Obviously putting it in series is better than the other two options. Putting in series, means that If you put two (perfectly calibrated) flow meters in series, then you will be able to make inferences on the actual value based on the calibration uncertainty, the reading uncertainty, and the sampling statistics.
Problems with the other setups
I'll start with the problems in the other two:
- if you put only one and leave the other as a backup:
In that case, you will only have only one measurement value. There is no added information from one of the flow meters.
- if you put two in parallel:
There is no way to make certain that the flow will be split equally, so you would still get only one value. If the flow meters are calibrated to measure the entire flow, you will have additional error due to discretization error (similar to taking a ruler with mm tick intervals and try to measure something that is close to a mm compared to a cm. The error will be much greater in the smaller object).
which of the bad setups is better
If you really can't put two in series, then probably you'd be (marginally) better off putting two in parallel. However this has nothing to do with the uncertainty, that the question is about .
The benefit is that you will be able to see if for some reason one of the flow meters has a problem. So if it breaks down or if a cable becomes loose, then you'd see a significant difference between the two. Additionally, you can set it up, so that you can replace a flow meter without stopping the process (since you essentially have a bypass).
On the other hand, if you only have two flow meters (no backup), then you might be better off, to use only one. Again this has nothing to do with uncertainty.