[Paper Review] Predicting basin and landfalling hurricane numbers from sea surface temperature

Stephen Jewson, Roman Binter+3 more|ArXiv.org|Jan 15, 2007

Tropical and Extratropical Cyclones Research19 references5 citations

TL;DR

This paper develops a three-stage statistical model to predict Atlantic basin and U.S. landfalling hurricane numbers using main development region (MDR) sea surface temperature (SST) as a predictor. It combines SST forecasts from Laepple et al. (2007) with two statistical models linking SST to hurricane counts—linear and exponential Poisson models—yielding 5-year predictions of 2.04 to 2.52 landfalling hurricanes per year, with uncertainty primarily driven by SST forecast variability.

ABSTRACT

We are building a hurricane number prediction scheme based on first predicting main development region sea surface temperature (SST), then predicting the number of hurricanes in the Atlantic basin given the SST prediction, and finally predicting the number of US landfalling hurricanes based on the prediction of the number of basin hurricanes. We have described a number of SST prediction methods in previous work. We now investigate the empirical relationship between SST and basin hurricane numbers, and put this together with the SST predictions to make predictions of both basin and landfalling hurricane numbers.

Motivation & Objective

• To develop a practical, three-stage prediction system for Atlantic basin and U.S. landfalling hurricane numbers using sea surface temperature (SST) as a predictor.
• To quantify the empirical relationship between main development region (MDR) SST and observed hurricane counts in the Atlantic basin.
• To reduce uncertainty in long-term hurricane forecasts by combining SST predictions with statistical models of hurricane frequency.
• To evaluate multiple SST forecast models and SST-to-hurricane number conversion models to assess prediction robustness and spread.
• To provide probabilistic forecasts of landfalling hurricanes by combining basin hurricane predictions with historical landfall probabilities.

Proposed method

• Uses HadISST data to compute a summer (July–September) MDR SST index over the region 10°–20°N, 15°–70°W.
• Applies two statistical models to relate SST to hurricane numbers: a linear Poisson model and an exponential Poisson model, both fitted to 1900–2005 and 1950–2005 data.
• Utilizes three distinct SST forecasts from Laepple et al. (2007), ranging from constant SST to rapidly increasing SST, to drive the hurricane number predictions.
• Converts basin hurricane predictions to landfalling hurricane numbers using historical landfall probabilities: 0.254 for all hurricanes (category 1–5), 0.240 for intense hurricanes (category 3–5).
• Employs analytic relationships from Jewson (2007) to propagate uncertainty from SST forecasts through to final landfalling hurricane predictions, estimating mean, variance, and standard error.
• Generates six distinct prediction scenarios by combining three SST forecasts with two SST-to-hurricane number models, enabling assessment of forecast spread and sensitivity.

Experimental results

Research questions

• RQ1 What is the statistical relationship between main development region (MDR) sea surface temperature (SST) and the number of Atlantic basin hurricanes?
• RQ2 How do different statistical models (linear vs. exponential Poisson) for converting SST to hurricane counts affect long-term predictions?
• RQ3 To what extent do uncertainties in SST forecasts propagate into uncertainty in predicted landfalling hurricane numbers?
• RQ4 How do the predicted numbers of landfalling hurricanes compare across different SST forecast scenarios and SST-to-hurricane number models?
• RQ5 Can a three-stage SST-based prediction system produce forecasts consistent with historical trends and alternative time-series methods?

Key findings

• The exponential Poisson model for SST-to-hurricane number conversion consistently outperforms the linear model in fitting historical data, though both are statistically plausible.
• Five-year predictions of U.S. landfalling hurricanes (category 1–5) range from 2.04 to 2.52 per year, depending on the SST forecast and conversion model used.
• The highest predictions (2.52/year) arise from rapidly increasing SST forecasts combined with the exponential Poisson model, indicating strong sensitivity to SST trends.
• The lowest predictions (2.04/year) come from constant SST forecasts paired with the linear Poisson model, suggesting a conservative baseline.
• Uncertainty in SST forecasts is the dominant source of uncertainty in final landfalling hurricane predictions, highlighting the need for improved SST forecasting.
• The model system produces predictions that are broadly consistent with time-series-based methods (e.g., Binter et al., 2006), but extend to higher values under optimistic SST scenarios.