During implementation of velocity model uncertainties it was found that there were differences between Emod3d v3.0.4 and v3.0.8 that were unrelated to perturbations.
During inspection of waveforms from otherwise identical simulations it was found that the v3.0.8 waveform has a delay of 400 timesteps when compared with the v3.0.4 waveform. When combined with the HF waveform, this resulted in the BB waveforms having different properties.
Investigation into the source code has revealed that the variable itsh is involved with this. It is responsible for creating a buffer at the start and end of a simulation to ensure causality.
In version 3.0.4 the variable was set to 1/(flo*dt), while in version 3.0.8 it is set to 3/(flo*dt), resulting in this buffer being 3 times larger.
After conversations with Rob Graves we now know that this is a timeshift intended to allow for seismic energy to be correctly expressed during a rupture event.
In light of this we intend to extend the EMOD3D sim duration by the length of the expected time shift, and change the LFSeis class to determine the timeshift used, and apply this during the conversion to the BB binary.
The other changes in the shape of the waveform have been attributed to the change in the seismic energy released during events.