Atlantic basin tropical storms—interannual variability and nonparametric simulation of tracks

There is a great deal of interannual variability in the occurrence of named storms in the Atlantic Basin. We use "bootstrap" technique to find regions of Atlantic basin that exhibit significant differences in average storm days among various phases of large scale climate indicators (El Nino/Southern Oscillation ENSO), the North Atlantic Oscillation (NAO), and Tropical North Atlantic (TNA) sea surface temperatures. We find that warm phase of ENSO substantially reduces the number of storm days over the Gulf of Mexico and eastern US relative to neutral or cold phase. We also find that the high phase of NAO in preceeding winters tend to favor lesser storm days in general over the eastern Atlantic basin. Furthermore, we find that ENSO significantlyaffects the number of storms "Generated" in the basin, while the preceeding winter NAO affects the number of storms over the land regions. These results are interesting and have significant potential for probabilistic forecasting of storm days over different parts of the basin.

The ability to realistically simulate potential hurricane storm tracks, and condition them on global climate conditions (e.g., ENSO and NAO, or atmospheric circulation indices) is important for assessing the hazard these storms present to coastal regions in North and Central America for specific low frequency climate conditions and under model generated climate change scenarios. To address this problem we develop a model that considers hurricane tracks to be a spatial Markov process over a regular 5x5 degree latitude longitude grid based on a directional probability distribution derived from the observed record of Atlantic tropical storms. The model (hereafter referred to as the Markov Model) also weighs in the probability for the termination of a storm's life as a function of the location, and is initialized based on the probability for storm inception, both calculated from historical data. A variety of statistics (spatial and life distributions) of hurricanes simulated from the Markov model were examined. Systematic biases were identified and potential improvements, including a bootstrap procedure are being explored.