Kempsey, a small rural city situated on the Macleay River in New South Wales Mid-North Coast region, suffered major flooding during March 2001. Flooding was a result of an intense East Coast low-pressure system located offshore in the Tasman Sea. Intense rainfall saw the river peak on Saturday the 10th, recording a peak height of 7.43m AHD at the Kempsey traffic bridge gauge. The 2001 flood is the 9th highest since 1838, with an estimated Average Recurrence Interval (ARI) of 12 years. Kempsey’s last major flood occurred almost 38 years before; as a result few residents had previously experienced major flooding.
Approximately 94 businesses in the CBD suffered overfloor flooding ranging from a couple of centimeters to three metres. Water velocities appear to have been low. The duration of inundation varied from several hours to about 90 hours.
The resulting total direct commercial damage was estimated at $2.5 million. Damage incurred by individual businesses ranged from several hundred dollars to almost half a million dollars.
Stage damage curves were constructed from survey data collected to establish the relationship between stage depth and direct damage. The scatter of data points in the figure is evidence of the commercial sector’s heterogeneous nature. The coefficients of determination for the fitted curves indicate that little relationship exists between overfloor depth and direct damage. This suggests that other factors may be more significant in determining the amount of direct damage, factors such as warning time, the extent of preparation, construction materials, flow velocities, sediment levels, etc.
The distribution of data illustrates that on average businesses experiencing longer durations of inundation also experienced higher overfloor flood depths. The fitted curves indicate that above 0.3 metres over-floor depth, higher direct damage is experienced at greater durations of inundation.
Floor-coverings were the most frequent items to be damaged, with carpet predominate. Other items to record high frequencies included office furniture, white goods and shop fittings. Though recording the highest frequencies these four categories total 36% of total direct damage value. Damaged stock (19%) and miscellaneous equipment (26%) contributed most to total direct damage value.
The flood’s indirect effects were felt long after the water receded. The majority of businesses closed for about one week, with some businesses remaining closed for periods up to a month. Franchises were able to re-open faster than other businesses due to the support given by businesses within their networks. Approximately 60% of businesses experienced a downturn in trade for a period after the event, with 30% of shop owners believing that trade had been lost to competition.
The initial research leads to the conclusion that due to the heterogeneous nature of the commercial sector there are many variables that explain the extent of both direct and indirect flood damages.
Severe thunderstorm cells crossed the Gold Coast – Brisbane – Sunshine Coast area on the afternoon and evening of 9 March 2001. While no hailfalls were reported and windspeeds do not seem to have been exceptional, the cells dumped high intensity rainfalls across much of the area. Intensities expected on average once in 100 years were exceeded for durations from tens of minutes to 11 hours at several Bureau of Meteorology stations in the area.
Insurance Council data indicate nearly 12,300 claims have been paid to a total value of $36.7 million. On our reckoning, this places the event about tenth on the ICA list (1967-2001) of storms (not including hailstorms or tropical cyclones). A 100-year rainstorm, but a fairly ordinary insurance loss – natural perils losses of this size have averaged one every 8 months or so for the last 35 years. Using 735 domestic claims files supplied by RACQ, Risk Frontiers examined several aspects of the March 2001 insured losses.
The first table summarises the proportions of $ damage that can be assigned to major building components. The summary emphasizes the importance of water penetration through roofs and into ceiling spaces, but the surprising feature is the high proportion of damage to garden and pool areas.
Loss assessors’ reports and other damage data allowed us to determine the causal thunderstorm elements for nearly 94% of the total building and contents’ damage recorded in the sample, though evidence for differentiating between runoff and flood is almost always suspect and between rainwater and wind sometimes slender. Nonetheless, the overwhelming impression is that more than half the damage was produced by water flowing over the ground surface while less than a third is the direct product of rainfall (see table below). There is nothing very surprising in this but perhaps we have, for the first time, sufficient data to quantify the view.
Comparison of the two columns in the table indicates that the runoff, flood and seepage claims tend to be above average in value while lightning claims and wind-related claims are relatively cheap.
Intuitively, we would expect more claims where rainfall (and, hence, runoff) were most intense. The figure illustrates persuasively that this is the case. However, preliminary analyses suggest limited relationships between rainfall amount and claim amount.
Risk Frontiers wishes to record the excellent assistance provided by Ian Norris, Brad Heath and other RACQ staff, and Cathy Muller of the Bureau of Meteorology in preparing this report.
One of the least developed tasks in risk assessment is vulnerability assessment, partly hampered by the lack of relevant spatial data, which should include physical vulnerability data on the built environment, and socioeconomic data such as population and incomes. This short article illustrates the extraction of useful building features such as roof areas, and other urban surface covers for potential damage assessment in risk assessment, by using high resolution imagery.
Air- and space-borne remotely sensed images with high spatial, temporal, and spectral resolutions hold promise of providing detailed spatial data on hazards, built environment, and their consequences (e.g., flooded areas, bushfire-destroyed properties) across various geographic scales. Such images include IKONOS satellite imagery (panchromatic image with 1m spatial resolution, multi-spectral 4 m, www.spaceimaging.com), AUSIMAGETM digital aerial photographs (colour, ortho-corrected, 0.2 m spatial resolution, www.ausimage.com.au), and hyper-spectral images generally with over 100 wavelength channels and 2-10 m spatial resolution. Below we demonstrate the use of AUSIMAGETM data for the extraction of roof areas and other covers.
Both pixel- and object-based image processing methods are used. First, in order to decrease the mixed pixels with high intensity variance and noise of ground objects associated with high resolution images, and to remove the strong “salt-and-pepper” effect in the classified image by using classic methods such as maximum likelihood classifier, image textures including means and energy with Red/Green/Blue colour bands are incorporated in a supervised artificial neural network classifier to achieve good results on different land surfaces. Next, within object-based image processing methods a hybrid of edge and region segmentations using colours, textures and shapes is employed to extract useful spatial information on ground objects. Finally, the extracted spatial information is used to refine the pre-classified image.
The figure and table show an image clip and extracted spatial data of individual buildings. The directly extracted data can include locations, areas and perimeters. Derived data could be distances from building centroids to street centre lines, distances between adjacent buildings, etc. Other surface covers of interest can include trees, lawns, roads and building shadows. There are advantages in generating these data layers by using a semi-automatic feature extraction approach. The approach is applicable to large areas in a cost-effective way, instead of using labourious digitisation and/or measurement. The detailed map of buildings is distinct from the conventional cadastral map, which often only indicates an overall land parcel area per address rather than the real physical locations of buildings.
It is evident that the above data have a great potential use in bushfire, hail, and flood risk assessment. For example, the vegetation amount and density can be quantified for those buildings in the interface between urban and bush lands; it is possible to estimate surrogate potential damages from hail based on delineated roof areas. Given the above detailed data on a property-by-property level, it is very straightforward to extract those parameters at coarse geographical scales, such as census collection districts and postcodes. A multi-scale and hierarchical examination on the damage assessment can be conducted. The extracted data are also complementary to those already existing in vector-based GIS, including StreetWorks and areal census data. As a result, it facilitates wider and in-depth data integration and spatial analysis.
The extraction of insurance-relevant data on residential buildings is an on-going project. Methods and implementations of image processing are critical to achieve rigorous results and produce accurate risk-related thematic layers. More importantly, practical case studies for the risk assessment of individual hazards and multiple hazards at large areas will be pursued in the future.
AAD is a convenient single damage statistic that incorporates the effect of both the magnitude and frequency of damages. Effectively the damage associated with a particular event is weighted by the probability of occurrence of the event. On a plot of damages against the associated probabilities, AAD is equal to the area under the curve joining the data points.
From an insurance perspective AAD may be considered to be the annual pure risk premium required to insure the properties included in the analysis.
Risk Frontiers – NHRC has developed loss estimation models for selected areas in Australia for flood, hailstorm and earthquake perils. These models (FloodAUS, HailAUS and QuakeAUS) can be used to estimate AAD and Probable Maximum Losses (PML) for specific residential property portfolios. This allows us to consider the following questions:
1. Which peril produces
the highest PML?
The results of analyses of a synthetic portfolio of approximately 90,000 houses spread across 208 postcodes in CRESTA Zones 41-43 are shown in Tables 1 and 2. The synthetic portfolio has an average building sum insured of about $184,000, an average contents sum insured of about $61,000 and a Total Sum Insured of $21.85 billion. About 38% of the dwellings are within the Hawkesbury-Nepean, Georges and Upper Parramatta River catchments, and only mainstream flooding from these rivers is included. The results, of course, assume that flood is fully insured.
Table 1 does not yield an unequivocal answer to the first question, particularly when factors such as the scatter around the estimates, the exclusion of local flooding and building collapse from the flood analysis, and the difficulty of allocating return periods to earthquake losses are considered. Nevertheless, based on this analysis, it seems that at commonly applied PML return periods of between 100 and 500 years, flood (if insured) and hail are just as likely to be the source of the defining loss as earthquake.
The answer to the second question depends on whether one considers the whole portfolio or only the dwellings at risk. When considering the entire portfolio, hail yields the highest AAD. Flood is different to the other perils since not all dwellings in the portfolio are at risk. Consequently the AAD per dwelling at risk is high. However properties that are vulnerable to high flood losses are easier to identify than properties that are vulnerable to high earthquake or hail losses. This means that, using information about the location, ground height and floor height of houses within a portfolio, flood losses can be relatively easily mitigated through prudent portfolio management. Strategies such as excluding properties located below the level of the ARI 20-year flood can dramatically reduce AAD estimates.
The loss estimates presented here are realistic as they are based on an amalgam of portfolios provided by four insurers. The results, however, are portfolio specific - and clearly Any Answer Will Not Do.
To find out which peril produces the highest PML or AAD for your portfolio contact Roy Leigh or Laraine Hunter on tel: 9850 9683, fax: 9850 9394 or email: firstname.lastname@example.org