Up to this point, the prediction methods that have been discussed assume that an estimate of current hurricane activity is a good estimate of future activity. This assumption, that the mean number of hurricanes will remain at the same level as it is now, clearly ignores the possible occurrence of further change-points in the next five years. This may be a reasonable approximation, considering that the last change-point was 10-13 years before our predictive period, which is only five years. Since hurricane change-points might well be related to the Atlantic Multi-decadal Oscillation (AMO), and according to Sutton and Hodson (2005), AMO phases currently last about 30 years, we might expect to wait another decade or so before worrying about future change- points. However, the possibility of a climate shift in the next five years may not be entirely negligible. Enfield and Cid-Serrano (2006) use Gray et al.'s (2004) tree ring reconstruction of the AMO to estimate the probability of regime shifts. The probabilities are derived using a Monte-Carlo resampling of the frequency space of the reconstructed paleo time-series and give us an approximation for the probability of a shift in the AMO over the next five years. There are a few issues with this analysis that need to be taken into consideration; namely (a) that the tree ring time-series is shown to be non-stationary, so the frequency distribution for the present and the future may be quite different from the past, and (b), although the tree ring time-series has been fitted to the recent AMO index, there is no certainty that the past part of the record also reflects only the AMO since tree rings can be influenced by other large-scale atmospheric features. Given these issues, in our climate shift prediction models, we let the hurricane experts determine both the probability of a shift by the next year and a probability of a shift within the next five years using their own best judgement (rather than taking these probabilities directly from the Enfield Cid-Serrano model). This model is restricted to downward shifts from the current activity rate. Using the probabilities that the experts provide we estimate the annual mean expected number of hurricanes for the next five years under the assumption that the probability of a shift increases linearly through the five years. The estimate for the annual mean number of hurricanes is then
where Jui is the estimated number of hurricanes for year i, given by
mi = i1 - pOm + i>i m mc and mi are the estimated annual hurricane numbers for the current state and for the low state of hurricane activity and pAi is the probability that the i-th year is in the low state. The low state estimate is made using the hurricane numbers between 1970 and 1994 and the current estimate of the annual number of hurricanes is derived using the mixed baseline model.
Predictions from the climate shift model vary according to the shift probabilities that the experts assign. If the probability of a shift is zero, then this model is just the mixed baseline model so, for the 2007 elicitation, this model was used in place of the mixed baseline model. For the 2008-2012 predictions in Table 4, the low state estimate of activity and the mixed baseline estimates are shown in rows 4a and 4b for the direct climate shift model and in rows 5a and 5b for the indirect climate shift model. These estimates give the possible range that experts could have chosen.
Was this article helpful?