## Short Baseline Predictions

In the previous section, we justify the CLP model of basin and landfalling hurricane numbers that assumes (a) the mean number of basin storms varies in time and

10

a No. Cat 1-5 Basin Hurricanes b

### No. Cat 3-5 Basin Hurricanes

Fig. 3 Relationship between landfalling hurricanes and basin hurricanes for (a) Cat 1-5 and (b) Cat 3-5 intensities from 1948 to 2007. The larger circles indicate higher frequency. The largest filled circles indicate six or more years with the same landfall to basin ratio

a No. Cat 1-5 Basin Hurricanes b

### No. Cat 3-5 Basin Hurricanes

Fig. 3 Relationship between landfalling hurricanes and basin hurricanes for (a) Cat 1-5 and (b) Cat 3-5 intensities from 1948 to 2007. The larger circles indicate higher frequency. The largest filled circles indicate six or more years with the same landfall to basin ratio

(b) the probability of storms making landfall can be modeled as constant. A consequence of these assumptions is that the mean number of storms making landfall varies in time, following the mean number in the basin.

There are two important implications of this model for the prediction of numbers of landfalling hurricanes. The first is that we can take the change-points that have been identified in the basin hurricane number time-series, and assume that they apply to the landfalling series, even though they cannot be detected in the land-falling series. This leads to a method for predicting the landfalling series, which we call the 'short baseline' method, and which involves making a prediction for future landfalling hurricane numbers which consists of the average number of landfalling hurricanes since the most recent basin change-point. Based on the CLP model assumptions, this is then likely to be a better prediction method than the long baseline method, because of the underlying non-stationarity in the series.

Based on the estimated levels between the change-points, the CLP model implies that we would expect the long-baseline prediction to be biased. If we assume that the level of hurricane activity over the next five years will remain at the same level that it has been at since 1995, then the short baseline prediction will, on the other hand, be unbiased. The assumption that the number of hurricanes will remain at the same level clearly ignores decadal oscillations and trends, and the possible occurrence of further change-points in the next five years. These approximations may be good ones since our forecast horizon is fairly short and the trends are apparently rather weak. Nevertheless, we do try to model these effects in other models (see below). Conveniently, the various change-point analyses listed above all give the same point in time for the most recent change-point (1994/1995), and so all lead to the same short baseline prediction. This prediction is given in row 3 of the summary tables.