Fragility curves may be sampled for every event or once per realization, depending on if the elevation to be sampled is a function of natural variability or knowledge uncertainty.  HEC-WAT Flood Risk Analysis (FRA) simulations sample inputs in two loops: an outer loop where parameters linked to knowledge uncertainty are sampled once per realization, while an inner loop samples parameters linked to natural variability once per event  – however the development of fragility curves for a structure may assume the probability of failure at a elevation is a function of one or the other, and so the specific system being modeled will dictate which to choose.  In both sampling by event as well as the lifecycle/realization sampling, the uncertainty distribution for the fragility curve is sampled every realization to generate a fragility curve for each failure mode at each structure.

Recommendations

If the system performance curves or fragility curves come from a previous study or analysis that identifies them as representing natural variability (per event) or knowledge uncertainty (per realization), the choice made for fragility curves in HEC-WAT should match this.

Choose randomize by realization (knowledge uncertainty) if one of the following is true:

  • The probability of failure by loading elevation is largely determined by unknowns in the system itself, not the conditions of the event.
  • Guidance for the type of study being conducted requires treating the failure elevation as knowledge uncertainty.
  • Sampling the same failure elevation for an entire realization is necessary to represent the same behavior across multiple events.

Choose randomize by event if:

  • Levee performance is expected to be unreliable or highly variable and expected to vary substantially by flood event.
  • Guidance for the type of study being conducted requires treating the failure elevation as natural variability.
  • The HEC-WAT FRA compute is configured to ignore the nested looping process, such as using the Stochastic Data Importer (SDI) plugin, and sampling by event is correct for the hydrology used.


Case 1: Randomize by Realization

At the start of the Realization (Outer Loop)

  1. Randomly generate a fragility curve from the user input distribution for every potential failure mode
  2. Sample an elevation from the curve for every potential failure mode
  3. Sample a duration, if applicable, for every potential failure mode

At the start of each Event (Inner Loop)

  1. Choose a potential failure mode according to assigned probability.
  2. Save elevation sampled in the outer loop and duration data to the Fragility Curve sampled object

Case 2: Randomize by Event

At the start of the Realization (Outer Loop)

  1. Randomly generate a fragility curve from the user input distribution for every potential failure mode

At the start of each Event (Inner Loop)

  1. Choose a potential failure mode according to assigned probability.
  2. Sample an elevation from the curve of the active potential failure mode
  3. Sample a duration, if applicable, for the active potential failure mode
  4. Save elevation and duration data to the Fragility Curve sampled object