Is there an specific reason to remove those empty lines?
Brainstorm will store the covariance matrix in the same size as total number of sensors, regardless their type that is why the rows and columns for A1 and A2 are included. On the further processing those lines will be ignored as they do not correspond to EEG sensors.
The way to remove those empty lines once for all is to remove the A1 and A2 sensors for the data. However, this is not trivial as you need to manually modify the structures of the data, see this post:
If it is for visualizations purposes, you could export the Noise covariance matrix as variable in Matlab, there you can modify it and plot it.