The question of how to predict the number and intensity of Atlantic hurricanes that make landfall in the United States is of great interest to the insurance and reinsurance industry. There is interest in predictions on time-scales from as short as a few hours to as long as 50 years. The physical processes that determine predictability, and the methods that one might use to make predictions, vary greatly according to time-scale. Here we are concerned with predictions on time-scales of one to five years. For these time-scales both natural variability and anthropogenic influences contribute to the climatic conditions which impact hurricanes. As a result, one to five year hurricane predictability depends on the extent to which current climate, decadal fluctuations and trends can be estimated in the hurricane numbers and associated parts of the climate system. This leads to the use of prediction methodologies that first attempt to define the current state, and then try to capture the variability and/or trends that might determine changes from this current state.
Our approach to making predictions of hurricane numbers on these interannual time-scales has two steps. First, we build a large number of forecast models based on different ideas for how one might predict future hurricane numbers. We try and include all reasonable and obvious approaches. Our models are based on inputs such as historical numbers of landfalling Atlantic hurricanes, historical numbers of hurricanes in the Atlantic basin, historical sea surface temperatures (SSTs) and predictions from dynamical model simulations. The second step is to present these
J.B. Elsner and T.H. Jagger (eds.), Hurricanes and Climate Change, 73
doi: 10.1007/978-0-387-09410-6, © Springer Science + Business Media, LLC 2009
models to a panel of experts, who combine the various predictions by weighting the models using their best judgement. This paper focuses on the first step (the modelling), and presents the models that were used as input for expert elicitation processes in 2006 and 2007. The details of the elicitation process itself are described in another paper (Lonfat et al., 2007).
Was this article helpful?