Characterizing Groundwater Dynamics Using ...

1 downloads 0 Views 360KB Size Report
In western Victoria, south east Australia, the land clearing during European ... short groundwater response times and the completion of most clearing long time ...
Characterizing Groundwater Dynamics Using Transfer FunctionNoise and Auto-Regressive Modeling in Western Victoria, Australia Yohannes Yihdego 1, John A. Webb 1 1Environmental Geosciences, La Trobe University, Melbourne, Australia

Abstract In western Victoria, south east Australia, the land clearing during European settlement has had significant impacts on the water table to rise. On the other hand, water table have been declining for at least the last 10-15 years, and this is attributed to the consistently low rainfall for these years, but over the same period of time there has been substantial change in land use, with grazing land replaced by cropping and tree plantations appearing in some areas resulting to decline the water table. The aim of the present study is to model groundwater level fluctuations, relating groundwater levels to precipitation excess, and the impact of landuse change could be assessed. To this purpose, a specific type of transfer function-noise model, the Predefined Impulse Response Function In Continuous Time (PIRFICT) model and an auto-regressive model, Hydrograph Analysis: Rainfall and Time Trends (HARTT) are applied to quantify the relative influence of climatic variable and land use change on the course of groundwater level and to characterize the groundwater system. In PIRFICT modelling, the spatial differences in the groundwater system are determined by the system properties, while temporal variation is driven by the dynamics of the input into the system. It became evident that the no evapotranspiration effect assumption of the HARTT becomes a serious drawback in areas with shallow groundwater levels, because of significant seasonal evaporation component as identified in PIRFICT modelling. However the average results of estimated trend using PIRFICT models and the HARTT models differ only slightly, despite having clear different theoretical stating point. A PIRFICT model was found a better way of estimating the non-climate trend because the climatic variables (rainfall and evapotranspiration) are accounted for. Most of the groundwater level fluctuations are explained by climatic variables and non-climatic trend (90%). The average trend is -0.029 m/yr which is insignificant. Though, the trend estimate from bores screened in Port Campbell Limestone is relatively significant (-0.3 m/yr), it was mainly due to groundwater pumping from the nearby production bores for irrigation purpose. The results are validated and cross checked using estimated parameters with the physical knowledge of the area. From this study there is no indication that the groundwater table was rising/falling due to changes in landuse, at least not during the observation period. From HARTT analysis, the impact of recharge was felt with no delay period (less than a month lag time) on the course of groundwater level. In continuity with this, PIRFICT modeling further identified how long the impact already felt stays in the system (estimated 5.7 years on average). Therefore, an explicit effect of massive clearing on the water table could not be detected across the study area, owing to the short groundwater response times and the completion of most clearing long time ago. This will have an implication for mitigation of dry land salinity in particular and water resources management in general. However, such short response times disagree with the currently assumed slow-reacting groundwater flow system concepts. The lag times and processes identified in this study are related to time needed for the groundwater storage to move to a new state of hydrologic (physical, pressurerelated) equilibrium. This will improve our knowledge of the hydrogeology and thus groundwater flow systems across the Victorian Volcanic plain. Keywords: Trend, system response, time-series modeling, groundwater dynamics, Time-series modeling, Spatio-Temporal Modeling, Hydrology, system response, groundwater dynamics, land use, climate variable

Introduction There has been land clearing of native vegetation across much of the western Victoria, Australia since European settlement since 160 years ago resulting increase in recharge, consequently the groundwater level rises. On the other hand, water table have been declining for at least the last 10-15 years, and this is attributed to the consistently low rainfall for these years, but over the same period of time there has been substantial change in land use, with grazing land replaced by cropping and tree

BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

1

plantations appearing in some areas. Hence, it is important to determine the relative effect the climate and land use factors on the water table changes. Fresh groundwater is a valuable resource while saline groundwater may be a threat to natural resources. In both cases, monitoring and interpreting changes in groundwater levels is essential for management in indicating the degree of threat faced to agricultural and public assets. Hydrographs show changes in groundwater levels over time and are often the most important source of information about the hydrological and hydrogeological conditions of aquifers. The pattern of waterlevel change in a hydrograph is governed by physical characteristics of the groundwater flow system, the rainfall pattern and the interrelation between recharge to and discharge from an aquifer. Water level changes in a hydrograph can also be caused by other management options such as extraction, irrigation and land use change. Information about hydrological systems and water table fluctuations is important for water management and is needed to assess choices for long-term water management policy. Knowledge about the spatio-temporal dynamics of the water table is important to optimize and balance the interest of economical and ecological land use purposes (Von Asmuth and Knotters, 2004). In hydrology, water table dynamics are modelled in several ways. Many authors refer to transfer function-noise (TFN) models to describe the dynamic relationship between precipitation and the water table depths (Box and Jenkins, 1976; Hipel and McLeod, 1994; Tankersley and Graham, 1994; Van Geer and Zuur, 1997). Basically, these models can be seen as multiple regression methods, where the system is seen as a black box that transforms series of observations on the input (the explanatory variables) into a series of the output variable (the response variable). The parameters of time series models address the temporal variation of the water table depths, while the spatial component can be accessed by regionalizing the outputs using ancillary information related to the physical basis of these models (Knotters and Bierkens, 2000). This approach can be used to describe the spatio-temporal variation of the water table depths. It is assumed that the spatial differences in water table dynamics are determined by the spatial variation of the system properties, while its temporal variation is driven by the dynamics of the input into the system. To link the response characteristics of the water table system to the dynamic behaviour of the input, Von Asmuth et al. (2002) presented a method based on the use of a transfer function-noise model in continuous time, the so-called PIRFICT-model. An important advantage of the PIRFICT model as compared to discrete-time TFN-models is that it can deal with input and output series which have different observation frequencies and irregular time intervals. The dynamics of the groundwater system in the western Victoria, mainly on the basalt plain, have been modelled to determine the climatic influence in water table fluctuations. Previously, linear regression analysis was used to estimate trends in individual bores in the study area and thereby predict areas most at risk from shallow or rapidly rising groundwater (Pillai, 2003, Leblanc, 2007). The aim of the present study is to quantify the relative influence of landuse and climate variable on the groundwater course in Glenelg-Hopkins Catchment Management Area (GHCMA). We applied two different types of model, with different theoretical starting-points. First, a specific type of transfer function-noise (TFN) model, the Predefined Impulse Response Function In Continuous Time (PIRFICT) model is applied. Precipitation and potential evapotranspiration are incorporated as exogenous variables into the model being the most important driving forces of water table fluctuation. Using the TFN model, the noise component of the models is used to identify any trend due to unknown cause including landuse change. In this study, a standardized computer package Menyanthes (Von Asmuth et al., 2002) was used for quantifying the influence of climatic variables on the groundwater level, statistically estimating trends in groundwater levels and identify the properties that determine the dynamics of groundwater system. This method is optimized for use on hydrological problems and is based on the use of continuous time transfer function noise model, which estimates the Impulse response function of the system from the temporal correlation between time series of groundwater level and precipitation surplus. Secondly, we applied an auto-regression model. As the historical landuse data are not readily available, the conclusions were based on which the fluctuation of the groundwater level can be explained by climatic variable, rainfall, using HARTT. We calibrated both types of models on 82 time series of water-table depths. As the historical landuse data are not readily available, the conclusions are based on which the fluctuation of the groundwater level can be explained by climatic variables. The results were validated and cross checked using estimated parameters with the physical knowledge of the area.

BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

2

Materials and methods Data Climatic data Rainfall data is available on a daily and monthly basis from rainfall gauges record for the Bureau of Meteorology. Rainfall records were chosen mainly on the length of the record with continuous data for a majority of the period 1960 to date and to be in a suitable location in relation to the monitoring bores. The numbers of recorded and estimated (patched) evaporation station data were obtained from the SILO website (www.nrm.qld.gov.au/silo/). Groundwater data Most of the groundwater bores were installed after 1994 corresponding to the period of below average rainfall. The groundwater data are from the Victorian Water Resources Data Warehouse. A set of 82 bores were selected. The selection criteria included consideration of monitoring period, aquifer monitored, spatial distribution, bores not affected by abstraction and data reliability. Time series series of observed water table depths of 6 to 33 years length are available from 82 wells in the GlenelgHopkins watershed (Fig.1). The surface elevation of the bore ranges from 36 to 437 m.a.sl while the depth to water table ranges from near surface to 56 m below surface. Study area The study area is predominantly the Glenelg-Hopkins catchment located in sub-humid region of southeastern Australia (Fig.1). The region experiences a Mediterranean climate, with hot, dry summers and cool, wet winters. Average annual rainfall in the area ranges from 500 - 910 mm, with higher rainfalls typically occurring in coastal regions. Most of the study region is dedicated to agriculture and regionally there has been a significant shift from grazing to cropping in recent years. These agricultural practices have replaced native grasslands since the 1930s and as a result less than 20% of the original vegetation remains in the region (Ierodiaconou et al., (2005); Ierodiaconou et al., (2008)). Land use in the Glenelg–Hopkins region has changed substantially over the last two decades, 4.84% from 1980 to 1995, and 13.85% from 1995 to 2002 (Ierodiaconou et al., (2005). Groundwater flow patterns have been altered following clearing of deep-rooted native vegetation and groundwater extractions. The region is considered to be one of the areas most at risk from rising water tables and dryland salinity. The study area comprises mainly the Victorian Volcanic Plain. The volcanic plains of western Victoria are topographically subdued with an average elevation of ~200 mAHD, gently increasing to the northeast. Drainage is typically poor with many ephemeral swamps/lakes and several volcanoes rise (eruption point) are found above the plain (Fig.1). The basement geology of the area consists of Early Palaeozoic volcanics and turbidites and Silurian sandstone, intruded by Devonian granites. These basement rocks outcrop over limited areas, and are overlain by an extensive cover of Cenozoic basalt, colluvium and alluvium. Underlying the Newer Volcanic Basalts (NVB) in the centre of the catchment are Miocene and Pliocene ligneous clays, sands and gravels (Calivil Formation), called deep leads, and representing the pre-existing stream system incised deeply into the highly weathered early Cainozoic palaeo-surface. The disruption of the drainage system by the basalt flows formed lakes across the surface of the basalt plain. The Port Campbell Limestone is a major aquifer in the southern west of the study area, mostly confined beneath the Newer Volcanics Basalt aquifer.

BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

3

Figure 1. Study area watershed and location of the observation wells ( )

The PIRFICT-model (TFN modeling) The behaviour of linear input-output systems can be completely characterized by their impulse response (IR) function (Ziemer et al., 1998; Von Asmuth et al., 2002). The response of water table depth to impulses of precipitation series can be modelled by a transfer function-noise (TFN) model (Box and Jenkins, 1976; Hipel and McLeod, 1994; Von Asmuth and Knotters, 2004). For water table depths, the dynamic relationship between precipitation and water table depth can also be described using physical mechanistic groundwater flow models. However, by using much less complex TFN models predictions of the water table depth can be obtained which are often as accurate as those obtained by physical mechanistic modelling (Von Asmuth and Knotters, 2004). The basic idea behind TFN modelling is to split the observed series (output) into a sum of transfer components related to known causes (inputs) that influence the temporal variation of the output and an unknown noise component. TFN models are often applied to distinguish between natural and maninduced components of groundwater series (Van Geer and Zuur, 1997). In TFN models one or more deterministic transfer components and a noise component are distinguished. These components are additive. Each transfer component describes the part of the water table depth that can be explained from an input by a linear transformation of a time series of this input. The noise component describes the autoregressive structure of the differences between the observed water table depths and the sum of the transfer components. The input of the noise model is a series of independently and identically distributed disturbances with zero mean, and finite and constant variance, i.e., white noise. The PIRFICT-model, introduced by Von Asmuth et al., (2002), is a specific type of TFN models and an alternative to discrete-time TFN models. In the PIRFICT-model a block pulse of the input is transformed into an outputseries by a continuous-time transfer function. The coefficients of this function do not depend on the observation frequency. The following single input continuous TFN model can be used to model the relationship between water table dynamics and precipitation surplus/deficit. For the simple case of a linear, undisturbed phreatic system that is influenced by precipitation surplus/deficit only (Von Asmuth et al., 2002): * h(t) = h (t) + d + r(t) (1) (2) (3) where: h(t) = observed water table depth at time t [T]; * h (t) = predicted water table depth at time t attributed to the precipitation surplus/deficit, relative to d [L]; d = level of h*(t) without precipitation, or in other words the local drainage level, relative to ground surface [L]; r(t) = residuals series [L]; p(t) = precipitation surplus/deficit intensity at time t [L/T]; θ(t) = transfer Impulse Response (IR) function [-]; (tφ= noise IR function [-]; 2 W(t) = continuous Wiener white noise process [L], with properties E{dW(t)}=0, E[{dW(t)} ]=dt, E[dW(t1)dW(t2)]=0, t1 ≠ t2. TFN models are identified by choosing mathematical functions which describe the IR and the autoregressive structure of the noise. θ(t) is a Pearson type III distribution function (PIII df, Abramowitz and Stegun, 1964). Because of its flexible nature, this function adequately models the response of a broad range of groundwater systems. Under the assumption of linearity, the deterministic part of the water table dynamics is completely determined by the IR function moments. After the selection of an IR function that represents the underlying physical process, the available time series have to be transformed to continuous series. First, in order to characterise the variability of precipitation and evaporation, we rely on a simple but effective method to estimate the average precipitation surplus intensity and its annual amplitude. Finally the accuracy and validity of the model are checked using the auto and cross-correlation functions of the innovations, the covariance matrix of the model parameters and the variance of the IR functions. The PIRFICT-model was applied in this study because the model can describe a wide range of response times with differences in sampling frequency between input series and output series. For the western Victoria situation, it is particularly interesting because different behaviours of water table can be found even in small catchments due to geological complexity. to explain the variation in actual vs simulated groundwater level due to climatic variables, precipitation and evapotranspiration are incorporated as exogenous variables input into the model at daily time scale, being the most important BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

4

driving forces of water table fluctuation and what is left unexplained is attributed to unknown causes including land use change.

The HARTT (Auto-regressive modeling Ferdowsian et al., (2001) presented a statistical approach for analysing hydrographs, called HARTT (Hydrograph Analysis: Rainfall and Time Trends). The method is able to distinguish between the effect of rainfall fluctuations and the underlying trend of groundwater levels over time. Rainfall is represented as an accumulation of deviations from average rainfall and the lag between rainfall and its impact on groundwater explicitly represented. Two forms of accumulative residual rainfall were used and compared. The first was accumulative monthly residual rainfall (AMRR): t

AMRRt = ∑ ( M i , j − M j )

(4)

i =1

where Mi,j is rainfall (in mm) in month i (a sequential index of time since the start of the data set) which corresponds to the jth month of the year, M j is mean monthly rainfall (in mm) for the jth month of the year, and t is months since the start of the data set. The second was accumulative annual residual rainfall (AARR): t

AARRt = ∑ ( M i − A / 12)

(5)

i =1

where A is mean annual rainfall (in mm). Because A is a constant, the fluctuations in Mi are not moderated as they are for AMRR, so AARR has higher within-year fluctuations. In other words, the AMRR variable tends to have relatively low within-year fluctuations because, in calculating AMRR, the fluctuations in actual rainfall tend to be offset by seasonal variation in average monthly rainfall. For this reason, it was expected to be well correlated with data from bores with shallow groundwater levels which typically have seasonally fluctuating watertables. For both AMRR and AARR, construction of the variables was based on data sets pre-dating the earliest recording of depth to groundwater. This allowed long lag effects of rainfall on groundwater to be detected, if they occurred. Lags of up to a few years were investigated. In order to draw conclusions about the relationship between groundwater trends and rainfall records, the HARTT method (equation 6) was used. The HARTT method uses the residual rainfall described above and the lag between rainfall and groundwater dynamics to define a correlation between climate and groundwater fluctuations: Deptht=k0+k1×AMRRt-L+k2×t (6) where Depth is the depth of groundwater below the ground surface, t is the months since observations commenced, L is the length of time lag (in months) between rainfall and its impact on groundwater, and k0, k1 and k2 are parameters to be estimated. AMRR is the accumulative monthly residual rainfall. Parameter k0 is approximately equal to the initial depth to groundwater, k1 represents the impact of above/below average rainfall on the groundwater level, and k2 is the trend rate of groundwater rise/fall over time. The rationale for using this model to explain groundwater levels is as follows. Prior to clearing of native vegetation in southeastern Australia, it is presumed that groundwater tables were in long-run equilibrium, meaning that average rainfall equalled average evaporation and discharge from a catchment. Deviations of rainfall levels from the average level would have resulted in short-term fluctuations in the groundwater level, centred around the stable long-run equilibrium level. Following clearing of native vegetation, rates of recharge increased, introducing an upward trend to the groundwater level. The t variable in equation (6) captures this upward trend, while AMRR captures the short run fluctuations around that trend. Similarly, the t variable captures the downward trend due to tree plantation as well. In cases where it produces models with a higher r2 (most of which are shallow bores), AARR is substituted for AMRR in the regression model. The value of L was estimated separately for each bore by selecting the value that resulted in the highest R2 for the regression. Thus L does not necessarily represent the lag until either the first impact or the largest impact of rainfall on watertable depth, but the lag that produces the highest statistical correlation. In many cases, L is longer than the first detectable impact.

BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

5

Simulating water depths PIRFICT modeling Time series models using precipitation surplus/deficit as input variable, calibrated on time series of water table depths with the observed period. These models contain a dynamic component, describing the dynamic relationship between the input and the output, either physically or empirically. But variation of the water table cannot be completely explained from the precipitation and evapotranspiration series. So, the models must contain a noise component, which describes the part of water table fluctuation that cannot be explained with the used physical concepts or empirically from the input series. The unexplained part (noise component) has to be taken into account in the simulation procedure, since we are interested in detecting trend. Here, the following steps are followed: The present study focuses on separation of the natural and man-induced components of the groundwater fluctuations in the western Victoria. The fluctuations are first modelled with a two input, single-output stochastic linear transfer function model, relating groundwater levels to precipitation excess. Any other influence on the groundwater table will form part of the residual terms from the time series analysis. This appears to be an effective technique to decompose measured groundwater level into various components. As a second step, groundwater levels are modelled using a triple-input, single-output transfer model, where the third input time series represents artificial influences of either landuse change or the cumulative effect of groundwater withdrawal. To model the groundwater level fluctuation using the climatic variables and estimate the unexplained variance due to unknown causes including land use change. Land use change will result in a trend in the residuals, when the other explanatory variables are accounted for. By using a trend as an explanatory variables, it was aimed to detect and quantify its effect in the groundwater level not due to climate variables.

HARTT The time trend identified in HARTT is the groundwater level trends if we had mean monthly rainfall of the selected rainfall period. The impact of rainfall shows how groundwater levels fluctuate because we do not have mean monthly rainfall. However, It required re-analyzing again with out impact of rainfall or time trend, if either of the parameter k1 (impact of rainfall) and K2 (underlying time trend) is statistically insignificant (p value is >0.05) respectively. HARTT finds the best delay that results in the highest explanatory power (R2). A review of the results with a range of lag estimates was carried out to choose the best that has high R2 and is meaningful.

Result PIRFICT modeling Results of the 82 time series models are summarized in table 1, where the minimum, median, and maximum values of the parameters are given, along with their 95% confidence interval. The median value of R2 points out that the fit of most models is good, although there are clearly outliers in the results, as the extremes of most parameter estimates are not within a range that is physically plausible. Most of the groundwater level fluctuations are explained by climatic variables and nonclimatic trend (90%). Low percentages might be caused by errors in the data or lack of data, or possibly other inputs that affect the groundwater dynamics are not incorporated into the model. Some problems with the calibration were diagnosed by checking the impulse response function for each well. After several calibrations, minimum RMSE and RMSI values were the found for most of the wells. Comparison of R2 for double- and triple-input models shows the improvement that was obtained by applying a triple-input model (i.e. trend as a third input variable) component.

BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

6

Table 1 Range of model results and parameter estimates for all 82 ground water head series (PIRFICT modelling)

Parameter

Minimum (±2σ)

Median

Maximum (±2σ)

R2adj (%) RMSE (m) RMSI (m) M 0,p (m/mm) f

23 0.04 0.04 0.25 (±0.12)

90 0.184 0.150 1.5

99 5 3 32 (±4.32)

0.02 (±0.32)

0.5

1 (±0.2)

D (m)

63

179

411

Response time (yr) =3*M1/M0 Non-climatic Trend (m/yr)

0.6 (±0.12)

8.42

5059 (±234)

-0.58 (±0.002)

-0.04

0.03 (±0.14)

R2adj = Percentage of explained variance; RMSE = Root Mean Squared Error (meters); RMSI = Root Mean Squared Innovation (meters); σ = standard deviation; D = Drainage Base (meters); f = evaporation factor. The evaporation factor (f) is the factor by which Menyanthes multiplies the values of the evaporation series from Meteorological Institute which are supposed to represent evaporation by short grass that has no shortage of water supply. It shows that most of the time the actual evaporation was less than the reference evaporation but with significant evaporation factor (0.5) (Table 1).The plausibility checks to guide in assessing whether the results of the transfer model are physically realistic include the model residuals, the evaporation factor f, the local drainage base d, and the moments of the impulse response functions and their standard deviations. The model residuals, uniting all factors that are not accounted for by the model, are an important aid in identifying possible unknown stresses that may be a source of model distortions. Non-random patterns of the residuals in space or time reveal the fact that there are still stresses missing in the model. The patterns themselves often give enough information to pinpoint the nature and location of the missing stresses. Regarding the drainage base, an estimate that is too low or too high may be caused by the influence of stresses that are not incorporated in the model or that are not well quantified. This can easily happen with stresses that do not show pronounced dynamics, such as more or less constant seepage or withdrawal rates. In addition, the moments of the impulse response functions of the different stresses provide relevant information. Moments can be used to characterize the functioning of the ground water system and can be related to its geo-hydrologic properties (Von Asmuth and Maas 2001; Von Asmuth and Knotters 2004). In contrast to physical parameters that are defined only in the context of a certain schematization, moments are related to common statistical terms and are more generally applicable. The zero-order moment Mo of a distribution function is its area and M1 is related to the mean of the impulse response function. The relation is M1/Mo. It is a measure of the system’s memory. It takes approximately 3 times the mean time (M1/Mo) for the effect of a recharge to disappear completely from the system. Autocorrelation functions of the innovation series of the models were also evaluated. Autocorrelation generally falls within the indicated confidence limits, and the innovation series can be accepted as being white noise. First, autocorrelation of the resulting innovations was examined, as the innovation series is expected to be white noise. Next, the correlation matrix of the estimated parameters was analysed, to verify independence of the separate models: correlation between parameters of different model components should be minimal. Only 52 time series modelling out of 82 were considered as a reliable trend estimate using a diagnostic criteria mentioned above to cross check the plausibility of the result. HARTT Changes in groundwater trends may be due to either changes in climate or changes in the process governing recharge. Using HARTT, the effect of unusual/atypical climatic conditions were identified from the observed hydrograph; the resulting residual time series thus represents the combined effects of both changes in recharge processes and non-average climatic conditions. BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

7

Correlation technique namely lagging was undertaken in an attempt to display an underlying trend. This produced mixed results which have proven inconclusive for this study, and as such, only the results of these analyses are presented in table 2. Lagging of the water level measurements (up to 12 months) against rainfall totals were conducted for all aquifer units. Changes in water levels due to high rainfall events may not be immediately visible in the hydrographic records due to physical constraints such as aquifer depth or permeability of overlying layers. However, the testing exercise did not produce greater indicators of correlation (higher r2 values) than obtained using data from the same monthly period (0 lag). Overall, the model fitted the data well, explaining 79% (median value of R2) of variation in groundwater level using the climatic variable (rainfall) left with less significant trend (-0.066 m/yr, on average) (Table.2). However, some limitations was identified with the HARTT method, in shallow bores, where the explanation of water depth less than 5 m requires the addition of further variables to the model, because of vicinity to the discharge zone; evaporation rate and hydraulic gradient are affecting the groundwater level fluctuations. The HARTT may therefore be compromised by factors such as geological and geographical characteristics of the monitored sites. Table 2 Range of model results and parameter estimates for all 82 ground water head series (HARTT)

Parameter

Minimum

Median

Maximum

R2 K0 (m) K1 (m/mm) Lag time (month) K2 (m/yr)Time trend

0.1 0 0 0

0.79 11 0.80 0

0.97 56 5.8

-0.335

-0.066

0.216

11

To analyse the result it needs to check the significance of each variable. The p-value indicates the level of significance of each variable by assessing whether the means and measure of dispersion of two variables are statistically different from each other. If the p-value is less than 0.05 then the variable is significant. If the trend (K2) is not significant then the rate of rise/fall is uncertain. Also, if the rainfall variable (K1) is not significant then there is lack of surety regarding the effect and delay period. Parameters with values which were not significantly different from zero were omitted to avoid model redundancy. Similarly, if the delay period estimate was not meaningful, like unrealistic lag estimate from the physical existing knowledge of the area for bores located near eruption points which were identified as areas of preferential recharge (Raiber et al., 2009; Bennetts et al., 2003), it implies that the effect of rainfall was not well simulated and hence the trend estimate was not reliable either. The estimated K0 was cross checked with the actual water table depth (usually initial water table depth of the observation period. Any significant deviation indicates uncertainty in the output parameter and the trend estimate was not acceptable. From the HARTT analyses, only 44 out of 82 bores modelling were considered as a reliable trend estimate using the procedure mentioned above to cross check the plausibility of the result.

Trend The auto-regressive model (HARTT) showed that the groundwater course can be explained by climate variable (79%) supporting the less significant non-climatic trend identified by PIRFICT model implying the land use change categorized among the non-climatic trend had no significance sign of influence on the groundwater level at least for the observation period. Both HARTT and PIRFICT modelling shows a downward trend, 43 out of 44 and 38 out of 52 respectively, indicating the effect of land clearing (supposed to result in upward trend) had already felt long time ago. Therefore, the drought has likely masked recent affects of tree plantations on groundwater discharge. Those time series PIRFICT modellings that were not considered as reliable trend estimate were subjected to further analysis to correlate with the depth to water table, which indirectly shows the thickness of the unsaturated zone which is often a reason for the none linearity of the system. However the data did not support this. This implies that the water table fluctuation was governed by pressure from distant recharge instead of direct infiltration. This is in agreement with the very short lag BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

8

time (Table 2) identified in HARTT and fast response time of the system identified in PIRFICT modelling (Table 1). Range of evaporation factor was employed to test how sensitive the result is to this choice. It showed that the range of evaporation factor falls within the estimated confidence interval which gives confidence on the trend estimate. The overall average trend estimated from PIRFICT modeling and HARTT show comparable result. However, the correlation between the trend estimate by PIRFICT modelling and HARTT is not strong (Fig.2).

Figure 2. Correlation of trend estimated using PIRFICT modelling (Menyanthes) and HARTT The deviation is significant for shallow bores because HARTT did not incorporate the effect of evaporation. It became evident that the no evapo-transpiration effect assumption of the HARTT becomes a serious drawback in areas with shallow groundwater levels, because of significant seasonal evaporation component as identified in PIRFICT modelling (Table 1). Hence, the method adopted by HARTT to predict water table depths using precipitation as input did not meet the linearity of the system because there are nonlinear relationships between water table depth and precipitation due to the presence of evaporation in shallow groundwater levels. Moreover, the trend identified by PIRFICT modeling refers to non-climatic trend, since both climatic factors (rainfall and evaporation) were accounted for as an explanatory variables and what was left includes effect of land use change in the residual. So its trend should better reflect land use changes. In this study the trend estimate by PIRFICT modeling was chosen for further analysis to derive relationship between different aquifer units (Table 3). Table 3 Statistical analysis results and model parameter estimates (median value) based on aquifer units

Aquifer

Non-climate trend (m/yr)

Response time (yrs)

R2 (%)

NVB Deep lead PCL All

-0.027 -0.029 -0.30 -0.039

5.6 12.3 3.4 5.7

89 94 90 91

Bores screened in The Newer Volcanic basalt (NVB) aquifer shows insignificant trend and relatively fast response and bores screened in the deep lead aquifer shows insignificant trend and slow response (Table 3). The slow response of the deep lead aquifer is in agreement with the previous conceptualization of the deep lead aquifer system in which the deep lead predominantly does have a regional groundwater flow system (up to about 60 km) and as a result it travels long way before it forced to move upward due to geological constriction (Raiber et al., 2009; Dahlhaus et al., 2002). There was a significant trend in bores screened in PCL compared to bores screened in NVB and deep lead aquifers. These bores might have been affected by nearby groundwater pumping for irrigation use, resulting in a reduction in groundwater level in the last decade (Fig.3) as the bores lie within the BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

9

Condah region irrigation area where water table is close to surface. Also over extraction of water from the underlying aquifers (namely the Clifton Formation) could contribute somehow to the trend. Another evidence for the effect of abstraction is that the bores estimated with significant trend have a drainage base estimate a bit high. The deviation of the drainage basin estimate from the minimum groundwater level gives additional information what other factors including the abstraction might contributed to the fluctuation of the groundwater level beneath the estimated drainage base, as drainage basin is referred groundwater level that eventually reaches long time no recharge. The higher value estimate of drainage base identified indicates groundwater level below the drainage base was not related to recharge reduction but most likely due to groundwater pumping. In addition, the moments of the impulse response function of precipitation provided relevant information that supports the effect of groundwater withdrawal on the course of groundwater level from these bores. Also, the physical plausibility of the results of a TFN model can be judge, by checking the IR functions. It is equivalent to the cross-correlation function. We check if the memory of the hydrological system, indicated by the time lag where the IR function approximates to zero, is covered by the monitoring period (De Gruijter et al., 2006). Most of the bores screened at PCL aquifer have short period of record. Some of the bores have only 6 years of observation period and the trend estimation might not be reliable as such if the length of the groundwater series is shorter than, approximately twice the time it takes for a unit recharge to disappear from the system which ranges from 0.7 to 9 yrs for the PCL bores, implying trend estimate is considered still acceptable for most of them. On the other hand the Port Campbell Limestone (PCL) aquifer shows fast response (Table 3). The fast response could be explained due to karstic features which allow fast water movement. Analysis of the trend rate can provide insight to the type of recharge mechanisms which are taking place across the catchment. Apart from detecting non-climate trend which is the main objective of the study, the approach developed as an attempt to remove all known effects of climate, leaving behind a residual stress could help to assess recharge processes and to display any changes in recharge.

Figure 3 Spatial distribution of estimated trend (from PIRFICT modelling) with respect to landuse map (2002) (Ierodiaconou et al., (2005)) and production bores. Groundwater usage/Withdrawal (usage 2006/7) data is courtesy of Southern Rural Water Trends in groundwater levels need to be assessed in the context of the absolute groundwater at any individual site. A zero trend could indicate equilibrium conditions through recharge equal to lateral groundwater flows or leakage down to an underlying aquifer. However, a zero trend in an already high water table environment most likely indicates equilibrium through a significant component of discharge being via evapotrasnpiration. These two cases need to be recognized when interpreting trends and any interference with respect to recharge and discharge. The average non-climatic trend in the groundwater levels was -0.04 m/yr. This is indicative of relatively steady of the system indicating it reached equilibrium at that time. The fast response time of the system (Table 1) supports this. Apparently, changing land use due to replacement of native vegetation by short grass vegetation BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

10

since 1860 caused the groundwater table to rise. It is unfortunate that most groundwater level series start only after 1991. Since the climate was so atypical since then, it may be hard to find information on a systematic rise of the groundwater table in the foregoing period. It is necessary to have long sequences of measurements (>20 years) to reliably detect long-term trends over the effects of drought and wet years. Identifying long-term trends in the study area data set is difficult, because only 7% of the bores have records longer than 20 years, and the current drought that begins in 1997 often imparts a non-representative negative trend to the data. Removal of the drought-stricken years leaves only a 10-15 years record from which to deduce the long-term trend. One way to approach this problem was to select the longer series. However there was no indication that the groundwater table was rising due to changes in land use, from the time series modelling carried out using the longer series available in the study area. In regions of southeast Australia where native vegetation included deep rooted trees, land clearance resulted in rising groundwater levels due to increased groundwater recharge (Allison et al., 1990). From this study, there is no indication that the groundwater table was rising/falling due to changes in landuse, at least not during the observation period. The recent multiyear drought has reversed this groundwater level rising trend inherited from land clearing, with observations in the Murray-Darling Basin showing that on average the water table has declined by ~1 m between 2001 and 2008 (Leblanc et al., 2009). Interpretation of the lag time (Rancic et al., 2009) postulated that the estimated lags between standing water level and rainfall consist of up to three components: Tr, the time needed for recharge to infiltrate and reach groundwater. Te, the time needed for the portion of the storage that recharges mainly via vertical flux to reach equilibrium. If a storage that had maintained a balance between inflow and outflow suddenly starts receiving additional inflow, it reacts by increasing its level until the outflow matches the new inflow and a new level is achieved. Te is not a constant, as the system needs more time to adapt to a large change than to a small one. Tl, the time needed for the pressure impulse to laterally propagate down the system, from the highest recharge point to the lowest discharge point. This component is pronounced in complex systems (intermediate or regional), where recharge areas are mainly spread over several eruption points, rocky terrain and thin soil cover. Each of the three components can contribute to the less variable lags within the study area. Shorter lags characterise the study area. Therefore, the recharge response happens faster and Tr is smaller. The size of the system is reflected through the Te component, as small systems move to a new equilibrium faster than large systems, which have more inertia. The longer lag times in few bores might be a consequence of the larger, intermediate systems found in these areas. In the complex intermediate or regional systems the flux propagates laterally through a sequence of unconfined and confined units. The response displayed by most hydrographs at the study area is clearly seasonal. It reflects maximum recharge during the winter months, when rainfall exceeds potential evaporation. The time lag estimated ranges from 0 to 11 months (average less than a month (0)) however the time lag estimated is statistical significant not the max impact or impact begins (Ferdowsian et al., 2001). Bores which takes more months to respond, indicates slow and diffuse recharge though the low permeability soil and sediment horizons. However the estimated time lag does not have any defined trend with depth to water table. HARTT analysis (using moving average) captures the overall system lag. The lag times suggested by the HARTT analysis are indicative of maximum time lag for the groundwater system as a whole, behind rainfall. There is a less variation between the different parts of the catchments in response times. Lags that are presented in this study should be seen as approximate and informative, not as accurate values. Even so, they are no more than one thirtieth of the longest lag (11 month) that are currently conceptualised under the Groundwater Flow System framework (Coram et al., 2000). Such short response times disagree with the currently assumed slow-reacting groundwater flow system concept. In this study, boreholes with the shortest time lags did not necessarily correspond to areas with shallow water tables. This implies that a rise in the potentiometric surface does not imply necessarily that infiltrating water has reached the water table; water tables may rise simply due to pressure transferral from hydraulically connected areas where recharge has occurred. For example, seasonal potentiometric surface variation occurs in boreholes where the groundwater is thousands of years old and no modern recharge is evident; seasonal potentiometric surface variability must be due to pressure transferral rather than direct addition of recharge to the groundwater. BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

11

The response time identified by PIRFICT modeling gives new information on the dynamical memory of the system, which is a continuity of the lag time identified by HARTT. The average estimated system response (memory to disappear) is 5.7 years. Apparently, there is no relation between groundwater age and the dynamical memory of the system. Groundwater age depends on how groundwater moves through the system while the dynamical memory identified by PIRFICT modelling depends on how pressure moves through the system. The rate of groundwater flow in the study area is quite slow and it can take hundreds or even thousands of years for the groundwater to travel only a few kilometres. This is reflected in the groundwater ages; the oldest NVB groundwater in the region are ~ 22, 000 years old while in underlying deep lead system it ranges up to ~ 35,000 years (Raiber et al., 2009; Bennetts et al., 2006). Although the groundwater moves slowly, changes in groundwater pressure can be transmitted rapidly. This means that within a few time an increase in the height of the water table at a high elevation end of the catchment can cause an increase in flow at a low elevation area tens of kilometres away. This is supported by the response time identified in PIRFICT modelling.

Impact of rainfall The impact of rainfall estimate from the paramater output using PIRFICT modeling (Mo,p) and HARTT (K1) show comparable result. These two parameter used for quick scenario calculations on the effect of rainfall on groundwater level as the multiplication of Mo,p and K1 and anticipated rise/fall in precipitation yields a prediction of the groundwater level rise/fall. The average M0,p (precipitation) is 1.5 m (Table 1), which means that a precipitation of 1 mm/yr will eventually lead to a ground water level rise of 1.5 m on the location. Of course, it requires to subtract the evaporation rate in order to get the 'real' stationary groundwater level. The average evaporation (1.004 m/yr) scaled by the estimated evaporation factor 0.5 (Table 1) equates to a groundwater level rise of 0.9 m/yr. Average value of K1 identified in HARTT (Table 2) is 0.8 m rise/low in groundwater level for 1mm of rainfall. The discrepancy between the estimated rates of rise/fall on groundwater level by PIRFICT modelling and HARTT could be due to the averaging of values and outliers, besides to their inherent methodological difference. However it still gives an insight on the range of estimate for future prediction. On the other hand, if k1 (impact of rainfall) is statistically insignificant it implies that the course of groundwater level is controlled by the underlying time trend. This will have an implication from water resource management point of view in general and from dry land salinity in particular.

Conclusions The analysis of groundwater trends is critical in the study of salinity risk and the effectiveness of preventative measures. In the study area, land clearing and drought have the opposite affect: land clearing results in rising water table levels, whereas the drought results in declining water table levels. Tree plantations and drought have the same affect: a reduction in groundwater recharge rates. The affects of the drought in the last decade coincide with increased tree plantations in the Glenelg Hopkins CMA region. Two programs for modelling of bore hydrographs were employed for estimation of the effect of climate variables and human intervention (including land use change), characterization of groundwater system and prediction. PIRFICT modelling accounts for climate variations (rainfall and evaporation), so its trend should better reflect land use changes than HARTT. Moreover the input variables in PIRFICT modeling were at daily time scales which allowed capturing the seasonal groundwater variation and improve the response time of the system. The general conclusion from this study is that lowering of groundwater levels can be explained by meteorological causes and non-climate causes (including groundwater withdrawal and land use change). Most of the groundwater level fluctuation is explained (R2=90%) by meteorological variables and non-climate trend. The average non-climate trend is less significant -0.027 m/yr) and -0.029 m/yr for the Newer Volcanic Basalt (VNB) and deep lead aquifers respectively but is relatively significant (0.30 m/yr) for the Port Campbell Limestone (PCL), however this trend is largely related to groundwater pumping from nearby production bores spatially located in the intense irrigation area. From HARTT analysis, the impact of recharge was felt with no delay period (less than a month lag time) on the course of groundwater level. In continuity with this, PIRFICT modeling further identified how long the impact already felt stays in the system (estimated 5.7 years on average). This will improve our knowledge of the hydrogeology and thus groundwater flow systems across the Victorian Volcanic plain. Therefore, an explicit effect of massive clearing on the water table could not be BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

12

detected across the study area, owing to the short groundwater response times and the completion of most clearing long time ago. This will have an implication for mitigation of dry land salinity in particular and water resources management in general. It will greatly help the impacts of land management change be more reliably predicted and allow both the best management of groundwater resources and salinity.It is important to understand that the lag times and processes identified in this study are related to time needed for the groundwater storage to move to a new state of hydrologic (physical, pressurerelated) equilibrium. Groundwater flow systems in the study area show much faster responses to changed recharge conditions than former estimates implicitly included in the previous Salinity Audit and current estimates as conceptualised within the GFS framework (Coram et al., 2000), which was used in the National Land and Water Resource Audit and the Australian Dryland Salinity Assessment 2000 (Natural Heritage Trust 2001).

Acknowledgements This work was conducted in collaboration with and funded by the Glenelg Hopkins Catchment Management Authority, Victoria, Australia. We acknowledge the assistance of Menyanthes developers, Kees Maas and Jos Von Asmuth from The Netherlands.

References Abramowitz, M., Stegun, I.A., 1964: Handbook of mathematical functions, Dover Publications Inc, New York, 1964. Allison, G.B., Cook, P.G., Barnett, S.R., Walker, G.R., Jolly, I.D., Hughes, M.W., 1990: Land clearance and river salinization in the western Murray Basin, Australia, Journal of Hydrology, Vol.119, 1990, pp.1–20. Bennetts, D.A, Webb, J.A, Gray, C.M., 2003: Distribution of Plio-Pleistocene basalts and regolith around Hamilton, wester Victoria, and their relationship to groundwater recharge and discharge. Proceedings from Advances in Regolith, CRCLEME, Perth, Australia, 2003, pp.11–15. Bennetts, D.A, Webb, J.A, Stone, D.J.M, Hill D.M., 2006: Understanding the salinisation process for groundwater in an area of southeastern Australia, using hydrochemical and isotopic evidence, Journal of Hydrology, Vol.323, 2006, pp.178–192. Box, G.E.P., Jenkins, G.M., 1976: Time series analysis: forecasting and control, Holden-Day, San Francisco, 1976. Coram, J.E., Dyson, P.R., Houlder, P.A., Evans, W.R., 2000: Australian groundwater flow systems contributing to dry land salinity, report by Bureau of Rural Sciences for the Dry land Salinity Theme of the National Land and Water Resources Audit, Canberra, 2000. Dahlhaus, P.G., Cox, J.W., Simmons, C.T., Smitt, C.M., 2008: Beyond hydrogeologic evidence. Challenging the current assumptions about salinity processes in the Corangamite region, Australia, Hydrogeology Journal, Vol.16, 2008, pp.1283-1298. Dahlhaus, P.G., Heislers, D., Dyson, P., 2002: Groundwater Flow Systems. Glenelg-Hopkins Catchment Management Authority report No. GHCMA 02/02,Victoria, Australia, 2002. De Gruijter, J.J., Brus, D.J., Bierkens, M.F.P., Knotters, M., 2006: Sampling for natural resources monitoring. Springer-Verlag, Berlin, 2006. Ferdowsian, R., Pannell, D., McCarron, C., Ryder, A., Crossing, L., 2001: Explaining groundwater hydrographs: Separating atypical rainfall events from time trends, Australian Journal of Soil Research Vol.39, 2001, pp.861-875. Hipel, K.W., McLeod, A.I., 1994: Time series modelling of water resources and environmental systems, Elsevier, Amsterdam, 1994. Ierodiaconou, D., Laurenson, L., Leblanc, M., Stagnitti, F., Duff, G., Salzman, S., Versace, V., 2005: Consequences of land-use change on nutrient exports: a regional scale assessment in Southwest Victoria, Australia, Journal of Environmental Management, Vol.74, No.4, 2005, pp.305-316. Knotters, M., Bierkens, M.F.P., 2000: Physical basis of time series models for water table depths, Water Resour. Res., Vol.36, 2000, pp.181-188. Leblanc, M., 2007: An investigation of the risk to natural assets from groundwater salinisation, Monash and Deakin universities and Glenelg-Hopkins Catchment Management Authority, Project: NAP272, 2007. Leblanc, M.J., Tregoning, P., Ramillien, G., Tweed, S., Fakes, A., 2008: Basin scale, integrated observations of the 21st Century multi-year drought in southeast Australia, Water Resources Research, doi:10.1029/2008WR007333, 2009.

BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

13

Natural Heritage Trust, 2001: Australian Dryland Salinity Assessment 2000: Extent, impacts, processes, monitoring and management options, National Land and Water Resources Audit, Canberra, Australia, 2001. Pillai, M. (ed.), 2003: Groundwater monitoring for environmental condition and salinity management in the Glenelg-Hopkins region. ISBN 1 74146 000 X, The state of Victoria, Department of primary Industries Research Victoria, 2003. Raiber. M., Webb. J.A., Bennetts. D.A., 2009: Strontium isotopes as tracers to delineate aquifer interactions and the influence of rainfall in the basalt plains of southeastern Australia, Journal of Hydrology, Vol.367, 2009, pp.188-199. Rancic, A., Salas, G., Kathuria, A., Acworth, I., Johnston, W., Smithson, A. and Beale, G., 2009: Climatic influence on shallow fractured-rock groundwater system in the Murray-Darling Basin, NSW, ISBN 978 1 74122 871 7, DECC 2009/108, Department of Environment and Climate Change NSW, Australia, 2009. Tankersley, C.D., Graham, W.D., 1994: Development of an optimal control system for maintaining minimum groundwater levels, Water Resources Research, Vol.30, 1994, pp.3171-3181. Van Geer, F.C., Zuur, A.F., 1997: An extension of Box-Jenkins transfer/noise models for spatial interpolation of groundwater head series, Journal of Hydrology, Vol.192, 1997, pp.65-80. Von Asmuth, J.R., Bierkens, M.F.P., Maas, C., 2002: Transfer function noise modeling in continuous time using predefined impulse response functions, Water Resources Research,Vol.38, No.12, 2002, pp.23.1-23.12. Von Asmuth, J.R., Knotters, M., 2004: Characterising groundwater dynamics based on a system identification approach, Journal of Hydrology, Vol.296, 2004, pp.118-34. Von Asmuth, J.R., Maas, C., 2001: The method of impulse response moments: a new method integrating time series, groundwater and eco-hydrological modelling. In: Impact of human activity on groundwater dynamics, Geherls, J.C. Peters, N.E., Hoehn, E., Jensen, K., Leibundgut, C., Griffioen, J., Webb, B., Zaadnoordijk, W.J. (Eds.), pp.51-58, Wallingford: IAHS Publication, 2001. Ziemer, R.E., Tranter, W.H., Fannin, D.R., 1998: Signals and systems: continuous and discret, Prentice-Hall, Upper Saddle River, 1998.

BALWOIS 2010 - Ohrid, Republic of Macedonia - 25, 29 May 2010

14