Interannual variations in U.S. tropical storm and hurricane landfalls are examined over the interval 1950– 2002. The 10 highest and 9 lowest U.S. storm and hurricane landfall years are highlighted. U.S. landfalls are 6 times more frequent during the 10 highest than during the 9 lowest years. Nine major hurricanes struck the United States during the 10 highest U.S. landfall years, one struck during the 9 lowest U.S. landfall years. There is a positive correlation between Atlantic basin storm and hurricane frequency and U.S. storm and hurricane landfall frequency, but U.S. landfall variability explained by that relationship is small. Years with high (low) U.S. landfalls have a high (low) frequency of storm and hurricane formation in the Gulf of Mexico and the western Caribbean Sea and a high (low) percentage of them landfall along the U.S. Gulf coast. U.S. landfall frequency of Atlantic storms and hurricanes is much higher (lower) during high (low) U.S. landfall years, implying that Atlantic steering currents are more (less) favorable for U.S. landfall. La Niña conditions occurred 19% more often during high U.S. landfall years than during remaining years. El Niño conditions occurred 10% more often during low U.S. landfall years than during remaining years. Skill of inferring how many storms and hurricanes will landfall in the United States from a forecast of the number of Atlantic basin storms and hurricanes explains an average of 18% or less of U.S. storm and hurricane variability in a hindcast setting. Results indicate that a large portion of U.S. storm and hurricane landfall variability is related to where the storms form and whether steering currents are favorable or unfavorable for bringing them to the United States.
Tropical cyclones affect U.S. tourism, transportation, life, property, military operations, the oil industry, and the coastal ecosystem. They can devastate the coast and bring massive flooding inland. The Federal Emergency Management Agency, civil defense personnel, as well as state and local emergency managers are challenged annually with evacuation decisions when tropical cyclones threaten. The U.S. coastline has never been more vulnerable to tropical cyclones, due to increasing coastal populations, widespread building in the coastal zone, persistent beach erosion, sea level rise, and coastal subsidence (Pielke and Pielke 1997; Douglas et al. 2001).
Much focus has been on U.S. major hurricane landfalls, because major hurricanes have been shown to cause a disproportionately large fraction of damage relative to their rarity (Pielke and Pielke 1997). This is due in part to the accumulation of wind damage with increasing wind speed, and the fact that structures have thresholds where wind damage begins. Since 1970 about 60% of U.S. tropical cyclone deaths have been caused by freshwater rainfall/flooding [see Rappaport (2000) and the National Hurricane Center Web site: http://www.nhc.noaa.gov/HAW2/english/inland_flood.shtml]. Historically, tropical cyclone flooding has proven to be, in general, relatively independent of intensity. Weak, slow-moving storms can produce huge flood damage and numerous deaths. For example, in 2001 weakening remnants of Tropical Storm Allison caused U.S. flood damage that exceeded $5 billion along with 41 deaths (see Beven et al. 2003). Because of ocean wave growth behavior, coastal beach flooding and erosion from ocean waves can be more extensive for large tropical storms than for small hurricanes. Storms and hurricanes of all intensities can be costly and disruptive to the United States. For these reasons examination of both storm and hurricane landfalls is undertaken.
There are large interannual variations in the number of U.S. storm and hurricane landfalls. Over the past 53 yr, an average of three storms and hurricanes have made landfall in the United States annually, about half of which have been hurricanes. Year-to-year variations in U.S. landfall frequency have been high, ranging from zero landfalls in 1962 to seven landfalls in 1985, 1998, and 2002. Storm and hurricane frequency across the Atlantic basin for the past 8 yr has been well above the long-term average, but the percent of storms (especially major hurricanes) that have struck the United States has been below average. For example, only 1 of the most recent 21 Atlantic basin hurricanes has struck the United States. Three of the last 27 major hurricanes in the Atlantic basin have struck the United States (the long-term average is one in three). This suggests that (at least episodically) the number of Atlantic basin hurricanes is not well related to the number that landfall in the United States.
Bove et al. (1998b) focused on U.S. hurricane landfall trends through time. Their results indicate that over the past 110 yr there has been no sign of an increase in hurricane frequency or intensity in the Gulf of Mexico. Smith (1999) indicates that trends in hurricane landfall frequency along the U.S. east coast are slightly positive, but no intensity trends were found. Landsea et al. (1999) expanded on this work by examining tropical cyclone development indices and their change through time.
Gray et al. (1992, 1993) has led research on tropical cyclone seasonal forecasts and their causes (information online at http://typhoon.atmos.colostate.edu). They have shown statistical relationships among weather parameter “predictors” and the frequency of intense U.S. hurricane landfalls and seasonal occurrence of storms and hurricanes across the Atlantic. From 1984 through 2002 their Atlantic basin storm and hurricane forecasts have shown skill (correlation of 0.68 for 1 June forecasts) above a forecast of average when attempting to predict damage potential to land across the Atlantic basin [see e.g., the Web site www.aoml.noaa.gov/hrd/project2000/ cl_proj1.html and Owen and Landsea (2003)]. They have shown that tropical Atlantic hurricanes are related to U.S. east coast landfalls, but have a poorer relationship to hurricanes that make landfall in the Gulf of Mexico.
Pielke and Landsea (1998, 1999) have shown a relationship between U.S. hurricane “damage” and El Niño and La Niña. When averaged through time, La Niña years were associated with higher U.S. hurricane damage than were El Niño years. Bove et al. (1998a) confirmed the findings of Pielke and Landsea. Namely, that there is a lower probability for a hurricane or a major hurricane landfall in the United States during an El Niño (especially a strong El Niño). They calculated the probability of two or more U.S. hurricane landfalls to be about 28% during an El Niño, about 48% during neutral conditions, and about 66% during a La Niña over the past 98 yr. Interestingly, 1 yr had just one U.S. named storm landfall, but the impact was huge. In 1992 (an El Niño year and quiet hurricane season) Andrew, the costliest hurricane in U.S. history, struck Florida.
Seasonal predictions of storms and hurricanes in the Atlantic basin (and other basins worldwide) reach the forefront of news at the beginning and at the peak of each hurricane season. The media and the public routinely infer the likelihood of U.S. landfalls from basin- wide seasonal forecasts (despite cautions not to). The most recent example of the failure of such an inference was 2002 (an El Niño year). Despite a seasonal forecast for a below average season, seven named tropical cyclones struck the United States and an eighth impacted the United States, making it one of the highest U.S. tropical cyclone landfall years on record.
3. Identification of interannual variability
Interannual variations in U.S. storm and hurricane landfalls are identified using the most recent portion of tropical cyclone records (1950–2002). This recent half- century is the most accurate of the records. Military aircraft have flown into tropical cyclones over this entire period, although not far out at sea. Satellites have monitored tropical cyclones over about 70% of this period. Once satellite information became routinely available (1960s) there was little doubt about the number of Atlantic basin tropical cyclones in any year, where they formed, or where they moved. Questions about tropical cyclone intensity and/or whether it is warm core, cold core, or a hybrid (subtropical), along with its size, are some of the remaining ambiguities. Recently, scatterometer wind measurements from polar-orbiting satellites have improved estimates of tropical cyclone size, at least around their rain-free peripheries. Prior to the satellite era records of where in the ocean tropical cyclones formed and where they moved were not as accurate. National Hurricane Center “best tracks” (see Neumann et al. 1999; Jarvinen et al. 1984) are used to separate U.S. storm and hurricane landfall years into three groups: higher than average, lower than average, and the remainder. Groups are compared and contrasted.
Atlantic basin and U.S. landfalling storm and hurricane averages, standard deviations, and ratios are shown in Table 1. Only storms and hurricanes whose circulation center crossed the U.S. coastline are considered landfalls. Close calls and tropical depressions are excluded. Landfall frequency is grouped as average (within one standard deviation of the mean), below average (one standard deviation or more below the mean), and above average (one standard deviation or more above the mean). Distributions are not exactly normal, but close. The average of three U.S. storm and hurricane landfalls with a standard deviation of 1.6 provides an estimate of 4.6 or more landfalls and 1.4 or less landfalls as exceeding plus and minus one standard deviation, respectively. The highest 10 yr (five or more U.S. landfalls) and lowest 9 yr (one or less U.S. landfalls) of this 53-yr study period come close to meeting these criteria.
5. High and low U.S. landfall years
Table 2 lists the 10 highest and 9 lowest U.S. storm and hurricane landfall years from 1950 to 2002 (hereafter referred to as H10 and L9, respectively). Examination of Fig. 1 illustrates that the outlier in the highest 10 yr is 1995 with 19 named cyclones and five U.S. landfalls. The year 1995 meets the criteria of five or more landfalls, but its ratio of total storms to U.S. landfalls is unlike the other nine high years. For example nine of the high years display a landfall ratio of 40% or higher, 1995 is down near 25%, which is close to the 53-yr mean. Conversely, all low landfall years have a basin to landfall ratio that is less than 0.2. It would have been possible to define “high” and “low” years based on the percent that made landfall. Primarily high years would have been defined slightly differently from the definition adopted here. However, percent of tropical storms and hurricanes that make landfall does not directly reflect U.S. landfall impacts and hence was not adopted. Removal of 1995 from the high-year group (to make like 9-yr samples of high and low landfalls) does little to affect the outcome of the statistics and results found in this paper, although it does increase the average number of storms and hurricanes within the 10-high- year sample by about one storm and hurricane per year.
Some storms and hurricanes made landfall more than once. Numbers in parentheses in Table 2 represent the total number of U.S. coastal crossings by landfalling storms and hurricanes each year. The average number of U.S. landfalls is about 6 times higher during H10 than during L9 [e.g., an average of 5.8 (0.9) landfalls occurred during H10 (L9)]. Major hurricanes (a subset of hurricanes) are also shown. Nine major hurricane landfalls occurred during H10 (five in the Gulf of Mexico and four along the East Coast). One major hurricane struck the U.S. Gulf coast during L9.
There were 58 (8) U.S. landfalls during H10 (L9), an average of 5.8 (0.9) landfalls per year. A total of 126 (77) storms and hurricanes, a frequency of 12.6 (8.6) per year, formed across the Atlantic basin during H10 (L9). This 47% difference in Atlantic basin storm and hurricane activity between H10 and L9 is mostly due to storm and hurricane formation differences in the Gulf of Mexico and the western Caribbean Sea. Linear scaling of U.S. landfall differences by this 47% difference in Atlantic basin storm and hurricane occurrences would suggest that H10 years might have an average of 1–1.5 more landfalls per year than L9 years, but not 4.9 more per year as was observed.
By comparing the percent of Atlantic basin storms and hurricanes that made landfall in the United States during H10 and during L9, normalized landfall differences (removal of differences of storm and hurricane occurrence and the number of years included within each group) are isolated (Table 3). For example 48% (11%) of Atlantic basin storms and hurricanes and 47% (13%) of Atlantic basin hurricanes hit the United States during H10 (L9). The percent of storms and hurricanes that made landfall during H10 is 4.4 times higher than the percent that made landfall during L9. The largest percent difference is evident in storm landfalls, 7.4 times higher during H10 years (52% of storms made landfall during H10 in contrast to 7% during L9).
Figure 1 plots numbers of Atlantic basin storms and hurricanes versus numbers that made landfall in the United States. Consistent with our definition L9 years, H10 years, and remaining years are clearly delineated. Interestingly, all 34 of the remaining years (those excluding H10 and L9 years) had 2, 3, or 4 U.S. landfalls while Atlantic basin storm and hurricane occurrences varied from 4 to 17. One or zero U.S. (five, six, or seven) storm and hurricane landfalls was observed when Atlantic basin occurrences ranged from 5 to 14 (8 to 19). Separately, H10, L9, and the remaining years display no trend in the number of U.S. landfalls relative to the number of Atlantic basin storms and hurricanes. Separation of each of these groups based solely upon Atlantic basin storm and hurricane totals is difficult to impossible. Even if we were to define L9 years as those years with the highest Atlantic storm and hurricane occurrences (excluding H10 years), the number of U.S. landfalls would still fall well below the number of U.S. landfalls observed during H10 years.
However, there is a small, statistically significant, positive correlation between Atlantic basin numbers and U.S. landfall numbers. The linear best-fit slope for Fig. 1 is 0.22. Thus for all groups combined, on average, 5 named storms per year would indicate 1.1 landfalls, 10 named storms per year would indicate 2.2 landfalls, and 15 named storms per year would indicate 3.3 landfalls per year. Unfortunately scatter is large.
Table 4 shows explained variances for combinations of Atlantic basin storms and hurricanes versus numbers that made U.S. landfall. Correlations are positive for all combinations listed and most are significant at the 5% level, but explained variances are small. Over the 53- yr period, 19% (81%) of U.S. landfalling storm and hurricane variability is (is not) explained by knowing how many storms and hurricanes occurred in the Atlantic basin. Explained variances are lowest for major hurricanes (10% or less). Variances shown are diagnostic, not prognostic. They were calculated already knowing the exact number of Atlantic basin storms and hurricanes that occurred each year since 1950. Uncertainty in predicting basin-wide storm and hurricane numbers as the season begins (1 June) typically results in forecast explained variances near 50% (based on a forecast correlation skill of 0.69 for 1 June). Hence, we might expect 10% (90%) of U.S. storm and hurricane landfall variability to be explained (unexplained) by a typical 1 June forecast of Atlantic basin tropical storms and hurricanes.
6. Variations in tropical cyclone origins
Formation locations (proximity to the U.S. coastline), steering currents (which favor tracks to the U.S. coastline), or both must explain much of the difference in U.S. landfalls between H10 and L9. Comparisons of U.S. storm and hurricane landfall locations during H10, L9, the remaining 34 yr (R34), and all years are shown in Table 5. About 65% of U.S. storm and hurricane landfalls occurred in the Gulf of Mexico during H10 years, R34, years and “all years” groupings. Ninety percent of U.S. landfalls occurred in the Gulf of Mexico during L9 years. Hence, Gulf of Mexico landfalls are critical when determining U.S. tropical cyclone impacts.
Differences in storm and hurricane formation areas between H10 and L9 are shown by plotting origins and end points (or first U.S. landfall positions) for H10 (Fig. 2a) and separately for L9 (Fig. 2b). Forty-nine storms and hurricanes originated in the Gulf of Mexico and Caribbean during H10 years, while only 14 originated there during L9 years. That equates to 4.9 origins per year during H10 versus 1.6 origins per year during L9. This difference is statistically significant at the 1% level, based on a sampling of storm and hurricane differences from random 9-yr groupings from 1950 to 2002 (of 50 random samples of 9 yr, none exceed a difference of greater than 2.1 origins per year). Thirty-six of 49 (73%) Gulf of Mexico and Caribbean Sea storms and hurricanes made landfall in the United States during H10. Five of 14 (36%) Gulf of Mexico and Caribbean Sea storms and hurricanes made landfall in the United States during L9. Hence, U.S. storm and hurricane landfalls during the 53-yr period are strongly modulated by the number of storms that originate in the Gulf of Mexico or Caribbean Sea and the fraction of them that finally hit the United States (which is related to steering currents favorable or unfavorable for U.S. landfall).
Close to the same frequency of storms and hurricanes formed in the Atlantic Ocean in H10 as in L9 (e.g., 7.7 origins per year during H10 compared to 7 origins per year during L9). During H10 years, 4.7 storms and hurricanes per year originated in the deep Tropics (at or south of 20°N). This compares to 3.2 origins per year there during L9. During H10, an average of three storms and hurricanes per year originated north of 20°N in the Atlantic. During L9 there was an average of 3.8 origins per year north of 20°N.
Most interesting are the formation locations for storms and hurricanes that eventually made U.S. landfall (Figs. 3a and 3b). Immediately obvious is the high frequency of landfalls originating in the Gulf of Mexico and tropical Atlantic during H10 compared to L9. For example, during H10, 36 (an average of 3.6 per year) U.S. storms and hurricanes originated in the Gulf of Mexico or the Caribbean Sea. Conversely, during L9 years, 5 (an average of only 0.6 per year) U.S. storms and hurricanes originated in the Gulf of Mexico or the Caribbean Sea. The frequency of U.S. landfalls originating in the Gulf of Mexico or the Caribbean Sea was about 6 times greater during H10 than during L9.
Eight (0.8 per year) storms and hurricanes that originated north of 20°N eventually made landfall in the United States during H10. In contrast 2 (0.2 per year) storms and hurricanes did so during L9. Fourteen storms and hurricanes that eventually hit the U.S. coast originated in the tropical Atlantic south of 20°N during H10 (an average of 1.4 per year). That was 30% of all tropical Atlantic storms and hurricanes during H10. In contrast during L9 only one U.S. hurricane originated in the tropical Atlantic, which was an average of 0.1 per year. That was only 3% of all tropical Atlantic storms and hurricanes during L9 years. This difference is also highly significant at least to the 1% level. Hence, the difference in tropical Atlantic storms that eventually hit the United States is also important (30%), but the total number of U.S. landfalls from this area does not compare to those originating in the Gulf of Mexico and the northwest Caribbean Sea (70%).
The number of storms and hurricanes that curved north of 35°N into the North Atlantic indicates that 42 did so during L9 and 35 did so during H10. The percent difference is 33% higher for L9 years than during H9 years. Obviously these results strongly imply that Atlantic basin steering currents favored U.S. landfalls during H10 years as compared to L9 years by a ratio of about 14 to 1. During H10 (L9), 10 tropical cyclones dissipated over open water south of 25°N in the Atlantic, in the Caribbean or in the Gulf of Mexico, indicating that the percent of systems decaying over water before having a chance to landfall in the United States was roughly the same for each group.
7. El Niño–Southern Oscillation connections
Table 6 displays ENSO phases as defined by the National Oceanic and Atmospheric Administration/National Weather Service/National Centers for Environmental Prediction (NOAA/NWS/NCEP) Climate Prediction Center for H10 years and for L9 years (information online at http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.html). Neutral and cool ENSO phases were equally frequent during H10 years. A warm ENSO phase was most frequent during L9. A neutral ENSO phase, followed closely by a warm phase, was most frequent during the remaining 34 yr. H10 (L9) years were 18% (25%) more likely than L9 (H10) years to occur during a La Niña (an El Niño). The largest percent separation among ENSO phases was the warm ENSO phase during L9 years. These percentages are based on already knowing the ENSO phase and amplitude. In a real-time forecast setting uncertainty in forecasting ENSO phase and strength through the Atlantic basin hurricane season can be problematic. These results are consistent with previous research on ENSO effects on U.S. landfalling hurricanes.
U.S. storm and hurricane landfalls were grouped into the highest 10 yr, into the lowest 9 yr and into remaining years within the most recent 53-yr record. U.S. storm and hurricane landfalls were about 6 times more frequent in high landfall years than in low landfall years. Major hurricanes struck the United States nine times during high U.S. landfall years (half originated in the Gulf of Mexico and the Caribbean Sea, and half originated in the Atlantic), but only once during low U.S. landfall years. Other differences between high and low U.S. landfall years are summarized below.
Storm and hurricane formation was about 6 times more frequent in the Gulf of Mexico and the western Caribbean Sea during high U.S. landfall years than during low U.S. landfall years, and a higher percentage of these storms made landfall in the United States during H10 years (75%) than during L9 years (36%). This result is significant at least to 1%.
Twenty-two of 77 (29%) Atlantic storms and hurricanes tracked to the U.S. coast during high U.S. landfall years, but only 3 of 63 (5%) did so during low U.S. landfall years. This strongly implies that Atlantic steering currents were more efficient at bringing storms and hurricanes to the U.S. coast during high landfall years.
Approximately 7.7 storms and hurricanes per year formed in the Atlantic Ocean during high U.S. landfall years, and similarly 7 per year formed there during low U.S. landfall years. The big difference was in the Gulf of Mexico and the western Caribbean Sea. During high U.S. landfall years an average of 4.9 origins per year were observed there, in contrast to 1.6 origins per year during L9 years.
La Niña (El Niño) conditions were 18% (25%) more frequent during high (low) U.S. landfall years than during low (high) U.S. landfall years.
U.S. storm and hurricane landfall frequency is positively correlated with Atlantic basin tropical cyclone frequency, but over the past 53 yr, level of success in inferring U.S. storm and hurricane landfall frequency from Atlantic basin storm and hurricane frequency is low.
Results strongly suggest that forecasting a high portion of variability in U.S. storm and hurricane landfalls requires at least 1) accurately forecasting variations in storm and hurricane origins, 2) accurately forecasting variations in steering currents favorable or unfavorable for moving storms and hurricanes to the U.S. coastline, and 3) accurately forecasting the phase and amplitude of ENSO through the Atlantic hurricane season.
For any specific year, it is “dangerous” to infer numbers of U.S. landfalling storms and hurricanes from numbers that occur or are expected to occur across the Atlantic basin.
Thanks to The Weather Channel for financial support for this research and to anonymous reviewers for helpful suggestions that improved this note.
Corresponding author address: Dr. Steven W. Lyons, The Weather Channel, 300 Interstate North Pkwy., Atlanta, GA 30339. Email: SLyons@weather.com