The simplest model run in this study uses area source zones only. The seismic activity within each source zone is homogenous – i.e. the seismicity is assumed to be spatio-temporally Poisson. The derivation of source zone geometries is detailed in Appendix 1. This model is also termed the reference model. In terms of inputs and approach, this model most closely follows the 2012 Australian Earthquake Hazard Map \cite{burbidge20122012}.
The passive faults model represents a hazard model where seismicity rates are entirely informed by the historical seismicity, but where the location of the modelled activity is partly ‘mapped’ to active faults. For example, in a zone with a number of active faults, we first determine the percentage of earthquakes occur within a 30 km buffer of those faults. We then limit this fraction of the seismicity to occur only on those faults, providing a geographical constraint. The rationale for this model is that it may help counter earthquake-location-accuracy, by locally concentrating seismicity closer to known faults; secondly it helps distribute seismicity onto faults which may be inactive on the timescale of seismic records, but appear to be active in the current tectonic environment.
In this model, geologically-derived fault slip rates are superimposed on a model containing only area source zones, i.e. no ‘passive faults. Because the ‘active faults’ represent additional sources, activity rates are higher than the Model 1 values, using only area sources. In this model the higher limit of the slip rate estimates are used. By adding both recorded and paleoseismic activity rates this model is extremely conservative.
The Final model assigns equal weights to the models described above. In PSHA, a preferred model is based largely on the professional judgement of the modeller. This choice of these weights is discussed below. We have not assembled a logic tree to try to quantify the uncertainty in the final model, as is common in PSHA. We are inclined to agree with \cite{bommer2006control} who views these weights as simply ratings to reflect the relative confidence of the analyst that the most appropriate model has been selected, rather than the “scientific uncertainty”.
The main difference between the models is the way in which the overall seismic activity is constrained. Models 1 & 2 preserve the seismic activity rate (the historical or catalogue rate) while modifying the spatial density of this rate to take into account locations of active faults. Hence for any fault where seismicity is reapportioned, the corresponding activity is subtracted from the area zone which contains the fault. This procedure is similar to that proposed by \cite{ninisdeveloping}.
Model 3 uses ‘active fault’ information to modify the seismic activity. The assumption is that the long-term slip rate on active SCR faults may not be well captured by the short seismic record. In general, the seismic record may over or underestimate the long-term. This model accounts takes into account the possibility that the seismic moment rate underestimates the long-term rate, and adds published fault slip-rate information on top of the model one seismicity rate estimation.
Combining all models into ‘average model’, as we have in this study, means effectively weighting different hypotheses about seismicity. These can be summarised as follows:
Some seismicity occurs where historical/catalogue seismicity occurred and at similar rates.
Some seismicity occurs on geologically-indentified active faults at rates determined by historical/catalogue seismicity.
Some seismicity occurs on geologically-indentified active faults at rates determined by geological slip-rate estimates.
In the absence of any firm constraints about the accuracy, or relative weighting of these models (hypotheses), we assign equal weighting for the Final model.
For all models, the Ground Motion Prediction Equation implemented is \cite{chiou2008nga}. This model appears to agree well with Victorian data collected in recent years by Victorian AGOS seismometers (Hoult and Sandiford, in preparation). No site amplification as a result of local conditions was included. This ensures that the model is consistent with the current map in AS1170.4. Intensity measures calculated are 5 % damped response spectra at periods of 0.01 s (PGA) 0.3 s and 1 second.