Computational tools to support decision making in flood risk management are often built up in layers. Components at each layer might be quite general, but the overall analysis is often highly customised for the particular location and type of decision.

We see this layering already as soon as we calibrate a model, which generally involves applying an search (optimisation) technique to solve the inverse problem:

• (search (simulation))

Then again if we use sensitivity analysis tools to explore the behaviour of our model:

• (sensitivity analysis (simulation))

Risk analysis is applied over a physical system simulator coupled to a damage estimation component:

• (risk analysis (simulation + damage estimation))

Given estimates of uncertainty regarding the actual values of inputs to this analysis, we can use [forward uncertainty projection]? techniques to estimate uncertainty in outputs:

• (uncertainty propagation (risk analysis (simulation + damage estimation)))

We might also wish to use sensitivity analysis to explore the properties of our risk analysis, for example in order to help target effort at reducing uncertainty in our risk estimates:

• (sensitivity analysis (risk analysis (simulation + damage estimation)))

For many decisions, proxies for risk such as Expected Annual Damage (EAD) are of more value to decision makers than raw simulation outputs. Further analysis can produce even more relevant information, however:

• (benefit/cost analysis (risk analysis (simulation + damage estimation)))

Given a space of intervention options, sufficient computer power and some clever search techniques, we might wish then to search the option space for options with a high benefit/cost ratio:

• (search (b/c analysis (risk analysis (simulation + damage estimation))))

Uncertainty projection allows us to be explicit about some of what we do not know:

• (uncertainty projection (b/c analysis (risk analysis (simulation + damage estimation))))

The common question now arises: what is the decision maker to do with uncertainty estimates? The concept of robustness allows us to answer this question. A robust option is one which performs well, according to some metric such as benefit/cost, over a wide range of variation of uncertain inputs. In a situation of high uncertainty, it may be wise to choose the more robust option over one with a higher performance around the "best guess" but a worse performance elsewhere.

• (robustness analysis (uncertainty projection (b/c analysis (risk analysis (simulation + damage estimation)))))