Now that EMOD3D v304 has the correct time shift to preserve causality we can compare the results of simulations performed with version 304 and 308 to see what differences there are, if any.
The simple end to end test for version 18p5p3p4 was run with the LF only workflow (EMOD3D, LF2BB, IM_Calc).
Waveforms
An example waveform comparison is available below, the black line is the v304 LF waveform, the red line is the v308 LF waveform (unfiltered, meaning frequencies above flo are present):
The increased amplitudes later on in the simulation can probably be explained by the increase in energy captured by the longer causality time shift in v308 (12s vs 4s for v304).
The other stations generally had similar results, with larger amplitudes for longer durations in the new versions.
Intensity Measures
The intensity measures were also investigated, as these are the primary outputs of the waveforms.
For intensity measure calculation the waveforms were converted from the EMOD3D results straight to BB without filtering.
The IM files were loaded using Python and Pandas, and the ratio of (v308 IMs)/(v304 IMs) observed. If the measures are approximately the same all the ratios will have a value near 1.
As there is more energy captured by the increased causality window we would expect the IMs for v308 to be generally larger than v304.
| Hossack_REL01 | PGA | PGV | CAV | AI | Ds575 | Ds595 | MMI | pSA_0.02 | pSA_0.05 | pSA_0.1 | pSA_0.2 | pSA_0.3 | pSA_0.4 | pSA_0.5 | pSA_0.75 | pSA_1.0 | pSA_2.0 | pSA_3.0 | pSA_4.0 | pSA_5.0 | pSA_7.5 | pSA_10.0 |
| Mean | 1.07092587046075 | 1.06641926175768 | 1.13937735221672 | 1.21300974380887 | 1.0422654862116 | 1.06406476279863 | 1.01675860042662 | 1.07092124230375 | 1.07091704856655 | 1.07081663453925 | 1.07022484365188 | 1.06995742962457 | 1.06828929943686 | 1.06667280996587 | 1.06312543764505 | 1.04067315107509 | 1.04229152194539 | 0.993409597593857 | 1.08534523843003 | 1.15635976481229 | 1.09244463382252 | 1.06306023343003 |
| Standard deviation | 0.063288101854903 | 0.104995224584679 | 0.111220887913525 | 0.142438699621476 | 0.131267088302554 | 0.133586322174042 | 0.01117802152697 | 0.063290632210506 | 0.063289538215637 | 0.063496624555496 | 0.064917424148519 | 0.063582578666072 | 0.063553639886157 | 0.068138814626744 | 0.068337787891797 | 0.082148812384953 | 0.078214379680128 | 0.071732655486479 | 0.071965102431381 | 0.082605907909773 | 0.097949297551634 | 0.08356819333254 |
| Min | 0.28966793 | 0.09806367 | 0.051653948 | 0.018004442 | 0 | 0 | 0.981561 | 0.28967208 | 0.28968304 | 0.2891708 | 0.28599125 | 0.2876601 | 0.29865447 | 0.26569054 | 0.251938 | 0.23234159 | 0.23937169 | 0.15354854 | 0.2310539 | 0.26135978 | 0.22558492 | 0.24895166 |
| Max | 1.7204375 | 2.2437246 | 1.492937 | 2.3731713 | 1.3378471 | 1.4565088 | 1.1468107 | 1.7205105 | 1.7204801 | 1.7200465 | 1.7183907 | 1.7064599 | 1.7135324 | 1.6979796 | 1.6955371 | 1.650316 | 1.4363729 | 1.2424793 | 1.2757007 | 1.6451819 | 2.7466543 | 2.3873415 |
As this was a 400m simulation with an flo of 0.25, the pSAs below 4s are not as significant as the ones above.
Stations of interest
2101bef
This station had the minimum ratio for PGV and all pSAs.
Investigating this station revealed other issues with the workflow detailed here: VM domain boundary station 0 value problem.
This station is situated near the southern edge and snaps to a point along the bottom of the velocity model domain. It has issues with the 000 component of its simulation results. Other stations near the edge have similar issues.
The waveforms are visible below:
1101b88
The station with greatest increase in PGV and pSAs 5-10s is 1101b88, which is situated near the trace of the fault.
As the station is so close to the fault the increased causality time probably causes this increase in movement.
