Projecting end-of-century North Atlantic Hurricane Activity using Forecast Statistical Models

For areas such as the Caribbean, for which the resources for dynamical modeling studies are not readily available, statistical models provide a useful, less computationally and resource demanding alternative. This study seeks to assess the skill of four statistical forecasting models as tools for evaluating future trends in hurricane activity. Three of the models are premised on standard hurricane predictors (SSTs and vertical wind shear) while the fourth model uses SST-gradient predictors, which have proven useful for Caribbean rainfall. The skill of the statistical modeling schemes at predicting annual North Atlantic hurricane frequency over the period 1950–2008 is first assessed. Backward regression is used to optimize the models while cross-validation is used to validate the schemes. Thereafter, changes in annual North Atlantic hurricane frequency are assessed for the period 2020-2100 under the A2 emissions scenario, where the future state of the predictors are extracted from the ECHAM5, HadCM3, MRI CGCM2.3.2a and MIROC3.2 global climate models run under the CMIP3 project. The results obtained show that for the first three models significant increases in the mean annual frequency of North Atlantic hurricanes, despite increasingly unfavourable conditions which should account for a decline in hurricane activity, e.g. significant increases in vertical shear. In these models, however, positive changes in SST magnitudes under global warming have a larger relative influence on the projections than negative changes in zonal wind predictors. The fourth model, however, showed only slight changes and a general decline in hurricane frequency compared to the present-day mean. The suggestion is that spatial variations in SSTs (as opposed to magnitude) may become the more dominant driver of hurricane (and convective activity) in the region in the future. Using the period of 1990-2008 as analogous of the future state, the first three models were reformulated and new future projections generated. The new projections show a decrease in, and in some instances, reversal of the linear increase similar to the results from the SST-gradient driven model. This is in line with recent dynamical models studies which suggest a possible decline in frequency.