Director: Professor Russell Blong

NHRC is kindly sponsored by:
Swiss Re
QBE Insurance
Benfield Greig Australia
Guy Carpenter & Co

In this issue:

In the months since the 1998 floods in Townsville, Katherine, Gippsland, the Namoi Valley, Bathurst, Wollongong and other parts of Australia, both the flood insurance and the regulatory scene have altered perceptibly in New South Wales and elsewhere.

Flood insurers have struggled to come to grips with the insurability of domestic property against flash floods and mainstream floods.  Some insurers believe flood insurance is the path to ruin, while others fear that not offering flood cover promises eternal damnation.  Some insurers have offered cover for flash floods but not mainstream floods, leaving the inevitable debates about rainfall where, the time from rainfall peak to flood peak [or should that be rainfall onset to the initiation of flooding?] to a later date.  A new definition of flood, at least as eye-glazing as the old one, has appeared.  The goalposts may have moved, but have they moved to positions that will please policy holders next time the waters rise?

Meanwhile, inexorably, the New South Wales Government has produced a Draft Floodplain Management Manual to replace the 1986 Floodplain Development Manual.  As the name change implies, the emphasis has swung from development to ecologically sustainable development.  An unforeseen consequence of the old manual was a perception by many, including many who should have known better, that land above the level of the 1-in-100 year flood [an Average Recurrence Interval - ARI - of 100 years in a very long record] is flood-free.  As a result of this misconception, many river valleys in New South Wales and elsewhere in Australia have hundreds of homes strung out across floodplains at levels just above the 1-in-100 year flood level.

In the new Manual the floodplain is defined as the area subject to inundation by floods up to the Probable Maximum Flood [PMF].  The ARI of the PMF [the Average Recurrence Interval of the Probable Maximum Flood] is generally regarded as indeterminate, but it is usually placed in the range 1-in-10,000 to 1-in-100,000 years.  Data from the floodplain [as defined] of the Hawkesbury - Nepean River west of Sydney provides an interesting illustration of the estimated present direct flood damage for domestic property, commercial/industrial buildings and motor vehicles. Figure 1 [after Molino Stewart, 1997].

Figure 1 impinges on issues regarding the insurability of flood.  The pool of flood-prone properties is larger than imagined, while the probability of inundation on the upper parts of the floodplain is suitably low.  The Probable Maximum Loss [PML] is in the several billion dollars range - the same range as the PML for a Sydney hailstorm, but several orders of magnitude less frequent.

Figure 1 also makes it pretty clear that for the Hawkesbury Nepean at least, PML estimation, calculations of Average Annual Damage, and efforts to rate properties for flood risk need to consider floods rarer than the 1-in-100 year ARI event.  In fact, the severity of the problem will vary with the relationship between water depth [or stage] and flood frequency.  For many rivers in inland New South Wales where the floodplains are so broad that the dramatic increase in flood discharge from, say, 1-in-100 years to 1-in-1000 years may increase inundation depth by only a few tens of cm and make little difference to the total damage.

Flood Risk Assessment

For insurance, planning or emergency response purposes, the flood risk assessment framework needs to be implemented on a property-by-property basis.  Risk assessment at the postcode level is unsatisfactory for most natural perils except perhaps hailstorms, but it is fallacious for flood risk.  There would be very few postcodes in Australia where more than 10-15% of land parcels are flood-prone even then many of the land parcels subject to inundation have no buildings on them.

Figure 2 illustrates the problem with flood risk assessment at the street level for a small portion of the floodplain of the Georges River in Sydney.  Background shading indicates water depth.  Circles, triangles and diamonds indicate the frequency of inundation of ground level, with land parcels grouped into four categories - below the ARI 20 year flood level, ARI 20 to 50 years, ARI 50-100 years and above ARI 100 years.  It is clear from Figure 2 that all four of these risk categories can exist in the one street.

The Natural Hazards Research Centre has developed a methodology allowing flood risk assessment on a property-by-property basis, using a 5 m or 25 m Digital Terrain Model, estimation of available flood surfaces, street addresses, and geocoding.  For each land parcel an estimate is provided of the ARI of inundation of the ground surface, and points 1 m and 2 m above the ground surface.  Where data on actual floor elevations are available, these can be substituted readily in the model to obtain average recurrence intervals for inundation of the floor level.

The Flood Inundation Risk Model [FIRM] has been developed with mainstream flooding in mind.  In theory, FIRM could be used for any area where a range of flood surfaces has been established.  However, we doubt that results from FIRM, or any other flood risk assessment model, are realistic in small catchments where floodplains are difficult to identify and where cars, trees and other debris can block channels and alter flowpaths.
Although NHRC hold excellent databases on flood damage, these date from the 1980s. These loss estimates now provide only a rough guide to current damage estimates; we await the opportunity to make new damage assessments.


  1. Flood risk assessment must be on a property-by-property basis
  2. FIRM provides a cheap, appropriate and flexible model for flood risk assessment
  3. Floods rarer than the ARI 100  year flood must be considered
  4. Future floods provide opportunities to collect the right damage information.


Molino Stewart, 1997, Hawkesbury - Nepean River - Impacts of flooding on communities and infrastructure, Molino Stewart, Parramatta.

Russell Blong & Laraine Hunter
Phone: +61-2-9850 9683
Fax: +61-2-9850-9394

Natural Hazards Quarterly, September 1999, Volume 5 Issue 3.

The NHRC's recently developed hail loss model is a probabilistic simulation model that estimates insured property losses based on thousands of synthetically generated hailstorms and relevant exposure and vulnerability information. Within the broad configurations of Sydney/Brisbane and houses/ motor vehicles, the model may be run in a number of modes. As shown on the model interface (Figure 1), individual company portfolios may be analysed, or in-built exposure data representing all insured units at risk in a particular city may be used.  Losses are calculated on a per-storm basis or as annual aggregates for a nominated number of storms or years.

Figure 1: Model interface showing different simulation modes

The amount and cost of damage caused by hailstorms depends on the physical characteristics of the storm and a range of factors relating to the exposure and vulnerability of the units at risk. These factors are incorporated into the first three elements of the model (Figure 2). The final (analysis) module uses input from the other modules to generate probabilistic outputs and presents the results in graphical and tabular formats.

Hazard Occurrence Module

This part of the model produces sets of stochastic hail events; each characterised by the following variables:

  • Direction and path of the storm
  • Areal extent of the storm
  • Hailstone size
  • Time of day
  • Day of week
  • Time of year
  • Annual frequency

The characteristics are varied in accordance with statistical distributions derived from the NHRC's hail data sets and other climatological information. Useful and reliable simulations require comprehensive, good quality hailfall data sets, a sound understanding of the climatology and meteorology of severe hailstorms, and careful statistical analysis. The NHRC has worked hard to develop these resources they form crucial inputs to the hazard occurrence module.

Exposure module

The output from the hazard occurrence module is then linked with data on the ground, whose distribution is modelled by the exposure module. The exposure module estimates the number of cars or houses in a range of vulnerability classes that are under the footprint of each storm. This is relatively straightforward for houses, depending purely on the size and location of the storm. Estimating the number of motor vehicles exposed to a particular storm is more complicated, depending on the number of cars in the affected region and whether they were under shelter or not. Both these factors vary with the time of day, day of the week and time of the year. The number of cars registered in different regions is a good starting point and reveals basic information about the risk profile for cars across the city (see Figure 3). 

Figure 3: Density of vehicles by postcode in Sydney (ICA zones 41-43)
based on registration address
Most cars are only at their garaging or registration address for a small proportion of their lives, the rest of the time they are parked elsewhere or are on the road. The exposure module captures these movements to build a picture of the vehicle populations in different areas throughout the day for different types of days (weekday or weekend). Similarly the proportion of cars parked under shelter (and therefore not exposed to hail damage) in a given area at a given time is variable.

Vulnerability Module

The information on physical and spatial distribution of the hail damage is integrated with loss data via the vulnerability module. This part of the model contains information relating the cost of damage sustained by cars or houses to some measure of hail intensity. Although there are numerous measures of hail intensity (such as kinetic energy, mean or modal size, stones/m2), damage appears to be well correlated with maximum hailstone size. Maximum hailstone size is also the characteristic about which we have the most information.

Houses and cars are divided into classes based on differing vulnerability to hail impact damage. This has allowed the development of a range of damage - hail size relationships that can be calibrated as more information becomes available. Loss curves exist for other hazards but have not previously been developed for cars and houses exposed to hail. The insurance industry is the best source for the damage cost information required for the vulnerability module. Close liaison between researchers and the insurance industry is necessary.  Improvements in the quality (and availability) of exposure and vulnerability data provided by insurers will enhance the performance of the hail loss estimation model.

Analysis Module

In the analysis module of the hail model the ensemble of the results are transformed into a probabilistic output format useful to the insurance industry. Results are presented on an Insurance Council of Australia accumulation zone basis in graphical and tabular format. Distributions of insured losses and loss ratios are presented in terms of frequency and probability of recurrence. Such loss profiles can provide a basis for probable maximum loss (PML) estimation, and also embody important information about the nature of the risk across a range of return periods. Files containing statistical parameters and detailed characteristics of individual storms may be exported for supplementary analysis.


The NHRC's hail loss estimation model for Sydney and Brisbane is a unique tool for analysing potential insured losses from hail damage to houses and motor vehicles in these urban areas. The outputs are probabilistic in nature allowing the risk to be analysed across a range of return periods. Given appropriate data, the framework developed for this model could be adapted to analyse the problem of commercial and industrial losses in urban areas, or crop losses in rural regions.

For more information please contact:

Roy Leigh or Ivan Kuhnel
Phone +61-2-9850 9683
Fax: +61-2-9850 9394

NHQ Homepage

NHRC Homepage