I would definitely recommend ...

I would definitely recommend jar testing as mentioned by Bob Foerster. I assume the feed turbidity is low if direct filtration is practiced. I would also attempt optimising the velocity gradients if possible to produce small compact floc as opposed to large flocs that would tend to result in shorter filter runs. Inefficient coagulation/flocculation (i.e. velocity gradient distribution and duration) would tend to result in larger flocs and incomplete destabilisation and coagulation before fitration.