CHAPTER 5 WATER DEMAND FORECASTING Related Strategy Objective: Adopt Integrated Resource Planning (IRP) (Determine and review annually a 2,5 and 10-year demand target goals). Scope The purpose of this chapter is to identify the factor affecting water demand, the general data requirements of various forecasting methods as well as the relevance of the various methods to the application of water conservation measures. Definitions, Terminology, Abbreviations A water demand forecast is a conditional prediction, it is a statement of an expected future condition (i.e. level of water use) provided that certain assumptions prove correct. The assumptions, therefore, are as important to the end result as the method. The various methods of forecasting water demand are described in 5.3. 5.1 Principles The provision of water supplies requires engineering intervention to the natural hydrologic cycle, which could be the construction of a dam on a river or drilling of a borehole into an aquifer. This water must then be treated and transmitted to those areas where the demand exists. The quantity of water supplied to an area for a particular period should be such that there is an equilibrium between the supply and demand. A state of non-equilibrium can be reached when demand exceeds supply or if there is a failure in the supply source with the resultant occurrence of water shortages. To restore the situation to equilibrium either the augmentation or storage, treatment and/or transmission facilities are required or, alternatively, the demand could be restricted, for example, by the introduction of conservation measures. The former solution requires the planning, design and construction of storage, treatment and transmission facilities to supply the required quantity of water. This is a slow process usually requiring a considerable amount of time before the increment in capacity can be provided. Therefore, decisions must be made in terms of of demand in the future. A forecast that estimates a low demand will lead to insufficient capacity being provided, or alternatively if the estimated demands are too high, considerable investment costs would be incurred in providing water facilities with unused excess capacity. Sophisticated design methods and sizing techniques for the design of water installations can only be considered as accurate as the forecast method used, which in turn is related to the accuracy of the data and the set of assumptions upon which the forecast is based. Historically, the most commonly adopted water demand forecasting method has been the single coefficient method which regards water use as a need which must be met by the supply system s capacity and with an associated high level of reliability. Examples of the single coefficient method are per capita, per connection and unit use coefficient. The assumption central to this method is that all the circumstances surrounding the consumption of water remain constant, for example the per capita or unit consumption methods water use at a future date by multiplying and extrapolated per capita or unit consumptions by a forecast future population. The extrapolation of the curve into the future assumes that the same trends will persist in the future. Losses are not separately considered. This method is inappropriate to the
consideration of WC/WDM measures and integrated resource planning (IRP) and should therefore not be used. Forecasting methods that regard water use as a need, which must be met by the supply, are unable to take into account the effects of price changes and conservation measures. These can generally lead to over-investment or underinvestment in water facilities, depending on the variables omitted from the model. 5.2 Approached Methodology To ensure that the forecasting process is executed properly, the following four steps need to be taken: (i) Determine the objective of the forecast and the moment when it is needed. In this way information is gathered about the necessary details, the amount of raw data involved and also the required accuracy. (ii) Determine the time horizon involved in the forecast. (iii) Choose a suitable forecasting technique. When choosing a suitable forecasting model, a number of factors need to be considered: the cost involved in the model; the accuracy demanded; the relevance of historical data; the projection horizon; and patterns occurring in the historical data. (iv) Gather and analyse relevant data, after which the forecast can be prepared. If not, review the method, assumptions and validity of the data and adapt the process to arrive at a revised forecast. 5.3 Determinants of water demand 5.3.1 General Multiple coefficient methods of water demand forecasting require a set of explanatory variables that reflect assumptions regarding causality, thus minimising the unexplained variance in the dependent variable, water use (i.e. for demand models). There are three characteristics that a variable should have if it is to be included in the water demand-forecasting model and they are as follows: It must be plausible as a cause of water use. The relationship between the variable and water use must be known or discoverable. Measurements of the variable must be reasonable available to potential users of the forecasting model.
5.3.2 Water Use/Consumption A water use forecasting system for developed communities includes the following demographic, socio-economic and climatic factors that effect water demand: Resident and seasonal population. Number, market value and types of housing units. Employment in service industries. Manufacturing employment and output (i.e. industrial water use) Water and wastewater prices and rate structures. Irrigated areas in residential and commercial/institutional use. Personal income. Climate (i.e. arid or humid) Weather conditions. Water-using appliances. Conservation activities The following factors influence water use and consumption in developing communities: The distance the consumer has to travel to the water source of the availability of water, Income, Education, Religion/culture, and Housing, Difference in the availability of water could also divide the population into two groups: Those households with a water connection inside their houses, and Households dependent upon public water sources outside the house (for example standpipes). Water requirements for food production should include the following major categories: Protein - Meat: The total amount of water required to produce 1 kg of meat Water requirements for meat consists of drinking-water requirements, slaughtering and feed production. - Milk: Per litre of milk produced including feed, processing and packaging assumptions are required as to the percentage of water needs are supplied by irrigation to produce feeds. Carbohydrates: - Bread: Amount of water required to produce a loaf of bread. - Maize: Amount of water required producing a kg of maize.
Assumptions are required as to the percentage of water needs are supplied by irrigation to produce the carbohydrates. Vegetables. The water requirements are linked to the type of vegetable. Liquid beverages. Litres of water to litres of beverage produced. Water requirements for consumer goods include the following: Energy (i.e. l/kwh) Paper (i.e. l/kg of paper) Textiles and clothing (i.e. l/kg of textiles) The water requirements for industry and commerce are usually determined from the number of employees and/or the unit mass of production. 5.3.3 Losses and Inefficiencies All real / physical losses and many types of apparent / non-physical loss can vary with the passage of time. Without any positive intervention by the body responsible for managing the system and its users, the infrastructure ageing process - effects of corrosion, ground movement, wear and tear etc. - combined with a passive approach to addressing non-physical losses, will result in a steady increase in losses. Each component of loss or water use may be seen as having an "as new" or lowest use under ideal conditions lower limit, and an upper limit that reflects an extreme laissezfaire or anarchic scenario. In the later case the total loss and inefficiency between the system and its users becomes resource limiting and the system rations itself according to its hydraulic configuration and the locations and magnitudes of demands. In demand forecasting however we start from the supposition that any additional supply needed will be provided and the upper limit is therefore not resource limited, in theory at least. An upper limit on infrastructure losses may be gauged by comparing the levels of loss existing in the oldest or poorest condition parts of the system with those of the newer parts, if flow data are available. Of course the factors affecting the losses in the older systems will not be the same as the newer parts, for example different pipeline materials, but it provides a useful baseline. In terms of consumer use, or more correctly abuse, taking municipal water supply as the example, an upper consumption limit is represented by the amount of water that can physically be drawn from the reticulation system. Obviously under that condition, if all consumers were to act in unison, the draw will be hydraulically limited even if not resource limited, so it is not simply a matter of multiplying the number of connections by a service pipe flow rate. Having an awareness of an upper limit is useful, however, in ensuring that sufficient cognisance is taken of the impact of a passive approach to water loss control. Particularly in the context of WC/WDM planning using an IRP approach, it is a good starting point to examine the "do nothing" WC/WDM scenario and make appropriate "high" projections before then looking at the impact of various active management options (refer Ch 6 for methodology). A good example to illustrate the dynamics of losses is to look at leakage in a reticulation network. We can have three levels of control: 1. no control 2. "passive" control 3. "active" control
Passive and active control methods are described in section 3.3.3 of the WSI guidelines manual. The impact of these three levels is illustrated diagramatically in Fig 5.1. no leakage control passive leakage control active leakage control Leakage unsustained intervention sustained intervention 0 10 20 30 40 50 60 Time, years Fig 5.1: Simplified Relationship Between Leakage and Pipe Age Box 5-1: Case Study City of Cork, Ireland A water conservation and network management project undertaken during 1997-99 included the sub-division of the existing 10 large trunk main areas into 47 metered districts (zones), average just under 1,000 connections per zone. The mean system age of the zones varied from 10 to over 60 years. The estimated losses in each zone before starting active leakage control were plotted against mean zone system age as shown in the graph below. The zones were also sub-divided into pressure bands to give a family of curves of loss/age/pressure. 7.0 6.0 5.0 Pressure >60m Pressure 51-60m Pressure 41-50m Pressure <41m All values Expon. (Pressure >60m) Expon. (Pressure 51-60m) Expon. (Pressure 41-50m) Expon. (Pressure <41m) Expon. (All values) m3/km.hr 4.0 3.0 2.0 1.0 0.0 0 10 20 30 40 50 60 70 80 Years The relationship between the loss (in this case under mainly passive control pre-project) and age is quite clear, as is the effect of pressure. A similar profile of loss/age was evident after leakage control, but with a flatter rate of rise.
The rate of rise of losses to be adopted in demand for the "do nothing" scenario will depend on the starting level and should be determined on a case by case basis in the light of local circumstances. In the particular case of leakage, the UK water industry's major work on the subject, Managing Leakage (ref), recommends a default value for the natural rate of rise of leakage (NRRL) under a passive control regime of 2.0 litres/hr per property per year. Lambert et al, in (ref), in reviewing a number of sources, report a wide variation in NRRL from almost nil to over 20 l/hr per property. In the same reference, Hirner quotes statistics from a German city - following a major defective pipe replacement programme, average leakage rose from a low of 3.5 l/hr at the rate of 0.4 l/hr per connection for each of the first 2 years, and 0.1 l/hr per connection per year for the next 4 years. Garrett (ref) made a study of xxx leakage zones under passive leakage control in South West Water (UK) and derived the relationship: NRRL = -0.015.Ln.SL + 0.01.Ln(L/N) where SL = starting rate of leakage l/hr.connection L/N = length in metres divided by number of connection For L/N in the range 10 to 40 this worked out at range 0.8 to 3.6 l/hr per property using the South West Water starting leakage levels. As a general guide, typically it can be expected that losses will rise at anything between 1.5% and 5% per year for an existing system under a passive control regime. If the system is being extended to cater for new centres of demand, then the losses in the new system should have a lower starting level but the rate of rise will also apply and needs to be allowed for. Under an active control regime, losses may be expected to rise at a lower rate, say between 1% and 2.5% per year, and from a lower starting level represented by the initial loss reduction campaign. But if active control measures are not sustained then the system will of course revert to its "natural" level, quite quickly. This is also illustrated in Fig 5.1. 5.4 Data availability Forecasting methods are usually chosen according to the forecast application, the characteristics of the study area, and the type and quantity of data which are available given appropriate data collection effort. Deficiencies in the data are a precursor to establishing deficiencies in the forecasting method, therefore, the proviso is that the various categories of data (i.e. levels) are clearly related to the various forecasting methods as illustrated in Table 5A.
Table 5A: Levels of data available required per forecasting method Level Data Availability Applicable forecasting methods Level 1 Water use data Readily available aggregate production and customer data Per capita methods Demographic data Aggregate data available from public records, such as Census reports Per connection methods Other data Qualitative descriptions of service area characteristics and trends Unit use coefficient methods (aggregate only) Level 2 Water use data As in Level 1, except data on industrial water use separately available Unit use coefficient methods (limited desegregation) Demographic data As in Level 1 Other data As in Level 1, plus climatic, demographic, economic and other aggregate data available from public records Level 3 Water use data Water use and customer data desegregated by user sector Multivariate requirements methods (limited variable list, limited desegregation) Demographic data As in Level 1, plus additional data on family sizes, housing characteristics, etc. Multivariate requirements method (limited variable list) Other data As in Level 2 Level 4 Water use data As in Level 3 Demographic data Comprehensive data is available regarding household numbers, sizes, compositions, including of key variables. Multivariate requirements method Other data Comprehensive socio-economic data is available, including employment characteristics and, water prices, family incomes, home values, commercial activities, etc. Demand model methods EJ/dg:CMA CHAPTER 5. Water demand forecasting Ch 5 : Water Demand Forecasting
5.5 Forecasting Methods: There is no single forecasting method, which is suitable for all applications. Applications include: Short-term used for operational planning, distribution system planning or revenue projections. The more common long-term used for capital improvement programmes or sources development plans. Water use forecasting models that are appropriate to WC/WDM and IRP approach include: (i) Multiple coefficient methods: These methods employ two or more explanatory variables. Multivariate requirements models These models incorporate more than one explanatory variable but exclude those based on econometric analysis of water use. Variables included are those which are observed to be significantly correlated with water use. Data collection efforts are considerable and must be balanced against potential improvements in the accuracy and usefulness of the forecast. Multivariate demand models: These models are based on economic reasoning and include only variables, which are causally related to and found to be significantly correlated with water use. Economic variables such as price income are included. These models permit the evaluation of the effectiveness of water conservation measures. The number and nature of explanatory variables can vary dependent on the application, data availability, required accuracy and local conditions. (ii) Probabilistic methods: One method is the contingency tree method which provides a means for considering uncertain factors in a water use forecast. A base forecast is prepared using one of the above methods. Possible sources of uncertainty regarding future water use levels are identified and subjective probabilities and subjective probabilities are estimated for each postulated outcome. The base forecast is modified to reflect the effects of all possible combinations of the uncertain factors, one combination at a time, and the joint probability of each of the combinations is associated with the forecast water use expected to result from that combination. (iii) Disaggregated water use : Disaggregated water use forecasting separately specifies water use for each sector, season or region, utilising the best available model for each type of water use. This method permits the use of explanatory variables unique to a given type of water use, and generally yields a more accurate composite forecast. In all of these methods, the impact of potential WC/WDM measures must be modelled. The characteristics of the various forecasting methods are given in Table 5B. EJ/dg:CMA CHAPTER 5. Water demand forecasting Ch 5 : Water Demand Forecasting
Table 5B : Comparison of Forecasting Approaches Facilities consistent with guidelines Facilities evaluation of water conservation measures Suitable for preliminary or reconnaissance studies Suitable for project planning applications Per Capita Single Coefficient Methods Per Connection Unit use Coefficient No No No No Multiple Coefficient Methods Requirements Model Demand Model Probabilistic Methods Contingency Tree Yes Yes Yes No, too complex No, too complex No, too complex No No Quantity of data needed Very little Very little Moderate to large Difficulty of obtaining needed data Low Low Low to moderate Forecasts should produce the following: Moderate large Moderate high to to Moderate large Moderate high to to Yes Yes Yes Depends application on Depends upon application Seasonal water use, maximum monthly use and average daily water use (for the design of surface water impoundments). Maximum daily water use for the design of treatment and conveyance works. Maximum 10-15 minutes water use, (or failing that, maximum hourly water use) and maximum daily water use for the design of water distribution systems and supply water mains. Average daily water use and average monthly water use for revenue estimates and operational planning. Water use can be desegregated in almost any way which water use itself can be broken down. The most common disaggregations are according to user sector (residential, commercial, industrial and so on) and geographic sub area. Losses must also be d as previously explained. Box 5-2 : Case History Magalies Water. The main deliverable of the project was computer based Strategic Planning System which will enable Magalies Water to evaluate scenarios for the development of its physical assets, viz abstractions works, waterworks, pumpstations, pipelines and reservoirs as well as to ascertain the resources (financial, water and human) required to support those assets. The system was developed in the form of several computer programs, computer models and spreadsheets. Data, assumptions and previous studies upon which the planning process was based were reported upon. The development of the dynamic Strategic Planning System for Magalies Water provides it with a friendly and flexible system for the preparation and updating of strategic plans. An initial strategic plan was developed with the aid of this GIS based dynamic system in conjunction with an extensive review of most of the relevant previous studies. The infrastructure that was expected to be required for the periods up to 2005, 2010 and 2020 were also determined. EJ/dg:CMA CHAPTER 5. Water demand forecasting Ch 5 : Water Demand Forecasting
Three growth scenarios were used for water demand. The scenarios were based on optimistic, moderate and pessimistic assumptions of prevailing political and economic conditions as follows: Scenario 1 - low rate of increase Scenario 2 - moderate rate of increase Scenario 3 - high rate of increase Average daily flows were determined as a product of the unit water demands for a particular scenario and a peak factor function was used to establish peak hourly flows from these average demands. The analysis of historic consumption records as well as the electronic logging of a sample of large water meters was undertaken to facilitate the determination of water demand profiles and peak demand functions. Where no meters existed a level of living index was used to determine theoretical water demands. The response of the existing infrastructure to these demands forecasted for the various time horizons was established together with the additional infrastructure requirements. Costs of An important consideration when selecting a suitable forecasting technique is the required accuracy of the forecast. However, accuracy is closely allied to the cost of the forecast. Figure 5.1 shows the relationship between cost and accuracy of water demand forecasting. Figure 5.2 : The relationship between accuracy and cost Total Cost Optimal region Cost Increases Econometric Models Regression and correlation Operating cost due to inaccurate forecast Sophisticated statistical models Simple statistical models Cost involved in implementing and maintaining forecast Intuitive methods Accuracy Decreases References and Suggested Further Reading Boland JJ,Moy W, Steiner RC, Pacey JI, (1983) Forecasting Municipal and Industrial Water Use: A Handbook of Methods. John Hopkins University. Gardiner V & Herrington P. (1986). Water Demand Forecasting. Geo Books UK. EJ/dg:CMA CHAPTER 5. Water demand forecasting Ch 5 : Water Demand Forecasting
Garrett. Managing Leakage, UK Water Industry. WRC 1994. Driving Down Leakage (conference). London, September 1999 Johnson E.H. (1993). Water Demand Forecasting for urban areas. Water Sewage and Effluent Vol 13 No 1. Lambert, Myers, Trow. Managing Water Leakage - Economic and Technical Issues. Financial Times Energy, 1998 Thompson B (1996). Comments on Water Demand and Population Growth by Schutte & Pretorius. Water SA Vol. 24 No. 3. Schutte C.F. & Pretorius W.A. (1997). Water Demand and Population Growth. Water S.A. Vol No 23. Twiss B.C. (1992). Forecasting for Technologists and Engineers A practical guide for better decisions. Management of Technology Series 15. Peter Peregrinus Ltd. Van der Walt JJ, Le Roux RM & Johnson EH (2000). The integration of GIS and hydraulic network analysis A dynamic strategic planning tool WISA Biennial Conference Sun City. Van Schalkwyk (1996). Guidelines for the estimation of domestic water demand of developing communities in the Northern Transvaal. WRC Report No 480/1/96. Managing Leakage (1994). UK Water Industry WRC. EJ/dg:CMA CHAPTER 5. Water demand forecasting Ch 5 : Water Demand Forecasting