Model-based decision-making under deep uncertainty

Levels of uncertainty

Not everything is equally uncertain. Peter van Thienen (KWR) took us through the four levels of uncertainty (Figure 1) between complete certainty and complete darkness. Deep uncertainty arises when people make a decision that doesn’t know or can’t agree on how actions lead to consequences, how likely certain outcomes are, what consequences are important, and how heavily those consequences weigh. These are often decisions that change over time and are influenced by the system they are dealing with.  Often, we count on a future based on what we have experienced in the past. This is the ‘continuity bias’ (Figure 2). The fact that it is wise to take other scenarios into account is evident from the pattern of drinking water demand in the Netherlands over the past 45 years: a succession of linear trends interrupted by sudden trend reversals that reversed the direction of development. But with an adequate understanding of uncertainty, this need not stand in the way of decision-making – we can deal with it.

Figure 1. The different levels of uncertainty (Marchau et al., 2019)

Figure 2. The continuity bias. Global Scenario Group (GSG) (1995)

Impact of decisions as a measure

David Gold (Utrecht University) uses models to calculate the impact of decisions. Unfortunately, there are not always good probability distributions for the input of the models (this is deep uncertainty) and sometimes events have never been observed, or may arise in the future. In those cases, you can move from a prediction to a decision-impact question: how likely is a scenario to influence strategy? With plausible assumptions, computational experiments can then be performed, so-called ‘exploratory modelling’. David uses Scenario Discovery to stress test for various decided strategies. Which factors are important for achieving the desired performance can also be found out. In the example David shows for a group of water utilities in Southeastern America , this is the water demand. If it is low, it will yield the best results in various decisions.

Fixed or flexible choices

Joeri Willet (KWR) has conducted research into aspects of uncertainty in the Dutch/Flemish drinking water sector. This showed that awareness of deep uncertainty does not match the scientific literature. The question is whether this is justified. Stakeholder needs, water quality and quantity were identified as deeply uncertain most often, while climate change and water demand are mentioned most often in literature. According to Joeri, decisions can be static (and robust) or adaptive (flexible). If there is little uncertainty, few flexible solutions available, and/or the time required for implementation is long, a static, robust solution can be chosen. With more uncertainty, flexible solutions, and/or a shorter implementation timescale, an adaptive strategy is smart. But, how flexible are the available solutions in the drinking water sector? DMDU methods make it possible to make a robust decision within these possibilities. There is a huge toolbox available for making decisions under uncertainty, such as ‘exploratory modeling’, ‘adaptive planning’, ‘decision support’. Joeri shows us a case study on water management in growing cities using dynamic adaptive policy pathways. In doing so, a strategy can be chosen in which the water level of an aquifer diminishes the least.

What is uncertain in the Dutch/Flemish drinking water sector

In the discussion that followed the three presentations, it emerged that the use of methods for dealing with uncertainty in decision-making is currently being tried out by Dunea, PWN and Vitens. The inclusion of several uncertain factors is not yet done. In order to make a good assessment, it may be necessary to involve other parties that need water (both aspects will be taken up in the Waterwijs study next year).

The representatives of the drinking water companies present discuss where their solutions generally sit on the yardstick from static to adaptive. In practice, the drinking water companies are already using modular solutions, such as an option for a second water production site instead of building it immediately. The timescale on which one looks is a big factor in being able to categorize the solutions as ‘static’ or ‘adaptive’. Methods for making and planning adaptive decisions can prevent a choice from being made that can no longer be deviated from, the usefulness of this is recognized.

At present, methods for dealing with uncertainty in decision-making are rarely used in drinking water practice. Decisions are now often made on the basis of an assessment of a limited number of scenarios. Participants agree that the use of methods for decision-making under deep uncertainty should be considered because the methods discussed can objectively underline the decision. Involving parties with sufficient knowledge of these methods is then of value.