Editor's Note: Homeowners along the Gulf and East Coasts of the United States breathed another collective sigh of relief as the 2007 hurricane season drew to a close on November 30. With just a single U.S. hurricane landfall—minimal Category 1 Humberto—2007 seemed almost anticlimactic given pre-season forecasts of well-above normal activity. Even the late-season upgrade of Tropical Storm Karen to Hurricane Karen—which brought the official hurricane count for 2007 to six—failed to engender much excitement. In this article Dr. Peter S. Dailey reviews the 2007 Atlantic hurricane season and addresses some of the questions it raises about near-term hurricane risk.

by Dr. Peter S. Dailey, Director of Research in Atmospheric Science, AIR Worldwide Corporation

The 2007 Atlantic hurricane season was certainly an interesting one. It was at once both record-breaking and average. Two hurricanes made landfall as Category 5 storms, albeit in Central America, and records were set for rapid intensification. Yet in the end, by several measures, including landfalls and insured losses, 2007 was an average or below-average hurricane year.

Figure 1 shows the season's 14 named storms. Other than Hurricane Humberto, which made landfall along the coast of Texas at minimal Category 1 strength, none of the other storms of 2007 made U.S. landfall as hurricanes. The two strongest events, Hurricanes Dean and Felix—which were also the only two that achieved anything more than Category 1 status—took more southerly tracks, wreaking havoc along the coasts of Mexico and Central America.

Figure 1. Named Storms of 2007

How did activity in 2007 stack up against long term averages?

Figure 2 shows the typical evolution of tropical cyclone activity in the North Atlantic compared to what we actually observed in 2007. Note that tropical cyclone activity generally accelerates in the months of August and September—historically the most active part of the season.

As Figure 2 shows, by the end of the season we should expect, on average, 10.7 named storms (tropical cyclones with wind speeds of at least 40 mph). In fact, there were 14 named storms this year, so basinwide activity was indeed above average as had been forecast.

It should be noted that Andrea, which formed in early May—well before the official June 1 start of the season—was actually a subtropical system. It is only in the last five years that the National Hurricane Center has implemented the policy of naming subtropical systems—with Nicole (2004) and Andrea this season falling into this category. In fact, there has been some discussion about whether various other of the 2007 storms would have been named in past years. Nevertheless, it is reasonable to say that overall activity was above average this year.

The number of hurricanes, however, was just average at 6 (the long-term average is 6.2) This indicates that the propensity of tropical storms to intensify was below average this year, despite Hurricanes Dean and Felix. Finally, the number of major hurricanes (Saffir Simpson Category 3 and higher) was about one short of the long-term average.

Figure 2. Comparison between the Typical Year and 2007

How did activity in 2007 stack up against the seasonal forecasts?

Based largely on the expectation that sea surface temperatures (SSTs) in the Atlantic would continue to be warmer than average, seasonal forecasters were predicting a very active season in 2007.

Figure 3 shows the forecasts issued before and during the course of the 2007 season. Note that only the UK Met Office, which developed a new type of forecast technology this year, issued a forecast that included the possibility of below-average activity.

Figure 3. The 2007 Seasonal Forecasts*

The forecasts were generally correct in predicting an above-average season in terms of tropical storm counts, though some were clearly too high. The forecasts did less well, however, in predicting the number of hurricanes that would form, which ranged from a low of 6 (WSI's July update) to as many as 10 (NOAA's pre-season May forecast).

Why didn't the seasonal forecasts perform well this year? For many of the same reasons that they did not perform well in 2006. In part, the 2007 forecasts over-predicted activity because sea surface temperatures themselves were over-predicted.

Figure 4 shows SST anomalies since 1948 (the period for which SST data is considered reliable). Some scientists believe that the Atlantic Ocean undergoes extended periods of cooler-than-average temperatures followed by warmer-than-average temperatures. The concept, known as the Atlantic Multidecadal Oscillation is a theory, not fact. Other scientists would argue that long-term fluctuations are more episodic than periodic. What is known is that every year since 1995, the Atlantic has been anomalously warm.

Figure 4. Atlantic Sear Surface Temperature Anomalies Since 1948

Though the observed ocean anomaly in 2007 was also positive, it was lower than expected—and in fact the lowest anomaly observed since 2002. Some scientists have attributed the lower-than-expected SSTs in 2007 to the effects of an influx of African dust, which can block sunlight from the ocean's surface, preventing it from warming to the temperatures needed to support hurricane formation and development.

Forecasters also expected a La Niña condition this year, which is known to suppress wind shear in the Caribbean and beyond, allowing more storms to both form and intensify. But the anticipated La Niña arrived slowly and late, and therefore wind shear was higher than expected over parts of the Atlantic.

Accumulated Cyclone Energy (ACE) Index

Most seasonal forecasters predict basin counts of tropical storms and hurricanes as they are a simple way of communicating seasonal activity. However, storm counts are a less than optimal metric. For example, Hurricane Karen, which was a borderline hurricane for just 12 hours, enters into the hurricane count on par with Category 5 Hurricane Dean. And had Humberto intensified only slightly more slowly, it would never have entered the hurricane count at all.

To capture the total wind energy (and therefore damage) potential of Atlantic storms, the Accumulated Cyclone Energy Index, or ACE, provides an improved measure of basin activity. ACE measures the total integrated wind energy over the complete life cycle of a storm. The seasonal ACE Index then sums the index for individual storms to provide an estimate of the potential for damaging winds from all storms over the entire season.

Figure 5 shows the individual ACE indexes for the 2007 storms and the accumulation of ACE through the season. The final seasonal ACE this year was 67.7. This about 33% below the annual average of 101.2, indicating that the total wind energy generated by the storms of the 2007 season fell well short of what one would expect in the typical year. In fact, two of the 14 events—Hurricanes Dean and Felix—account for about three quarters of the seasonal ACE; the remaining storms were generally weak and short-lived. While it is certainly unusual for two hurricanes in a season to reach Category 5 strength and to maintain that intensity to landfall, the total wind energy generated by all storms combined—including these two CAT5s—was still well below what's expected in a typical season.

Figure 5. ACE by Storm and Seasonal Total

How much does the ACE index fluctuate from year to year? To get a sense for annual variability, we can rank order the ACE values from 1950 to present, and look at the 25th and 75th quintiles. The results are shown in Figure 6, along with the seasonal ACE indexes for the last four seasons. Half of the historical ACE indices fall within the range from 54 to 145, with a median of around 101. The ACE index of 67 in 2007 and of 79 in 2006 both fall well within this range, though both were below average.

Figure 6. Variability of ACE

However, as is seen in Figure 6, the wind energy generated by tropical cyclones in 2004 and in 2005 was well above the average and fell well outside the 75th percentile observed historically. In fact, the ACE values for 2004 and 2005 are among the three highest observed since 1948. The last two seasons, 2006 and 2007, demonstrate that there are wide fluctuations in wind energy generated from year to year, and that despite the outliers of 2004 and 2005, we have recently returned to more typical values.

Hurricane Landfall Proportion

While ACE is a good measure of the destructive potential of storms in the open ocean, insurers are primarily interested in what proportion of that power reaches the heavily exposed U.S. coastline.

Since 1995, SSTs have been warmer than average and these elevated SSTs have produced a higher than average number of tropical storms in the Atlantic basin. This is clearly shown in the topmost time series of Figure 7, where each red dot indicates a year in which the observed tropical storm count exceeded the long-term average of 10.7 per year.

Forecasting the number of storms that will form in a season is the most skillful among the statistics we've discussed so far because there is a relatively strong relationship between sea surface temperatures and genesis frequency. Predicting how many of them will reach hurricane strength and further how many of those will go on to make landfall involves assumptions about where they form and how they move—and these complications make predictions about hurricanes and landfalls much more challenging for the forecasters.

The middle graph in Figure 7 shows what proportion of tropical storms intensified to hurricane strength. The long-term average is 58%, while in 2007 the proportion was 43%. Note that during the period starting in 1995, the proportion has been below average in 8 of 13 seasons. So, while there is a link between elevated sea surface temperatures and basin activity, the relationship is less clear when it comes to intensification. (You can also see the danger in extending short-term trends into the near-term. Using the years 2004 to 2007 to derive a trend would suggest that few if any hurricanes would develop in subsequent years.)

Figure 7. Variability of Hurricane and Landfall Proportion

Finally, a computation can be made of the proportion of tropical storms that intensify to hurricane strength and go on to make landfall in the U.S. As the bottom time series in Figure 7 shows, the percentage of basin storms that make U.S. landfall as hurricanes has only exceeded the long-term average of 14% in four of thirteen seasons. Note that 2004 is an outlier from the perspective of landfall proportion. In 2004, not only did the tropics produce 15 named storms, but fully one third of them made U.S. landfall as hurricanes. In 2007, just 1 of 14 storms, or just 7%, made U.S. hurricane landfall—well below the long-term average.

Again, it is evident that seasonal forecasts of tropical storm counts generally perform the best. It is the easiest statistic to predict because it requires the fewest assumptions—namely, elevated ocean temperatures will produce more storms. Forecasting the percentage of storms that intensify to hurricanes is more uncertain, and forecasting landfall proportion is even more difficult, requiring assumptions about where storms will form and where they will track. It is not surprising, therefore, that forecasters are generally silent on these points.

The main lesson of this analysis is that if forecasters simply apply long-term average intensification and landfall rates to their estimates of elevated storm counts, their forecasts can go very wrong. The assumption that elevated activity in the basin translates proportionally to elevated hurricane landfall activity (not to mention insured losses) is simply not borne out by the data.

What can be expected for 2008 and beyond?

Each of the last four hurricane seasons had unique features and it is this uniqueness that presents the real challenge for seasonal forecasters. In 2004, the proportion of hurricanes that made landfall was unusually high, as seen in Figure 7. In 2005, it was basin counts that were high, while the proportion of storms becoming hurricanes and making landfall was quite typical. In 2006, landfall proportion was unusually low, with no hurricane landfalls. Finally, this year the most unusual aspect of the season was the low proportion of the basin storms that went on to become hurricanes.

Clearly there is a danger in assuming that one or two single seasons are indicative of a paradigm shift in hurricane risk. While 2004 and 2005 were both very active seasons, they were not good predictors of activity in 2006 and 2007. What combination of seasons will be the best predictors of activity in 2008 and beyond? AIR would suggest that companies continue to take a broader look at all the data and science rather than rely on a few recent seasons and broad extrapolations from what goes on in the basin to what goes on along the U.S. coast.

AIR's current research is focused on Atlantic genesis and steering, which is providing valuable insights into why some seasons produce more landfalls than others. One of the more appealing aspects of the work AIR is doing on landfall risk is the transparent nature of the research. The datasets and analysis techniques are publicly available and the results are entirely reproducible.

The 2007 Atlantic hurricane season (along with 2006) highlights the uncertainty of seasonal and near-term forecasting—particularly as it relates to landfall risk. However, as in the past two years, AIR will continue to provide a near-term catalog as a supplement to, rather than a replacement for, the standard catalog for our 2008 U.S. Hurricane Model. The decision as to which catalog to use will depend on the client's business use case, risk tolerance and other factors. It is important to understand, however, that the near-term catalog continues to have a higher level of uncertainty than the long-term catalog.

Whichever AIR catalog you choose, you can be assured that it is the product of cutting edge research on landfall risk by a team of highly-credentialed meteorologists, climate scientists and statisticians at AIR.

*TSR=Tropical Storm Risk, CSU=Colorado State University, NOAA=National Oceanic and Atmospheric Administration, UKMet=Met Office