Adam S. Lea

Large contribution of sea surface warming to recent increase in Atlantic hurricane activity.

Atlantic hurricane activity has increased significantly since 1995 (refs 1–4), but the underlying causes of this increase remain uncertain. It is widely thought that rising Atlantic sea surface temperatures have had a role in this, but the magnitude of this contribution is not known. Here we quantify this contribution for storms that formed in the tropical North Atlantic, Caribbean Sea and Gulf of Mexico; these regions together account for most of the hurricanes that make landfall in the United States. We show that a statistical model based on two environmental variables—local sea surface temperature and an atmospheric wind field—can replicate a large proportion of the variance in tropical Atlantic hurricane frequency and activity between 1965 and 2005. We then remove the influence of the atmospheric wind field to assess the contribution of sea surface temperature. Our results indicate that the sensitivity of tropical Atlantic hurricane activity to August–September sea surface temperature over the period we consider is such that a 0.5 °C increase in sea surface temperature is associated with a ∼40% increase in hurricane frequency and activity. The results also indicate that local sea surface warming was responsible for ∼40% of the increase in hurricane activity relative to the 1950–2000 average between 1996 and 2005. Our analysis does not identify whether warming induced by greenhouse gases contributed to the increase in hurricane activity, but the ability of climate models to reproduce the observed relationship between hurricanes and sea surface temperature will serve as a useful means of assessing whether they are likely to provide reliable projections of future changes in Atlantic hurricane activity.

Seasonal prediction of hurricane activity reaching the coast of the United States

Much of the property damage from natural hazards in the United States is caused by landfalling hurricanes—strong tropical cyclones that reach the coast. For the southeastern Atlantic coast of the US, a statistical method for forecasting the occurrence of landfalling hurricanes for the season ahead has been reported, but the physical mechanisms linking the predictor variables to the frequency of hurricanes remain unclear. Here we present a statistical model that uses July wind anomalies between 1950 and 2003 to predict with significant and useful skill the wind energy of US landfalling hurricanes for the following main hurricane season (August to October). We have identified six regions over North America and over the east Pacific and North Atlantic oceans where July wind anomalies, averaged between heights of 925 and 400 mbar, exhibit a stationary and significant link to the energy of landfalling hurricanes during the subsequent hurricane season. The wind anomalies in these regions are indicative of atmospheric circulation patterns that either favour or hinder evolving hurricanes from reaching US shores.

Replicating annual North Atlantic hurricane activity 1878–2012 from environmental variables

Statistical models can replicate annual North Atlantic hurricane activity from large-scale environmental field data for August and September, the months of peak hurricane activity. We assess how well the six environmental fields used most often in contemporary statistical modeling of seasonal hurricane activity replicate North Atlantic hurricane numbers and Accumulated Cyclone Energy (ACE) over the 135 year period from 1878 to 2012. We find that these fields replicate historical hurricane activity surprisingly well, showing that contemporary statistical models and their seasonal physical links have long-term robustness. We find that August–September zonal trade wind speed over the Caribbean Sea and the tropical North Atlantic is the environmental field which individually replicates long-term hurricane activity the best and that trade wind speed combined with the difference in sea surface temperature between the tropical Atlantic and the tropical mean is the best multi-predictor model. Comparing the performance of the best single-predictor and best multi-predictor models shows that they exhibit little difference in hindcast skill for predicting long-term ACE but that the best multipredictor model offers improved skill for predicting long-term hurricane numbers. We examine whether replicated real-time prediction skill 1983–2012 increases as the model training period lengthens and find evidence that this happens slowly. We identify a dropout in hurricane replication centered on the 1940s and show that this is likely due to a decrease in data quality which affects all data sets but Atlantic sea surface temperatures in particular. Finally, we offer insights on the implications of our findings for seasonal hurricane prediction.