[Paper Review] Statistical Modelling of the Relationship Between Main Development Region Sea Surface Temperature and \emph{Landfalling} Atlantic Basin Hurricane Numbers

Roman Binter, Stephen Jewson+1 more|ArXiv.org|Jan 15, 2007

Tropical and Extratropical Cyclones Research3 references3 citations

TL;DR

This study investigates the statistical relationship between Main Development Region (MDR) sea surface temperature (SST) and landfalling Atlantic hurricanes from 1900–2005 and 1950–2005. Using multiple models—including linear, damped linear, Poisson, and negative binomial frameworks—it finds no significant relationship between MDR SST and total or intense landfalling hurricane counts in the 1900–2005 period, though weak evidence emerges in the 1950–2005 subset, with poorly estimated slopes and non-significant model improvements over a flat Poisson baseline.

ABSTRACT

We are building a hurricane number prediction scheme that relies, in part, on statistical modelling of the empirical relationship between Atlantic sea surface temperatures and landfalling hurricane numbers. We test out a number of simple statistical models for that relationship, using data from 1900 to 2005 and data from 1950 to 2005, and for both all hurricane numbers and intense hurricane numbers. The results are very different from the corresponding analysis for basin hurricane numbers.

Motivation & Objective

• To assess whether Main Development Region (MDR) sea surface temperature (SST) can predict landfalling Atlantic hurricane numbers.
• To evaluate the statistical significance of the relationship between MDR SST and landfalling hurricane frequency using historical data from 1900–2005 and 1950–2005.
• To compare multiple statistical models (e.g., linear, Poisson, negative binomial) for their ability to capture SST-hurricane number relationships.
• To determine whether the relationship differs for all landfalling hurricanes versus intense landfalling hurricanes.
• To assess whether improved data quality in the 1950–2005 period reveals a detectable signal where none was found in the longer 1900–2005 period.

Proposed method

• Empirical modeling of the relationship between MDR SST and landfalling hurricane counts using data from 1900–2005 and 1950–2005.
• Application of six statistical models: flat Poisson, linear normal, damped linear normal, linear Poisson, exponential Poisson, and exponential negative binomial.
• Model evaluation via out-of-sample root mean square error (RMSE), log-likelihood scores, and pairwise statistical comparisons using p-values.
• Use of rank correlation and linear correlation to assess bivariate relationships between SST and landfalling hurricane counts.
• Parameter estimation with standard errors to test whether slope coefficients are significantly different from zero.
• Model comparison using winning counts from repeated cross-validation, with p-values indicating statistical significance of model superiority.

Experimental results

Research questions

• RQ1 Is there a statistically significant relationship between Main Development Region (MDR) sea surface temperature (SST) and the number of landfalling Atlantic hurricanes from 1900 to 2005?
• RQ2 Does the relationship between MDR SST and landfalling hurricane counts differ when analyzed over the shorter 1950–2005 period compared to the full 1900–2005 period?
• RQ3 Do non-trivial statistical models (e.g., linear, Poisson) significantly outperform a flat Poisson model that assumes no SST dependence?
• RQ4 Is there a detectable relationship between MDR SST and the number of intense landfalling hurricanes, given the smaller sample size for this category?
• RQ5 How well are the parameters of the SST-hurricane relationship estimated, and are the estimated slopes statistically distinguishable from zero?

Key findings

• For the 1900–2005 period, the linear correlation between MDR SST and total landfalling hurricane counts is only 0.16, and rank correlation is 0.12, indicating no strong empirical relationship.
• The best-performing model (exponential negative binomial) shows only a slight improvement in out-of-sample RMSE over the flat Poisson model, and all non-trivial models are statistically indistinguishable from the flat model.
• In the 1950–2005 period, some non-trivial models (e.g., linear Poisson) show statistically significant improvement over the flat Poisson model in RMSE, but the slope estimates (0.99–1.2) have large standard errors (up to 0.63), indicating poor precision.
• For intense landfalling hurricanes, no non-trivial model significantly outperforms the flat Poisson model in either 1900–2005 or 1950–2005 data sets, and the flat model even defeats some linear models in probabilistic comparisons.
• Slope parameters across all non-trivial models are not statistically different from zero, suggesting no detectable sensitivity of landfalling hurricane numbers to MDR SST changes.
• The damping parameter in the damped linear model is 0.73, indicating a weak or negligible signal, consistent with the absence of a strong relationship.