Modeling advice, inflow tunnels.

Hello, ;)

I am like in a stage of my project where I need to calibrate by trial and error my model. It's a region, with twin tunnels. The model has also discrete features (vertical faults), and the tunnel was modeled by deactivating elements. The mesh is fully non-structured.

I also have 10 obs points in the surface, and I ran FePEST and I got the best parameters (zonally constant) for one (out of two) lithologies modeled.

The thing is that, while the water heads are somewhat acceptable, the inflow in the tunnel is just too damn high for some reason, and it's far away from reality. FEFLOW is giving me that the water budget in the tunnels is like 170 lt/s, but the actual flow is around 100 times less this quantity...  also, 170 lt/s in the tunnels is like almost 60% of the net recharge (!!!!). How is this possible?

It must be said that the recharge was set by using the In/Out flow top/bottom, by selecting the whole region a value of q=9.75E-9 [m/s], and the topographical area is 30761350 [m2] which is equivalent to 300 lt/s over the whole area.

The 'seepage face' boundary condition was set in the nodes of the tunnels. On surface, there are some small creeks as well and they were modeled by using the same BC.

Any comment or advice? I don't know what to do have some reasonable water budgets in the tunnel, there must be something wrong somewhere but I'm unable to detect it. I attach the geometry of the model, so you can see that it's unreasonable to have 60% of the recharge flowing in the tunnel.

While it is likely impossible to troubleshoot model remotely, I would check the following:
1) water balance error for the model. If it is higher than few % one would need to tighten convergence requirements
2) distribution of the modelled tunnel inflows. Is it uniform / through faults / within one unit / within few nodes? It should give understanding of what is going wrong
3) how calibrated K compares with back of the envelope estimates . There are analytical solutions for inflows into tunnels (and even more solutions for horizontal wells).

Was the model calibrated to both heads and measured tunnel inflows? FePest allows to have multiple types of observations during calibration. It also would provide information about which parameter a specific observation (e.g. tunnel inflows)  is particularly sensitive to.

Thanks for your answer,

Well, I think it's not related to convergence criteria or something like that. It's just unreasonable that (see picture attached above) that a tunnel like that gets almost 60% of the recharge as a water budget, by considering its relative size with the modeled area.

The distribution shows indeed that the water budget in the tunnel is somewhat homogenous with specific zones of higher inflows due to the faults, but it's not that different from the model without faults, so I discard this to be the cause of the wrong results.

And the model was only calibrated by considering the water heads on surface, not the water inflows into the tunnel.

Kind regards,

Indeed is is hard to say what the actual cause of the model, however if the only observations PEST can calibrate against are hydraulic heads, its likely that the predictive performance for fluxes is poor.
If you can exclude that the model itself is causing the problem (especially water balance), these would be my steps:"

Check the balance items of the water balance. If 60% of recharge is going to the tunnel, where do the remaining 40% go to? These are likely the creeks; if so, these probably show less discharge than you would observe in the fields. One solution would then be to use the discharge in the creeks as an additional observation to PEST.

When working with PEST keep in mind, that it sees the FEFLOW model as a black box. The only thing it does is changing the parameters until the observatiosn (hopefully) match the observed values. So if the only observations are heads, this is the only thing it optimizes. If you expect PEST to achieve reasonable water balance, defining budget items such as river discharge will tell PEST what this actually means.

Does this help?