So far we have presented landfalling hurricane-number prediction methods that are based on historically observed hurricane numbers alone. These methods have the advantage that they depend on relatively few assumptions. On the other hand, they have the disadvantages that (a) there seems to be no very satisfactory way to represent possible climate-related changes in hurricane numbers in these methods, and (b) the signals that we are trying to predict tend to be obscured by noise due, in large part, to the scarcity of hurricanes. In this section we now present a different set of models for predicting future hurricane numbers, based on the idea of first predicting SSTs in the tropical North Atlantic, and then converting these SST predictions to predictions of hurricane numbers. The rationale for these methods is that (a) the effects of climate trends and climate variability may be more robust, and hence easier to predict, in the SSTs than they are in the hurricane number time-series, and (b) there is a clear historical relationship between SSTs in the subtropical North Atlantic and the number of hurricanes in the Atlantic basin. All of the SST prediction methods we use assume that this relationship between SST and hurricane numbers will continue into the future. This is an assumption that is needed in order to make these predictions: however, it is not necessarily the case that this assumption is true. For instance, it may be that the historic relationship between SSTs and hurricane numbers will not apply if patterns of SST in the future are different from patterns experienced in the past, and it may be inappropriate to extrapolate the relationship between SST and hurricane numbers to levels of SST that are higher than those experienced in the past.
The first part of the development of our SST-based hurricane number prediction schemes is to predict future SSTs. We consider four different classes of models for making such predictions: one-step models, two-step models, models based on numerical model output, and climate-shift models. The one-step models predict future MDR SST based on past MDR SSTs. The two-step models predict future MDR SSTs by first predicting future Northern Hemisphere temperatures, and then predicting MDR SSTs from the predicted Northern Hemisphere temperatures. The climate-shift models attempt to model changes in SST as a combination of gradual climate change and climate shifts. We describe each of these model classes in more detail below, followed by a description of how these SST predictions are then converted to hurricane number predictions.
The methods we use are described in detail in a series of technical reports (Meagher and Jewson, 2006; Laepple et al., 2007, Jewson, 2007, Binter et al., 2007a, b).
Was this article helpful?