Assessing Model Structure Uncertainties in Water Distribution Models

Awareness of uncertainties is an important issue in every modeling task. In literature, extensive discussion of uncertainties regarding e.g. measurements in general, water consumption, pipe roughness coefficients, etc., can be found. Depending on the modeling aim and the desired outcome (i.e., required accurateness of the results), the question arises what is the impact of not available or poor quality structural data on model simulation results (e.g., if position and properties of pipes and other components are unknown). Therefore, a spanning tree-based algorithm to automatically assemble many possible water distribution models by using free available street network data as input is applied. The obtained uncertainty bandwidth of 180 created models with respect to the pressure head is approximately within ±20m, but can be much lower (±4m) for well performing systems. A qualitative visual comparison of the best performing created system indicates a similar and comparable pressure distribution as in the real system.