How can you tell whether a flood, storm, earthquake or heavy hailstorm is likely or not? Matthias Klawa explores this question at his job every day. The expert, with a doctorate in meteorology, is responsible for assessing natural risks at Germany's largest insurance company. He and his team of 15 other specialists – seismologists, meteorologists, a hydrologist, an agricultural engineer and, of course, mathematicians – help Allianz calculate the probabilities of natural disasters.

Granted, researching natural disasters is not overly spectacular. A lot of what's done here is nothing more than applied science; but also nothing less. It entails lots of mathematics, and history too. Because the figures from past weather disasters – like the Elbe flood of 2002 or the winter hurricanes Kyrill and Emma – make a very significant contribution toward calculating future ones.

What's new is "predictive research": With only a few exceptions, reinsurers only began systematically simulating disasters in the nineties. Primary insurers like Allianz have also been applying in-house expertise for the purpose since around 2000. There's a technical reason why the discipline is so very young – information technology only recently became able to run such complicated computations within a reasonable time.

By now, it's possible to make fairly accurate predictions of the probability that a natural disaster will happen within a certain interval of time. But note – what's predicted is the probability it will happen, not when it will happen.

So short-term weather conditions still retain their ability to surprise even storm expert Klawa. He points out the limitations of predictability: "When was the last time you got irritated because you didn't bring your umbrella along on the way home from work because the weather forecaster on the radio that morning said everything would be sunny?"

The primary duty of the Allianz natural risks team is to produce long-term estimates of losses so that the insurer can set aside reserves and calculate premiums. But another point is to make preventive use of the knowledge they develop. Nobody can entirely prevent a disaster – but if you take precautions, for example when there's a storm warning, you at least have a chance of reducing the risk. And if things get serious, it's essential to respond fast – to run triage, make repairs and pay for losses – as the appalling situation after Hurricane Katrina in the USA recently proved.

The team regularly gathers data from the portfolio of insured clients and runs a variety of crisis scenarios. Some of the tools they use for the purpose are vulnerability curves, which describe what percent of the insured total will be destroyed in an event of a given intensity. These curves are derived from historical loss events. For example, for Kyrill, storm damage was assessed in relation to the insured total and the wind speed. Now, if it's assumed that a potential storm will hit Germany at certain wind speeds, the associated loss can be calculated.

The latest findings about worldwide climate change also play an important role in the computations run by Klawa and his colleagues. "Here in Germany too, we're already seeing alarming shifts away from the weather figures we're used to. The number of days at high temperatures has increased substantially. There's less precipitation in the summer than before – and if it does rain, it rains harder."

Of course, what we'd like to have most from Klawa is a forecast of what weather's ahead. But, as he's said – predictions like that aren't possible. Predictive meteorology is mainly a calculation of probabilities. "It keeps you on your toes," he concludes, looking with a smile out his office window – in the sky over Munich, dark clouds are gathering and strong gusts have set the treetops waving.

As with all content published on this site, these statements are subject to our Forward Looking Statement disclaimer, provided on the right.