The inputs to a hazard model include various types of spatial date that broadly characterises the dynamic geophysical state of the area of interest. The section below details the way in which different datasets were used to create the area source zones, one of the inputs to the EZ-Frisk PSHA.
Earthquakes in continental interiors happen rarely compared to those at plate boundaries. Nevertheless, such ‘stable’ region seismicity is the critical component of hazard in Australia, as well as in many other regions. One aspect of intra-plate seismicity is the high spatial and temporal variability of seismic activity in continental interiors compared with plate boundaries (Vasudevan, 2010). When combined with generally low seismicity, this poses a number of modelling challenges. In particular, low seismicity hampers deeper analysis of the statistical properties of seismicity. For instance, evaluation of changes in seismicity (changes in a and b-value) are difficult in many areas because of insufficient data density.
Earthquakes in continental interiors happen rarely compared to those at plate boundaries. Nevertheless, such ‘stable’ region seismicity is the critical component of hazard in Australia, as well as in many other regions. One aspect of intra-plate seismicity is the high spatial and temporal variability of seismic activity in continental interiors compared with plate boundaries (Vasudevan, 2010). When combined with generally low seismicity, this poses a number of modelling challenges. In particular, low seismicity hampers deeper analysis of the statistical properties of seismicity. For instance, evaluation of changes in seismicity (changes in a and b-value) are difficult in many areas because of insufficient data density.
The GGcatalogue (Gary Gibson, 2013, pers. Comm.,) contains 3392 earthquakes larger than M 2.5. The record includes some larger events from the pre-instrumental era.
Declustering is the name given to separating earthquakes into the subcategories of foreshocks, main shocks, and aftershocks. Declustering is done to isolate ‘background’ earthquakes, i.e. those events that are independent of all preceding earthquakes. The declustering process used here was a simple forward-looking space time window method, without no magnitude dependence. The parameters used were 50 days and 20 kilometres, informed by the recent aftershock sequence of the 2012 Moe Earthquake. Frequency-size relations were defined using the following completeness parameters: M 5.5 up until 1960, M 3.0 for 1960-1970, M 2.5 1970-present. Completeness magnitudes (\(M_c\)) were considered homogenous across the study domain. After removing aftershocks, 2512 events remained.
A key input to the PSH analysis is the choice of source zones, seismicity model and the statistical and physical parameters (a, b, \(M_{max}\)) that define these. The choice of source zones implies a seismicity model, e.g. a smoothed seismicity approach implies that seismicity is stationary (spatially and temporally Poisson). Under this assumption the historical catalogue will predict future seismicity, out to return periods that greatly exceed the catalogue duration (Leonard, 2012b). The methodology for creating area source zones is summarised beneath. The main driver for this methodology was to minimise the arbitrary ‘drawing’ of lines, but rather have the data create the zones itself. In principal this methodology could be fully automated. The method is similar to previous studies using spatial smoothing of historical seismicity, e.g. Somerville (2008) and the “regional source zones” in the Australian Earthquake Hazard Map (Burbidge, 2012).
Create a point density estimate of seismicity (here we used all seismicity above M 2.5, i.e. not a ‘complete’ catalogue) and contour this density estimate (performed here using standard tools on Quantum GIS, with an arbitrary 30 km smoothing kernel)
Work from zones of highest intensity outwards, ensuring that each zone (contour line) encloses enough earthquakes to make earthquake magnitude calculation viable (50 events was the target).
Draw buffers around the active faults (here a 30 km buffer on all sides was used)
Where active faults and area contours intersect, use the fault buffers to adapt the area source zones so that he intersection is avoided (using the principle that the initial area is most closely preserved
If areas have significant ‘necks’, i.e. where a region is made up of two ‘fat’ areas connected by a ‘thin’ corridor, then cut the regions at the thinnest point (only if this leaves enough events for EMR in each sub region.)
The area source zones created following this process are shown in Figure 1. It should also be noted that the technique used here breaks down somewhat in areas of very low seismicity, i.e. it is recognised that with the current earthquake catalogue, the technique is not suitable for Australia-wide seismic source zone definition.
Frequency-magnitude parameters were estimated using a method similar to Leonard (2012), the scripts are written in R., and, along with an example, can be found on the website http://vicquakehazmap.org/. Completeness magnitudes (\(M_c\)) were considered homogenous across the study domain. These were M 5.5 up until 1960, M 3.0 for 1960-1970, M 2.5 1970-present. We have only attempted to calculate earthquake magnitude relationships (EMR) where there were at least 30 independent events in the zone. If there were fewer than 30 events, the zone parameters were taken from a representative zone (“background east”).
Over the last couple of decades, knowledge of Australian intraplate faults has increased significantly (Clark, 2006). The location of faults is defined by field observation and geophysics/DEM analysis. The recent or ongoing activity of a fault is basically a binary decision made by the scientist on the basis of offset geology and or geomorphology. Occasionally, trenching has been performed to constrain total lip. Clark (2006) defines an “active fault” as one which has hosted displacement under conditions imposed by the current Australian crustal stress regime, and hence may move again in the future. This implies that an active fault is one that accumulated slip in the late Neogene to present. In some cases, a fault scarp cannot be observed on the surface but folding of the landscape allows deduction of a blind fault. These are typically described as monoclines.
Fault slip rates in Australia’s active faults are typically less than a 0.1 mm/y. This compares to 10s of mm/y for plate boundaries, and 1 mm/y for intermediate regions, for example, the Canterbury Plains which hosted the 2011 Darfield Earthquake (Reyners, 2013).
Active faults and fault properties were mainly taken from the Neotectonic Features Database (Clark et al., 2012), comprising 131 faults in the model domain. Some minor adjustments to these faults were made to that they had appropriate geometry for EZ-FRISK input.
Active faults added to Neotectonic Features database (12 added in total) were generally taken to be those that had some evidence Neogene activity. These were mostly sourced from literature and discussion with geologists, in particular Ross Cayley of the Victorian Geological Survey. The Neotectonic Features database is the primary input to the GEM neotectonic features (for Australia). Thus, active fault inputs to the current study should be similar to future hazard estimates performed with GEM.
Apart from high uncertainties associated with fault slip rates, a number of other assumptions in adding faults need to be made:
All slip measured across the fault is assumed to be seismic slip (creep has not been recognised on Australian faults)
The slip rate is an average that makes no account for short term fluctuations
Measurements of slip rate along the surface is assumed to be representative of slip rates at seismogenic depths
Individual faults all have a b-value equal to the b-value of the area zone that contains them.
Assuming an exponential distribution on the activity model, the activity rate, N, is constrained by the upper bound magnitude, \(M_{max}\), the b-value for the region and the fault slip rate, S. EZ-Frisk uses the relationship derived in Youngs and Coppersmith (1985). Also needed is a model of fault rupture area as a function of magnitude. The functional form for this model is log-linear in EZ-Frisk. Here, typical values were used (e.g. Wells and Coppersmith). The equation is as follows:
\[\log_{10}(A) = A + Bm\]
The parameters used were \(A = -4.0\), \(B = 1\).
The primary GMPE used was Chiou and Youngs (2008). The SE Australia-specific relation of Allen (2012) was also tested on number of models. Although no explicit weights are attached to the models, it is noted that the 2012 Australian earthquake Hazard map weighted the above equations equally (25 % percent) along with two other equations used. Hoult (2013) shows that Chiou and Youngs (2008) provides a reasonable average of Australian-specific GMPEs, i.e. that that weightings used in the 2012 AEHM would result in an averaged GMPE close to Chiou and Youngs (2008).
All models were calculated on a rectangular grid that enclosed the state of Victoria. Sources were defined in a polygon that extended to a 300 km buffer around Victoria. The grid resolution was 0.1 degree. While the hazard outputs extend beyond the state of Victoria (due to the ease of computing on a square grid rather than a complex polygonal grid) the results are only valid for within Victoria.