Your acf plot is ambiguous. The paradigm you are trying to follow to identify the SARIMA model sometimes works when 1) there are no one-time pulses in the data 2) there are no seasonal pulses in the data 3) there are no level/step shifts in the data 4) there are no local trends in the data 5) the underlying (waiting to be discovered ) model generates errors that have constant variance 6) the underlying model has constant parameters over time.
If you post your data (which is always less ambiguous) , I will try and help further.