Thank you very much Toby, your advice were usefull.
I have allmost no more balance mass issue in my model.
Splitt cell across lateral structures generated the bug.
You can easily spot these cells by looking carefully your Ras Mapper with the 2D mesh.
I had 8 cells like that along my banks so it was quite a huge bug.
Splitt cells don t bug everytime, and when they do, sometimes I had also to have at least one whole cell between my LSs to avoid the bug.
I’ve noticed that the normal 2D Overflow Computation Method was stabler than the weir equation