The event generation component of AIR natural hazard models determines the frequency, magnitude, and other characteristics of potential catastrophe events by geographic location. This requires, among other things, a thorough analysis of the characteristics of historical events as well as an understanding of the region-specific features, whether seismotectonic, geological, topographical, or atmospheric, that are likely to influence the behavior of future catastrophe events.

AIR catastrophe models can be roughly classified into two types. Some are parameterized models, as in the case of AIR's tropical cyclone, earthquake, and severe thunderstorm models. Others are dynamical models designed to capture the complexities of extratropical cyclones, such as European winter storms, California's Santa Ana winds, and the so-called Nor'easters that affect the East Coast during winter months.

All catastrophe events are extremely complex and their characterization requires the use of large numbers of variables. In the case of parameterized models, event generation begins by collecting the available scientific data pertaining to these variables from many different sources. The data are cleaned and verified.

After a rigorous data analysis, AIR researchers develop probability distributions for each of the variables, testing them for goodness-of-fit and robustness. The selection and subsequent refinement of these distributions are based on statistical techniques, on well-established scientific principles and on an understanding of how catastrophic events behave. The resulting distributions are used to produce a large catalog of simulated events. That is, by sampling from these distributions, the model generates simulated "years" of event activity. Many thousands of these scenario years are then generated to produce a stochastic catalog that represents the complete and stable range of potential annual aggregate and occurrence experience of catastrophe event activity.

The AIR winter storm models, on the other hand, employ advanced dynamical weather modeling techniques. The complexity of these highly non-linear atmospheric structures, with their multiple and frequently changing areas of relative low and high pressure, makes them difficult to model using a parametric approach. Such complexity calls for the use of state-of-the-art Numerical Weather Prediction (NWP) technology. Small differences in the initial conditions that spawn such storms can result in large differences in storm evolution. Event generation begins by perturbing, both temporally and spatially, the initial conditions of a set of "seed" storms drawn from the historical data. The pressure fields of each of these storms are then moved forward in time through the application of a set of partial differential equations governing fluid flow. The European Centre for Medium-Range Weather Forecasts says that this technique, called ensemble-forecasting, "has much greater economic value than single deterministic forecasts." AIR is the only modeler to incorporate advanced NWP technology into its catastrophe models.