EDIT: if you use the new MODFLOW-NWT, you probably don't need any of these because NWT does not use wetting-drying. So go for NWT if you can.
The biggest pain in the ass of groundwater modeling is getting your solution to converge.
This is the due to the uniqueness of groundwater system - the aquifers can be unconfined, in which case part of the aquifer is not fully saturated with water and the rest is saturated with water. In realty, the unsaturated part still has some residual water content and is called vadose zone. However, MODFLOW does not simulate water content in the vadose zone. It treats it as completely dry. This caused a huge problem when solving the model's linear system because cells has to be constantly switched between dry and wet to satisfy the boundary conditions. And this is often accompanied with abrupt changes making it even harder to converge. The 'rewetting' option can be turned off in the input file but that means once water level in a cell drop below the bottom in a solver iteration it will never come back. You may get a solution in the end but it will not be the correct solution.
Here are a couple of tricks/techniques to better assist you with model convergence.
1. First run it with rewetting on, if the residual fluctuates but the head change is small (around 1 m) then run it again using the previous solution as initial head with rewetting of this time. You should get a solution with good mass balance and correctly reflect the groundwater system you are trying to simulate.
2. If the above does not work, change all layer type to confined (laycon=0), then run the model. You should be able to obtain a solution with good mass balance easily. (If not, then there is something wrong with the model, check boundary conditions etc). Then using the solution as initial head, progressively change laycon of top layers to 3. For example, change layer 1 to type 3, run it, get a solution, use the new solution as initial head, change layer 1 and 2 to type 3...
thanks, Sheldon, this is very useful tricks.
ReplyDeleteGreat info! I found that my convergence problem may be related to rewetting issues. When I turn a convertible layer into a laycon=0 type, it converges. There may indeed unsaturated from between my layers... I am looking at the connection between a river alluvium (layer 1) and an aquifer that is generally confined expect in the area of the river, where it goes unconfined according to head data (layer 2).
ReplyDeleteI really think that a laycon=3 is most appropriate for the system, so are there any hints to test where this is falling apart? Turning on rewetting did not help convergence.
Hi Terrance,
DeleteAny chance you can migrate the model to MODFLOW-NWT? It would work a lot better.
Which solver are you using? The above method works with PCG mostly. Also try another solver like Geometric Multigrid (GMG). Also, your solver settings are also important in getting convergence, like for PCG number of inner/outer iterations, dumping factors etc. And it is a lot of trial and error. Try NWT.
If you can't use NWT. have you tried using solution from laycon=0 for all layers as initial head and re-run model with laycon changed form 0 to 3 for layer 1 only. Are you able to get a convergence for that case?
It's hard to say what the problem is without looking at it. It could be layer configurations like sudden change form very thing to very thick that keeps some cells constantly dry/wet, or some crazy boundary condition somewhere or even one dangling active cell messing up the convergence.
Laycon= 3 is realistic scenario most of the time. But one thing to keep in mind is that MODFLOW and groundwater model in general are powerful enough to represent every detail of the real world but merely advanced data processing. So depending on what you are trying to do with the model, you could simplify the conditions a little bit. For instance, if you are trying to predict flow into an open pit, then a fully confined model would be more conservative and sometimes that's enough information.