The Effect of the North Atlantic Oscillation On Atlantic Hurricanes Michael Barak-NYAS-Mentors: Dr. Yochanan Kushnir, Jennifer Miller Abstract Tropical cyclone behavior in the Gulf of Mexico (GM) and East Coast (ECB) was analyzed with respect to the North Atlantic Oscillation (NAO). Significant behavioral differences were found in the region off the East Coast of the United States and in the Gulf of Mexico following a High NAO winter. The NAO did not appear to significantly alter Cape Verde or GM births. Analysis showed that Cape Verde storms (those developing off the African Coast) were less likely to enter region ECB following a High NAO winter than in a Low NAO winter. Dating back to 1944 when hurricane records are relatively accurate, the mean number of storms to enter ECB that originated east of 81W was 2.57, 3.63, and 3.73, following High, Neutral and Low NAO winters respectively. A t-test calculates significant results to around 3 percent. Original data analysis of activity in region GM dating back to 1886 showed significant results for storms that were either born or traveled through the region. However, an analysis of only births revealed virtually zero difference in region GM following High or Low NAO winters. Consequently, an analysis of only storms that traveled through the region GM was completed and revealed highly significant results. Following Low NAO winters, tropical storms were twice as likely to enter region GM than after a High NAO winter. Results were significant to around 3 percent. Also of interest was an extreme difference in variance for the same test. There are several possible mechanisms responsible for decreased tropical activity near the United States following High NAO winters. First, an analysis of northerly steering winds shows an anomalous decrease of northerly winds off the East Coast and an anomalous increase of northerly winds in the far East Atlantic. This is indicative of the breakdown of the Bermuda high, which would allow for a more zonal flow across the Atlantic Ocean, thus carrying any tropical activity away. Moreover, SST over much of the tropical Atlantic is unusually cold following High NAO winters. Introduction Past studies of hurricanes in the Atlantic basin have concentrated on such factors as the El Nino- Southern Oscillation (ENSO) and rainfall in Africa. Gray (1984) attributes decreased hurricane activity in the Atlantic basin to processes that are caused by El Nino. Although much is known about factors that contribute to the frequency of hurricanes, little research has been done to predict the behavior of tropical systems once they develop. Climatologically, tropical systems are known to embark on general paths depending on the region in which they develop. For example, systems that form near the Cape Verde islands will travel west towards the Caribbean Islands and coastal United States. Forecasters can predict with relative certainty the tracks of hurricanes on a day to day basis by analyzing steering currents, but little is known why during some seasons the systems follow different general paths than during other seasons. The mechanism controlling much of the weather patterns encompassing the Atlantic Ocean must be responsible for seasonal variations of hurricane paths. The North Atlantic Oscillation (NAO) is the most dominant determinant of climate variability in the Northern Atlantic. For centuries, scientists have been aware of the impact of the NAO, but only in recent years has there been increased attention given to the NAO.
The NAO seesaws between two distinct phases and is measured by sea-level pressure (SLP). During the positive (High) phase, there exists a substantial difference in SLP between Iceland and the Portugal. During the negative (Low) phase, the difference in SLP is much less noticeable. The NAO is therefore measured by taking the anomalous difference in SLP between Iceland and the Portugal. (Stephenson, David, 1999) Because the NAO index is largest in the winter, scientists use a December to March average for assigning a value. Each winter, the NAO has a different index. Each phase is accompanied by distinct weather patterns. During the positive phase, northern Europe is more susceptible to strong storms while the eastern half of the United States experiences a mild, but wet winter. During the negative phase, storms follow a more linear path because of the breakdown of the subtropical high. The result is more rain in the region of Portugal and Spain and colder, snowier winters on the East Coast (Stevenson, David, 1999). Although the major affect of the NAO occurs during winter, the NAO affects the location of hurricanes in the following summer (Kushnir, Rajagopalan et. al, In preparation). For example, during the winter of 1996, the NAO index was strongly negative. During the following hurricane season, tropical systems tended to track very close to the coast of the United States. On the other hand, the winter of 1995 experienced a strongly positive NAO, and as a result, tropical systems during the 1995 hurricane season tracked at much safer distance from the East Coast of the United States. The following graphs depict this well.
1995 1996
Clearly, the 1996 hurricane season experienced many more hurricanes that either entered or hugged the East Coast than did 1995. This paper will focus on the regions of the Atlantic Basin in which tropical systems are affected by the NAO. Data/Methodology All tropical storm data is taken from the National Hurricane Center's Atlantic Tracks File. For a tropical storm to be counted, maximum sustained winds within the center of circulation must be at least 17ms -1. Position, latitude, longitude, wind-speed, and minimum pressure are given every 6 hours. Statistical tests use data after 1944 but analysis is made for all years since 1886 to show continuous trends. The rationale for not relying on data before 1944 is that prior to 1944 reconnaissance aircraft were not sent to investigate tropical systems, thus making the data less accurate. The index for the NAO is provided by Jim Hurrell. His method calculates the anomalous difference in SLP between Lisbon, Portugal and Stykkisholmur, Iceland during the winter months (DJFB). The index is ascribed to the year of January. Regions in the Atlantic Basin where analysis was made were the Gulf of Mexico (GM) and the region off the East Coast of the United States (ECB). These areas were selected based on their proximity to the United States. For purposes of this research, the coordinates 15-30N and 81-100W bound the GM region. In addition, the coordinates bounding the ECB region were 25-40N and 81-70W. Using the MATLAB (version 5.3) software package, restrictions were placed on how tropical storm data was recorded in each region using specific coding. One restriction remained constant for all trials: if a storm was plotted through the same region more than once, it was counted only once. This could occur if a storm made landfall and then tracked through the same region again. Data was recorded one hurricane season at a time. All trials recorded data from 1886 to 2000. In the first trial, all named storms to have positions within the ECB region were recorded. In a second trial for the ECB region, all named storms that developed east of 81W and traveled through ECB were recorded. This last experiment was performed to isolate Cape Verde and Bahamas originating storms. For the GM region, several trials were run with different restrictions. In the first experiment, all storms that had positions in the GM region were recorded. In the second trial, only storms that appeared in the GM box during the months of August and September were recorded. In the third trial, storms that were
born in GM were recorded. In the final experiment, all storms that originated east of 81W and traveled into GM were recorded. In addition to the suite of experiments with the two boxes, all storms originating east of 81W were also recorded to see if the NAO caused any significant change on Cape Verde or Bahamas births. As stated above, because data is not reliable before 1944, statistical evidence in each experiment was only used for years between 1944 and 2000, but analysis was conducted for years since 1886 on the assumption that the data was sufficiently accurate. In order to divide the NAO into three phases (High, Low, and Neutral), all years from 1944-2000 were divided into thirds based upon their NAO index value. A High NAO year is defined as any year where the value of the NAO index is greater than 0.7363 based on Jim Hurrell's scale. A Neutral NAO year is defined as any year where the NAO index is between 0.7363 and -0.3654. A Low NAO year is defined as any year where the NAO index is less than -0.3654. For each experiment, that data was placed into Excel version 2000. Column one was the MATLAB output of the individual experiment, column two the corresponding year, and column 3 the NAO index for the winter before the hurricane season. All the data was then sorted with ascending NAO index values. The following diagram displays this method for one of the experiments.
Two sample t-tests assuming unequal variance were then performed for each experiment. For each experiment, three statistical tests were performed. One test was done for High years vs. Low years, another for High years vs. Neutral years, and a final for Low years vs. Neutral years. All significant and nonsignificant results were recorded. Results Several of the experiments revealed significant results. For all named since 1944, the mean number in ECB following a High, Neutral, and Low NAO winter was 3.42, 4.57, and 4.52 respectively. The resulting t-tests when comparing High vs. Low and High vs. Neutral years revealed p-values near 5 percent. Statistical evidence became even stronger when data since 1886 was included. For the experiment in which storms originated east of 81W and traveled through ECB, significant results were once again found for High vs. Low and High vs. Neutral years. The mean number of storms to enter ECB that originated east of 81W since 1944 was 2.57, 3.63, and 3.73 for High, Neutral and Low NAO years respectively. In total, 94, 134, and 130 storms entered ECB from east of 81W since 1886 following High, Neutral, and Low NAO winters. P-values for this experiment were more significant as they were closer to 3 percent. Once again, the statistical evidence became stronger when data since 1886 was included.
Gulf of Mexico results were only significant for storms entering GM that originated east of 81W. Since 1944, the mean number of storms to enter GM during a High NAO year was 0.894, while during a Low NAO year, the mean was 1.736. The resulting p-value for this t-test was near 3 percent. There was little difference for the same trial after High NAO and Neutral NAO winters. Significant results were found for births in GM between Low and Neutral years. However, the data did not hold true for analysis that included all years since 1886. Also, borderline significant statistical evidence was found for storms that were recorded in region GM during the months of August and September. In the last trial for storms that were born east of 81W, results did not fluctuate significantly with the changing NAO phases.
Discussion The analysis of the data shows that regardless of the winter s NAO phase, the hurricane season s number of births in any region is not significantly impacted. Therefore, the significant results found for ECB and GM can be attributed to the general change in tracks of the storms once they develop. There is decreased tropical storm activity in the region off the East Coast of the United States following High NAO winters. Even more significant results were found for storms that originated east of 81W. Most of these storms were formed off the coast of Africa near the Cape Verde Islands. Because analysis of the data revealed that the NAO has no effect on Cape Verde or Bahamas births, it is reasonable to say that there are significant behavioral differences with these storms once they develop. Their behavior is dictated by physical factors that accompany the different phases of the NAO.
Following a High NAO phase, there is a dramatic drop in SST over much of the Atlantic Basin (Kushnir, Rajagopalan et. al, In preparation). However, because there appears to be no effect on births, the drop in SST must cause the storms to either become weakened and die when they encounter the colder water, or they may take a more northerly track which. If tropical systems were following more northerly tracks, they would become more susceptible to the upper level westerlies. The colder than normal SST is not nearly as strong during a Neutral NAO year. This can explain why the ECB region experiences approximately the same number of storms following Neutral and Low phases, as opposed to High phases. Region GM also experiences a significant drop in the number of storms entering from east of 81W after a High NAO winter. Once again, the drop in SST over much of the tropical Atlantic is probably responsible for this. The question as to why following Neutral NAO winters, Gulf traveling storms behave similarly to a High NAO winter, it may be that the time scale is to short. When including data from 1886, this trend dissipates. However, the drop in the number of storms following High NAO winters continues.
Another more important factor that may explain decreased tropical cyclone activity in ECB and GM is due to a significant variation in the steering currents following a High NAO winter. These changes show an anomalous decrease in northerly winds near the East Coast, and an anomalous increase of northerly winds in the Far East Atlantic. Because high-pressure systems spin clockwise, this change in steering currents indicates a breakdown of the Bermuda High pressure system. Without the dominant pattern created by the Bermuda High, weather patterns crossing the Atlantic would follow a more east to west path. Consequently, tropical systems would experience a much more difficult task making landfall on the US. Kushnir and Rajagopalan have found that there is a significant decrease in landfalls on the East Coast and Gulf of Mexico following a High NAO winter. Figure C depicts a combination of the Bermuda High shifting and strengthening following a Low NAO winter and is most likely responsible for the increase in storms entering GM after a Low NAO winter. As for storms in the Gulf of Mexico during the months of August and September, borderline significant results were found probably because although those are the months of the Cape Verde season, births were also included.
Conclusions The NAO affects the behavior of Atlantic hurricanes. The winter NAO appears to impact the following summer hurricane season. Fewer storms can be expected in the region off the East Coast and Gulf of Mexico following High NAO winters. Subsequently, more storms can be expected near the East Coast following High and Neutral NAO winters, and more storms can be expected to enter the Gulf of Mexico following Low NAO winters. Although the correlations are not completely definitive, the trends undeniably exist. Also, this research uses boxes to bound specific regions, but hurricanes do not behave in a linear fashion. In addition, hurricanes are rare events that will never be predicted with certainty. The distribution of storms was normal for some experiments and not normal for others. This could potentially impact the significance of some of the tests. Hopefully this research will spawn new studies on how the NAO during winter or perhaps other seasons impacts the hurricane season. Also, it may be interesting to look not only at tropical storms, but also their predecessors, tropical waves. Because this study analyzes the effect of the winter months, appropriate predictions can be made for the following summer season. The ability to forecast the destructive forces of nature over the long term is beneficial to every segment of society, and is especially useful to insurers, urban planners, land developers, and government organizations that must properly allocate limited resources. It is my belief that it is of significant value to not only predict the number of storms that will develop each year, but also to predict where those storms will go once they are born.
References Bell, Ian, North Atlantic Oscillation Web Page, Lamont Doherty Earth Observatory of Columbia University, Palisades, New York 10964-8000 (Site name: http://www.ldeo.columbia.edu/nao) Gray, W. M., 1984; Atlantic seasonal hurricane frequency. Part I; El Nino and 30mb quasi-biennial oscillation influences. Mon. Wea. Rev., 112, 1649-1668 Hurrell, J. W., 1995 Decadal trends in the North Atlantic Oscillation: regional temperatures and precipitation. Science, 269, 676-679 National Hurricane Center Web Page (Site name: http://www.nhc.noaa.gov) Stephenson, David B., 1999 The North Atlantic Oscillation Thematic Web Site University of Reading, Early Gate, PO Box 243, Reading RG6 6BB UK Yochanan Kushnir, Balaji Rajagopalan, Relationships Between Large Scale Climate Indices and Named Storm Frequency in the Atlantic Basin In Preparation