9D.5 Hurricane Intensity Probability Prediction in a Changing Climate: A Multiple-Linear Regression Modeling Approach

Wednesday, 2 April 2014: 11:30 AM
Regency Ballroom (Town and Country Resort )
Chia-Ying Lee, Columbia University, New York, NY; and M. K. Tippett, S. J. Camargo, and A. H. sobel

As the climate changes, the environmental conditions that influence the formation and evolution of hurricanes are expected to change as well. Although the potential influence of climate change on tropical cyclones has been the subject of number of recent studies, it is still difficult to confidently assess the magnitude of future changes in storm intensity. One of the reason is that the present global climate models are too coarse to adequately capture the storm structure and therefore to model the storm intensification. Here we explore this issue with a statistical approach. Although global climate models themselves fail to accurately simulate storm intensity, previous studies using the Statistical Hurricane Intensity Prediction Scheme (SHIPS) have shown that a multiple-linear regression model with predictors from the Global Forecast System (GFS), in addition to those from storm persistence and from climatology, is capable of predicting storm intensity with some skill. Hence, our first step is to develop a SHIPS-like model with only essential predictors calculated from global models or reanalysis and from storm persistence. Monthly and daily 2.5o X 2.5o NCEP reanalysis data from 1981 to 2012 are used here.

Data from 1981 to 1999 is used for model training while the period 2000-2012 is used for testing model performance. A forward-selection procedure is used to identify the best 6 predictors: the difference between the maximum potential intensity and current maximum wind speed (dPIVmax), changes in the maximum wind speed in the past 12 hours (dVmax), the magnitude of 850-200 hPa deep-layer vertical shear (SHRD), storm translation speed (trSpeed), 200 hPa divergence (200mbDiv), and atmospheric stability (S). The performance of the regression models using varying number of these six predictors is tested with daily and monthly data. Preliminary results suggest that 1) the regression model with all 6 predictors utilizing either daily or monthly data has a skill comparable to the official NHC forecasts as well as SHIPS forecasts; 2) the skill of the regression model with only dPIVmax and dVdt as predictors is close to that of the model with all 6 predictors, although adding the other four predictors does further improve the model performance; 3) adding additional predictors (such as relative humidity) neither improves nor hurts overall model performance; 4) the improvement of model performance from using daily data instead of monthly data is limited. Windspeed probability predictions utilizing the regression model will also be presented.

Given these results, monthly data from global climate models with only the most essential predictors might be sufficient for us to understand the response of hurricane intensity to a changing climate from a statistical perspective.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner