J3.3 Predicting Atlantic tropical cyclone activity

Monday, 24 January 2011: 4:45 PM
2A (Washington State Convention Center)
Roshanak Nateghi, Johns Hopkins University, Baltimore, MD; and S. M. Quiring, S. D. Guikema, and A. Schumacher

An objective statistical methodology was employed to develop models for predicting Atlantic tropical cyclone (TC) activity using a large suite of climate variables. First, a Classification and Regression Trees (CART) is used to select the most appropriate variables. These variables are used to develop a Poisson-Generalized Linear Model based on Atlantic hurricane activity (1949–2009). The resultant models were evaluated using cross-validation and there are generally small errors (<10%) in the predictions of annual TC. A comparison with previously published models suggests that this methodology can lead to stronger fits and improved predictive accuracy. The utility of these models for quantifying the uncertainty in future Atlantic hurricane activity is demonstrated using climate scenarios developed for the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4). This approach can also be used to forecast Atlantic TC activity prior to the start of the hurricane season.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner