At the moment the remove_nans filter generates a mask based on the location of NaNs in the first field it sees, and then applies this mask to all subsequent fields (regardless of parameter, level etc). This feels like behaviour that could potentially surprise a new user.
We should understand how our current users make use of this filter and see if there are improvements relating to either the interface (how a mask-generating field is selected); or applied (e.g. separate masks for different parameters, levels etc).