Forecasting models and scenario development for the drinking water infrastructure

Taking uncertainties into account is essential

Forecasting models and scenario development can help water utilities better manage their current water infrastructure and also better design their future infrastructure. Infrastructural investments must after all prove their value over several decades. On 18 April, the Hydroinformatics Platform discussed how one can effectively and meaningfully implement forecasting models and scenario development. The essential elements are the need to pay attention to the uncertainties, and to have an approach that takes utmost account of them.

Water utilities have to plan for the short- and long-term future: the infrastructure underground has to last a very long time. But how can you know what will be expected of this infrastructure in one, twenty, fifty or even a hundred years from now? The future is after all increasingly difficult to predict, changes are occurring at an accelerating pace, and the direction the changes are taking are more and more unforeseeable.

Forecasting models and scenario development can help water utilities better manage their current water infrastructure and also better design their future infrastructure, but what is the most effective way of implementing these tools? That was the subject addressed on 18 April by the

The combination of water research and information technology makes it possible to translate data into knowledge, after which well-founded decisions can be made.

Platform at its second meeting in 2019. Within the Hydroinformatics Platform, which was set up in the framework of the Joint Research Programme with the water utilities, about 20 individuals from the water utilities and KWR meet four times a year to share their knowledge and experience regarding water-oriented ICT applications at water utilities and other companies.

Forecasting the short- to the very long-term


A model that solves the differential equations that describe the physical, chemical and / or biological processes.

– both deterministic and black-box – have been employed for decades in the water sector, their use is on the increase. Forecasting models can be applied, among others, to support operational and management decisions. They can also be used to develop and assess several scenarios for an uncertain future; for instance, water utilities can select the concepts and designs that offer the best chances of meeting the demands of an uncertain future. Forecasting models can be used for several objectives and the different associated time scales. Operational management decisions call for instance for real-time horizon planning: looking ahead one day. Support for the daily management of water systems demands short-term planning, middle-term planning is needed to develop management and adaptation strategies, while robust infrastructure design requires long-term planning that spans decades. A few examples of the use of forecasting models and scenario development in the drinking water sector were highlighted on 18 April.

Effective and meaningful use of forecasting models and scenario development requires that attention be paid to the uncertainties and an approach that takes utmost account of them (after Nekkers, 2006: Wijzer in de Toekomst. Werken met toekomstscenario’s).

Resilient urban water cycle

A method has thus been developed within the Joint Research Programme for a resilience analysis of the urban water cycle, from source to tap. This method makes it possible to compare the water company’s different strategic planning options. The decision-makers gain insight into the robustness and resilience of different urban water-system configuration options, and can appraise them in terms of their associated costs. The method can draw on existing Watershare tools like the Urban Water Optioneering Tool (UWOT), as well as new KWR tools like the Scenario Builder.

Climate-robust estimates

Erwin Vonk (KWR) and Jurjen den Besten (Oasen and Spatial Insight) presented a methodology that has been developed to determine both the average as well as the extreme water demand, which can statistically occur once every ten years until the reference years 2050 and 2085. The method, named EDWARD, makes it possible for the water utilities to take the effects of climate change and changes in holiday-staggering into account, when planning the infrastructural investments in their production and storage capacity.

Predicting mechanical excavation damage

Kristina Arsova made a presentation on how Brabant Water predicts excavation damage to transport mains, using self-learning (machine-learning) algorithms, including Random Forest, SVM and Logistic Regression, combined with different

A selection from a total population for the purpose of measuring certain characteristics of that population.

. Based on the available planning information (e.g., KLIC data), those projects are identified in which the risk of excavation damage to the transport mains is the highest.

Curtailing uncertainties

Predicting the future is never simple. But for KWR’s Christos Makropoulos, the inescapable uncertainties one faces when dealing with forecasts can be partially curtailed by always bearing in mind the importance of:

  • generating (many) alternative, plausible forecasts;
  • taking into account the intrinsic uncertainty of variables like precipitation;
  • exploring alternative future possibilities;
  • employing methods that are based on probability and reliability, so as to render the decision-making process for the design and operation of a water system ‘uncertainty-proof’;
  • quantifying the impact of different decisions/options.

Uncertainty is part of it

Are our forecasts good enough? This question is always pertinent, because no model works flawlessly. This does not mean that models are unusable, but to make meaningful use of them you have to be clear about their flaws and their uncertainties. We have to accept that forecasts are flawed and that they always contain an element of uncertainty that we have to learn to deal with. You should regularly update your forecasts on the basis of the latest and best data. By applying our forecasting models, and continually testing them in practice, we will be able to develop increasingly improved models. But it is an illusion to think that we will ever be able to predict the future exactly. But what hydroinformatics can do is provide direct support for an integrated and resilient management and design of drinking water infrastructure. This was one of the conclusions reached in the meeting.