We are modeling the flood of record on a moderate size system(140m wide channel). It is a gravel bed channel that has an n value something like 0.04 at bankfull flows or below. But n values decrease with increasing Q, as shown in the roughness characteristics book from New Zealand (Hicks and Mason1998). Do you know of any literature about n values at extreme flows in the channel?
In practice, I usually will keep n values constant when going from bank full on up. I will often times increase n values for lower stages though. If you really want to see how roughness changes as you increase above bank full, you can always calculate Chezy C values, which are depth dependent, and then convert your C value to an equivalent Mannings n value.