How do I perform a sensitivity analysis?

  Applies To 
WaterGEMS, SewerGEMS, CivilStorm, StormCAD, PondPack, SewerCAD, HAMMER, WaterCAD
  Version(s): 08.11.XX.XX, 10.XX.XX.XX
  Original Author: Mark Pachlhofer, Bentley Technical Support Group


How do I perform a sensitivity analysis?

What is a sensitivity analysis?


A sensitivity analysis consists of running multiple scenarios in order to determine how model results turn out from trying slightly different ways to model an element that may not have an explicit way to be constructed using the default elements or to model different properties of elements slightly differently to determine, which results look more similar to what you are expecting the outcome to be. 

This is a method that was developed because the engineer was looking for the best way to model an element that may not be as straight forward as using a single element from the software package. Sometimes you need to combine elements or try different combinations or ideas to determine the best solution. This is why the hydrology and hydraulics products have scenarios and alternatives. The OpenFlows line of hydraulics and hydrology products allows the engineer to easily construct and test these different solutions then compare the results.

Being familiar with the active topology alternative and how it works will provide a good start to using this method. More information about the active topology alternative can be found in this wiki.


Example 1

Question of what is desired: 

I am working on some headloss calculations for a hydro scheme, which I will eventually model in HAMMER. The scheme includes a section of tunnel supplying the station where it is partially rough cut (D&B) and partially shotcrete lined (~120 degree arc). I only have roughness values for shotcrete lined and rough cut - not a combination.

So the question is, is it valid to take an average of the roughness values weighted by the perimeter each value is applied to?

Solution using a sensitivity analysis:

It may or may not be alright to combine the roughness values and go with an average, but it certainly seems like there is a case in which a good argument could be made for trying it. Here are a few suggestions to try 1) Add multiple pipes to break the tunnel up into sections where you know the lining and it's roughness value. For example, if you know the first 1500 ft. of the tunnel is shotcrete lined, followed by 1500 ft. of partially rough cut D&B, followed by another 1500 ft. of shotcrete lining, then you might want to try making it 3 separate pipes, each with their respective roughness values. Since it may not be this simple here is a second suggestion 2) Use the roughness value that is the most conservative. This will assure that you design your system to handle the worst case scenario or at least know what the worst case scenario results 3) Try to model it a few different ways by taking advantage of creating some child scenarios and then compare the results to see the differences, if any. It would be a good use of the scenarios and would allow you to assess the results side by side then use your engineering judgment to determine which solution works best for meeting the study's goals.

Example 2 - Modeling elements that are not explicitly provided with software

The wiki link here titled: How do I model devices, control structures, or other elements the software might not explicitly provide when modeling my system? , shows another example of how you might use a sensitivity analysis to  determine how to model a clean out on a force main in a combined sewer model. There are a few different configurations that could be run in different scenarios to determine which provides the results the engineer would most expect to be correct.