Using catastrophe loss models to improve decision making in disaster management

By Andrew Gissing and Ryan Crompton

Catastrophe loss models

Catastrophe loss models are decision support systems used extensively in the (re)insurance industry to assist in pricing risk and aggregate exposure management. They also offer significant benefits in improving disaster risk reduction decision making.

Figure (1): Catastrophe model framework

Risk Frontiers over the last 25 years has developed a sophisticated suite of Australian probabilistic catastrophe loss models to quantify the impacts of flood, bushfire, hail, tropical cyclone and earthquake. These risk models have nationwide coverage and are comprised of the following modules (see Figure 1):

Figure (2): Risk exposure comparison
  • Hazard – estimates the hazard intensity footprint for a specific event. For example, flood extent or ground shaking intensity.
  • Exposure – provides location-based information about relevant assets.
  • Vulnerability – estimates the level of financial loss to different types of property as a function of hazard intensity.

Risk Frontiers’ catastrophe loss models provide scientifically based damage estimates to insurable assets such as residential, commercial and industrial properties and provide users with information about possible financial losses and associated average recurrence intervals (ARIs). Standard outputs from the financial module include exceedance probability (EP) curves (return periods) and average annual losses (AALs).

Using a suite of models enables the comparison of possible losses between hazards at various ARIs for a given geographic area (Figure 2). Loss estimates can also be used to inform benefit cost estimations of different disaster risk reduction investments by varying the vulnerability module1.

In addition to estimating financial losses, model outputs can be combined with vulnerability functions that enable the estimation of loss of life and infrastructure disruption.

Catastrophe loss models can be used to inform disaster planning and capability analysis by enabling the development of ‘what if’ scenarios. For example, to estimate the impact of a magnitude 7 earthquake occurring underneath Melbourne (Figure 3). The models can also be used before or during actual events to forecast possible impacts.

Risk Frontiers combines output from its hazard modules with other data sources to maintain a multi-hazard database for Australia (Figure 4). This database provides national address-based risk ratings for flood, bushfire, earthquake, severe storms, storm tide, tropical cyclones and other hazards. It can be used to assess risk across national asset portfolios, identify community risk profiles and inform property owners of their natural hazard risk profile.

Figure (3): Damage estimate for magnitude 7 earthquake in Melbourne
1 Walker, G. R., M. S. Mason, R. P. Crompton, and R. T. Musulin, 2016. Application of insurance modelling tools to climate change adaptation decision-making relating to the built environment. Struct Infrastruct E., 12, 450-462.

Understanding future risk

The catastrophe loss modelling framework is ideally suited to consider influences on future risk such as climate change, mitigation investment, increased development and changes to building codes. The Geneva Association, a peak insurance industry think tank, concluded that by combining catastrophe models with latest climate science an enhanced understanding of future weather-related risk impacts could be developed. Such use provides greater insights into the impacts of climate change on natural hazards not currently possible using Global Climate Models.

Figure (4): Multi-hazard address-based risk rating

Risk Frontiers’ Australian catastrophe loss models

FloodAUS. FloodAUS is based on the National Flood Information Database. The scope of the model is further extended using Risk Frontiers’ Flood Exclusion Methodology. Correlations between catchments are modelled to provide estimates of total event losses.

FireAUS. The upcoming release uses MODIS Burnt Area Products along with other data sources, machine learning models and fire-tracking algorithms to derive a national synthetic event set from which losses are calculated.

QuakeAUS. QuakeAUS is a national earthquake model for Australia. Starting from a record of historical seismicity, it uses a ground motion prediction model developed specifically for Australia. A major update of the model has been completed to incorporate Geoscience Australia’s recent revision of the National Seismic Hazard Assessment, including a revision of the Australian Earthquake Catalogue. It also includes for the first time an active fault model.

HailAUS. HailAUS is a loss model for hail with nationwide coverage. It includes a catalogue of hailstorms reflecting the frequency and severity of ‘high storm potential days’ derived from reanalysis data and the observed historical record. In addition to calculating damage to property the model includes a motor vehicle damage estimation module.

CyclAUS. CyclAUS is a tropical cyclone wind loss model for Australia. It covers the entire region at-risk of tropical cyclone. Detailed vulnerability functions enable the estimate of loss.


For further information contact Andrew Gissing at


The 14 July 2019 Mw 6.6 Offshore Broome and Mw 7.3 Halmahera Earthquakes

by Paul Somerville

A magnitude Mw 6.6 earthquake occurred about 200 km west of Broome on 14 July 2019 (Figure 1). It is the second largest earthquake to have occurred in or near Western Australia in historical time. This earthquake was followed about 3.5 hours later by a magnitude Mw 7.3 earthquake in Halmahera, Indonesia (Figure 2).  Both earthquakes occurred at shallow depths of about 10 km, and both had strike-slip focal mechanisms, which involve the horizontal movement of one side of the fault past the other side. This is presumably why no tsunami warning was issued (and none was observed) for either event, because tsunami generation requires uplift or subsidence of the sea floor, or some other form of volume change that could be caused by submarine landsliding or volcanic eruption.  As described below, the focal mechanisms indicate that the earthquakes were caused by similarly oriented stress fields, but the large distance separating them suggests that their close time of occurrence was coincidental.

The Offshore Broome earthquake occurred at about 1:30 pm local time and was felt widely in Western Australia, from Esperance to Darwin.  The duration of the shaking was consistently reported to be between 45 seconds and one minute. There are no known reports of structural damage, but there was some non-structural damage to ceilings, and objects fell from supermarket shelves in Broome and other towns. Based on these reports, it is likely that the peak ground accelerations were in the range of 2-10%g. Several people reported that the initial shaking was accompanied by a roaring noise. This may have been caused by the acoustic coupling of the compressional (P) waves, which are sound waves in rock, into the air.  Although no tsunami warning was issued, the Shire of Broome announced just after 4pm that it would be closing Cable Beach, Town Beach, Entrance Point and Reddell Beach for the time being as a temporary precaution against any potential tidal surge.

The Halmahera earthquake occurred at about 4pm local time, and Indonesia’s earthquake monitoring agency, the BMKG, estimated that 4000 people were exposed to “very strong” effects from the earthquake.  It is reported that people were seen fleeing a building on the neighbouring island of Ternate, about 168 kilometres north-west of the epicentre.  Ternate is the largest city in the province of North Maluku, and home to about 200,000 people.  One person is reported to have been killed in southern Halmahera.

Figure 1. Left: The small white star shows the location of the Offshore Broome earthquake, and the beach ball shows a map view of the two possible fault planes, one oriented northeast and the other oriented northwest. Right: Historical earthquakes in northwestern Western Australia, showing the Offshore Broome earthquake as a red star. Source: European-Mediterranean Seismological Centre.
Figure 2. Left: Location map of the Halmahera earthquake. Centre: Detailed location map. Right: Focal mechanism showing possible northeast and northwest oriented fault planes. Source: USGS.

Causes of the earthquakes

As shown in Figure 3, the Indo-Australian Plate is subducting (diving down beneath) the Sunda Plate along the Java Trench (western part of the map), because oceanic crust is thin and dense and easily subducts.  However, the Indo-Australian Plate is colliding with the Sunda Plate in the Timor region (eastern part of the map), because in this region the Indo-Australian plate consists of thick, buoyant continental crust that cannot be subducted.  Consequently, the part of the Indo-Pacific plate that is subducting beneath the Sumatra Plate is moving to the northeast at a higher rate than the part that is colliding with Timor. This causes strike-slip earthquakes, like the Offshore Broome earthquake, to occur along the Western Australia Shear Zone, as shown by the numerous offshore earthquakes with a northeast-southwest alignment on the right side of Figure 1. The focal mechanism of the earthquake, shown on the left side of Figure 1, indicates faulting on either a northeast or northwest oriented fault plane; the northeast plane is consistent with the northeast orientation of the Western Australia Shear Zone in Figure 3. In either case, the earthquake was caused by local crustal shortening in a north-south direction and extension in an east-west direction.

Unlike the Offshore Broome earthquake, which occurred within the Indo-Australian Plate, the Halmahera earthquake occurred within the Sunda Plate, and was caused by the collision of those two plates in Timor.  As for the Offshore Broome earthquake, the focal mechanism of the Halmahera earthquake, shown on the right side of Figure 2, indicates faulting on either a northeast or northwest oriented fault plane. In either case, the earthquake was caused by local crustal shortening in a north-south direction and extension in an east-west direction, due to the plate collision shown in Figure 3.

As shown in Figure 3, the Western Australia Seismic Zone extends onshore in the vicinity of Dampier. The largest earthquake known to have occurred in historical time in Western Australia is the Mw 7.25 Offshore Geraldton earthquake of 11 November 1909. The location of that event is shown in the bottom left corner of Figure 3; the Ms of 7.8 is the surface wave magnitude. The Mw 5.58 1968 Meckering earthquake had a marginally lower magnitude that the Offshore Broome event, and practically destroyed the town of Meckering. The Offshore Broome earthquake is also larger than the 1941 Mw 5.52 Meeberrie earthquake.

Figure 3. Tectonic map of northwestern Western Australia showing the locations of the Western Australia Shear Zone (the zone outlined by white lines; mapped faults shown by red lines). The Offshore Broome earthquake occurred near the eastern edge of the zone, southeast of the letters “RS” (Rowley Shoals). The purple line, which lies just north of Timor, shows the edge of Australian continental basement. The Indonesian islands are located on the Sunda Plate to the north, and Australia is located on the Indo-Australian plate to the south. Source: Hengesh & Whitney, 2016.


Hengesh, J. V., and B. B. Whitney (2016). Transcurrent reactivation of Australia’s western passive margin: An example of intraplate deformation from the central Indo-Australian plate, Tectonics, 35, 1066–1089, doi:10.1002/2015TC004103.


The 4-5 July 2019 M 6.4 and 7.1 Ridgecrest, California Earthquakes

Paul Somerville, Chief Geoscientist, Risk Frontiers

An M 6.4 earthquake occurred near Ridgecrest, Southern California, about 180 km north of Los Angeles,  on July 4th, 2019, preceded by a short series of small foreshocks (including an M 4.0 earthquake 30 minutes prior), and was followed by a strong sequence of aftershocks, whose epicentres aligned with both possible fault planes (NE-SW and NW-SE) of the focal mechanism solution of the M 6.4 event, as shown in Figure 1. On July 6th UTC (July 5th 20:19 locally) an Mw 7.1 earthquake at the northwest extension of the M 6.4 event was preceded by 20 seconds by a magnitude 5.5 earthquake.

Figure 1. Location of the M 6.4 July 4 earthquake and aftershocks (left) and the M 7.1 July 5 earthquake and aftershocks (right), also showing in green the M 6.4 event and its late aftershocks. Source: Temblor.

M 6.4 Earthquake Ruptured Two Orthogonal Faults

The epicenter of the M 6.4 earthquake is located near the intersection of its two possible fault planes, and the distribution of aftershocks on two orthogonal planes, shown on the left of Figure 1, suggests that it ruptured both of them. As shown on the left side of Figure 2, an earthquake is represented by a shear dislocation on a fault, shown by the opposing thick arrows, which has a force representation consisting of two equal but opposing couples, represented by the two pairs of thin arrows shown near the outer edges of the cloverleaf pattern. The opposing couples maintain dynamic equilibrium by yielding no net force and no net torque. This means that, observed from a distance and only using information on the direction of first motion of the seismic waves (up or down), we cannot identify on which of the two possible fault planes the earthquake occurred.

Figure 2. Map views of the double couple representation of a vertical strike-slip earthquake mechanism.

As shown in the centre panel of Figure 2, the same force system – north-south compression and east-west extension, shown by the wide double arrows, can produce strike-slip movement on either the northwest (left) or northeast (right) striking fault plane.  This explains why the M 6.4 event could rupture both of the two possible fault planes in the same earthquake. Seen from afar using first motions of P waves, the two potential fault planes are demarked by up (compressional) and down (rarefactional) quadrants that cover the earth’s surface, as shown on the right of Figure 2.

To resolve which plane or planes hosted the earthquake, we need to look at aftershock locations (Figure 1), look for surface faulting (Figure 3), or analyse the mainshock waveforms to identify where the seismic waves actually came from (Figure 4).  The latter has been done of the M 7.1 event (Figure 4), which shows horizontal rupture of up to 2.5 metres over a length of 60 km of the fault extending from the ground surface to a depth of about 10 km.  The road that was ruptured by both earthquakes is State Highway 178, whose location is shown partially in Figure 5; it extends eastward from Ridgecrest.

Figure 3. A few cm of left-lateral surface rupture of the M 6.4 earthquake (left) and about one metre total of right-lateral surface rupture of the M 7.1 earthquake (right). Here and in Figure 2, left lateral means the other side of the fault from the one on which you are standing has moved to the left; and right-lateral means it has moved to the right.
Figure 4. Slip on the vertical fault plane of the M 7.1 earthquake, right end is southeast. Source: USGS.
Figure 5. Map of earthquake epicentres and roads. Source: USGS.

Relationship to the San Andreas Fault

The San Andreas fault, which runs diagonally from northwest to southeast on the left side of Figure 6, forms the main and long-established boundary between the North American plate to the east and the Pacific plate to the west. It runs through San Francisco and lies close to Los Angeles and other Southern California cities, so news of a large earthquake in California raises alarm among the general public.

Figure 6. Fault map of central and southern California (CGS, left), with Ridgecrest located above the first “M” in “Mojave Desert”(see also Figure 7 for location), and map of the 1872 Owens Valley earthquake showing the location of the Ridgecrest earthquakes in the lower right (Temblor, right).

However, another incipient part of the boundary is forming to the east of it, running in a north-northwesterly direction, and it hosted the M 7.6 Owens Valley earthquake of 1872 and the July 2019 Ridgecrest earthquakes, shown on the right side of Figure 6. The orange lines in Figure 6 indicate faults that have not ruptured in historical time, and the red lines are ones that have. The three big historical earthquakes in California are the M 7.9 1857 Fort Tejon, south-central San Andreas earthquake (roughly from the latitudes of San Luis Obispo to Riverside, Figure 6, left), the 1872 M 7.6 Owens Valley earthquake (Figure 6, right), and the 1906 M 7.8 San Francisco earthquake on the Northern San Andreas fault (roughly from the latitude of San Luis Obispo past San Francisco, off the map in Figure 6, left). Since the 1872 Owens Valley earthquake, other earthquakes that have occurred recently on this incipient eastern plate boundary are the 1992 M 6.3, Joshua Tree, 1992 M 7.2 Landers, 1992 Big Bear, 1995 M 5.8 Ridgecrest, and 1999 Hector Mine earthquakes. The rupture zones of the Landers and Hector Mine earthquakes are shown by red lines east of Victorville on the left side of Figure 6.

Reported Damage and Future Warnings

The MMI Intensity shakemap of the earthquake is shown on the left side of Figure 7.  The main damage to the towns on Ridgecrest (population 29,000) and Trona (population 1,900) appears to have been incurred by older houses and trailer homes (which readily topple from their foundations, right side of Figure 7); some house fires also started but were soon extinguished.  There were no reported deaths or serious injuries.

Figure 7. Shakemap (USGS, left), and damage to a trailer home and a fire (right).

The Governor of California announced that the estimated losses are about $US100 million, but the USGS made a preliminary estimate of at least $1billion. It is possible that this larger estimate may include damage to the Naval Air Weapons Station China Lake, a 1.1 million acre weapons testing facility, the Navy’s largest, whose location is shown in Figure 1 and within which the epicentres of both the M 6.4 and 7.1 earthquakes were located.  The facility’s Facebook page announced that it is “not mission capable until further notice,” but officials said that security protocols “remain in effect.”

According to the current USGS forecast, over the next one week, beginning on July 6, 2019 at 2:20 p.m. Pacific Time (5:20 p.m. ET), there is a 2% chance of one or more aftershocks that are larger than magnitude 7.1. The number of aftershocks will drop off over time, with the largest expected to have a magnitude of about 6, but a large aftershock can increase the numbers again temporarily.

The recently installed early earthquake warning system ShakeAlert, which detects earthquakes and announces that they have occurred in near real time, worked as planned but not as some people would have liked.  The designers of the ShakeAlert LA app decided that most people would not want to be notified of large earthquakes that are too distant to cause damaging shaking near them.  Consequently, the ShakeAlert LA app was designed to only send an alert if the magnitude is above 5 and the MMI intensity is 4 or higher somewhere in Los Angeles County.  The Ridgecrest earthquakes met the magnitude criterion but not the intensity criterion in Los Angeles, and so did not result in alarms.  However, many people were alarmed because they clearly felt the long “rolling” motions (surface waves) of the distant large earthquakes and were concerned that the alarm was not working properly.  The designers of the app may now consider providing alarms, perhaps nuanced, of large distant earthquakes that people may feel but that do not present local damage potential.