We have presented a set of models that produce predictions for the number of US landfalling hurricanes likely to occur in the next five years. The models are based on a variety of plausible assumptions and methods. In this article we have emphasized the assumptions, the benefits and the shortcomings of each model.
In developing our prediction schemes we first address the problem of estimating the current state. We solve this problem in a variety of ways. Initially we make the assumption that the rate at which hurricanes are generated and subsequently hit land is a stationary random process. In this case, it makes sense to use all the available reliable historical data equally. However, since the time-series of basin hurricane numbers is not stationary, we go beyond such straight historical averages to consider alternative methods to estimate the current state which involve giving more weight to the more recent years in the data. We also consider models that use related information, such as information from the basin hurricanes and from sea surface temperatures, to predict landfalling hurricane numbers. These variables are less noisy than the landfalling hurricane number time-series and there are good reasons to believe they may allow us to make more accurate predictions of landfall numbers. The next issue we consider in our model development is how to model potential changes from the current state over the next five years. We include models that account for both gradual trends and jumps in the climate.
To keep things simple, the goal we set for our predictions is to minimize the root mean square error (RMSE) between the predicted and actual numbers of hurricanes. This provides a useful metric of comparison for parameter choices (like window lengths for calibration and extrapolation) as well as for model comparison. We also note that most of our analysis uses simple classical statistical methods, which have the benefits of simplicity and transparency. This directly addresses one of our primary goals, which is to introduce methods for the 5 year prediction of hurricane numbers that can be widely understood by meteorologists, climate modelers, and insurance industrypractitioners. Our suite of models achieves this goal, as well as providing a broad range of predictions representing the various relevant scientific theories and ideas. We then rely on a panel of international hurricane experts to weight these models in an expert elicitation process to give us a prediction of future hurricane activity.
Looking to the future, we plan to continue to update our model set as new scientific ideas and understanding relevant to the question of how to predict land-falling hurricane numbers appear.
Was this article helpful?