The Institution of Engineers, 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW Development of Stochastic Multisite Rainfall and Urban Water Demand for the Central Coast Region of

Peter J. Coombes1, Mark A. Thyer2, Pavel Kozaroski 3, Andrew Frost4, George Kuczera5 and Ian Grimster6

1 ARC Post Doctoral Fellow, School of Engineering, University of Newcastle, Callaghan 2308. 2 Hydrologist, Tonkin Consulting, Adelaide, South Australia. 3Hydrologist, Snow Mountains Engineering Corporation, North 4 PhD student, School of Engineering, University of Newcastle, Callaghan 2308. 5Associate Professor, School of Engineering, University of Newcastle, Callaghan 2308. 6 Headworks Engineer, Wyong Councils’ Water Authority

Abstract:

A novel approach for concurrently simulating daily catchment rainfall, streamflow and urban demand for water supply headworks modelling of the Central Coast Region of New South Wales is presented. A methodology for simulating daily multi-site rainfall based on rainfall at a single site generated using the DRIP (Disaggregated Rectangular Intensity Pulse) rainfall model by Heneker et al. [2000] was developed. This approach satisfactorily reproduces observed daily, monthly, annual rainfall statistics at the multiple sites. This demonstrates it is able to capture the inter-annual persistence and spatial variability that exists within the Central Coast water supply catchments. Rainfall series generated by the multi-site method are used in a non-parametric regional water demand model that incorporates water balances in households to estimate daily water demand in the Central Coast region of New South Wales. No metering data was available for this study other than daily water demand at the water treatment plants. Nonetheless the regional demand method was able to adequately estimate regional water demand including the day to day variation and strong seasonal trends of water demand in the Central Coast region.

Keywords: Synthetic rainfall, multi-site, urban water demand, headworks security, socio-economics

1. INTRODUCTION extractions on the ecology of streams daily water The Central Coast region of New South Wales demand and headworks models were required. includes the Gosford and Wyong local Many authors including Maidment et al. [1985], government areas. The region is situated Kuczera and Ng [1994], Zhou et al. [2000] and between the Sydney and Lower Hunter regions Coombes et al. [2000; 2002] have established with the adjacent to the Sydney that urban water demand is dependent on region and extends to Lake seasonal and climatic variables. To accurately Macquarie in the North. The Gosford-Wyong determine the impact of urban water demand on Councils’ Water Authority (GWCWA) supplies streamflow the correlation between climatic water to the Central Coast region. In the year conditions, domestic water use and streamflow 2002 the region had an estimated population of must be maintained. In particular the correlation 300,400 people with a growth rate of 1.37% per between hot dry conditions, high domestic water annum that is expected to slow to 0.24% per use and low stream flow and vice versa must be annum in 2030. preserved because it is likely to have significant The region’s water supply headworks system impact on the viability of streams and the presently includes a major surface reservoir on reliability of water supply headworks systems.

Mangrove Creek with additional water harvested The spatial variance of urban water demand is downstream from Mangrove Creek’s residual often ignored in analysis of water supply catchment, , Ourimbah headworks systems in preference to the use of Creek and Wyong catchments. Mangrove total regional demand. Maidment and Miaou and Mooney Mooney Creeks discharge to the [1986] established that water use in different . and Wyong locations was dependent on spatially averaged River discharge to . rainfall series, ambient air temperatures and GWCWA are evaluating the ecological established water use patterns. Coombes et al. significance of stream flows in their water supply [2000; 2002] found that urban water demand catchments and the impact of water supply and was spatially dependent on local climatic and demand management measures in urban areas. socio-economic variables. The use of demand In order to evaluate the impact of water and supply management measures such as rainwater tanks and AAA shower roses was The Institution of Engineers, Australia 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW expected to further increase the spatial and day catchment simulation of daily rainfall at three to day variation of water demand. other sites was required. These sites included

A modified version of the network linear program another coastal site at Wyong and two sites for headworks simulation WATHNET by Kuczera situated further inland at Kulnura North and [1992] was employed to analyse stream flows . A methodology was required to and headworks security. To preserve the climatic transfer the DRIP simulated daily rainfall at Gosford to the three other sites. correlation between the urban and water supply catchments the DRIP synthetic rainfall model by A stepwise approach was adopted where at Heneker et al. [2001] was used in a multi-site each step daily rainfall at one secondary site framework to generate daily rainfall series. The was simulated conditional on rainfall data from rainfall series was used at each time step in the a number of other primary sites. Therefore, at headworks model to determine stream flow using the first step Wyong rainfall was based on the a rainfall/runoff model and urban water demand Gosford rainfall, in the second step Kulnura using the non-parametric regional demand North rainfall was conditional on Gosford and method by Coombes et al. [2002]. Wyong rainfall and in the third step Mogo Creek

The formulation of a multi-site synthetic rainfall rainfall was based on the rainfall from the other three sites. model and the subsequent generation of urban water demand are discussed in this paper. Each step in this procedure had two stages. In

2. MULTI-SITE DAILY RAINFALL the first stage the secondary site daily rainfall GENERATION occurrence is simulated using the following conditional probability: In this study a new approach was developed to generate multi-site daily rainfall data within the i (1) GWCWA’s water supply catchments. The P(Rs {S p ,i =1, N p }) approach is described as follows: The DRIP where Rs is the rainfall occurrence (wet or dry) i model was used to generate daily rainfall at a at the secondary site and S p is the daily rainfall single primary site and then a transfer scheme state of primary site I = 1, …, Np primary sites. was applied to simulate daily rainfall at several The daily rainfall state at each of the primary other sites conditioned on the rainfall at the sites was based on daily rainfall depth. This primary site. The DRIP model was chosen approach was used because it was found that because it is an event based rainfall model the conditional occurrence probability was which can reproduce inter-annual persistence strongly dependent on the rainfall depth at the evident in rainfall data from sites such as primary site(s). Five daily rainfall states and Sydney (Frost et al, 2000). There is strong their rainfall ranges were arbitrarily chosen. The evidence of an inter-annual persistence similar simulation results given in the following section to the Sydney data in the rainfall data within the show that this approach was able to Gosford-Wyong water supply catchment (Thyer satisfactorily reproduce the observed daily and Kuczera, 2003). rainfall occurrence probabilities at each of the Recent research has investigated the secondary sites. The scheme summarized in regionalisation of the DRIP model so that it can (1) exploited the observation that there was a be used to simulate rainfall at sites where only low probability (<4%) of rainfall at the two inland daily rainfall data are available for model sites if there was no rainfall at either of the two calibration [Jennings, 2003]. This technique coastal sites. essentially scales the DRIP model parameters Table 1: Daily Rainfall States derived for a key site to reproduce the daily Daily Rainfall State, Daily rainfall depth, rainfall statistics at another target site. In this S y study the key site was the Sydney Observatory 1 y = 0 Hill gauge and the target site was the Gosford 2 0 < y ≤ 5 gauge. The length, accuracy and availability of 3 5 < y ≤ 20 rainfall data at the Observatory Hill gauge 4 20 < y ≤ 50 motivated its selection as a key site The DRIP 5 y > 50 model was able to adequately reproduce the In the second stage the secondary site daily daily rainfall statistics (mean, standard rainfall depth was simulated using the following deviation, probability of a dry day) for the approach: If there was a strong correlation majority of months and the annual rainfall between the secondary and primary site’s daily distribution for Gosford site (Figure 1). rainfall then a regression of inverse normal 2.1 Multi-site Daily Rainfall Transfer Scheme transformed amounts was used to simulate To provide a suitable representation of the secondary site daily rainfall. The details of this rainfall regime within GWCWA’s water supply daily rainfall depth regression model are given The Institution of Engineers, Australia 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW by Charles et al. [1999]. the results were considered to be reasonably

An inverse normal transformation was used on good. the daily rainfall depths because it produced 25 residuals which did not strongly violate the Observed Daily Mean Observed Daily SD assumption in the regression model that the 20 Median Simulated errors are normally distributed. Two sets of 90% Confidence Limits regression parameters, one for the summer ) 15 m m

(Nov. to Apr.) and one for the winter (May to ( ll fa

Oct.) were required to capture the seasonal in a differences in spatial rainfall variability due to R 10 the different weather patterns that produce rainfall in each season. If there was no rainfall 5 at any of the primary sites or the daily rainfall correlation between the secondary and primary 0 sites was low then the secondary site rainfall Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec was sampled from its observed historical Month distribution for the particular primary site rainfall (a) Daily Mean and SD occurrence pattern. 0.8

2.2 Multi-Site Daily Rainfall Simulation Results 0.75 To evaluate the performance of the approach

outlined above 1000 simulations of daily rainfall ) 0.7 y Da

time series with the same length as the y r D historical record were generated for each site. If P( 0.65 the observed values of the relevant statistics Observed P(Dry Day) were within the 90% confidence limits Median Simulated 0.6 90% Confidence Limit calculated from these multiple simulations then the model is considered to be not inconsistent with the observed values. 0.55 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month In this study the relevant statistics compared were the daily and monthly mean and standard (b) Dry Day Probability deviation (SD), probability of a dry day, the 3500 annual distribution and monthly and annual Observed correlations. A full analysis of these results is 3000 Median Simulated

) 90% Confidence Limit m

presented in Thyer and Kozarovski [2001]. m 2500 ( l a t

For all four sites the observed daily means were o

T 2000 ll a

within the simulated 90% confidence limits for a f majority of months. The results for the in 1500 Ra l a

simulated daily SD’s were reasonable, although u

n 1000 n not as good as the daily means. This is likely A because the results for the three other stations 500 are based on the rainfall simulated at Gosford. 0 Therefore any deficiencies in the rainfall .01 .1 1 5102030 50 70809095 99 99.999.99 simulated at Gosford will also be evident at the Percentile other sites. Figure 1 and Figure 2 shows the (c) Annual Distribution results for Gosford and Mogo Creek respectively. Figure 1 – Comparison of observed and simulated statistics for Gosford. The observed dry day probabilities were within the simulated 90% confidence limits for Gosford At all four sites a majority of the observed and Wyong for all months. For Kulnura North annual distribution was within the simulated and Mogo Creek there are some months where 90% confidence limits. Figure 1(c) shows these the observed values fall just outside the results for Gosford. The results for Mogo Creek simulated confidence limits. Similarly for the show that the upper tail of the distribution monthly means and SD’s the observed values seems to be underestimated [Figure 2 (c)]. This for Gosford and Wyong were well reproduced, is likely because the DRIP model was while for Kulnura North and particularly Mogo calibrated to Gosford rainfall from a long record Creek the number of months outside the 90% covering 1886-1992, whereas the observed confidence limits was higher. However, overall record for Kulnura North was from a more recent and comparatively wetter period, 1963- The Institution of Engineers, Australia 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW 1992. Hence, as these simulated values are (1) was conditioned on the previous day’s based on a longer time series, they may rainfall then improved results for the inter-site provide a better representation of the long-term correlations may be possible. rainfall characteristics. The results show that the daily rainfall transfer 15 scheme was able to adequately reproduce the

Observed Daily Mean majority of relevant statistics from the daily, Observed Daily SD Median Simulated monthly and annual time scales for all four 90% Confidence Limit rainfall sites. The results for the two coastal 10 sites at Gosford and Wyong were better than ) m

m the results for the inland sites at Kulnura North (

fall and Mogo Creek. Given the challenge of in

Ra simulating daily rainfall at multiple sites based 5 on the rainfall at a single site in a different climate zone this is considered to be an excellent result.

0 It is reiterated that one of the major advantages Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec of this approach is that the simulated time Month series can reproduce the inter-annual (a) Daily Mean and SD persistence and spatial variability that is evident 0.92 in the Central Coast region. For water resource

0.9 planners this greatly improves the confidence in simulated drought risks. In addition, using a 0.88 stochastic approach eliminated the need to infill 0.86 any missing data which can reduce the spatial y) variability in daily rainfall. Instead, has a

y Da 0.84 r D ( mechanism which exploits the spatial variability P 0.82 in GWCWA’s water supply catchment. This

0.8 methodology represents a significant Observed P(Dry Day) Median Simulated improvement in the previous efforts to simulate 0.78 90% Confidence Limit hydrological time series for this catchment. 0.76 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 3. ESTIMATION OF WATER DEMAND Month A method was required to simulate regional

(b) Dry Day Probability water demand in the Gosford and Wyong local government areas using synthetically generated 1800 daily rainfall data and observations of regional Observed 1600 Median Simulated daily demand taken at the water treatment

) 90% Confidence Limit plant. m 1400 m (

l 3.1 Estimating Household Water Demand

ta 1200 o

T GWCWA does not have a monitoring program ll

fa 1000 that disaggregates household water use into in a indoor and outdoor components. Indeed R l 800 a

u distributed monitoring or metering data for n

n 600 A urban water demand was not available for this

400 study. Accordingly, methods were developed to transfer results from the Lower Hunter region to 200 the Central Coast region that were verified .01 .1 1 5 10 2030 50 7080 90 95 99 99.9 99.99 Percentile using daily bulk water use measured at the water treatment plants and the proportion of (c) Annual Distribution domestic household water use (64%) estimated

from billing records. Figure 2: Comparison of observed and simulated statistics for Mogo Creek. A set of methods to estimate domestic water demand have been developed by Coombes et For the monthly and annual inter-site al. [2000] using monitoring data from the Lower correlations the simulated values are generally Hunter region of New South Wales. These below the observed values. This highlights a methods are based on data from 130 possible deficiency in the approach. One of the households monitored for indoor and outdoor assumptions was that the daily rainfall water use in different socio-economic zones occurrence probability was independent of the over a 12 year period. previous day’s rainfall. If the probability given in The Institution of Engineers, Australia 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW A linear regression was developed to estimate monthly average daily water use results shown daily household water use in the Lower Hunter in Tables 3 and 4 were used in the PURRS zones using climate, socio-economic and water model by Coombes and Kuczera [2001] in use data. The long-term average daily indoor combination with historical climate data to water use for a particular month (inDem) in create resource files of domestic water demand L/day was estimated using: in a variety of dwelling types for use in the inDem = 27.79 + 145.69 P− 0.422 M− 10.579 AveR regional demand model. The dwelling types used in the study are shown in Table 5. + 6.74 AveRdays − 0.162 Inc − 12.28 G + 0.49 AveTemp (1) Table 3. Estimated average daily household water use for the Gosford area where P is the number of occupants in a Average water use (Litres per day) dwelling, M is a seasonal index ranging from 1 Indoor (number of to 6 (the index for January and December is 1, Outdoor occupants) and the index for June and July is 6), Inc is Month 1 2 3 4 5+ average weekly income per person ($), G is annual population growth (%), and for the January 271 151 296 442 588 734 month in question, AveR is long-term average February 254 140 286 431 577 723 daily rainfall (mm), AveRdays is the long-term March 232 147 292 438 584 730 average number of days with rainfall and April 207 141 287 432 578 724 AveTemp is the long-term average daily May 162 137 283 428 574 720 maximum temperature (°C). June 119 133 279 424 570 716 Equation (1) yielded a coefficient of July 121 137 282 428 574 719 2 determination (R ) of 0.81. The long-term August 173 143 289 434 580 726 average daily outdoor water use for a particular September 224 144 290 435 581 727 month (exDayDem) in L/day was estimated using: October 272 148 294 440 586 731 November 292 151 296 442 588 733 exDayDem = − 251 .5 + 7.53M − 11 .3AveR December 331 152 297 443 589 734 − 0.025 Inc − 0.816 AveRdays Table 4. Estimated average daily household + 24 .44G + 19 .08 AveTemp (2) water use for the Wyong area 2 Equation (2) yielded an R value of 0.69. These Average water use (Litres per day) equations show that indoor and outdoor water Indoor (number of use from the Lower Hunter region was is Outdoor occupants) strongly dependent on the climatic and Month 1 2 3 4 5+ demographic variables [Coombes et al., 2000]. Monthly daily average domestic water demand January 317 165 311 456 602 748 for different dwelling types within the Gosford February 302 152 297 443 589 735 and Wyong zones was estimated using climate March 285 163 309 454 600 746 and socio-economic data from the Australian April 256 153 299 444 590 736 Bureau of Statistics [ABS, 2000] (shown in May 188 156 302 447 593 739 Table 2) in Equations (1) and (2). June 136 143 289 434 580 726 Table 2: Climatic and socio-economic data for July 157 150 296 442 587 733 the Central Coast region August 197 152 298 443 589 735 Zone Gosford Wyong September 252 154 300 446 592 737 Average weekly income 293 251 October 309 162 308 453 599 745 ($/person) November 344 161 307 453 598 744 Average daily rainfall 3.67 3.33 (mm/day) December 392 159 305 450 596 742 Population growth rate 1.84 2.35 3.2 Compiling Regional Urban Water (%/year) Demand Average maximum daily 22.9 24.1 The Central Coast region was divided into two temperature (°C) zones with different climatic conditions (rainfall, Average rain days per 12 10 temperature and number of days since last rain month (days) day): namely Gosford and Wyong.

The estimated monthly average daily indoor A method of non-parametric aggregation was and outdoor water demand for the Gosford and used to generate daily domestic water use for Wyong areas for different household sizes is each dwelling type in each climate zone using shown in Tables 3 and 4 respectively. The the historical resource files. The daily water The Institution of Engineers, Australia 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW balance results from the PURRS model was cluster C1 (C1W and C1T), cluster C2 (C2W compiled into daily water use totals and and C2T), house H1 (H1W and H1T), house H2 combined with the historical daily climate data (H2W and H2T), house H3 (H3W and H3T), (rain depth, number of days since last rain day house H4 (H4W and H4T) and house H5 (H5W and maximum ambient air temperature) to and H5T) are selected from the resource file. i create resource files of water demand. If there The domestic demand Demm (ML) for record m are NR years of daily climate data, the resource in zone i is: file has NR replicates for each day of the year. F C1W + F C2W  Table 5 Dwelling types  1 2    Item Occupants Dwellings i + F3H1W + F4H2W Dem m = ()1 − inst m .Pop m .  H1 1 1  + F5H3 + F6H4W  H2 2 1    + F7H5W  H3 3 1 F C1T + F C2T  H4 4 1  1 2  H5 5+ 1  + F3H1T + F4H2T  + inst .Pop .  3 C1 11 4 ()m m ()  + F5H3T + F6H4T  C2 24 9   + F H5T H  7  In the headworks modelling let AveTempm be H where the factors F1 to F7 indicate the the daily maximum temperature (ºC), Rdepthm H proportion of the population (Popm) in a zone be the daily rainfall depth (mm) and DryDaysm be the number of days since last rain event for that occupies a particular dwelling type derived day m for each zone used in the regional from the data shown in Table 6 and the demand model. The climate variables in the installation fraction for rainwater tanks or demand management measures is instm. headworks model were derived using the synthetic rainfall series generated by the DRIP Population data used in this study was provided model. Importantly they are temporally and by the Australian Bureau of Statistics [ABS, spatially consistent with the rainfall and stream 2000], Planning NSW and Wyong City Council flows in the water supply catchments. and is shown in Table 7.

The difference between climate data used in Table 7: Population statistics the headworks simulation and the resource file Number of people ,j Diffm for replicate j of day m in the resource file Year Gosford Wyong for a particular demand zone is shown as 1996 150,220 120,185 follows: 1997 153,024 123,108

j R,j 1998 154,946 125,946 = H − Diffm w1|AveTempm AveTempm | 1999 158,172 129,309 2001 159,412 134,424 R,j + w |DryDaysH − DryDays | 2 m m The total regional domestic demand DomDemm R,j (ML) on day m is: + H − (2) w3 |Rdepthm Rdepthm | where w , w and w are weights, which were  n If Month ≠ Dec, 1 2 3 Demi set to 1 in this study.  ∑ m Jan or Feb i = 1 The dwelling statistics from the Central Coast   n i region used in this study are shown in Table 6. DomDemm = Hol ∑ Dem If Month = Jan m Table 6: Dwelling statistics for the Central  i = 1  Coast region n i If Month = Dec Hol1 ∑ Dem Item Wyong Gosford  m or Feb (4)  i = 1 Houses (%) 85.8 85.8 Units (%) 14.2 14.2 WE.DomDem If Day = Sat or 1 25 23.9  m DomDemm =  Sun 2 35 34.3 DomDem Occupants per  m dwelling by area 3 15 15.2 Otherwise (5) population (%) 4 15 16 where n is the number of zones in the region 5+ 10 10.6 Hol and Hol1 are multipliers of water demand For each day m in each zone the index j in the during summer holiday periods and WE is a j resource file which minimizes Diffm is identified multiplier of water demand during week ends. A and the corresponding values for water use for daily outdoor multiplier Ex is also used in the The Institution of Engineers, Australia 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW calibration. However given that domestic, commercial and

Non-domestic monthly demand for a region industrial metering data was not available for also needs to be accounted for. The simple this study the regional demand model has model of commercial and industrial water use provided an acceptable estimation of regional developed by Kuczera and Ng [1994] was water demand. altered to produce a model of daily non- 170 Predicted demand Observed demand domestic demand defined as follows: ) y 150 da Q = Α + Β Year− yr / IC []() L 130 M ( Pow 110

 2π  nd 1.0+ Amp* Sin ()mth+ Phase  a   m 90

 12  e D

r 70

−C * Rd+ D()T − T Rcount (6) e

ave t

a 50

If Mth = Jan W Hol.QIC  30 QiC = Hol1.QIC If Mth = Dec or Feb 0 300 600 900 1200 1500 Q Otherwise (7)  IC Days since 31 March 1996 where QIC is the non-domestic demand for Figure 3: Calibration of the regional demand month mth and year Year, A and B are model to observed data parameters defining the long-term trend in consumption, Amp and Phase are parameters 170 defining the seasonal cycle, Pow is an ) Predicted demand y 150 a Observed demand exponant used to control the magnitude of d / demand, C is a parameter accounting for the L 130 effect of daily rain depth Rd, T is the daily (M

d 110 maximum temperature, T is the long term n ave a daily average daily maximum temperature and m 90 e Rcount is the count of days since the last day r D 70 with rainfall. te a 50 Because no metered domestic data was W available the non domestic water use model 30 had to be calibrated to the difference between 410 450 490 530 570 610 650 690 730 770 810 850 household and total consumption and accounts Days since 31 March 1996 for about 36% of total mains water use for the Figure 4: Comparison between observed and Central Coast region including unaccounted-for predicted data for the summer of 1998 consumption and losses. The regional demand model implements equations (2) to (7) to Water demand in the Central Coast region was produce total domestic and non-domestic found to be strongly dependent on the climatic demand for the Central Coast region. variables daily rain depth, days since last rain event and daily maximum temperature. A 4. RESULTS selection of model parameter values used to Calibration of the regional demand model to calibrate the regional demand model are shown daily demand data recorded at the central water in Tables 8. treatment plants during the period April 1996 to December 2001 provided a reasonable Table 8: Model parameters agreement to observed demand with an R2 WE Ex Hol Hol1 1.1 0.21 1.3 1.1 value of 0.55 and a SD of ±15.5 ML/day.

The regional demand model was able to Table 8 shows that domestic water demand reproduce the strong seasonal variation in increased by about 10% on weekends (WE). water demand experienced in the Central Coast During the peak holiday period in January region (Figure 3). The monthly and annual regional water demand was estimated to water use volumes were also accurately increase by 30% (Hol). In December and reproduced (R2 = 0.78 and 0.92 respectively). February the holiday period was estimated to Figures 3 and 4 show that the model was able increase regional water demand by 10% (Hol1). to estimate summer peak demands and the day The result for the outdoor water use multiplier to day variation in water demand but in Figure 4 (Ex) of 0.21 is of interest. This indicates that it is shown the model could not fully describe peak outdoor water demands at each dwelling the variation in summer water demand. in the region do not occur simultaneously. The Institution of Engineers, Australia 28th International Hydrology and Water Resources Symposium 10 -14 November 2003 Wollongong, NSW Indeed the results indicate that only 21% of Geophysical Research, 104(D24), 31,657- domestic dwellings contributed to the peak 31,669. 1999. water demand on any day. This result also Coombes P.J., and G. Kuczera. Rainwater tank suggests that not all domestic dwellings will design for water supply and stormwater simultaneously have no outdoor water use on a management. Stormwater Industry low demand day indicating that probabilistic Association 2001 Regional Conference. Port behavioural distribution of outdoor water Stephens, NSW. 2001 demand [see Coombes et al., 2000] will also Coombes P.J., G. Kuczera, J.D. Kalma and need to account for natural variation across J.R. Argue. An evaluation of the benefits of different scales. source control measures at the regional scale. 5 CONCLUSIONS Urban Water. 4(4). London, UK. 2002. This study describes the development of a multi- Coombes P. J., G Kuczera and J.D. Kalma. A site synthetic rainfall series that was used to behavioural model for prediction of exhouse simultaneously estimate stream flow in water water demand, 3rd International Hydrology and supply catchments and water demand in urban Water Resource Symposium, 793-798, Perth, areas. Thus the method was able to maintain a Australia. 2000. correlation between stream flow in the water Frost, A. J., Jennings, S., Thyer, M., Lambert, supply catchment and urban water demand. In M. and Kuczera, G. Incorporating long-term particular the correlation between hot dry climate variability into a short-timescale rainfall conditions, high domestic water use and low model using a hidden state Markov model, streamflow and vice versa is more likely to be Australian Journal of Water Resources, Vol 6, preserved. The methodology will allow improved 1, 63-70. 2002. understanding of the security of water supply. Jennings S., Regionalisation of a high

Multi-site production of synthetic daily rainfall resolution point rainfall model. Ph.D. thesis, using the DRIP model was able to simulate daily University of Adelaide, Adelaide. 2003. rainfall at multiple sites based on rainfall at a Jennings S., Lambert M., Frost A. and Kuczera single site and reproduce the inter-annual G., Regionalisation of a high resolution point persistence and spatial variability that exists rainfall model. Proc. 27th Hydrology and within the Central Coast water supply Water Resources Symposium, IEAust, catchments. The multi-site rainfall method Melbourne. 2002. described in this study has potential to Kuczera, G. Water supply headworks significantly improve the simulation of simulation using network linear programming. hydrological time series in catchments. Advances in Engineering Software. Vol. 14. Nonetheless the method experienced some 55-60. 1992. difficulty in reproducing statistics at the inland Kuczera, G., and W.S. Ng. Stochastic economic sites that may be improved with a different approach to headworks augmentation timing. configuration of the multi-site daily rainfall Research report No. 72. Urban Water conditioning scheme. Research Association of Australia. 1994. Maidment, D.R., S.P. Miaou and M.M. The multi-site rainfall series generated using Crawford. Transfer functions models of daily DRIP was used to generate the climatic urban water use. Water Resources Research. parameters daily rain depth, days since last Vol. 21, 4, 425-432. 1985. rainfall event and maximum temperature that Maidment, D.R., and S.P. Miaou. Daily water were employed in a regional demand method to use in nine cities. Water Resources Research. produce daily water demand in the Central Vol. 22, 6, 845-851. 1996. Coast region. No metering data was available Thyer, M. and G. Kuczera. A Multi-Site for this study other than daily water demand at Framework for Modelling Long-term the water treatment plants. Nonetheless the Persistence in Rainfall Time Series: 2. Real regional demand method was able to Data Analysis, submitted to Journal of adequately estimate regional water demand Hydrology. 2001 including the day to day variation and strong Thyer, M. and P. Kozarovski. Simulating multi- seasonal trends of water demand in the Central site daily rainfall time series in the Gosford- Coast region. Wyong council water supply catchment, 6 REFERENCES SMEC report. 2001. ABS. Regional statistics, NSW. Australian Zhou, S.L., McMahon T.A., and Wang Q.J. Bureau of Statistics Catalogue no. 1304.1. (2001). Frequency analysis of water 1999. consumption for metropolitan area of Charles, S. P., B. C. Bates and J. P. Hughes. A Melbourne. J. Hydrology. Vol 247, 72-84. spatiotemporal model for downscaling precipitation occurrence and amounts, J.