Start from the decision, not the model
The required accuracy sets the method. Choosing the model first is how projects end up with a beautiful answer to the wrong question.
Most flood modeling projects begin with a model. Someone has ADCIRC experience, or the last study used HEC-RAS, or the proposal named a package because proposals have to name something. The method is chosen, and then the question is fitted to it.
Doing it in that order is how a project produces a technically excellent result that nobody can act on.
Different decisions need different accuracy
Consider four questions a flood study might have to answer:
- What elevation should this structure be built to?
- Where should the regulatory floodplain boundary be drawn?
- When should this road be closed?
- Which of these forty sites should we survey first?
The first needs vertical precision at a point. The second needs a defensible, reproducible boundary across an area. The third needs timing and lead time far more than it needs depth. The fourth needs relative ranking and can tolerate large absolute error.
Those are four different models. Sometimes four different physics.
What this looks like in practice
Before anything is set up, we establish what the result has to support, what would make it wrong enough to matter, and what accuracy the decision actually tolerates. That fixes the method, the resolution, the forcing, and the validation targets — in that order.
It also, often, makes the project smaller. A screening question answered by a surrogate model in an afternoon is a better outcome than the same question answered by a two-month coupled simulation, if the decision only ever needed a ranking.
The uncomfortable version
Occasionally this process concludes that the modeling is not the constraint at all — that the data does not support the question being asked, or that the decision would be the same across the entire plausible range of results.
Saying so is part of the work. It costs a scoping study and saves a project.