Agent-based models for water demand – an exploration

The design of a future-proof drinking water distribution network requires a vision of the evolution of the demand for drinking water. There is a need for a method to calculate drinking water demand scenarios, which takes into account sociological, economic, political, technological, ecological and demographic (SEPTED) developments. Agent-based models (ABMs) are typically used to tackle these kinds of questions and might also prove useful in this case.

This exploratory research shows that there is adequate access to all the ingredients required to develop an ABM that can simulate water demand over time, enabling the inclusion of the combined effects of policy, climate change and, for example, increased home-working. Such an ABM could then be implemented for the design of future-proof drinking water infrastructure.

Future water demand is biggest uncertainty factor

Since the drinking water infrastructure is designed for the long term, future developments must be taken into account. Drinking water demand is the most important big unknown in this context. Besides the total demand for drinking water, this also refers to the variation over time (day, week, year, decade) and within the supply area. Moreover, the demand for drinking water does not evolve autonomously, but is subject to all kinds of influences.

We distinguish the sociological, economic, political, technological, ecological and demographic developments (SEPTED) – for example: (S) the trends in daily showering; (E) more frequent home-working, resulting in a spatial and time shift in water consumption; (P) policy that attempts to limit water demand, for instance, by subsidising rainwater tanks; (T) the development of efficient washing machines; (E) climate change leading to more frequent garden watering; and (D) relative increase in number of senior citizens.

Because there are so many key factors, the project investigated whether an agent-based model would offer an appropriate means of elaborating future drinking water demand scenarios. Agent-based models (ABMs) operate bottom-up, that is: the behaviour at the system (macro) level is studied on the basis of the behaviour of and interaction between individual agents at the local (micro) level.

Conceptual design

This exploratory research drew up a picture of what an ABM for water demand involves, what data are needed and what models are available. The KWR-developed SIMDEUM water consumption model served as a reference. As an illustration, students from TU Delft built a relatively simple ABM to estimate daily water demand today and over the next 50 years, as it is affected by people’s time-allocation patterns (at home, at work or school, and in their leisure activities).

The research showed that it is not simple to build an ABM for water demand. Moreover, there is as yet no model that is ready for use. KWR does however dispose of all the ingredients necessary to develop such a model: detailed knowledge of water consumption, knowledge of effective policy strategies and communication with drinking water customers, knowledge about assimilating these strategies in an ABM, and access to geographical data (for example, demographics and mobility). A conceptual design based on different sub-models has been set up; the details should then be elaborated in a follow-up project.

Applicable, but further development required

An ABM for drinking water demand with different SEPTED components is suited for the design of the drinking water infrastructure. It would enable the inclusion of the combined effects of policy, climate change and, for example, increased home-working. SIMDEUM is for example not as suited as a tool for translating the effects of policy and urban changes to changes in the level and location of drinking water demand in a supply area.

An ABM for drinking water demand can offer a great deal, but it also requires a great effort to bring about. Ideally, sub-modules developed elsewhere in areas outside of KWR’s expertise – such as one for demographic developments –, would be plugged into the model. Possible collaboration partners still need to be sought.