Pipeflow not returning to 0

[jiwaszko 0]

Hi,

I am modelling a 460ha urban catchment with a large number of pits and pipes (~2000). The model is running as expected however between bursts the pipeflow in most pipes does not return to 0, and instead plateaus at a number greater than 0. By the time that flows are reaching the bottom end of the pipe network the cumulative effect of this is significant (see attachment). I've cross checked flows entering the pipe network via the pits and they don't seem to match. i.e. pit inflow =0, pipeflow > 0 (even at the most upstream pipe).

The model is rainfall on grid and being run using GPU HPC.

Does anyone have any ideas about what might be causing this?

Many thanks,

Josh

[peteraylett 0]

Hi Josh,

It sounds like it could be a mass balance issue in ESTRY.

When you run on GPU, you're normally pushed into running single precision, and sometimes that's not quite good enough. This could be on one of those times. You can test for this easily enough but simply running you model with the double precision engine and see if it comes out different! You'll need to add the command "GPU DP Check == OFF" to run this on GPU.

Another possibility is that you need to run with a smaller ESTRY timestep (which would again manifest as mass balance problems). If you've got small elements then your timestep needs to be smaller. Often ESTRY timesteps seem to be a neglected consideration, and set at 1 (default) and ignored. You may very well need to go smaller.

Finally though, are you sure it's not correct? A small trickle slowly draining off such an area would accumulate through a pipe network into something non-negligible.

I'd start by looking at your mass balance outputs and see if there's anything untoward in there. Let us know how you get on, I'm sure other folk are likely to run into similar and would like to know the outcome!

PHA.

[Pavlina Monhartova 0]

Hi Josh,

Recently we had model (not rain on grid) with similar behaviour and double precision was the fix. Due to not enough significant numbers in single precision there was a trickle flow remaining throughout the pipes.

The rest of Peter's suggestions are also worth checking, e.g. 1D timestep and justification for the trickle flow, especially with your model being rain on grid.

If you can't get to the bottom of this, please send the model to support@tuflow.com for investigation. Thank you.

 

Kind Regards,

Pavlina

[jiwaszko 0]

Hi Peter and Pavlina,

Thanks for your help. It turns out that this too was fixed by running the model in double precision. Happy to have got to the bottom of it.

Many thanks,

Josh