throughout the year.

4 Table of Contents 1 INTRODUCTION BACKGROUND AND LITERATURE REVIEW HYDROLOGICAL PROCESSES Hydrologic cycle Streamflow in the Hydrologic Cycle Rainfall-Runoff Relationships Runoff Generation Mechanisms CLIMATE CHANGE AND CLIMATE VARIABILITY Defining Climate Change Measures of Climate Variability Climate Variability in the South-West region of Western Australia Relating Climate Change and Streamflow HYDROLOGICAL MODELS Modelling Concepts CATCHMENT DESCRIPTION Physical Environment Gauging Stations STOCHASTIC RAINFALL SIMULATION MODEL Model Description Model Inputs Model Process Model Outputs LARGE SCALE CATCHMENT MODEL MODEL DESCRIPTION LASCAM Inputs LASCAM Process LASCAM Output RAINFALL-RUNOFF ANALYSIS RAINFALL AND RUNOFF ANALYSIS TECHNIQUES... 38

5 7 RESULTS AND DISCUSSION PRECIPITATION Statistical Comparison Wet and Dry Cycles Seasonality and Distribution Distribution of Rainfall Events STREAMFLOW Statistical Comparison Seasonality and Distribution Intensity of Streamflow Events Interannual Variability Contribution of Surface and Subsurface Runoff to Streamflow Discharge Hydrograph Analysis COMPARING RAINFALL AND STREAMFLOW Hypotheses for Rainfall-Runoff Decline Difference Rainfall-Runoff Relationship RUNOFF COEFFICIENTS FLOW DURATION CURVES IMPLICATIONS OF CLIMATE CHANGE ON STREAMFLOW CONCLUSION RECOMMENDATIONS FOR FUTURE WORK REFERENCES... 77

6 Table of Figures FIGURE 2.1: THE HYDROLOGIC CYCLE... 3 FIGURE 2.2: PORTION OF THE HYDROLOGIC CYCLE RELATING TO STREAMFLOW... 4 FIGURE 2.3: TYPICAL STREAMFLOW HYDROGRAPH SHOWING THE RELATIONSHIP BETWEEN RAINFALL AND RUNOFF FIGURE 2.4: TYPES OF RUNOFF GENERATION MECHANISMS FIGURE 2.5: SATURATION EXCESS GENERATION DUE TO DPS AND RF FIGURE 2.6: THE SOUTHERN OSCILLATION INDEX DEPICTING EL NINO, LA NINA AND NEUTRAL EVENTS FIGURE 2.7: RELATIONSHIP BETWEEN SOI AND RAINFALL PATTERNS IN NORTHERN AND EASTERN AUSTRALIA FIGURE 2.8: ANNUAL RAINFALL IN PERTH FROM FIGURE 2.9: CSIRO MODEL RAINFALL FOR JUNE-AUGUST RAINFALL FOR MANJIMUP FIGURE 2.10: TOTAL ANNUAL INFLOW INTO PERTH DAMS (GL) FROM 1911 TO FIGURE 3.1: LOCATION OF THE SERPENTINE CATCHMENT FIGURE 3.2: RAINFALL PATTERN FOR THE SERPENTINE CATCHMENT FROM FIGURE 3.3: STATION LOCATION MAP OF THE SERPENTINE CATCHMENT FIGURE 4.1: FIGURATIVE ILLUSTRATION OF THE STOCHASTIC RAINFALL MODEL FIGURE 4.2: VARIOUS TIMESCALES INCORPORATED BY THE STOCHASTIC RAINFALL MODEL FIGURE 4.3: KEY STORM PARAMETERS FIGURE 4.4: (A) STATISTICAL COMPARISON OF SIMULATED RAINFALL FOR THE TEN REALIZATIONS AND (B) A COMPARISON OF REALIZATION 1 WITH REALIZATION AVERAGE FIGURE 5.1: SCHEMATIC DIAGRAM OF AN IDEALISED HILLSLOPE SHOWING THE THREE CONCEPTUAL MODEL STORES AND THE PRINCIPAL WATER FLUXES FIGURE 6.1: PLOT OF CUMULATIVE RAINFALL VS. CUMULATIVE DISCHARGE FOR BLOCK 1 REALIZATION FIGURE 6.2: CURVE FIT TO CUMULATIVE RAINFALL VS CUMULATIVE RUNOFF FOR BLOCK FIGURE 7.1: STATISTICAL COMPARISON FOR OBSERVED RAINFALL WITH REALIZATION FIGURE 7.2: SIMULATED ANNUAL RAINFALL FOR REALIZATION FIGURE 7.3: WET AND DRY CYCLES FOR REALIZATION 1, 5 AND FIGURE 7.4: MEAN MONTHLY COMPARISON FOR OBSERVED AND SIMULATED RAINFALL FOR REALIZATION FIGURE 7.5: MONTHLY RAINFALL PERCENTILES FOR REALIZATION FIGURE 7.6: ANNUAL SIMULATED STREAMFLOW FOR REALIZATION FIGURE 7.7: MEAN MONTHLY DISCHARGE FOR OBSERVED DATA AND REALIZATION FIGURE 7.8: MONTHLY STREAMFLOW PERCENTILES FOR REALIZATION FIGURE 7.9: INTERANNUAL VARIABILITY OF STREAMFLOW FOR REALIZATION

7 FIGURE 7.10: DISCHARGE HYDROGRAPHS FOR REALIZATION 1 FOR THE YEARS 2001, 2035, 2065 AND FIGURE 7.11: SIMULATED RAINFALL FOR REALIZATION 1 ( ) FIGURE 7.12: SIMULATED DISCHARGE FOR REALIZATION 1 ( ) FIGURE 7.13: RELATIONSHIP BETWEEN ANNUAL RAINFALL AND ANNUAL RUNOFF FIGURE 7.14: COMPARISON OF MEAN COEFFICIENT OF VARIATION FOR RAINFALL AND RUNOFF FOR REALIZATION FIGURE 7.15: RUNOFF COEFFICIENT FOR REALIZATION FIGURE 7.16: FLOW DURATION CURVES FOR OBSERVED DISCHARGE FIGURE 7.17: FLOW DURATION CURVES FOR REALIZATION

8 List of Tables TABLE 7-1: MONTHLY RAINFALL MEAN FOR OBSERVED RAINFALL AND FOR SELECTED DECADES FOR REALIZATION TABLE 7-2: DECADAL STATISTICAL COMPARISON OF OBSERVED DISCHARGE WITH REALIZATION TABLE 7-3: MEAN MONTHLY STREAMFLOW COMPARISON FOR OBSERVED AND SELECTED DECADES OF REALIZATION TABLE 7-4: SURFACE AND SUBSURFACE RUNOFF RATIOS TO TOTAL STREAMFLOW TABLE 7-5: COMPARISON OF RAINFALL AND RUNOFF OVER A 30 YEAR PERIOD... 61

9 Introduction 1 Introduction Over recent decades, the south-west of Western Australia has been experiencing a decline in mean annual rainfall, which has resulted in associated reductions in inflows into Perth s surface reservoirs. The dry trend currently observed in the climate of the south-west region has been mainly attributed to the influences of natural climate variability, and the effect of El Nino- Southern Oscillation events. Many studies have been conducted which have looked particularly into the impacts of climate change related to global warming on environmental and socioeconomic systems. The effects of a natural change in climate however, and its impacts upon streamflow in the south-west of Western Australia, has to date been cursory. The purpose of this research therefore, is to investigate how natural climate variability coupled with a declining rainfall will impact upon streamflow in the South-West region of Western Australia. This is achieved by examining the relationship between rainfall and streamflow and by providing further insights into the processes of runoff due to a change in climate. Consistent with the approach of other studies (Chiew et al. 1995; Viney and Sivapalan 1996), this study does not attempt to predict a future climate, or a future hydrological regime. Rather, as an indicative study, the focus was to look at the potential effects of a climate change scenario on catchment hydrology. The investigation was undertaken through hydrological modelling. The research undertaken for this thesis involved investigating the effect of a climate change scenario on the hydrology of the Serpentine catchment. A synthetic sequence of rainfall, generated by a stochastic rainfall model, was used to represent a future climate. The climate sequence provided the input for a conceptual hydrological model, which simulated the catchment s streamflow in response to the climate change scenario. Rainfall-runoff analysis techniques were then conducted to examine changes in the trends of simulated streamflow relative to current observations. The study of the impact of climate change upon streamflows is crucial to enable greater understanding of catchment processes, as well as having important implications for systems dependent upon streamflow behaviours. In particular, the future availability of surface water resources and traditional flood prediction analyses will be affected by changes in the magnitude 1

10 Introduction of streamflow. Therefore to conclude the study, the implications of the rainfall-runoff analyses were extended to consider the implications of climate change on surface water resources and future water availability for the South-West of Western Australia. 2

11 Background and Literature Review 2 Background and Literature Review 2.1 Hydrological Processes It is well known that because hydrology and water resources are inextricably linked to climate, climate change will have a lot of impacts on various components of hydrological systems and water resources aspects of the south-west region of Western Australia, including the characteristics of rivers, streams, groundwater and the catchment as a whole (Merritt and Schreider 2000; Ojo et al. 2000). It is therefore essential to have an understanding of the underlying nature of the hydrological cycle and its systems. As this thesis is concerned with the effect of climate change on streamflows, particular attention will be focused upon the runoff component of the hydrological cycle and the relationship between precipitation and runoff Hydrologic cycle The circulation and conservation of earth's water through different mediums of the earth is called the hydrologic cycle. An illustration of the hydrologic cycle can be seen in Figure 2.1. Figure 2.1: The Hydrologic Cycle The hydrologic cycle begins with the evaporation of water from the surface of the ocean. As moist air is lifted, it cools and water vapor condenses to form clouds. Moisture is transported 3

12 Background and Literature Review around the globe until it returns to the surface as precipitation. Once the water reaches the ground, one of two processes may occur; 1) some of the water may evaporate back into the atmosphere or 2) the water may penetrate the surface and become groundwater. Groundwater either seeps its way into the oceans, rivers, and streams, or is released back into the atmosphere through transpiration. The balance of water that remains on the earth's surface is runoff, which empties into lakes, rivers and streams and is carried back to the oceans, where the cycle begins again (Bramer 1997; Fetter 2001) Streamflow in the Hydrologic Cycle Figure 2.2 illustrates the portion of the hydrologic cycle that is pertinent to streamflow. The water present in streams is mainly a function of runoff, baseflow, throughflow, interflow and direct precipitation (Watson et al. 1993). Runoff (the surface water contribution to a stream) and baseflow (the groundwater discharge to the stream) are the main contributors to streamflow. To a lesser extent, streamflow is derived from throughflow, interflow and direct precipitation into streams (Watson et al. 1993). Figure 2.2: Portion of the Hydrologic Cycle relating to Streamflow There are four main processes that occur when precipitation reaches the surface. These are interception, depression storage, infiltration and overland flow. Overland flow and infiltration are 4

13 Background and Literature Review processes that are primarily responsible for stimulating runoff generation processes and will be described in section Interception Interception occurs when precipitation is stopped, intercepted or temporarily held by any form of vegetative canopy or vegetation residue. Interception affects the amount, timing and spatial distribution of water added to a catchment, for example, by reducing the eventual amount of rain reaching the ground, or by reducing the intensity of the rainfall event (Fetter 2001). Therefore, in highly forested areas the amount of rainfall reaching the ground and being made available to contribute to runoff may be lessened. In tropical forests, over 30% of the annual precipitation may be lost via canopy interception. Interception losses are less in more arid or semiarid environments with more sparse vegetation. Precipitation caught by interception may partly be evaporated directly back into the atmosphere or drain into the soil surface (Brooks et al. 1992). Depression Storage When precipitation falls on the land surface it can be held in depression storage. Depression storage refers to the accumulation of puddles of water at the land surface (Fetter 2001; Berti 2002). These puddles can occur across the landscape, producing scattered bodies of water that may reduce the total amount of streamflow occurring through a catchment due to their unconnected nature. The water held in depression storage ultimately evaporates or infiltrates into the soil (Hornberger et al. 1998) Rainfall-Runoff Relationships Runoff, which can also be referred to as streamflow, stream, river discharge or catchment yield, comprises the gravity movement of water over the surface of the earth. In general, this water represents the excess of rainfall over evapotranspiration, when allowances are made for storage on and under the ground surface. However, whereas rainfall on the land is irregular in space, time and volume, runoff from the land surface is comparatively constant (Ward 1990). This contrast between runoff and the rainfall from which it is directly or indirectly derived, results mainly from the great storage capacity of the surface layers of the earth, through which excess rainfall is detained and gradually released into streams. Therefore, the relationship between rainfall and runoff is not linear. The relationship that precipitation has with the surface 5

14 Background and Literature Review upon which it falls, will therefore affect the character, amount and distribution of runoff (Manning 1987). A number of studies (Kohler and Linsley 1951) have attempted to develop rainfall-runoff relationships that could apply to any region or catchment under any set of conditions. However, these methods must be used with great caution because of the variable factors that affect the calculation of runoff from a known volume of rainfall (Bedient and Huber 1992). One of the simplest rainfall-runoff formulas is called the rational method, which predicts the amount of peak flow (Q) as a function of rainfall intensity and duration, area of the catchment and a runoff coefficient. Q = CIA Equation 2-1 where C = runoff coefficient I = intensity of rainfall of chosen frequency A = area of catchment The assumption behind this method is that a steady, uniform rainfall rate will produce maximum runoff when all parts of a catchment are contributing to outflow (Bedient and Huber 1992) Factors Affecting Runoff Factors affecting runoff can be classified into factors that affect the total volume of runoff and factors that combine to influence the distribution of runoff. Factors Affecting the Total Volume of Runoff Factors affecting the total volume of runoff will significantly affect the areal distribution of runoff and occurs over a period of several years. These include climatic and catchment factors. Climatic factors are the most effective influence on the total volume of runoff (Q) in the longterm water balance, in the form of water gained by a catchment from precipitation (P), and lost through evapotranspiration (E) (Q = P-E). In this case, the climate of a catchment area sets the upper limits to the total volume of streamflow leaving the area. However, this link between annual totals and the means of rainfall and evapotranspiration may be modified by short-term factors, such as sudden changes in vegetation cover as a result of changes in land use (Ward 1990). 6

15 Background and Literature Review Short-term variations and seasonal distribution of rainfall will also affect the volume of runoff within a catchment. For example, an amount of rain falling in the form of small periodic episodes will contribute less to streamflow than the same amount of rain falling as one continuous storm (Manning 1987). It is also argued by Ward (1990) that vegetation cover of the catchment area is likely to affect the balance between rainfall and evapotranspiration losses, especially if rapid changes in vegetation density occur. However, there is evidence which conflicts with this viewpoint, as it is believed that even a complete change from a forested to a sparsely vegetated condition will only alter the timing and not the total amount of runoff. Catchment properties, particularly the catchment area (all other factors being constant) will determine the total amount of rainfall that is captured. However, this is dependent upon the climate regime of the area. For example, if the climate of a catchment has potential evaporation that is greater than rainfall, a larger catchment is just as likely to have low runoff as a smaller catchment (Ward 1990). Factors Affecting the Distribution of Runoff in Time Factors affecting the distribution of runoff in time tend to be more temporally variable over a short period of time. These factors include meteorological, catchment and anthropogenic factors. Meteorological Factors The variation of meteorological factors with time tends to be closely related to variations in runoff. Meteorological factors include rainfall intensity, duration and distribution. Rainfall intensity is one of the most important factors in determining the proportions of rainfall that will result in either surface or groundwater runoff. Therefore, heavy rain which falls in excess to the infiltration capacity of the soil (see section ) will contribute largely to surface runoff, and therefore reach water bodies faster. In contrast, rain falling at lower intensities will be largely absorbed by the soil. While this may eventually reach groundwater, its contribution to streamflow will be greatly delayed (Manning 1987). 7

16 Background and Literature Review Rainfall duration becomes a significant factor when related to the mean travel time from when rain falls onto the catchment to its exit from the catchment as streamflow. If rainfall duration is greater than the mean travel time, then the whole catchment is likely to contribute to runoff during the latter stages of the storm. Potential runoff is therefore at a maximum. If rainfall duration is less than the mean travel time, then potential runoff will be lower than the maximum as only part of the catchment will be contributing to runoff before the rainfall ends (Ward 1990). Rainfall duration is also important in terms of the infiltration capacity of runoff. A longer period of rainfall will lead to a lower infiltration capacity and therefore, a greater volume of surface runoff is produced. The temporal relationship between rainfall and runoff may also be greatly influenced by the distribution of rainfall over the catchment area. For a given volume of rainfall which is uniformly distributed over the whole catchment, lower intensities is experienced and therefore there is less chance of producing surface runoff compared to the same volume of rainfall onto a small, localized section of the catchment (Bedient and Huber 1992). Uniform distribution of rainfall will tend to result in an increase in groundwater runoff, and consequently a long-term increase in streamflow, whereas variable distribution of rainfall will tend to give larger volumes of surface runoff and therefore more sudden but shorter increases in streamflow (Bedient and Huber 1992). Although the intensity, duration and distribution of rainfall can be reasonably measured, assessments of catchment properties are less precise. In addition, some catchment factors (such as topography, shape and soil properties) will remain fairly constant over a long time, while others such as those impacted by land use, may change drastically and therefore strongly affect runoff (Bedient and Huber 1992). Catchment Factors Topography refers to the general shape of the catchment, and the steepness of the slopes which will affect the way precipitation reaches the streams. The shape of the catchment area is known to influence runoff through its effects on flood intensities and on the mean travel time of water from when rain falls into the catchment to the time it exits in the form of streamflow. Long, narrow catchment shapes tend to have shorter stream networks and are more likely to join the main 8

17 Background and Literature Review stream at intervals along its length (Ward 1990). This implies that after a heavy rainfall over the area, the runoff peaks of streams situated in the lower catchment will have left the catchment before those of the streams in the upper catchment have moved very far down the main stream. This means that elongated catchments are less subject to high runoff peaks. This is in contrast to a uniformly shaped catchment (such as a circular catchment) where stream networks tend to congregate near the centre of the catchment area (Manning 1987). Consequently, runoff peaks are likely to reach the main stream in approximately the same region at about the same time, resulting in a large and rapid increase in the discharge of the main stream. The slope of the catchment area will mainly tend to affect the amount of infiltration which is able to take place, and the rate at which water moves over the surface toward stream channels. Steeper slopes imply that there is a greater chance that water will move off the surface before it has time to infiltrate, so that surface runoff is large. Furthermore, movement of water on steep slopes are more rapid, so that runoff reaches streams more quickly resulting in a rapid increase in streamflow (Ward 1990). Soil type of a catchment will have an influence on infiltration capacity. Sandy soils will tend to have higher infiltration capacities than fine grained, closely compacted soils such as clay, and therefore will result in less surface runoff. The moisture content of the soil may also have an influence. When the soil moisture content is high, infiltration capacity is low so that surface runoff will tend to occur (Fetter 2001). The vegetation cover of a catchment area may affect runoff in a number of ways and compared to other catchment properties, its importance is considerable. In water limited areas, more evapotranspiration occurs from deeper-rooted species such as trees, than shallower-rooted vegetation. Runoff will therefore be reduced in a forested catchment than in a grass-covered catchment (Manning 1987). However, the most important effect of vegetation cover is its ability to slow the movement of water over the surface after a rain event, therefore allowing more time for infiltration to take place. In this way, the timing of runoff after rainfall may be considerably modified and peak stream flows will tend to be much lower, although more prolonged. Vegetation has effectively 9

18 Background and Literature Review resulted in a greater proportion of streamflow being contributed by groundwater runoff than surface runoff (Manning 1987). Anthropogenic Factors In many areas of the world, runoff is been increasingly affected to some extent by the influence of human activity. Construction of water supplies, hydro-electric power generators, clearing for agriculture, re-forestation, and increasing urbanization and development are just some of the activities that have steadily impacted upon the response of catchment areas to rainfall (Bedient and Huber 1992). Consequently, the pattern and distribution of runoff has been changed. Whether runoff is increased or decreased depends upon the type of human activity. For example, urbanization tends to result in an increase in impermeable surfaces such as roads and pavements. Over large areas, infiltration capacity is greatly reduced. Falling precipitation is captured by drains and other artificial structures, which serve to direct runoff into streams as rapidly as possible. The result is a rapid increase in surface runoff below large urban areas (Ward 1990) Runoff Generation Mechanisms As mentioned previously, various processes and pathways determine how excess water becomes streamflow. Excess water represents the portion of total precipitation that runs off the land surface plus that which drains from the soil and is lost through neither evapotranspiration nor leaks to deep groundwater (Brooks et al. 1992). This water eventually reaches streams and causes discharge to increase, that is, storm flow (Dunne & Leopold 1978). Figure 2.3 shows a typical streamflow hydrograph. It illustrates that there is a lag time between the occurrence of a storm and subsequent peak discharge into a water body. A perennial stream is most likely fed by groundwater (baseflow component) as shown in Figure 2.3, which sustains streamflow between periods of precipitation. Because of the long and complicated pathway involved, baseflow does not respond quickly to moisture or rainfall input (Brooks et al. 1992). 10

19 Background and Literature Review Figure 2.3: Typical Streamflow Hydrograph showing the relationship between rainfall and runoff Runoff into a receiving water body after a storm event can be caused by two broad runoff mechanisms, surface flow and groundwater flow. This is illustrated in Figure 2.4, where overland flow is seen as 1 and 4, and groundwater flow is considered to be 2 and 3. The water that flows over the surface reaches the stream channel in a shorter period of time than that flowing through the soil. The peak flow measured in the receiving water body is determined greatly by the manner in which rainfall is distributed into these two broad runoff mechanisms (Chow et al. 1988). 1. Horton Overland Flow 2. Groundwater (Baseflow) 3. Subsurface storm flow 4. Saturation overland flow Figure 2.4: Types of Runoff Generation Mechanisms 11

20 Background and Literature Review Surface Runoff Mechanisms Surface runoff, also referred to as overland flow, was described in Horton (1933) as that part of the rainfall which is not absorbed by the soil by infiltration. There are two mechanisms that generate surface runoff: i) Infiltration Excess and ii) Saturation Excess Infiltration Excess Runoff Infiltration excess runoff (IER) or Hortonian overland flow, is the excess precipitation that is not infiltrated by the soil as a result of the rainfall intensity (p) being greater then the infiltration capacity (f). It was considered by Horton (1933) that surface runoff is equivalent to the rainfall intensity minus the soil infiltration capacity (p-f). The soil infiltration capacity is the maximum rate at which the soil absorbs the rainfall. This runoff mechanism is dependent upon hydrological characteristics such as antecedent moisture content, soil properties, hydraulic conductivity, porosity, and vegetation cover, as well as meteorological characteristics such as rainfall intensity, duration, and inter-storm period (Xie et al. 2003). The most important determinant in the generation of IER is storm intensity. Storm intensity must be adequately high in order to generate overland flow. As previously mentioned, interception by vegetation reduces the intensity of rainfall. This combined with the spatial coverage of the storm itself combines to determine the area of the catchment over which IER is generated. Horton (1933) assumes that during a given rain event, the infiltration capacity will decrease exponentially. This decrease in infiltration capacity is attributed to spatial variability of soil properties. These structural changes include compaction of soils due to raindrops and the filling of pores by fine particles, which together form a surface seal on the soil limiting infiltration (Kirby 1988). An underlying assumption in the application of this mechanism is that IER occurs simultaneously across the entire catchment. Many studies (Bonell and Williams 1986; Merz and Bardossy 1998, Xie et al. 2003) have showed the existence of localised surface runoff, indicating the spatial and temporal heterogeneities inherent in runoff generation. The estimation of the area of the 12

21 Background and Literature Review catchment that contributes to streamflow is critical in the calculation of the quantity of surface runoff produced and is a large source of uncertainty in modeling. Horton overland flow is not common and does not occur everywhere. This conceptual model for surface runoff generation is most useful for urban runoff conditions as well as in natural environments where the infiltration capacity is low or close to zero, such as arid and semi arid regions (Chow et al. 1988). Saturation Excess Runoff Saturation excess runoff (SER) is generated when precipitation falls upon a saturated portion of the catchment. The soil moisture capacity is filled as a result of the water table reaching the surface, and no rainfall can infiltrate. This results in excess rainfall which generates overland flow (DPS). In many cases, return flow (RF) can occur where subsurface water returns to the surface. This process is illustrated in Figure 2.5. SER is dependent upon aquifer characteristics, the antecedent conditions of the catchment, and the duration of the storm event (Chow et al. 1988). RF = Return Flow DPS = Direct Precipitation onto saturated areas Figure 2.5: Saturation Excess generation due to DPS and RF SER tends to be generated at the bottom of valleys where the soil profile is completely saturated due to discharge out of a hillslope and into a water body (Brooks et al. 1992). A strong seasonal component of saturated area can be witnessed with saturated areas peaking towards the end of the winter and contracting during the summer months (Chow et al. 1988) Subsurface Runoff Generation If the major proportion of precipitation of streamflow originates from subsurface flow, the manner in which this subsurface flow mechanism transports and discharges water into the receiving body is of great importance to the peak of the stream flow (Davie 2003). 13

22 Background and Literature Review A large proportion of flow that enters a stream in a forested catchment is not overland flow but subsurface flow. Groundwater flow is the water that infiltrates through the soil profile until it recharges the underlying aquifer. This groundwater may eventually discharge into a receiving body of water or in some cases outside catchment boundaries (Wittenberg and Sivapalan 1999). This form of runoff is considered delayed due to the slow transmission of water through the soil matrix. This discharge often occurs well after the rain event has ceased (Brooks et al. 1992). The subsurface contribution can be divided into a perched aquifer component and a deep groundwater component. The perched aquifer is highly seasonal, and when present, contributes the majority of flow from groundwater to the receiving stream. The formation of this aquifer occurs during winter and through summer the aquifer depletes as a result of continual discharge and evapotranspiration. In contrast the deep groundwater component is relatively constant with relatively little variation (Davie 2003). 2.2 Climate Change and Climate Variability Recently, and especially over the last three to four decades, the issue of climate change and variability has been a subject of scientific discussions and public debate. Not only are the processes of climate complex and not well understood, increasing concern has been raised over the impacts of climate change on environmental dynamics, and subsequent implications on socioeconomic activities. Various literatures provide different definitions of climate change and this is discussed in section Measures of climate variability and its inherent processes are presented in section The impact of climate change in the south-west of Western Australia and its relationship and effects upon streamflow are reviewed in subsequent sections Defining Climate Change Like with many other disciplines, there are difficulties inherent in arriving at generally acceptable definitions of climate, climate variability and climate change. Variations in definitions exist which is dependent upon the context. One simple definition of climate which is sufficient for our purposes is that climate is: A synthesis of weather conditions in a given area, characterized by long-term statistics (mean values, variances, probabilities of extreme values, etc.) of the meteorological 14

23 Background and Literature Review elements (such as temperature, pressure, humidity, winds) in that area (Arctic Climatology and Meteorology 2003). Pittock et al. (1978) and the Bureau of Meteorology (2003) provide more succinct definitions of climate. Definitions of climate change (Pittock et al. 1978; Bureau of Meteorology 2003) illustrate the many meanings implied by the term. In the most general sense, the term climate change encompasses all forms of climatic inconstancy (that is, any differences between long-term statistics of the meteorological elements calculated for different periods but relating to the same area) regardless of their statistical nature or physical causes. Climate change may result from such factors as changes in solar activity, long-period changes in the Earth's orbital elements, natural internal processes of the climate system, or anthropogenic forcing (for example, increasing atmospheric concentrations of carbon dioxide and other greenhouse gases) (Arctic Climatology and Meteorology 2003). In a more restricted sense, the term climate change denotes a significant change (such as a change having important economic, environmental and social effects) in the mean values of a meteorological element (in particular temperature or amount of precipitation) in the course of a certain period of time, where the means are taken over periods of the order of a decade or longer (Arctic Climatology and Meteorology 2003). Hence, inherent in the definition is that climate change may occur as a result of anthropogenic change or natural variability in climate. Subsequently, for the purposes of this study, the term climate change will now imply change due to natural variability in climate. Generally, the term natural climate variability denotes the inherent characteristic of climate which manifests itself in changes of climate with time. The degree of climate variability can be described by the differences between long-term statistics of meteorological elements (such as rainfall) calculated for different periods (Arctic Climatology and Meteorology 2003). Climate variability is also often used to denote deviations of climate statistics over a given period of time (such as a specific month, season or year) from the long-term climate statistics relating to the corresponding calendar period. In this sense, climate variability is measured by those deviations, which are usually termed anomalies. 15

24 Background and Literature Review Measures of Climate Variability Many studies have been conducted which have considered the relationship between Australian rainfall and the Southern Oscillation (Drosdowsky and Williams 1991; Nicholls and Kariko 1993; Clewett et al. 2000). In particular, these studies have established a strong correlation between natural climate variability due to El Nino-Southern Oscillation (ENSO) events and its impacts upon streamflow in Australia. As a result, this section will briefly examine the mechanisms of ENSO and the Southern Oscillation Index (SOI), and its significance for Australian rainfall El Nino-Southern Oscillation (ENSO) Climate has been shown to fluctuate on seasonal, annual, decadal and much longer time scales. In some years, there are more extreme meteorological events such as drought and floods. It has been suggested that some of these climatic extremes might have had a common geographical origin changes in sea surface temperatures in the tropical Pacific Ocean (El Nino) and changes in atmospheric pressure at sea level across the Pacific basin (the Southern Oscillation). These combined changes have come to be commonly referred to as ENSO events (Glantz 2001). The ocean-atmospheric system tends to vary between two extremes, either El Nino or La Nina states. This cycle between El Nino and La Nina is termed the ENSO cycle and occurs aperiodically, between every 2 to 7 years on average (Glantz 2001). It should be noted that individual El Nino or La Nina events are never exactly the same and may vary with respect to magnitude, duration and spatial extent (Allan et al. 1996) Southern Oscillation Index As previously mentioned, the Southern Oscillation is a major air pressure shift between the Asian and east Pacific regions whose best-known extremes are El Nino events. The strength and direction of ENSO is measured by a simple index, the Southern Oscillation Index (SOI). The SOI as seen in Figure 2.6 is calculated from the monthly or seasonal fluctuations in the air pressure difference between Tahiti and Darwin. A low SOI, typically with an index of less than -5 (indices below the red line in Figure 2.6), implies a small pressure difference, and is associated with an El Nino event. A high SOI (greater than 5 above the red line in Figure 2.6) is associated with a La Nina event. Pressure differences occurring between -5 and 5 (between the red lines in Figure 2.6) 16

25 Background and Literature Review describe neutral or normal conditions observed over the Pacific Ocean, that is, the SOI is close to the long-term average state (Bureau of Meteorology 2003). Figure 2.6: The Southern Oscillation Index depicting El Nino, La Nina and Neutral events During El Nino events, warming of tropical regions of the Pacific and Indian Ocean leads to the displacement of major rainfall-producing systems from the continents, which causes massive redistributions of climatic regimes. In Australia, there appears to be an association between El Nino events and a fall in rainfall, leading to an increased likelihood of droughts, particularly in the north-eastern parts of Australia (Glantz 2001). This can be seen in Figure 2.7, which shows that a strongly negative SOI value coincides with an El Nino event. This is associated with a dry year of low winter and spring rainfall and the occurrence of a late monsoon. The end of the El Nino is associated with a break in the drought, followed by heavy rain events (Bureau of Meteorology 2003). Figure 2.7: Relationship between SOI and rainfall patterns in northern and eastern Australia 17

26 Background and Literature Review Climate Variability in the South-West region of Western Australia Over recent decades, the south-west region of Western Australia has experienced exceptional and mostly unpredictable climatic shifts. In particular, drier than average winters over the last 30 years have resulted in a mean decline in annual rainfall of about 10%, and a fall in winter rainfall of approximately 15-20% (IOCI 2002). This fall in mean annual rainfall can be seen in Figure 2.8. These have placed water systems under stress and have profoundly altered the basis for future planning of climate affected industries. The Indian Ocean Climate Initiative (IOCI) was established by the Western Australian state government in 1998 as a result. Its aims are to improve the understanding of climate variability and climate change in order to support informed and effective climate adaptation (IOCI 2002). Currently, the IOCI is unable to conclusively identify the cause of the observed rainfall decline in the south-west. As Figure 2.8 shows, speculations exist over the future behavior of rainfall over the next few decades. That is, will annual rainfall begin to increase, remain constant or continue to decrease? Figure 2.8: Annual Rainfall in Perth from The IOCI (2002) believes that the cause of the observed rainfall decline may be a result of either the enhanced greenhouse effect or natural climate variability. The most credible explanation is 18

27 Background and Literature Review that the rainfall decline is a result of a combination of these two factors. A summary of IOCI investigations have revealed that: The observed run of dry years in south-west Western Australia over the last three decades is unusual in a historical and global context. This trend is discernible through declines in both the number of rain days and the amount of rain during extreme events Climate model simulations run over 1000 years suggest that natural climate variability can give rise to decadal or even longer dry spells, and that they may occur without any obvious external factors While natural climate variability may be the primary cause for the run of dry years, factors relating to human development and global warming have not been eliminated Rainfall projections are difficult to approximate. Most models however, suggest a decline of autumn and winter rainfall in southern Western Australia over the next few decades The ENSO phenomenon has a considerable effect on the climate of south-west Western Australia Although there is a relationship between SOI values and interannual rainfall in south-west Western Australia, ENSO can only partly account for the decrease in rainfall seen in the south-west of W.A. There is a need for research into the various factors that could account for the rainfall decrease (natural variability, greenhouse effect, ozone depletion) Figure 2.9: CSIRO model rainfall for June-August rainfall for Manjimup 19

28 Background and Literature Review. Rainfall models developed by the CSIRO predict that rainfall in future decades will continue to decline for the south-west region of Western Australia. This is illustrated by Figure 2.9, which predicts that winter rainfall in Manjimup will exhibit a steady decrease over the next 50 years (Centre for Water Research 2002). These observed changes in rainfall have resulted in an even sharper fall in observed streamflow in the south-west, which is illustrated in Figure The figure demonstrates that total annual inflows into Perth dams have been estimated to have decreased by about 50% since 1975, and since 1997, this figure has dropped to a further 62% decline in mean inflows from 1975 (Water Corporation 2003). Figure 2.10: Total Annual Inflow into Perth Dams (GL) from 1911 to 2002 The implications of natural climate variability upon rainfall and runoff in the south-west of Western Australia are therefore considerable, especially with respect to the reliability of current and future availability of Perth s water resources, and the associated economic, environmental and social impacts Relating Climate Change and Streamflow There have been numerous studies conducted which have examined the impact of a potential climate change due to global warming and its implications upon runoff and water resources in 20

29 Background and Literature Review Australia, particularly in eastern Australia (Stewart 2000; Evans et al. 2000). Fewer studies have focused upon the impact of climate variability (and the correlation between ENSO and rainfall patterns) upon runoff and water resources, particularly in the south-west of Western Australia, with one notable exception by Ruprecht and Rodgers (1999) which examined the impact of observed climate variability on the surface water resources of the south-west region. The study suggests that although there is clear evidence, from analysis of historical annual flow records, that there has been a change in median annual flow volumes for some rivers in the south-west, there is no evidence to suggest that these changes are linked to greenhouse induced climate change. Studies of the potential implications of climate change for many river flow regimes, in different environments have shown that the effects of climate change on river flows depend not only on the extent of change in climatic inputs to the catchment, but also on the characteristics of the catchment itself. However, virtually all of these catchment studies have assessed the effects of a change in mean climate (Hulme et al. 1999). Whilst it is possible from such studies to determine potential changes in the frequency with which extremes are exceeded, changes in the variability in climate from day-to-day and year-to year will also influence changes in the frequency of extremes. Also, human-induced climate change will be superimposed on the effects of natural multidecadal climatic variability due to patterns and rhythms within the climate system (Hulme et al. 1999), and over the next few decades the effect of this variability may be greater than that of climate change. 2.3 Hydrological Models Hydrologic information is needed for catchment management planning and other types of analyses. Sometimes, there may not be adequate hydrologic, meteorological and geomorphological data available at locations of interest, and even in the case where there is, it can be difficult to decide the most appropriate method to use. Hydrologists must apply the appropriate tool or method for a given situation. This section therefore presents an overview of various hydrologic methods, types and applications (Clarke 1994). The hydrological system (Figure 2.1) is very sensitive to changes in climate (IPCC 1995). Changes in precipitation affect the magnitude and timing of runoff. The flux of water through 21

30 Background and Literature Review each pathway in the hydrological cycle is a function of the coupled relationship that exists with each of the other flux paths and the catchment s moisture storages. The catchment s hydrological cycle can only be effectively simulated using models that operate at spatial and temporal scales appropriate to those that exist in practice (Klemes 1985). The complex and coupled nature of the hydrological cycle means that unless all of the key processes that represent the storage or flow of water are included, then a model will not capture the full impact of a climate change on the hydrology of a catchment. For example, numerous studies (Bates et al. 1994; Chiew et al. 1996) have shown that catchment response to changes in rainfall depends upon hydrological and meteorological properties such as antecedent soil moisture conditions and the timing of rainfall events. It has also been shown that dry catchments are far more sensitive to climate changes than humid catchments. The hydrological characteristics of individual catchments can only be interpreted by models that are suitable for local catchment and climate conditions (Klemes 1985). As hydrological modelling can simulate the relevant hydrological processes at the appropriate scales, hydrological models have become the accepted approach for examining the impact of climate change on catchment hydrology Modelling Concepts Hydrologic models are simplified representations of actual hydrologic systems that allow the study of the functions and responses of catchments to various inputs and therefore to enable a better understanding of hydrologic events. For the most part, hydrologic models are based on the systems approach, and differ in terms of how and to what extent each component of the hydrologic process is considered (Beven 2001). There are numerous types of hydrological models available, and many different ways of classifying hydrological models (Wheater et al. 1993; Clarke 1994; Singh 1995). Simply however, hydrological models can be generally classified as using either a lumped or distributed modelling approach. Lumped models treat the catchment as a single unit, with state variables that represent averages over the catchment area, such as average storage in the saturated zone. Distributed models make predictions that are distributed in space, with state variables that represent local averages of 22

31 Background and Literature Review storage, flow depths or hydraulic potential, by discretizing the catchment into a large number of elements or grid squares and solving the equations for the state variables associated with every element grid square (Clarke 1994). A second consideration is whether to use a deterministic or stochastic model. Deterministic models permit only one outcome from a simulation with one set of inputs and parameter values. Stochastic models allow for some randomness or uncertainty in the possible outcomes due to uncertainty in input variables, boundary conditions or model parameters (Clarke 1994). The vast majority of models used in rainfall-runoff modelling are used in a deterministic way, although the distinction is not so clear-cut, since there are examples of models which add a stochastic error model to the deterministic predictions of the hydrological model. Finally, hydrological models can be empirical, conceptual or physically based. Empirical models generate statistical relations between the runoff, precipitation, temperature and actual or potential evaporation data in the observed record. Empirical models usually treat the catchment as a lumped system, and operate on larger spatial and temporal scales than other model types (Sivapalan et al. 1996). A limitation of the empirical approach is that because there is no physical basis to the model structure and algorithms, the parameters of the statistical relationship are only appropriate to the conditions under which the model is calibrated. Conceptual models refer to models that represent the physical processes in the hydrological cycle by equations governing flow rates that are loosely based upon the physics of the particular flow paths concerned. Conceptual models employ a range of arbitrary structures and mass balance equations to represent the flux paths and moisture stores in the hydrological cycle. The importance of any flux path or moisture storage depends on the parameter values in the relevant equations. As each equation is hypothetical, model parameters cannot be measured in the field. Therefore parameter values are determined by calibrating the model s simulated response to observed data. A major consideration in using conceptual models is the calibration process, and the number of parameters requiring calibration (Beven 2001). 23

32 Background and Literature Review Physically based models operate at the finest spatial and temporal scales of all model categories. Physically based models attempt to simulate the exact hydrological processes that define the passage of water through a catchment, and so they are required to operate at a higher resolution than empirical or conceptual models. Because the equations of the model are derived from the physics of the process, little calibration should be required. In theory, the majority of the physical parameters can be measured in the field (Beven 2001). There are a number of limitations associated with fully distributed physically based models. Due to the highly non-linear nature of most of the governing equations, and the considerable variability of catchment physiographic parameters, these models tend to be very complex and computationally expensive. Therefore, limitations of computing requirements and parameter measurement have meant that the application of physically based models for anything but small research catchments is largely impractical and unnecessary (Clarke 1994). 24

33 Catchment Description 3 Catchment Description As mentioned previously, this thesis is an indicative study to provide further insights into the role of climate change impacts on the key controls governing streamflow. As a result, the Serpentine catchment, situated in the south-west region of Western Australia, was selected as the study site. Although details of the Serpentine catchment such as rainfall, streamflow and catchment properties were used, this study is not meant to model the effect of climate change on the Serpentine catchment per se Physical Environment The Serpentine catchment is located on the Darling Scarp, approximately 60km south-east of Perth (Figure 3.1). The area of the Serpentine catchment which flows into the main Serpentine dam is 664 km 2 (Water Authority of Western Australia 1994). Figure 3.1: Location of the Serpentine Catchment The climate of the Darling Range is described as Mediterranean, with dry hot summers and wet, cool winters usually occurring from May to September (Water and Rivers Commission 1997). Average annual rainfall in the area ranges from 800mm to 1200mm (Water and Rivers Commission 2003) and average potential evapotranspiration is approximately 1700mm. 25

34 Catchment Description Observed rainfall data for the Serpentine catchment has been shown to exhibit a general downward trend over the last decade (Figure 3.2), which appears to be consistent with the declining trend seen in the south-west region. Rainfall Pattern for Serpentine Catchment Rainfall (mm) Year Figure 3.2: Rainfall pattern for the Serpentine Catchment from The natural vegetation in the catchment is classified as open forest, which are dominated by jarrah (Eucalyptus marginate) and marri (Eucalyptus calophylla). Land use is primarily state forest reserve and mining tenement. There is no permanent clearing within the catchment, although there are small areas of open cut bauxite mining, which may have the potential to have a major impact on the hydrology of the catchment (Water Authority of Western Australia 1994). Much of the Serpentine River has been modified including a dam in the upper reaches and rivertraining on the middle reaches to enable high volume flows. A large drainage network has also been constructed on the coastal plain to rapidly drain water from rural and semi-rural land adjoining the Serpentine River (Water and Rivers Commission 2003) Gauging Stations The catchment has a good network of meteorological stations, with several pluviograph recording rainfall stations available within the catchment. Three main streamflow gauging stations currently operate upstream of the reservoir. For the purposes of this study, streamflow gauging station S at the outlet of the Bigbrook catchment was used as the main basis for calibration and streamflow simulation. Data from the rainfall gauging station M509569, which is situated at Cameron West in the Bigbrook tributary, was used as calibration for the stochastic rainfall model. 26

35 Catchment Description Figure 3.3 shows the location of the stations within the catchment, with S station highlighted in red. Figure 3.3: Station Location Map of the Serpentine Catchment 27

36 4 Stochastic Rainfall Simulation Model Stochastic Rainfall Simulation Model Model Description A stochastic rainfall model was employed to simulate climate change. The model was developed to capture multi-annual and interannual variability inherent in rainfall as well as incorporating a steady decline in precipitation. The model was an attempt to mimic the main climate change processes observed in the south-west region of Western Australia, that is, rainfall variability as a result of ENSO cycle variations and the declining rainfall discerned in the last 30 years. A figurative illustration of the model can be seen in Figure 4.1. It should be noted that the rainfall model is an attempt to produce one scenario of climate change. INPUT Rainfall Data -hourly rainfall data -daily rainfall data -Sea surface temperature -Mean Sea Level Pressure PROCESS Modeling Events Between Events Seasonality Inter-annual Inter-decadal Boundary Random Random Deterministic Random Deterministic OUTPUT Rainfall Model CHECK Figure 4.1: Figurative Illustration of the Stochastic Rainfall Model The stochastic model is an adaptation of one used previously in Sivandran (2002) and was further modified by Jos Samuel, a PhD student at the Centre for Water Research. Previous stochastic rainfall modelling (Robinson and Sivapalan 1997; Sivapalan et al. 2002) incorporated the characteristics of storm events (including rainfall duration, inter-storm period and rainfall intensity), within-storm patterns and seasonal variability. The modified stochastic rainfall model 28

37 Stochastic Rainfall Simulation Model used for this study has in addition, integrated interannual and interdecadal variability. An overview of the time-scales featured in the stochastic rainfall model can be seen in Figure 4.2. Figure 4.2: Various timescales incorporated by the Stochastic Rainfall Model There are two essential qualities exhibited by the rainfall model a stochastic nature and time series characteristics. In this case, the stochastic processes in the model consider chronological sequences of climatic events with the aim of attempting to explain the irregularities of occurrence and in particular, of forecasting the incidence of outstanding important extremes (Shaw 1994). The measurements or numerical values of any variables that changes with time constitute a time series Model Inputs The model is adapted for conditions in the Serpentine catchment, and the parameters of the model are calibrated to mimic the observed hourly and daily rainfall records for the Cameron West rain gauge station in the Big Brook Tributary of the Serpentine catchment. Other inputs into the model are details of sea surface temperature obtained from the Interdecadal Pacific Ocean (IPO) index to delineate information about interdecadal variability. Details about the mean sea level pressure were also acquired from the SOI index to obtain information about the interannual variability of rainfall events as a result of ENSO. In other words, inputs from the IPO and SOI are used to calibrate the rainfall model to match observed climate variability Model Process The rainfall model was programmed in FORTRAN, and its graphical properties are displayed using Excel. The main purpose of the model is to generate synthetic rainfall data reflecting a number of rainfall behaviours. These include incorporating seasonal and interdecadal variability, 29

38 Stochastic Rainfall Simulation Model assuming random variations exist between and within rainfall events, interannual variability and the influence of ENSO which is modelled through Markov Chain are included and finally, a steady decline in rainfall is produced. Details about these rainfall behaviours are explained below Storm Characteristics Essentially, the model consists of four key storm characteristics. These characteristics, interstorm period, storm duration, within-storm pattern and storm intensity are randomly generated from specific statistical distributions (Figure 4.3) (Sivandran 2002). Spatial variability of rainfall intensities has been ignored, and therefore rainfall is assumed to be point-source. Calculations of these characteristics are outlined below. Figure 4.3: Key Storm Parameters Inter-storm Period The period between storm events is assumed to follow a Weibull probability distribution with parameters that vary seasonally. The probability distribution function for inter-storm period (t b ) is given by: f Tb ( t γ ) b β β b tb = γ γ b 1 tb exp γ β b t b > 0 Equation 4-1 where: t b - Inter-storm period γ - Mean inter-storm period Seasonality in the inter-storm period is calculated by the following: γ 2π + α ( ) b cos τ τ Equation 4-2 ω = γ b b 30

39 Stochastic Rainfall Simulation Model where: γ b - Annual average inter-storm period α b - Amplitude of seasonal variation ω - Total time units τ - Time of year τ b - Phase shift Storm Duration Storm duration (t r ) is the length of time that a storm event lasts. Likewise, storm duration is assumed to follow the Weibull Distribution with parameters that vary seasonally, which is given by: f Tr ( t δ ) r β β r tr = δ δ rb 1 tr exp δ β r t r > 0 Equation 4-3 where: t r - Storm duration δ - Mean storm duration Seasonality in the storm duration is calculated by: δ 2π δ + ( ) r α r cos τ τ Equation 4-4 ω = r where: δ r - Annual average storm duration α r - Amplitude of seasonal variation ω - Total time units τ - Time of year τ b - Phase shift Average Storm Rainfall Intensity The average storm rainfall intensity (i) refers to the rate at which the rainfall event occurs. It is calculated by the following equation: f I ( i t ) r = Γ λ κ ( ) ( i) κ λ 1 exp ( λ i) Equation 4-5 where: κ,γ - gamma distribution parameters as functions of storm duration Seasonal variations in average storm intensity is given by: 31

40 Stochastic Rainfall Simulation Model 2π a = + ( ) 1 a10 α a cos τ τ a Equation 4-6 ω where: a 10 - Average coefficient value α a - Amplitude of seasonal variation ω - Total time of units τ - Time of year τ a - Phase shift Within-storm Temporal Patterns Within-storm patterns were obtained from the mass curves exemplified by Huff (1967) and Chow (1988) and were determined through a method of disaggregation of rainfall depth and duration (Sivandran 2002). Interannual and Interdecadal Variability Interdecadal variability was calculated through the following equation: a 2π ω ( ) + a a cos ( τ τ ) = a2m t2 1 a 1 Equation 4-7 which assumes that variability between decades are deterministic. Interannual variability on the other hand was determined by: a 2π ω () + a a cos ( τ τ ) = a1m t 1 a 1 Equation 4-8 where a 1 m was modified to reflect randomness rather than remain a constant. This randomness was determined through the Markov Chain which simply allows the determination of variability between years to occur as a random outcome Model Outputs The stochastic rainfall model produced a series of rainfall simulations in the form of various time scales daily, monthly and annually over a 100 year period, where the year 2001 represents the start of the simulation period and the year 2100 corresponds to the end of the simulation period. It was considered that at least ten realizations of simulated rainfall were required to ensure some measure of statistical validation of the simulation results (Figure 4.4). 32

41 Stochastic Rainfall Simulation Model Rainfall Statistical Parameters for the Ten Realizations Comparison of Realization 1 with Realization Average Mean, Std ( m m ) Mean Std Rainfall (m m ) R1 Rav Realization Year Figure 4.4: (a) Statistical Comparison of Simulated Rainfall for the Ten Realizations and (b) a Comparison of Realization 1 with Realization Average The ten realizations of the rainfall series were created by altering the randomness of ENSO occurrences, while maintaining a decrease in the rainfall trend. Figure 4.4a illustrates that the statistical parameters of mean and standard deviation for each of the ten realizations are statistically similar. Calculations derived from each realization yielded similar results. This means that it is reasonably valid to employ one realization as an example to illustrate analysis and discussion of the simulated results. As a result, Realization 1 has been drawn upon for results and discussion in section 7. An attempt was also conducted to obtain an average over the ten realizations. However, it was discovered that this tended to smooth the natural variability of the data, and therefore produce less meaningful results (Figure 4.4b). Therefore, for the purposes of analysis and discussion, there will be no analysis conducted at examining realization average. 33

42 Large Scale Catchment Model 5 Large Scale Catchment Model 5.1 Model Description Simulated rainfall generated from the stochastic rainfall model was subsequently incorporated into the Large Scale Catchment Model (LASCAM), in order to investigate the Serpentine catchment s response to a change in climate. LASCAM is a complex, conceptual hydrological model which was developed to predict the impacts of land-use and climate change on daily streamflows and salinity in large catchments of the south-west of Western Australia. The model was developed to contain enough physical meaning to be applicable in modelling land use change, but simple enough to warrant its practical application in terms of run time and data requirements. The eventual framework of LASCAM is that of a distributed model, at modest resolution, with processes lumped at the subcatchment scale in conceptual form (Sivapalan et al 1996). As discussed in section 2.3.1, LASCAM, as a conceptual model, relies upon conceptual, constitutive relations between store levels and water fluxes that represent physical processes in the hydrology of the catchment. As a result, model parameters cannot be measured in the field. Therefore parameter values have to be determined by calibrating the model s simulated response to observed data. The model has been applied in several instances across the south-west of Western Australia, and has been proven to produce accurate predictions in these conditions. In its initial development, the model was applied at a small scale on experimental catchments in the Collie River basin (Sivapalan et al 1996). The first application of the LASCAM was in the Conjurrunup catchment, again in the southwest, as part of a study on the effects of Bauxite mining (Sivapalan et al. 1996). Since this time, the model has been applied on a larger scale to the Avon and Ellen Brook catchments LASCAM Inputs The inputs required by LASCAM include daily rainfall which was represented by the 100 year simulated rainfall produced by the stochastic rainfall model. Other necessary inputs include 34

43 Large Scale Catchment Model annual pan evaporation, yearly evolution of the leaf area index (LAI) and details of the subcatchment topographical attributes such as stream length and area. Daily observed streamflow from the Serpentine catchment was also required for calibration LASCAM Process A general description of the processes operating within LASCAM is outlined in this section. The LASCAM model was run with the help of Dr. Christian Zammit from the Water and Rivers Commission Calibration Initially, the model LASCAM was adjusted to reflect the physical and hydrological conditions of the Bigbrook catchment. The Bigbrook catchment, with an area of 149km 2, is situated within the larger Serpentine catchment and was essentially used to represent the processes of the Serpentine catchment. The Bigbrook catchment was disaggregated into a series of 32 interconnected subcatchments of areas between 1 and 5 km 2. These subcatchments are basically the building blocks of the model and are derived so that surface outflows occur at exactly one location on the subcatchment boundary and surface inflows at no more than one location on the subcatchment boundary. Next, leaf area index values for each of the subcatchment were specified for each year of interest, followed by details of the subcatchment properties for Bigbrook. Mean annual rainfall for each subcatchment is used to adjust initial storage values to an approximate equilibrium value. Mean annual potential evaporation is used as a scalar for daily evaporation calculation from each store in each subcatchment. LASCAM assumes that daily potential evaporation values follow a sinusoidal trend in time according to a predetermined harmonic distribution. Calibration of the model was then conducted by adjusting model characteristics to reflect daily observed streamflow values for the Bigbrook catchment, and this was performed until the year

44 Large Scale Catchment Model Operations LASCAM operates at a daily timestep and hydrological processes are modelled at each subcatchment scale. The hydrology of each subcatchment is modelled in terms of three interconnected, conceptual stores of soil water (Viney & Sivapalan 2000). These are the: A Store, representing a perched, near stream aquifer B Store, the permanent groundwater store; and the F Store, intermediate unsaturated store (Figure 5.1) For each subcatchment, a range of constitutive relationships is used to direct water fluxes between the stores and to distribute incoming rainfall, either into the stores or directly into the stream (Viney and Sivapalan 2000). Figure 5.1: Schematic Diagram of an idealised hillslope showing the three conceptual model stores and the principal water fluxes Streamflow is assumed to be generated by three processes: infiltration excess runoff (q ie ), saturation excess runoff (q se ) and baseflow from the A store (q A ). The groundwater store is assumed not to discharge directly into the stream. Infiltration excess runoff is governed largely by the amount of vegetation cover, which has a relationship with the infiltration capacity of the soil, while the other two mechanisms are mainly dependent on the level of the A store (Sivapalan et al 1996). 36

45 Large Scale Catchment Model These hydrological processes modelled at the scale of the individual subcatchments are then aggregated via a routing subroutine to generate the response for the entire Bigbrook catchment. The calibrated LASCAM model was essentially fed the simulated rainfall data produced by the stochastic rainfall model. Since the stochastic rainfall model assumed rainfall data to be point source, it was required by LASCAM to firstly disaggregate the precipitation to produce individual rainfall for each of the 32 subcatchments. This disaggregation was assumed to follow the rainfall disaggregation pattern for the year LASCAM Output The required LASCAM output was the modelled catchment s response to the simulated rainfall data. This was the production of daily simulated streamflow data for the 100 years simulation period. 37

46 Rainfall-Runoff Analysis 6 Rainfall-Runoff Analysis The results from both the stochastic rainfall model and LASCAM were subsequently analysed to investigate the relationship between rainfall and runoff, and to provide greater insights into changes that occur in the mechanisms of runoff processes in catchment hydrology, as a result of climate change. Greater emphasis has been placed upon the qualitative aspect of analysis rather than the quantity. The understanding of hydrological processes gained from this study provides a basis for predicting the nature and scale of the hydrological response to other changes in climate. 6.1 Rainfall and Runoff Analysis Techniques The rainfall and runoff analyses carried out in this section were extensive. Many techniques were employed to study and measure the responsiveness of streamflow due to rainfall decline and variability. As a result, analysis was carried out to largely examine any trends or significant changes in rainfall and the way in which this has been translated to streamflow. One of the most important tasks was to analyse the observed and simulated streamflow data. Analysis of these data provides us with three important features: Description of the flow regime Potential for comparison between different catchments; and Prediction of possible future river flows There are well-established techniques available to achieve this, although they are not universally applied in the same manner. Some of the important methods applied in analysis of the results include: Statistical flow analysis Graphical analysis of interannual and seasonal variability Hydrograph analysis Ratio of surface to subsurface runoff to the total streamflow Changes in the runoff coefficient; and Flow duration curves for streamflow 38

47 Rainfall-Runoff Analysis The majority of these techniques are commonly used in statistical or hydrological studies, and need no explanation. However, the methodology used to determine some of the statistical parameters, flow duration curves, annual rainfall and streamflow probability distribution curves, runoff coefficient and the ratio of surface to subsurface flow will need further clarification. Manipulation of rainfall and streamflow data was conducted either in Excel or programmed in Matlab. Analysis and calculations conducted for each of the realizations produced very similar results, which may be attributed to their statistical similarity Statistical Analysis In some instances, the complexity of the hydrologic system defies a complete and rigorous physical analysis. An alternative to a complete analytical solution is to utilize some simultaneous observations of the independent and dependent variables, and then to apply accepted and wellknown statistical procedures to develop relationships between the variables (Schulz 1994). The main purpose of statistical analysis is to therefore analyse uncertainties, trends or changes in a hydrologic variable. In this case, the hydrologic variables are rainfall and runoff. The various statistical parameters employed for this study and their methodology, are explained below. Mean The mean ( x ) is regarded as an estimate of the variable which would be the most likely observed value next in the array of observations. The mean is a measure of central tendency (other measures of central tendency include median and mode). The mean is calculated by the equation: Mean = x = N i= 1 N x i Equation 6-1 Standard Deviation The standard deviation (s x ) is a measure of the dispersion of the various values of the variable x i, about the mean value x. The standard deviation is defined by the equation: StdDev = s x = ( x x) i N 2 Equation

48 Rainfall-Runoff Analysis For this study, a 95% confidence level has been specified as the case where any random measurement of x lies within two standard deviations from the mean value. Coefficient of Variation The coefficient of variation (Cv) is a measure of the dispersion of the variable about the mean expressed in dimensionless units. Generally, a high coefficient of variation indicates increasing variability of the data. It is calculated by dividing the standard deviation by the mean. Cv Stationarity s x = Equation 6-3 Stationarity in a hydrologic time series exists when the statistical parameters (mean and standard deviation) do not change relative to the time origin. A time series is simply a set of observations usually taken at regular time intervals. The analysis conducted assumes non-stationarity, as the mean and standard deviation has been shown to change over the 100 years. A nonstationary time series results when a trend or cyclic component is usually present (Schulz 1994). x Percentiles Percentiles were calculated for each of the Realizations for both simulated rainfall and streamflow. Percentiles are defined as the percentage of events that are below the given value. For example the 0.5 percentile for January is the median monthly rainfall over the 100 year simulation. Percentiles that were calculated include the minimum, maximum, median, upper quartile (75%) and lower quartile (25%). Percentiles were calculated by sorting the monthly data into ascending order over the 100 year simulation and then assigning a percentage rank. Autocorrelation Test The lag-one autocorrelation coefficient (r k, k=1) gives an indication of data independence. The coefficient is given by the following for sample size n: r k = n k i= 1 ( x x)( x x) i n i= 1 i+ k 2 ( x x) i Equation

49 Rainfall-Runoff Analysis A data series can be tested for short term dependence by checking whether r k is significantly different from the expected value E(r 1 ). Where: 1 E( r 1 ) = Equation 6-5 n 3 2 ( n 2n + 2) 2 n ( n 1) Var ( r1 ) = Equation [ r ] 1 E( r1 ) z-statistic = [ Var( r )] Equation 6-7 If this lies within the critical z-statistic value (with a 95% confidence limit) then the hypothesis that the sequence results from a random process is accepted (Grayson et al. 1996). Autocorrelation was conducted for simulated rainfall to statistically test the randomness of the data. Single Factor Anova Significance Test The single factor Anova test determines whether two data series have similar statistical properties, or if the difference between the two series is so significant, that the two data series represent two distinct families of data. An assumption employed in the use of the Anova test is that both data series were normally distributed (Grayson et al. 1996). For each data series pair, the test calculates a parameter F, where: Variancebetween Samples F = Equation 6-8 Variance within Samples This parameter (F) is compared to another parameter F critical, which is a function of the length of the data series and the confidence level used to determine significance. If F > F critical, the variability between the samples is greater than the variability within each sample set, which means that there is a significant difference between the data sets (Grayson et al. 1996). In this study, the Anova test measured the significance of the difference between simulated rainfall and streamflow between Realization 1 and Realization 10. An F critical value of 3.89 was used which was determined using a data series length of 100 years and a confidence level of 95%. 41

50 Rainfall-Runoff Analysis Ratio of surface to subsurface flow Simulated subsurface and surface flow data were obtained from the LASCAM model. A comparison of the ratio of surface to subsurface flow is important as the proportion gives an indication of the major runoff generation mechanisms operating in the Serpentine catchment, and how these mechanisms adjust to a change in climate. The values of the surface and subsurface flow were compared to the value of the total streamflow q T, at the end of each daily timestep. The ratio of surface to subsurface runoff under climate change was determined as a percentage using the following equations: q q s T = q s qs + q ss 100 Equation 6-9 q q ss T q qs + q ss s = Equation 6-10 ss = q q T Flow duration curves Flow duration curves are concerned with the amount of time a certain flow is exceeded. Flow duration curves convert the cumulative frequencies of all daily streamflows into a percentage of time that the flow equals or exceeds that flow value. The shape of the flow duration curve indicates a particular catchment s historical flow record. A steep curve indicates variable flow, while a flat slope indicates little variation in flow characteristics (Hornberger et al. 1998). Flow duration curves were calculated for daily observed and simulated discharge data for the Serpentine catchment, for various years of the simulation. Computation of flow durations involved determining the probability that an event of a certain magnitude will be equalled or exceeded in any one year. This was established by ranking each of the flow data and assigning a probability to each flow value. Pr obability Equation 6-11 = Rank N + 1 where N = number of data points to be evaluated Flow duration curves were then plotted on a log-normal graph for comparison. 42

51 Rainfall-Runoff Analysis Runoff coefficient The runoff coefficient depicts the relationship between rainfall and runoff. It is simply a measure of the percentage of runoff resulting from rainfall. Calculation of the runoff coefficient was programmed in Matlab and involved a number of steps: i. A plot of cumulative simulated rainfall against simulated discharge (Figure 6.1) data was separated into 20 blocks of five years an average cumulative rainfall versus discharge was obtained for each block Figure 6.1: Plot of Cumulative Rainfall vs. Cumulative Discharge for block 1 Realization 1 ii. Polynomial curve fit of order 5 was matched to the curve calculated in i). The polynomial curve fit was the best curve fit available, however, as evident from Figure 6.2, the curve fit was not able to exactly capture the cumulative curve. Figure 6.2: Curve fit to Cumulative Rainfall vs Cumulative Runoff for Block 20 43

52 iii. Rainfall-Runoff Analysis The runoff coefficient was then computed by deriving the curve fit obtained in ii) that is Runoff Coefficient ( RC) = dq dp ( P) iv. The resulting runoff coefficient was then evaluated as a function of time - dq RC = ( P() t ) and expressed in terms of a trend over the 20 blocks representing the 100 dp year sequence 44

53 Results and Discussion 7 Results and Discussion A catchment s hydrological balance may be modified as a result of changing climate conditions. Catchment response in this case indicates the balance and timing of the partitioning of rainfall into streamflow. This section analyses the change in the statistical properties and identifies patterns in the characteristics of the rainfall and streamflow of the modelled Serpentine catchment under the chosen climate change scenario. Throughout this section, it must be remembered that, as in any climate change study, the results depend heavily upon the climate change scenario chosen, and the appropriateness and consistency of the modelling strategy used. Autocorrelation and single factor anova test was conducted for each of the rainfall simulations. It was found that the results produced were very similar. The tests showed that rainfall data between simulations were independent of each other, and that variability between the data sets was insignificant. As a result, realization 1 has been used to represent rainfall simulation as produced by the stochastic rainfall model. 7.1 Precipitation Historical rainfall data from the period was obtained for the Serpentine catchment. Data for periods before 1993 was unable to be obtained; therefore any comparisons made with the observed data were constrained and could only occur in terms of decadal averages Statistical Comparison Rainfall Year Mean (µ) % change Std (σ) Cv Observed Realization Figure 7.1: Statistical Comparison for Observed Rainfall with Realization 1 Figure 7.1 illustrates statistical comparisons of observed rainfall with decadal averages of simulated rainfall for Realization 1. The table indicates that there is a smooth transition between 45

54 Results and Discussion observed rainfall into the first decade of simulation. The coefficient of variation demonstrates that variability in simulated rainfall alters very little over the simulation period. This is an indication that interdecadal variability, as reproduced by the stochastic rainfall model, appears to be deterministic. The average coefficient of variation for Realization 1 is approximately 0.18, which is consistent with the observed value and similar to the coefficient of variation for the south-west of 0.2 (Water and Rivers Commission 2003). Figure 7.2 shows simulated annual rainfall for Realization 1. The observed rainfall can be seen in the first ten years of the plot. The figure shows that there is a decline in decadal averages over the 100 year sequence, and that the difference in the decadal averages over this period is approximately 20%. While only Realization 1 has been shown, rainfall for the other Realizations have also declined, although the magnitude of the decrease is diverse, depending upon the variability inherent in the simulated rainfall (Appendix A). Rainfall Trend with Realization Rainfall (mm) Rainfall Average Year Figure 7.2: Simulated Annual Rainfall For Realization Wet and Dry Cycles A moving mean was used to damp or smooth some of the random variations in the simulated rainfall. This basically allowed any trends in the data to be revealed. Typically, precipitation data are analysed by means of a five-year moving average, which can be seen in Figure 7.3. It appears 46

55 Results and Discussion Simulated Rainfall for Realization 1 showing wet and dry cycles Rainfall (mm) Year Simulated Rainfall for Realization Rainfall Mean Moving Average 1000 Rainfall (mm) Rainfall Mean Moving Average Year Simulated Rainfall for Realization Rainfall (mm) Rainfall Mean Moving Average Year Figure 7.3: Wet and Dry Cycles for Realization 1, 5 and 10 47

56 Results and Discussion that five years are sufficient to damp out most of the random component, leaving the effects of longer term wet and dry cycles in the record (Schulz et al. 1974).The wet periods can be recognized by comparing the five-year mean line with the mean for the entire simulation. During a wet period, the five-year moving mean line is always above the long-term mean. During a drought period, the five-year moving mean line is always below the long-term mean. The figure shows that for Realizations 1, 5 and 10 there are significant fluctuations between wet and dry periods, which are a reflection of the variability of ENSO, as generated by the rainfall model. Although the pattern of wet and dry periods differ between the realizations, it appears that there is a general trend towards a greater number of dry periods occurring relative to the number and extent of wet periods. This is an indication that there may be a continuation of more intermittent drought periods experienced over the next 100 years. In terms of Realization 1, there is a major wet period from 2015 to This is followed by successive cycles of smaller wet and dry periods, however there is a general trend in rainfall becoming dryer, as depicted by the declining moving average Seasonality and Distribution The hydroclimate of South-West Western Australia is described as being in a threshold state (Water and Rivers Commission 1996). That is, streamflow is very sensitive to changes in rainfall because of the climate, soils and vegetation that characterize the region. In the context of this study, changes to the mean and variability of rainfall provide the first indication of how climate change may affect the seasonality of the catchment s hydrological cycle. Month Mean (mm) Observed January February March April May June July August September October November December Table 7-1: Monthly Rainfall Mean for Observed Rainfall and for selected decades for Realization 1 48

57 Results and Discussion Table 7-1 illustrates the monthly rainfall distribution for observed rainfall and Realization 1. The results in the table indicate that the observed rainfall has a more pronounced seasonal variation, with low summer rainfall and higher winter rainfall compared to that predicted by the simulation. There is also a general overall decrease in rainfall for Realization 1, as the simulation progresses over the 100 years. This is evident in the decrease in seasonal rainfall as seen in Figure 7.4. Comparisons with the seasonal coefficient of variation indicate that seasonal variations for the simulation within the decade are low. Synthetic winter rainfall (Figure 7.4) has declined by approximately 14% over the 100 years, calculated as the difference between the decadal differences in winter rainfall. This figure is compared with the observed decrease in winter rainfall over the last 30 years of approximately 15 20% (IOCI 2000). The figure also shows that there is a strong seasonal cycle with higher rainfall occurring during the winter months. Mean Monthly Rainfall Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Rainfall (mm) Observed Month Figure 7.4: Mean Monthly Comparison for Observed and Simulated Rainfall for Realization 1 It has been suggested (Ruprecht et al. 1996) that a change in rainfall will have an accentuated impact on streamflow. It is important to consider the factors affecting streamflow generation. As mentioned in section , streamflow is a function of short-term variations and seasonal distribution of rainfall and the volume of water stored in the perched aquifer Distribution of Rainfall Events Besides variation in seasonality, the other potential impact of climate change on rainfall is a change in the intensity distribution of rainfall. Streamflow generation is not only a function of 49

58 Results and Discussion rainfall volume, but also intensity. The rainfall intensity function in the stochastic rainfall model has been altered to reflect random variability, therefore, in order to quantify any changes in rainfall intensity, percentile tables were computed for each realization. Since LASCAM and the stochastic rainfall model both generate daily series of data, it is unfortunately not possible to determine if there is an increase in rainfall events of duration less than one day. However, apart from the daily timesteps constraint, percentile distributions are the most appropriate means to review changes in intensity distribution patterns. Monthly Rainfall Percentile for Realization Rainfall (mm) Lower Quartile Min Median Max Upper Quartile Month Figure 7.5: Monthly Rainfall Percentiles for Realization 1 As an indication, the rainfall distribution data for Realization 1 have been plotted in Figure 7.5 which depicts box plots of the percentiles. Box plots are graphical displays of the character of percentile distribution. The edges of the boxes mark the upper and lower quartiles (75% and 25%), while the green line within each box depicts the median. The end points attached to the boxes denote the extremes of the data. The box plot demonstrates that there has been a small increase in the number of higher intensity events, but overall there has been no gross shift in the rainfall intensity distribution. 50

59 7.2 Streamflow Results and Discussion Historical discharge data for the outlet of the Big Brook Catchment was obtained for the period As with precipitation, any comparisons made with the observed data had to be made in terms of decadal averages Statistical Comparison Streamflow Year Mean (µ) % change Std (σ) Cv Observed Realization Table 7-2: Decadal Statistical Comparison of Observed Discharge with Realization 1 A comparison of the coefficient of variation (Table 7-2) for each decade of the simulated streamflow for Realization 1, shows that compared with rainfall, there is slightly greater interdecadal variability. The transition from observed streamflow to the first decade of the streamflow simulation is less smooth compared to rainfall. This can be seen by the percentage change from to of approximately 57%. For the same period, the percentage change of rainfall is only about 1%. This discrepancy may be attributed to the fact that the LASCAM model was not given a warm-up period in which to adjust to the sudden change in climate. Essentially, LASCAM had been calibrated with climatic and rainfall data until the year Inputting the simulated rainfall data with a different climate scenario into the model system without an adjustment period is equivalent to splashing cold water on your face. The system has basically experienced a shock. This discrepancy may subsequently have been passed onto the streamflow simulation. Figure 7.6 shows a plot of annual simulated streamflow for Realization 1, with observed discharge representing the first ten years of the plot. By taking the difference between decadal averages, the model predicts that streamflow will decrease by approximately 95%. Results from 51

60 Results and Discussion the other Realizations also indicate a similar decrease in streamflow over the 100 year simulation period (Appendix A). Discharge trend with Realization Discharge (mm) Discharge Average Year Figure 7.6: Annual Simulated Streamflow for Realization Seasonality and Distribution Besides providing the most reliable measure of a catchment s hydrological response to climate change, the importance of changes to streamflow seasonality and distribution lie in its direct impact on the reliability of water reservoirs. Many studies have observed significant non-linear changes in streamflow in response to moderate changes in rainfall (Murphy and Lodge 2001), however very few studies have specifically examined the impact of climate change on the process of streamflow generation. This section therefore examines the seasonality and distribution of the modelled Serpentine catchment, and focuses on the impact of climate change on the hydrological processes that generate streamflow. Table 7-3 denotes the monthly mean distribution of simulated streamflow for the Serpentine catchment, averaged over the simulation period for Realization 1. The table shows that for both the observed and simulated data, there is no flow occurring during the summer to early autumn months. Peak discharges appear to occur during the winter season. Simulated streamflow 52

61 Results and Discussion seasonality can be seen to have significantly decreased over the simulation period. Winter streamflow has decreased by approximately 95%, which is consistent with the decrease seen in annual simulated streamflow. This suggests that streamflow during the winter months is the dominant contributor to annual streamflow. Month Mean (mm) Observed January February March April May June July August September October November December Table 7-3: Mean Monthly Streamflow Comparison for Observed and selected decades of Realization 1 Figure 7.7 shows a comparison of mean monthly discharge between observed and simulated data. As the diagram reveals, although the volume of streamflow has decreased over the simulation period, there is still a strong seasonal cycle of very low summer flows and high winter peaks. However, this seasonal variation decreases over time, clearly as a result of the decrease in streamflow volumes and an increasing decline in rainfall. Mean Monthly Discharge for Realization Mean Discharge (mm Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Observed Month Figure 7.7: Mean Monthly Discharge for Observed Data and Realization 1 53

62 Results and Discussion The figure also shows that there is a slight change in the seasonal distribution of the streamflow, with peak flows occurring earlier in the winter season for the simulation period, compared with the observed data which shows peak flow occurring in late winter Intensity of Streamflow Events Important features of catchment response can also be observed through the percentiles of the simulated streamflow series. Figure 7.8 depicts the monthly streamflow percentiles for Realization 1 over the simulation period. As with the rainfall percentiles, the edges of the box plots indicate the upper and lower quartiles (75% and 25%) for monthly streamflow. The whiskers attached to the boxes indicate the extreme values. Monthly Streamflow percentiles for Realization Runoff (mm) 5 4 Lower Quartile Min Median Max Upper Quartile Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month Figure 7.8: Monthly Streamflow Percentiles for Realization 1 The figure shows that there is very little difference in the distribution of streamflow during the summer months, however there is a very large difference in distribution during the winter months. The distance between the upper quartile (75%) and the maximum streamflow during the winter months is extremely large, compared with simulated rainfall, which is an indication that streamflow variability is much greater than rainfall variability. Therefore, there are extreme 54

63 Results and Discussion events occurring during the winter season. It is interesting to see that there is little change in extreme summer streamflow, given the large change in extreme summer rainfall events this may be related to the effect of climate change on soil moisture storage and seasonality of flow events Interannual Variability Variability of a data set may be measured by the coefficient of variation. In general, a higher coefficient of variation indicates increasing variability in the data. Interannual variability of simulated streamflows for Realization 1 was measured by plotting the coefficient of variation for the 100 years within a 20 year block (Figure 7.9). The figure shows that interannual variability for streamflows increases over time, with increasing variability occurring with a greater decline in rainfall. This variability is significant in the last 20 years of the simulation period. Figure 7.9: Interannual variability of Streamflow for Realization Contribution of Surface and Subsurface Runoff to Streamflow To further quantify the processes dominating the catchment response, the ratio of surface and subsurface runoff to total streamflow were assessed. Subsurface and surface flow ratios were 55

64 Results and Discussion provided by LASCAM. Table 7-4 shows that for the ten realizations, the proportion of subsurface runoff (Q ss ) contributing to total streamflow (Q T ) is considerably larger than that of surface flow (Q s ). The table essentially implies that streamflow in the Serpentine catchment is largely a result of groundwater runoff. Q Q Q Q ss T s T R1 R2 R3 R4 R5 R6 R7 R8 R9 R % 95.5% 95.6% 95.5% 94.3% 95.6% 95.2% 95.3% 94.5% 96.0% 5.0% 4.5% 4.4% 4.5% 5.7% 4.4% 4.8% 4.7% 5.5% 4.0% Table 7-4: Surface and Subsurface Runoff ratios to Total Streamflow As discussed in section 2.1.4, surface runoff can be generated in two ways - through infiltration excess or saturation excess mechanisms. Infiltration excess implies that the infiltration capacity of the soil is exceeded by the rate of rainfall and saturation excess implies that soil moisture capacity is exceeded. Both mechanisms imply that either there must be an excess of rainfall or that soil moisture capacity is exceeded. Table 7-4 indicates that both of these conditions do not occur for the Serpentine catchment. In other words, there is not enough rainfall to stimulate surface runoff and in addition, soil moisture capacity is low. There are a number of reasons for the lack of water reaching the surface and this may be a result of a combination of factors. The most apparent reason is that a declining rainfall over the simulation period will not produce enough volume of water falling into the catchment. Although there is not enough information about whether the decline in precipitation is a result of a decline in rainfall intensity, duration or the number of events, it is clear that a decrease in any one of these factors will be insufficient to generate surface runoff. That is, the rainfall rate will almost always be less than that of the infiltration capacity, so that any water reaching the surface will be immediately infiltrated to the perched aquifer or groundwater. Another reason for the lack of water reaching the surface is due to the role of interception and depression storages. Interception of rainfall by vegetation canopy, by either reducing the volume of water falling directly onto the surface or by reducing rainfall intensity is an important factor. It has been shown that in highly forested areas, 30% of annual precipitation may be lost as a result of canopy interception (Brooks et al. 1992). Since Serpentine is considered to be a fully forested catchment, this occurrence is extremely probable. Therefore, a threshold due to the presence of 56

65 Results and Discussion vegetation may exist, which needs to be overcome before the mechanisms of surface runoff can be generated. Water held in depression storage reduces the amount of surface runoff. This water eventually either infiltrates through the soil or is evaporated back into the atmosphere. This is particularly significant for the Serpentine catchment, as potential evaporation in the region is greater than precipitation. Any water that is held on the surface will be quickly evaporated. One other outcome is that soil moisture capacity is low. In other words, the catchment is dry, so that any moisture that falls onto the surface is either quickly infiltrated or lost to the atmosphere. This situation would be particularly true for the Serpentine catchment, as annual potential evaporation exceeds that of annual precipitation. Deep-rooted vegetation such as Jarrah and Marri which are predominant in the catchment may also draw upon water in the soil, preventing the water table from rising and stimulating saturation excess. Therefore, it appears that subsurface runoff is the main runoff mechanism occurring in the Serpentine catchment which is consistent with other studies (Wittenberg and Sivapalan 1999). Water infiltrates to the underlying aquifer, and is eventually discharged to a stream or water body Discharge Hydrograph Analysis A hydrograph is a continuous record of stream or river discharge. It is a basic working unit for a hydrologist to understand and interpret. The shape of a hydrograph is a response from a particular catchment to a series of unique conditions, ranging from the underlying geology and catchment shape to the antecedent wetness and storm duration. The spatial and temporal variations in these underlying conditions make it highly unlikely that two hydrographs will ever be the same. Figure 7.10 denotes plots of daily discharge hydrographs for the years 2001, 2035, 2065 and 2100 for Realization 1. The figure is a useful way to assess the daily variability of discharge which occurs within a year. Consistent with the previous findings, the figure indicates that streamflow variability is very high within a year. 57

66 Results and Discussion In particular, the hydrographs reveal that streamflow becomes increasingly ephemeral and highly episodic. According to Ward (1990), the episodic, erratic patterns signified by ephemeral streams are those which must comprise of surface runoff and which therefore flow only during and immediately after rainfall. Owing to the nature of groundwater, subsurface flow patterns are depicted in a hydrograph as a smoother, continuous trend. Figure 7.10: Discharge Hydrographs for Realization 1 for the years 2001, 2035, 2065 and 2100 This result seemed to contradict with the prior discovery that groundwater accounts for 95% of total streamflow. However, it can be discerned that over the long-term, groundwater contribution to total streamflow becomes less important and surface flow becomes the dominant runoff process. This is because total streamflow over the simulation period has decreased by such a vast extent, that groundwater and surface runoff must also decrease. The only possible peak flows in discharge that can now occur are surface flows due to a heavy rainfall event. This is confirmed in Figure 7.10, which shows that peak discharges occur during the winter months, possibly in response to a storm event. The diagram also depicts that groundwater flow has not entirely disappeared, which is evident by the small fluctuations in discharge seen during the length of the year. 58

67 7.3 Comparing Rainfall and Streamflow Results and Discussion The results from sections 0 and 7.2 are extremely significant in a number of ways. They are summarized below: Precipitation: Interdecadal variability is slight There has been a 20% decrease in rainfall over the simulation period There is great variability in rainfall decline between different realizations Wet and dry cycles show that there will be greater occurrences of intermittent drought For Realization 1, one major wet period during 2015 and 2030 has been identified and there is a trend towards dryer periods There is a strong seasonal rainfall cycle. Seasonal rainfall decreases with an overall decrease in annual rainfall over the simulation period Intensity distribution of rainfall shows a small change Streamflow: Greater interdecadal and interannual variability compared to simulated rainfall There has been a 95% decrease in streamflow over the simulation period There is less variability in streamflow between different realizations Seasonal variability decreases in conjunction with a decrease in annual streamflow over the simulation period, and an increasing decline in rainfall There is a slight shift in the distribution of seasonal streamflow Intensity distribution of streamflow shows large distribution during the winter months Extreme summer rainfall has not resulted in extreme summer streamflow Subsurface runoff is the dominant contributor to streamflow at the beginning of the simulation, which is surpassed by surface runoff as the dominant runoff mechanism towards the end of the simulation Streamflow becomes increasingly ephemeral and highly episodic The results indicate that the key processes and factors governing streamflow, which was discussed in the literature review (sections and 2.1.4), appear to have come into effect. 59

68 Results and Discussion One significant finding is that a decrease in rainfall over the simulation period of approximately 20% will lead to a subsequent decrease in streamflow of 95%. The implications of this are considerable. It implies that by the end of 2100, there will be very little runoff flowing out of the Serpentine catchment. This suggests that future availability of water from surface water resources in the south-west may be vastly diminished. Accordingly, it was thought that comparisons of the model results with other similar studies had to be conducted, in order to provide some measure of validation for the accuracy of these results. As previously mentioned in the literature review (section 2.2.3), it has been observed that mean annual rainfall in the south-west region has decreased by approximately 10% in the last 30 years. According to the Water Corporation, this has lead to a decrease in mean annual inflow into Perth s water supplies of approximately 51% since 1975 and by a further 10% decline since In other words, within the last 30 years, a 10% decline in rainfall has lead to a 62% decline in streamflow. CSIRO have also conducted studies on future changes in rainfall for Australia (Centre for Water Research 2002). For the south-west region, the worst-case scenario predicts that by the year 2030, rainfall will have decreased by approximately 20%. Data analysis carried out by the 2002 final year design team predicted that this would lead to a decrease in streamflow on the order of about 75% (Centre for Water Research 2002). Simulated Rainfall for Realization Rainfall (mm) Year Figure 7.11: Simulated Rainfall for Realization 1 ( ) 60

69 Results and Discussion Simulated Discharge for Realization 1 ( ) Discharge (mm) Year Figure 7.12: Simulated Discharge for Realization 1 ( ) Figure 7.11 shows the model results for simulated rainfall from Within the next 30 years as predicted by the model, rainfall will have declined by approximately 8%. Runoff response to this fall in rainfall is a decline by about 94% (Figure 7.12). The comparisons over a 30 year period are summarized in Table 7-5. % fall in Rainfall % fall in Runoff Model 8 94 Observed CSIRO Table 7-5: Comparison of Rainfall and Runoff over a 30 year period One important question is - why has the model produced such a large decrease in streamflow in response to a small decrease in rainfall? This issue will be examined in the next section Hypotheses for Rainfall-Runoff Decline Difference There are a number of explanations which may account for this large difference between rainfall and runoff decline. In fact, this difference may be the result of a combination of these influences. The underlying factor behind these hypotheses however, is that the use of a conceptual model to predict catchment response to climate change may not be the best option. The hypotheses are presented below Vegetation Cover It is believed that climate change over the long-term will lead to a change in vegetation density of the catchment area and in fact, some studies have indicated that this will occur (Ward 1990). 61

70 Results and Discussion Natural vegetation tends to adapt itself to a changing climate. Generally, a dryer climate will lead to sparser vegetation cover, as water becomes a limiting factor. Deep-rooted vegetation, which dominates the forest of the Serpentine catchment, tends to use all available water. As the rainfall decreases, the biomass of the trees must adjust to limiting amounts of water. This implies that less water will be taken up and lost through evapotranspiration, and may also lead to a relative increase in surface runoff, due to less interception by vegetation canopy. The LASCAM model however, assumes that vegetation density (LAI) remains fully forested over the 100 year sequence. This means that for the simulation, as rainfall declines, groundwater resources are gradually depleted, as the model will continue to pump water out of the system. Other factors affecting runoff as described in section will also impact upon the system at the same time. As a result, data generated by LASCAM may produce streamflow values that are lower than what could be expected to occur under changing vegetation conditions due to climate change Runoff Processes Another possibility which might explain the drastic reduction in streamflow is the behaviour of runoff processes. A decrease in rainfall not only produces a decrease in streamflow, but it also implies a fall in recharge to the soil. This means that over a long time period, the catchment gradually becomes dryer. Subsequent streamflows through the catchment will be quickly absorbed into the system, with no opportunity for runoff out of the catchment. Runoff processes are highly non-linear and are threshold dominated. This means that as the water level drops, the water deficit in the ground must first be filled before water can overflow. In other words, the model results may indicate that some kind of runoff threshold exists. No runoff will be generated until this threshold is exceeded, which may be another reason that has contributed to the drastic reduction in runoff Model Assumptions There were a number of assumptions that were held by both the stochastic rainfall model and the LASCAM model, which may have limited their ability to predict catchment response to a change in climate with greater accuracy. 62

71 Results and Discussion It was considered that the simulation period of 100 years was overly long to enable an accurate prediction of a future climate. A long time frame allows the potential for many physical parameters and hydrological variables to change within the catchment boundary. As demonstrated earlier, vegetation density is most likely to adjust in response to a change in climate over the long-term. A simulation period of 20 years for example, is a more realistic time frame for vegetation parameters to be ignored or assumed to be constant. The stochastic rainfall model incorporated temporal variability, but did not account for spatial rainfall variability. Therefore, rainfall was assumed to be point source. As a result, before incorporation into LASCAM, there was a need for rainfall data to be disaggregated into rainfall patterns for each of the subcatchments. These patterns were based on calibrated data for the Serpentine catchment. As LASCAM is a conceptual model, there is a danger in attempting to extrapolate data beyond the calibration period. The model is dependent upon historical data and patterns of the current catchment. Therefore, changes in hydrological and physical parameters such as changes in the dominance of runoff processes which may occur under long-term climate change have not been accounted for. In contrast, the use of a physical model, which attempts to simulate the exact hydrological processes through equations which describe the physics of the process, require little to no calibration. Hence, a physical model may be more suited to predicting climate change impacts Rainfall-Runoff Relationship More importantly than quantification are the insights that can be gained by examining the qualitative relationship between rainfall and runoff for the simulation. Figure 7.13 depicts a plot of simulated annual rainfall against annual runoff for Realization 1. The graph illustrates that if a curve was fitted to the data points, it would show that the rainfall-runoff relationship as modelled by the simulation is strongly non-linear. This outcome is consistent with many other studies (Water and Rivers Commission 1996). The strong non-linear relationship between rainfall and runoff may be explained by the lag in runoff that occurs from the time rain falls onto the surface. As we have found, runoff in the Serpentine catchment is predominantly subsurface. This form of runoff is considered delayed due 63

72 Results and Discussion to the slow transmission of water through the soil matrix. Therefore discharge often occurs well after the rain event has ceased, leading to an accentuation of non-linearity. A study by Woolridge et al. (1999) suggested that the factors affecting the non-linearity of runoff production may also include the nature of ENSO influenced rainfalls, such as rainfall duration and intensity, antecedent soil moisture conditions and also evaporative losses. It was also implied that runoff production was reduced as a result of the combined effects of ENSO-related reduction in rainfall intensities, drier antecedent soil conditions and higher evaporative losses. More significantly, Figure 7.13 illustrates that a moderate decrease in rainfall is translated into a substantial decrease in streamflow. However, a different picture of catchment response emerges when a monthly timestep is applied. Flow behaviour during winter months indicates that modest changes in rainfall are accentuated in the streamflow volume. In contrast, decreases in rainfall during the summer months had a slight effect on streamflow. As predicted from the rainfall analysis, the modelled catchment response appears to be dependent upon the season and the time of year in which rain occurs. More specifically, the response is reliant upon the volume of water stored in the catchment s soil moisture storages and the seasonality of rainfall. Simulated Rainfall and Runoff Relationship for Realization Runoff (mm) Rainfall (mm) Figure 7.13: Relationship between Annual Rainfall and Annual Runoff 64

73 Results and Discussion The simulated streamflow is directly dependent on the volume of water stored in the near-stream perched aquifer (Figure 5.1). During winter months, the perched aquifer contains sufficient water for changes in rainfall volume to be expressed in the catchment streamflow. The effect on streamflow is a non-linear accentuation of the change in rainfall. During summer months however, the aquifer depletes as it is subjected to continual discharge and evapotranspiration. Therefore the perched water levels are so low that any increase in rainfall is likely to have a negligible effect on streamflow. Additional rainfall is either evaporated or infiltrates into the soil where it recharges the surface soil moisture storage. Based on this analysis, it appears that for the Serpentine catchment, streamflow response to changes in rainfall will only be significant if there is a sufficient volume of water stored in the surface soils. If the soil water storage is low (dry), the effect of moderate changes in rainfall may be negligible. Therefore, unless rainfall intensities are high enough to cause flooding, significant summer streamflows can only be generated if there is sufficient rainfall to reduce soil storage deficits. Unfortunately, this may not occur, due to the decreasing trend in rainfall as simulated by the model. An additional consideration is that soil moisture storage can only increase if the effect of evapotranspiration does not negate the effect of rainfall volumes. Evapotranspiration is the only mechanism predominantly responsible for the depletion of soil moisture during summer. Unfortunately, the effect of a climate change upon evapotranspiration has not been accounted for within the LASCAM model. Evaporation calculation by the model is assumed to be constant since calibration. Studies have found that in relatively dry areas where catchment moisture is rarely high enough to allow evaporation at the potential rate, streamflow response is highly variable (Evans et al. 2000). Factors affecting the rate of evapotranspiration however, are quite complex, and are dependent upon net solar radiation, surface area of open bodies of water, wind speed, density and type of vegetative cover, availability of soil moisture, root depth, reflective land-surface characteristics, and season of year (Hanson 1991). Generally however, keeping other factors constant, a decrease in vegetation cover will reduce evapotranspiration even further under the long-term climate 65

74 Results and Discussion change scenario. This would imply a relatively greater availability of water and therefore increase the potential for greater runoff Co-Variation of Rainfall with Runoff In addition to the direct relationship that simulated rainfall has with runoff, an assessment of their variability can provide insight into their properties. Figure 7.14 depicts a comparison of the mean coefficient of variation between interdecadal rainfall and runoff for Realization 1.The figure demonstrates that rainfall variations are low compared with the more variable streamflow. This result is not unusual for Australia. According to McMahon et al. (1992), Australia has the largest interannual variability of streamflows across the world. As previously described, this outcome is due to the highly non-linear, threshold dominated relationship between rainfall and runoff. The slight variation in interdecadal rainfall is a result of the stochastic rainfall model, which assumed a deterministic outcome for interdecadal variability. In comparison, the interdecadal variability evident in simulated runoff is due to the fact that apart from rainfall, streamflow is a function of other hydrological and catchment parameters, and it is due to these parameters that the differences in variability can be seen. Mean Cv for Rainfall and Runoff for Realization Mean CV Rainfall Runoff Figure 7.14: Comparison of Mean Coefficient of Variation for Rainfall and Runoff for Realization 1 66

75 Results and Discussion 7.4 Runoff Coefficients A report by the Water and Rivers Commission (1996) observed that only a relatively low proportion of rainfall becomes streamflow in the wetter areas of the South-West region. Typical proportions are about 10 to 20%. The runoff coefficient depicts the relationship between rainfall and runoff. It is simply a measure of the percentage of runoff resulting from rainfall. Figure 7.15 shows the runoff coefficient for Realization 1 on the vertical axis against days. The coefficient has been calculated for 20 blocks over the simulation period. Each of the 20 blocks represents runoff coefficient averaged over 5 years. The model shows that approximately 5% of runoff occurs as a result of rainfall, although this figure exhibits a downward trend in the runoff coefficient over time. This indicates that catchment properties such as vegetation and soil moisture content and hydrological characteristics such as evaporation play an important role in influencing the generation of runoff mechanisms. 5% of runoff resulting from rainfall implies that 95% of rainfall is lost either through interception, evapotranspiration or infiltration into the deeper groundwater layer. It appears likely that interception and evapotranspiration may be primarily responsible for reducing the amount of rainfall available for runoff. Keeping in mind that LASCAM assumes vegetation density remains constant, a decrease in rainfall results in less water falling into the catchment over the 100 year sequence. However, the same rates of evapotranspiration and interception by vegetation apply over the simulation period. A simple water balance calculation shows that an increasing proportion of rain will be lost through these factors over time, as confirmed by the diagram. 67

76 Results and Discussion Trend of Runoff Coefficient for Realization Runoff Coefficient Day Figure 7.15: Runoff Coefficient for Realization 1 Figure 7.15 also demonstrates that a greater proportion of runoff will result from winter rainfall. This outcome is consistent with previous findings. One note of interest is that the variability in the runoff coefficient (the up-down cycles) reflects the natural variability of rainfall produced from the ENSO events as predicted by the rainfall model. 7.5 Flow Duration Curves Flow duration curves convert the cumulative frequencies of all daily streamflows into a percentage of time that the flow equals or exceeds that flow value. The shape of the flow duration curve gives a good indication of a catchment s characteristic response to its average rainfall history. A steep curve indicates variable flow usually from small catchments with little storage where the streamflow reflects directly the rainfall pattern, while a flat slope indicates little variation in flow regime, which is the resultant of the damping effects of large storages (Hornberger et al. 1998). Figure 7.16 depicts the flow duration curve for observed discharge from 1984 to The figure shows that the observed flow for 1993, 1996 and 1999 had a relatively flat slope compared to the 2001 curve. This indicates that there was very little variation in flow characteristics. By 2001, this 68

77 Results and Discussion Figure 7.16: Flow Duration Curves for Observed Discharge Figure 7.17: Flow Duration Curves for Realization 1 69