Our sewer models are all wrong – how can we make them more useful?

The statistician George Box is reported to have said:

“All models are wrong, some are useful”.

UKWIR recently published the report, ‘Quantifying, managing and communicating the differences in storm overflow spill data between EDM outputs and hydraulic model predictions’.

In a LinkedIn post on this report, Martin Osborne set me thinking that perhaps we should fundamentally reconsider how we undertake hydraulic design.

Our society now values water quality somewhat higher than it has done in the past, and the financial and reputational consequences of failure of sewer designs are consequently becoming more severe. The harmful effects of pollution on people who use rivers and the sea for recreation are now in the headlines. If the consequences are becoming so severe, perhaps we should now consider what we can learn from structural design where the consequences of failure have always been seen as serious – financially, reputationally and in injury to people

Reliability in structural design

In structural design, engineers have applied factors of safety for as long as they have been doing structural calculations. Initially these were largely arbitrary but over time these have become quite sophisticated based on statistical models of the variation in applied loads and strength of materials, some even consider the potential for inaccuracies in the model used to undertake the structural analysis.

The Eurocodes, EN 1990 ‘The basis for structural design’ lists four types of partial safety factor in two groups (see below from EN 1990, Figure C3).

How might we apply these ideas to hydraulic design?

The ‘actions’ are the inputs to the model and instead of material properties we have the hydraulic parameter of our design.

• Uncertainty in representative values of actions - and we would need to consider the uncertainty in rainfall (including climate change) and the uncertainty in other inputs such as infiltration and urban creep.
• Model uncertainty in actions and action effects - this might consider the uncertainties in runoff models. This might take into account the results of studies such as the one mentioned above undertaken by UKWIR.
• Model uncertainty in hydraulic properties - these are largely known but in areas such as modelling low side weirs there are known limitations to our models.
• Uncertainty in hydraulic properties - This might include uncertainties due to roughness, siltation of the sewers and performance of flow control devices.
  

The performance of hydraulic design of sewer systems is under scrutiny as never before and the consequence of failure is become financially and reputationally severe. In view of this we should be considering the use of safety factors in hydraulic design to ensure that these designs achieve the required performance.