Accidental Ammonium Nitrate Explosions

Paul Somerville and Ryan Crompton, Risk Frontiers

Insurers are gearing up for what is likely to be one of the most expensive insured cargo and port infrastructure losses ever from the Beirut explosion, on a scale at least as large as the one resulting from the explosions at the Chinese port of Tianjin in 2015 (800 tonnes, 173 deaths). It is expected that Lebanon’s second port of Tripoli, believed to be operating at just 40% capacity on account of COVID-19, will become the country’s main gateway for both emergency supplies and normal trading.

The accidental ammonium nitrate explosion in Beirut serves as a reminder of how frequent and deadly these events are. A timeline and description of events since 2000 is shown in Figure 1. The Wyandra, Northern Territory event of 2014 is shown in Figure 1 and is one of three Australian events, described later, that appear in the Han (2016) catalogue.

Timeline of accidental ammonium nitrate explosions.
Figure 1. Timeline of the largest accidental ammonium nitrate explosions in the world since 2000.  Source: VisualCapitalist.

We analysed Han’s (2016) catalogue which lists 79 events since 1896, 42 of which occurred in the United States, to assess their frequency of occurrence. Since 1900, they have occurred at a uniform rate of about 0.75 per year, with an increase to about 1 per year since 2000. To the extent that Han’s list is incomplete, these rates are underestimated. Han (2016) notes that the ammonium nitrate that exploded in the 1947 Texas City (Galveston) event was coated with wax to prevent caking.  Practices introduced in the 1950s eliminating the use of wax coatings yield ammonium nitrate, used in fertilisers, that contain less than 0.2 percent combustible material. This practice does not appear to have impacted the frequency of events.

We also analysed the Wikipedia catalogue of 36 events, which lists both size (tonnes) and deaths, to assess the relation between size (tonnes) and number of deaths, shown in Figure 2. To first order, Log10 Deaths = 0.85 log10 Tonnes. Four notable events on the left panel of Figure 2, clockwise from top left, and labelled by numbers of deaths, are the 1921 Oppau, Germany event (450 tonnes, 561 deaths), the 1947 Texas City (Galveston, U.S.) event (2906 tonnes, 581 deaths),  the 2020 Beirut event (2750 tonnes, 220 deaths), and the 1947 Brest, France event (3000 tonnes, 29 deaths).

The relationship between size and deaths from ammonium nitrate explosions.
Figure 2. Relation between size (tonnes) and deaths from accidental ammonium nitrate explosions on linear (left) and log (right) scales. Several zero values on the axes of the log plot actually represent zero values: the 2004 North Korean event of 162 tonnes had no reported deaths.

Australian-based company, Orica, the world’s largest provider of commercial explosives and blasting systems to the mining, quarrying, oil and gas and construction markets, has a stockpile of ammonium nitrate up to four times the size of the one in Beirut. There are many stockpiles in Australia but Orica’s Kooragang Island plant has received a lot of media attention.

Between 6,000 to 12,000 tonnes are currently stored at Orica’s Kooragang Island plant in the Port of Newcastle, which produces approximately 400,000 tonnes each year. This plant is located 3 km from Newcastle’s CBD and 800 m from residents in Stockton. Up to 40,000 people live in what would be the ‘blast zone’ if there were to be an explosion. Orica state that they follow strict safety protocols and ensure that the ammonium nitrate storage areas are fire resistant and built exclusively from non-flammable materials, with no flammable sources within designated exclusion zones. The operations on the Kooragang Island site, which has been in operation for 51 years, are highly regulated to state and federal standards. The site’s safety management systems, security arrangements, and emergency response procedures undergo a strict auditing and verification process by SafeWork NSW. The Kooragang Precinct Emergency Sub Plan can be found here.

The safety of ammonium nitrate was previously highlighted in South Australia in 2013, when concerns were raised about the location of the Incitec Pivot fertiliser plant at Port Adelaide following the West, Texas, explosion of 2013 involving 240 tonnes of chemical that killed 15 people in a 50 unit apartment block. In 2013, the South Australian Government made an agreement with Incitec Pivot to move its plant away from the heart of Port Adelaide, because it posed an unacceptable risk to residents of a proposed major development there. The company moved its operations to a location further from the centre of Port Adelaide to Gillman in 2018. According to SafeWork SA, all 170 of the ammonium nitrate storages in the state are heavily regulated, heavily controlled and monitored.

Figure 3. Left: Incitec Pivot plant, Port of Adelaide; Right: Orica Kooragang Island plant.

In the remainder of this briefing we describe three Australian accidental explosions, all involving trucks.

Taroom, Queensland, 30 August 1972

A truck explosion occurred near Stonecroft Station on Fitzroy Development Road in August 1972. The truck and trailer were carrying 12 tonnes of ammonium nitrate. The truck experienced an electrical fault and caught fire north of Taroom. After the driver stopped and parked the burning truck, two brothers from a nearby cattle property who saw the fire rode up on motorbikes to assist. The three men were killed when the truck exploded at around 18:15. The explosion destroyed the prime mover and trailer, leaving a crater in the road 2 m deep, 5 m wide, 20 m long. Parts of the truck and trailer were scattered up to 2 km away. The explosion burnt out more than 800 hectares (2,000 acres) of surrounding bushland. The explosion was heard and shook houses 88 km away in Moura and 55 km away in Theodore.

Wyandra, Queensland, 5 September 2014

On September 5, 2014, an ammonium nitrate truck explosion (Figure 4) occurred near Wyandra, about 75 km south of Charleville in south-west Queensland, Australia. The truck carrying 56 tonnes of ammonium nitrate for making explosives rolled over a bridge and exploded, injuring eight persons including the driver, a police officer, and six firefighters. Rescue crews were trying to extract the driver from the truck when they found out there was ammonium nitrate inside. They were making a mad dash from the truck when it exploded.

The prime mover caught fire about 9.50pm and the driver steered off the highway, causing it to hit a guard rail near the Angellala Creek Bridge and roll onto its side in the dry creek bed. The crash led to two explosions occurring at 10.11pm and 10.12pm. The blast was so powerful that the truck disintegrated, destroying two firefighting vehicles along with it and causing catastrophic damage to the Mitchell Highway. Two road bridges were destroyed (Figure 5), one of the railway bridge spans was thrown 20 m through the air, and a major section of the highway was missing. Geoscience Australia recorded the explosion as a magnitude 2.0 event, and coincidentally, 20 minutes after the explosion, a magnitude 2 earthquake was recorded 55 km south of Charleville.

Emergency vehicles damaged by the Wyandra truck explosion. Queensland Police Service
Figure 4. Emergency vehicles damaged by the Wyandra truck explosion. Queensland Police Service.

The dangers posed by the remaining ammonium nitrate led to a 2km exclusion zone around the site for a number of days. The large crater formed by the blast closed the highway necessitating detours of up to 600 km, including a 100 km detour to Cunnamulla along the Charleville-Bollon Road. In April 2015, the $10 million tender to reconstruct the highway and bridges were awarded and the construction work took place between June and November 2015.

Damage to bridges caused by the Wyandra explosion.
Figure 5. Damage to bridges caused by the Wyandra explosion. Queensland Police Service

Queensland Transport Minister Scott Emerson noted that there are rules in place relating to signage and the particular routes that are allowed to carry dangerous goods and that he would be talking to police about whether anything was done wrongly. However, Assistant Fire Commissioner Dawson dismissed concerns that such a volatile material was being carried in trucks. “Not so much a worry; this product – and trucks like this very same truck – travel these roads every day,” he said. “Every day they’re out there and they don’t go bang. Something’s happened to bring this truck in a situation, which has possibly mixed the product on the back of the truck – maybe with the diesel fuel, the impact of the initial [crash] when it goes off the road – so those circumstances have had more of a connection to the end result. You’d be surprised – there’s a lot of these trucks – they do it very safely and very effectively.

On January 10, 2019, the Queensland State Government launched a lawsuit in the Brisbane Supreme Court claiming more than $7.8 million in damages, the estimated cost of building a temporary detour, and inspected the area to ensure it was safe as well as replacing the road and railway bridge. It was holding the trucking company, Kalari Proprietary Limited, road train driver Anthony David Eden and insurer Dornoch Limited responsible for the repair bill.

Ti Tree, Northern Territory, 18 November 2014

A road train consisting of three flat-bed trailers carrying ammonium nitrate fertiliser exploded in Ti Tree, NT on November 19, 2014 (Figure 6). Witnesses at the Ti Tree roadhouse, 200 km north of Alice Springs, saw a fire igniting on the left-hand side of the rear axle of the rear trailer. The road train driver inhaled fumes as he desperately unhooked the burning trailer of explosive ammonium nitrate from his truck on the Stuart Highway at Ti Tree. Moments later the trailer exploded with a loud bang, startling residents more than several hundred metres. The driver had towed away the two other trailers of ammonium nitrate. No-one was injured.

Police went door-to-door to evacuate residents to the school and establish a 1 km exclusion zone. Sixty to eighty people were evacuated to the school at the northern end of town at 10:30 pm, and were allowed to go home at 1:30 am but there was still a 300 m exclusion zone. At 2:00 am the fire crew declared the fire ‘safe’ and Stuart Highway was reopened.

Figure 6. Ti Tree explosion (left, Nicolai Bangsagaard) near the Ti Tree Roadhouse (right, Olivia Ryder).


From the Vault: What we knew about a future pandemic in 2005

Paul Somerville, Briefings editor

The following briefing is a reproduction of the article entitled “A Future Pandemic” that was published in our Quarterly Newsletter Volume 5 Issue 2, December 2005. It was written by Risk Frontiers’ former employee Jeffrey Fisher and Peter Curson, Emeritus Professor at Macquarie University, and edited by John McAneney. In the light of the current coronavirus pandemic, the article was very insightful and needs no further introduction. Additional Briefing Notes on this theme are numbers 121 and 173 (available on request). From time-to-time we will return to our Insights “vault” to assess how well our understanding of natural hazards and other extremes stands up to the test of actual events.

It is difficult to pick up a paper or watch television today without seeing some reference to bird flu, H5N1, or a possible influenza pandemic and the world’s lack of preparedness for it. One thing seems clear: in an increasingly interconnected world, where 1.5 billion people cross international borders by air every year, a virus could circle the globe very rapidly, possibly even before it was detected. In contrast, the so-called 1918-19 ‘Spanish Influenza’ took some 18 months to circle the globe and about four to six months to do its damage in any one country. This article examines some implications of such an event for the life insurance business and the wider economy.

Some insurance and reinsurance companies have prepared for the eventuality of a pandemic, assessing their risk and taking steps to offset expected losses. Financial instruments such as mortality bonds, the life insurance equivalent of catastrophe bonds, have been used to transfer some of the risk to the capital markets. However, catastrophe modelling, now standard for non-life lines of business, seems far less sophisticated in the case of life insurance. Many companies still appear to be working out what their losses might be.

Some Basic Numbers for Australia

So how many people are likely to die if a flu pandemic reaches Australia? If there were a repeat of the ‘Spanish Influenza’ pandemic, the death toll in Australia could be somewhere between 60,000 and 80,000. To put this in some context, some 130,000 Australians die in any one year, a sum that includes about 2,000 from influenza. Thus, a repeat of the 1918-19 scenario would represent an increase in the annual death toll of over 50%.

Circumstances today, however, are very different from 1918. Medical and community health standards have improved dramatically. In 1918, intensive care wards had yet to be developed; there were no effective drug therapies for pneumonia; knowledge of viruses was rudimentary; and doctors had no antibiotics or antiviral drugs. Taken together, these factors could reduce the death rate considerably below that experienced in 1918-19.

On the other hand, the mobility of people today could allow the disease to spread very rapidly causing a dramatic increase in patient numbers in a short space of time. This has the potential to overwhelm the health system and reduce the benefits of modern medicine because of a shortage of drugs and hospital beds. Thus, while the death rate is unlikely to be as high as for the 1918-19 pandemic, it nonetheless remains a good benchmark as a plausible worst-case scenario.

An Optimisation Problem

The current strain of bird flu has killed over half the people known to have become infected with it. While this is cause for concern, a flu pandemic could not develop with a mortality rate this high. The mortality rate is defined as the proportion of the infected population that dies from the disease.

Influenza viruses have an initial period when an infected person exhibits no symptoms. This is the virus’s window of opportunity to spread; once symptoms appear, it is fairly easy to isolate cases and prevent further transmission. Furthermore, approximately half of all people who catch the virus get only a mild case with no obvious symptoms while still being infectious. These two attributes allow a flu virus to spread throughout a population. An influenza virus that kills its host too quickly will die out before it can cause a global epidemic.

Figure 1: Modelled deaths in a population of 10,000 as a function of mortality rate.

Risk Frontiers has a simple simulation model to examine this issue. A typical simulation deals with a population of 10,000 people. They are assumed to be a fairly homogenous group in the same geographic area – imagine a small Australian town or suburb. One way of incorporating the viral attributes mentioned above is to assume some degree of negative correlation between the length of the infectious period and the mortality rate. Negatively correlating these variables means that, on average, as the mortality rate increases, the length of the infectious period decreases and so less people catch the disease. Figure 1 shows this trade-off. Initially, as the mortality rate increases, more people die. Beyond a rate of around 1.6%, however, the tide turns as the likelihood of someone dying after catching the disease increases but the total number of people infected goes down.

The real concern is that the current strain of bird flu will combine with human flu viruses or develop the ability to jump directly to humans. If this were to happen then the global death toll could be very high. The 1918-19 flu virus killed, depending upon different reports, somewhere between 1.2% and 2.8% of people who contracted it, i.e. close to the optimum shown in Figure 1. A very well-designed bug!


Clearly, targeted vaccination at the source of an outbreak is likely to be the best means of avoiding its wider dissemination. In a recent article, Ferguson et al. (2005) explores the efficacy of using such targeted preventative medicine. These authors argue that if good detection measures are in place, if anti-viral drugs are stockpiled appropriately and deployed quickly, then the chances of containing an outbreak at source by treating everyone in the vicinity would be greater than 90%. While this conclusion is encouraging, neither sufficiently rapid detection nor efficient implementation of preventative measures can be taken for granted in the countries where outbreaks are most likely to occur.

Australia is an unlikely source of the disease and it is far more likely that the general population would have to be vaccinated. Let’s assume for the moment that this is possible. The proportion of the general population that must be vaccinated to stop an epidemic depends on the Basic Infection Rate (BIR), essentially a measure of how easily transmissible the virus is. The BIR is the average number of new cases caused by each virus-infected person in a population with no immunity to that virus.  For the current strain of bird flu, we might prudently assume that no one has immunity. According to Ferguson et al. (2005), a typical pandemic strain of influenza would likely have a BIR of around 1.8, a figure that implies the need to vaccinate roughly one half of the population in order to arrest the spread of the disease (see Inset). In other words, about nine million people in Australia.

All this presupposes the availability of a vaccine. In fact, it would take about six months to isolate a particular strain and produce a vaccine in sufficient numbers.  By this time the pandemic would be over.  So, will there be enough vaccine to go around? Given current global development and production capabilities, the answer is no.

Insurance Costs

What would a modern-day pandemic cost the Australian life insurance industry? If we assume a very rough estimate of $300,000 for a life insurance payout, then it is simply a matter of counting the dead or, at least, those with life insurance.  The current level of life insurance penetration is around 30% of the adult workforce, who in turn comprise about 65% of the entire population. This being the case, a pandemic comparable to the 1918-19 influenza outbreak would lead to a total insured loss of around $4.1 billion. This sum is of the same order as a repeat of Cyclone Tracy that destroyed Darwin in 1974 (see the next issue of Risk Frontiers’ Quarterly Newsletter.)

There are, however, other complications not considered in the above calculation. For reasons that are still not entirely clear, the 1918-19 epidemic preferentially killed people between the ages of 25 and 40, i.e. those normally at the lowest risk of dying from influenza.  Thus, usual actuarial assumptions about expected age at death may not apply in the case of a pandemic.  Moreover, people in this target age group are more likely to have life insurance and will tend to be insured for relatively higher amounts.

There may be other calls on insurance caused by the failure of some businesses to fulfil critical supply contracts due to workers being afraid to, or prevented by Government decree, from turning up to work. Private medical insurance could be another source of losses for the insurance industry.

Social and Economic Consequences

While our analyses suggest that the implications of a 1918-19-type pandemic could be significant for the insurance industry, insured losses will represent only a tiny fraction of the wider economic losses borne by society.

The recent SARS epidemic gives us some clues to the likely magnitude of these losses.  The province of Ontario, for example, suffered an estimated loss of more than C$2 billion due to reductions in tourism, including lost income and jobs. Hotels in Toronto remained two-thirds empty during the peak of the epidemic and cost the hotel industry more than C$125 million. More than 15,000 people were quarantined at home for at least 10 days. If nothing else, SARS demonstrated the impact that a short-lived epidemic can have on consumer confidence, investment and consumer spending. Some sources have estimated the total global economic cost of SARS at $US 30 – 50 billion (Financial Times, 14/11/05).

A major flu pandemic would be much more significant than SARS. Businesses could be confronted by 25-30% absenteeism as home quarantine removed many from the workforce for up to two months; people would avoid shops, restaurants, hotels, places of recreation and public transport. There would be a run on basic foodstuffs, medications, masks and gloves. As there is little surge capacity in our hospitals, temporary hospitals would need to be established. Schools, childcare centres, theatres, not to mention pubs and race meetings – the fundamental heartbeat of our nation – would be closed or cancelled. Government imposed quarantine and absenteeism would severely disrupt interstate and international trade. All this would produce a decline in consumer confidence leading to significant reductions in consumption spending.

Some Other Issues

Let’s return now to the question of preventative medicine. As has already been explained, there is simply not going to be enough anti-viral drugs, vaccines and other preventative measures to go around. The current stockpile of anti-viral drugs could be insufficient even for just all essential health care workers, emergency service workers – and politicians? And what about me?  Yes moi!

Assuming Australia has the luxury of time to become better prepared, then difficult choices still remain.  For example, who will get the extra supply after the needs of essential workers are met? Would they be handed out by lottery, should they go to the elderly and young, would people be able to buy them?  Public outcry might prevent a scheme where they were sold to the highest bidder, but it is easy to imagine somebody risking the small chance of personal death and selling their vaccine shots on e-bay for large sums of money.  The problem could be an administrative and ethical nightmare.

Final Thoughts

So, where does all this leave us? As far as preventing an outbreak, the only place that this can be done is in the place of origin, most probably somewhere in Asia. If a pandemic does occur, then it is going to inevitably affect Australia. Quarantine measures that the government will feel obliged to put in place might delay its development but are unlikely to prevent it from reaching us. Given a lead-time of six months to develop an effective vaccine, society and the government will be faced with some difficult choices about who gets access to limited supplies of anti-viral drugs. And for the life insurance industry, our admittedly rough calculations suggest that it is not good news. However only a minor proportion of the economic costs will be borne by the insurance sector.  And underlying all this is a fundamental truth – a healthy population represents the human capital necessary for productivity, innovation and economic growth.

Calculating the proportion of people to vaccinate

The relationship between the Basic Infection Rate (BIR) and the proportion of people who need to be vaccinated to contain or prevent an epidemic is a relatively simple one. In order for the virus to propagate through a population, an infected person must infect at least one other person.  Thus for a vaccine program to be effective, it must lower the effective BIR of the virus to below 1.0. Assuming no immunity within population, the proportion that needs vaccination is given by the formula:

Proportion = (BIR – 1.0)/(BIR)

Given a BIR > 1.0, then vaccinating this proportion of the population will stop an epidemic from gaining hold, although small outbreaks are still possible. With a typical value for a pandemic-type strain of 1.8 (Fergusan et al., 2005), the formula suggests 44% of the population will need to be vaccinated.

If a virus is currently in circulation, then people with it already or having low level infections can be assumed to be immune and not require vaccination. This will reduce the quantity of vaccine needed.

If the virus is sufficiently widespread, however, it will still take a long time to die out and so vaccinating as large a proportion of the population as is feasible is the best defence. Moreover, we will not know the actual BIR for some time and so once again assuming a 1918-19-like worst-case scenario may be the only prudent policy.


Ferguson, Cummings, Couchemez, Fraser, Riley, Meeyai, Lamsirithaworn and Burke, Strategies for containing an influenza pandemic in Southeast Asia, Nature, Volume 437, 2005, pp 209 – 213.

Harris, Melling, Borsay, ed. The Spanish Influenza Pandemic of 1918 – 1919: New Perspectives. Routledge, 2003.

Crosby, America’s Forgotten Pandemic: The Influenza of 1918. Cambridge University Press, 1989.