The EnergyPlus Reference specifically states that breaking up a continuous vertical volume into different zones is not the correct way to model stratification (even when using a fully detailed airflow network). Not only will this method be unable to model of stratification but you are killing the radiant heat transfer between different surfaces of your atrium by putting air walls in between surfaces that should be exchanging radiant heat.
If you need to get a sense of stratification, the Honeybee Microclimate maps have built-in methods that estimate air temperature variation over individual zones from certain utputs of E+. You will see that stratification is usually only a degree or two different and typically doesn’t drastically change the results that you get when you assume that the air is well mixed in your zone.
If you need to get even deeper into this, EnergyPlus also has several objects that you can add to the model to separate the single air node of the atrium into several sub-nodes:
Whatever method you choose, you should be modeling the atrium as a single continuous zone. You might still want to break the walls of the atrium into different sub surfaces but breaking it up into separate zones is not correct.