Demand management network:

UK Water resources in the Loddon are under increasing stress and the plan to build a large number of houses in the region over the next 15 years will only add to the pressure. The water supply companies operating in the area are legally obliged to provide a safe and reliable water supply to all houses, so more water will be needed to meet this extra demand. An obvious solution is to increase the supply by installing more boreholes or increasing reservoir capacity, but this is expensive and encounters environmental concerns. An alternative option is to concentrate on reducing domestic demand, which accounts for the bulk of consumption in the region. The question is how best to achieve such a reduction.

An initial stakeholder meeting was held in December 2001, at which a number of stakeholder concerns were raised. Among others, these included:

• Probable expansion of housing in the area, and the effects on the adjacent rural and greenbelt areas. Water management plans have to identify local sources of extra water for new housing, or consider long-distance water transport to meet the extra demand.
• Water companies have a statutory duty to supply domestic water users, but are often constrained by restrictions on abstraction from local sources to meet these requirements.
• Most of the stakeholders present raised concerns about the abstraction management strategy process being non-predictive. For example, the water companies frequently operate on a 25 year planning period, and need to be able to consider aspects such as increases in population, and matching high demand during periods of low supply. There was generally an optimistic response to the idea of using a Bayesian network to complement the abstraction management strategy process, if this could lead to some extra degree of forecasting.

  With a better understanding of individual stakeholder concerns, we were then in a position to develop a network that links these concerns to the need for improved demand management. A number of potential strategies for managing domestic water demand are available. These strategies have been included within a demand management network, which examines the effectiveness of each strategy to reduce consumption, either individually or in combination.

  The overall structure of the draft network is shown in the figure below (although the detailed design cannot easily be seen at this scale). Nodes are drawn as ellipses, which denote different factors, and the arrows show the links between them. The network identifies 4 potential types of action for controlling domestic water consumption: pricing, awareness-education campaigns, grey water reuse, and leak reduction. These actions have been included following discussions with the company responsible for supplying water to most of the catchment. Further refinements to the network have been made at the suggestion of the other two stakeholder groups involved in the network construction.

The section of the network relating to the determination of house numbers in the catchment can be used to illustrate the way in which uncertainty is included as an explicit element of the Bn approach. To estimate total house numbers a GIS survey based on post code areas was undertaken from which a figure of 280,000 dwellings was obtained. Although confident the result was reasonably accurate it was recognised that because of a slight mismatch between post code and catchment boundaries, an error of up to +/- 10% might be involved. Given this uncertainty the network was used to convert the estimated number to a probability distribution, based on our knowledge of the possible scale of error. The result is shown in the following diagram:

From the single figure of 280,000 entered into the node 'NumExHouses', and assuming a normal distribution, the 'GISestimate' node generates a probability distribution based on the information entered into the conditional probability table linking the two factors.

From the distribution, the probability of there being 280,000 houses is 40%, but there is a 1.4% chance it might be as high as 305,000 or as low as 255,000. This distribution is subsequently used as the basis to calculate the number of metered and un-metered houses, and ultimately the total water consumption for the catchment. The explicit recognition and display of uncertainty with available data is one of the strengths of Bns, making them a particularly open and transparent tool with which to work.

Project progress:

Work is continuing with the demand network, while initial stakeholder consultations and network development have commenced for another study - looking at the likely effects of a change in abstraction volume from chalk streams by the water company (in the adjacent catchment to the Loddon) on the population and diversity of bird species where the freshwater reaches the sea.