Predicting Atlantic hurricane paths using monthly surface pressure data more

Predicting Paths of Atlantic Tropical Cyclones Using Monthly Surface Pressure Data P. Grady Dixon Department of Geosciences Mississippi State University Mississippi State, Mississippi 39762 E-mail: grady.dixon@msstate.edu Michael E. Brown Department of Geosciences Mississippi State University Mississippi State, Mississippi 39762 W. Michael Carter Department of Geography University of Georgia Athens, Georgia 30602 W. Scott Gunter Jared S. Allen Ashley M. Hayes Laura E. Becker Heather S. Eschete Ryan P. Aylward Department of Geosciences Mississippi State University Mississippi State, Mississippi 39762 Kelsey N. Scheitlin Department of Geography Florida State University Tallahassee, Florida 32306 ABSTRACT Previous research has had some success in predicting likely tracks of tropical cyclones in the Atlantic basin using North Atlantic Oscillation (NAO) anomalies (above or below average values) during preceding months . This paper expands on this research by incorporating other surface pressure values as independent variables as alluded to by some of the earliest NAO research . We examine monthly sea-level-pressure (SLP) data from Reykjavik (Iceland), Cape Hatteras (North Carolina), and Nassau (Bahamas), along with NAO index anomalies, to see if they can be used to predict future paths and landfall locations of Atlantic tropical cyclones during the period 1970–2005 . Average SLP from Nassau during the preceding May and June shows the strongest correlations with tropical cyclone tracks, with increased values correlated with more cyclones impacting the southeastern United States, but fewer landfalls along the Gulf of Mexico . Results also show virtually no relationship between the storm variables tested and the NAO index . Key Words: hurricanes, NAO, storm track INTRODUCTION Historically, there has been much interannual variation in the number of tropical cyclones that form in the North Atlantic Ocean as well as the number that make landfall in the United States (Xie et al . 2005) . The past ten years have experienced notably more Atlantic storms than previous decades (Goldenberg et al . 2001), with Hurricane Katrina being the most infamous (Bell et al . 2007) . Hence, it has become particularly important to prepare coastal communities as far in advance of landfall as possible . There has been some success in long-term forecasting of the number and intensity of North Atlantic tropical cyclones using various global climate measures, but prediction of storm tracks seems more difficult (Klotzbach and Gray 2004; Xie et al . 2005) . 77 The Geographical Bulletin 49: 77-86 ©2008 by Gamma Theta Upsilon P. Grady Dixon et al. Elsner and Jagger show that variations in both the El Niño–Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO) can be used seasonally to predict North Atlantic tropical cyclone frequency and tracks (2004) . More specifically, Elsner et al . suggest that North Atlantic tropical cyclone activity at low latitudes is below normal whenever activity is above normal at high latitudes, and vice versa (2000) . The NAO is one of the longest-researched large-scale climate patterns (Walker 1924; Walker and Bliss 1932) . The term refers to an oscillation of surface atmospheric pressure between the subtropical high-pressure belt (near the Azores) and the subpolar lowpressure belt (near Iceland) . Positive phases of the NAO are characterized by above-average high pressure at Ponta Delgada (Azores) and below-normal pressure near Reykjavik (Iceland), resulting in an enhanced pressure gradient across the northern Atlantic Ocean (Fig . 1) . Negative phases result in the opposite pressure anomalies (i .e ., above-average pressure near Reykjavik and below-average pressure near Ponta Delgada), and therefore, a weaker pressure gradient between the two regions (Lamb and Peppler 1987) . The purpose of this study is to explore variables for predicting seasonal tracks of tropical systems that are less ambiguous than the commonly used NAO index, which does not necessarily represent the atmospheric pressure field across the Western Atlantic . Previous research has shown some success in discriminating between northern and southern storm tracks of Atlantic tropical systems by using the NAO index (Elsner et Figure 1 . Locations of interest for this study . Pressure-based variables include the NAO index (calculated using pressure data from Reykjavik and Ponta Delgada) and SLP data from Reykjavik, Cape Hatteras, and Nassau . 78 Predicting Paths of Atlantic Tropical Cyclones Using Monthly Surface Pressure Data al . 2000; Elsner 2003; Elsner et al . 2004; Xie et al . 2005), but this seems counterintuitive, since the NAO index is most commonly calculated using pressure data from Iceland and the Azores, which are both across the Atlantic from the primary area of interest . A decrease in SLP near the Azores could be due to a weakened subtropical high pressure, or it could be due to a shift in this high-pressure center . A strong high-pressure center that is located nearer the eastern Atlantic Ocean is likely to cause more storms to curve northward prior to reaching North America . Conversely, a southwestward shift of the high pressure would certainly force more storms into the Gulf of Mexico . A weakened high-pressure center should have fewer effects on storm tracks, but there is virtually no way to distinguish between a weakened high pressure and a southwestern shift based solely on the NAO index . Therefore, based on previous NAO research that suggested the use of western action points (locations used as points of measurement for determining the NAO) (Walker and Bliss 1932), we used pressure data from Hatteras (North Carolina) and Nassau (Bahamas) in this study . HISTORy OF THE NAO Since its discovery, the NAO has been measured in a variety of ways . Walker originally defined the NAO as the simple difference in pressure between Iceland and the Azores (1924) . A more precise definition was developed by Walker and Bliss, who defined a formula to calculate the NAO that was based on the surface pressure and temperature data during the months of December, January, and February for several different locations in Iceland and western Europe (1932) . However, Walker and Bliss stated that the Azores center should not be included in the formula due to its normally small range in pressure values (1932) . Nonetheless, several authors have included the Azores as an action point for the NAO (Wallace and Gutzler 1981; Rogers 1984) . Hurrell (1995) and Hurrell and Van Loon (1997) designated the action points to be Lisbon (Portugal) for the southern node and Stykkisholmur (Iceland) for the northern node, so that the NAO record could be extended prior to 1895 . Using Lisbon and Stykkisholmur, Hurrell (1995) was able to obtain NAO indices dating back to 1865 . Upon comparing the NAO index calculated using the nodes of Hurrell (1995) and the nodes of Walker and Bliss (1932), Hurrell found that this change in location did not significantly alter the original NAO index (Hurrell 1995) . In an effort to further lengthen the NAO index record, Jones et al . used pressure data from Gibraltar and Reykjavik, Iceland (1997), whose more complete records of pressure data allowed the NAO index record to be extended back to 1821 . They also compared the three southern points of Ponta Delgada, Lisbon, and Gibraltar to determine if either had a stronger correlation with the NAO, and it was found that during the winter months, Lisbon and Gibraltar provided index values as reliable as Ponta Delgada . This shows that the derivation of the NAO index is not necessarily limited to two specific locations, but that variances in the southern centers (throughout the Iberian Peninsula) also give similar outcomes for the NAO index . One possibility that has not been extensively explored is using a southern center on the western side of the Atlantic Ocean, as opposed to the eastern side . Walker and Bliss (1932) included the temperatures from Cape Hatteras, North Carolina and Washington, D .C . in their original formula for calculating the NAO index, but pressure data from this part of the world have not been consistently used in this calculation since their formulation . Therefore, it is not known whether there is any correlation between sea-level pressure on the western side of the Atlantic Ocean, the NAO index, and hurricane paths . It is this idea that led to the current study . DATA This study is testing the utility of some pressure-based variables for predicting future 79 P. Grady Dixon et al. paths of named tropical systems . Our data include information about past Atlantic storms (named tropical systems) and sea level pressure (SLP) from specific locations . NAO index data, which are partially based on SLP data, are also used as a standard to compare other independent variables, since the NAO index has already been shown to be a useful tool for predicting tropical system variables . The NAO index is based on leading rotated (varimax) principal component analyses of mean 500-hPa heights over the northern (20º–90º N latitude) Atlantic Ocean (Barnston and Livezey 1987) . We obtained monthly NAO index anomalies (based on the 1950–2000 climatological mean) for the study period from the Climate Prediction Center (available online at [http://www . cpc .noaa .gov]) . For pressure data, we relied on three weather stations that recorded daily average sea level pressure: Cape Hatteras (North Carolina), New Nassau Airport (Bahamas), and Reykjavik (Iceland) (Fig . 1) . We obtained data for Cape Hatteras and New Nassau Airport through the National Climatic Data Center encompassing the years 1973–present and 1973–2000, respectively . Cape Hatteras’ station is located at 35 .27° N, 75 .55° W . New Nassau Airport is located at 25 .05° N, 77 .47° W . Reykjavik’s data are from the Icelandic Meteorological Office and encompassed the years 1970–2005 . The station is currently located at 64 .13° N, 21 .90° W . Before 1973, the station was located nearby, 40 meters lower in elevation . As is standard for most climate reporting stations, the pressure data used in this study are reported as sea-level pressure, thus the differences in elevation will not have any effect on the data . This study makes use of tropical cyclone data to test the reliability of different variables as predictors of seasonal tropical cyclone paths . These data contain information associated with only named tropical cyclones over a thirty-six year span from 1970 through 2005 . We included storm track, pressure, and intensity data for the 382 named storms 80 during the study period from the National Hurricane Center (NHC) . METHODS Average NAO index values from each May– June period (2-month) and January–June period (6-month) have both been used successfully to determine whether storms during the following hurricane season will tend to track toward the north or the south (Elsner et al . 2000; Elsner 2003; Xie et al . 2005) . Therefore, we employ both time periods (2-month and 6-month) for the analyses in this project . We use twelve pressure-based variables in this study . In addition to the two NAO variables, the variables are meant to represent the standard northern action point (Reykjavik) of the NAO index as well as potential southern action points that are located along the western Atlantic coast (Hatteras; Nassau) . All of these are based on monthly composites of sea-level pressure (SLP) at each location, and four of the variables actually represent the pressure gradient (PG) between Reykjavik and the two southern locations . The variables used in this study are listed in Table 1 . The storms were categorized by their location relative to 35°N and 75°W, similar to Elsner (2003) . We established two categories for storms that had a northern track or a southern track . A storm was considered “north” if it crossed 35°N latitude before reaching 75°W longitude . “South” storms are Table 1 . List of pressure-based variables used in this study . SLP = sea level pressure; PG = pressure gradient . Study Variables NAO Index Reykjavik SLP Hatteras SLP Reykjavik–Hatteras PG Nassau SLP Reykjavik–Nassau PG Predicting Paths of Atlantic Tropical Cyclones Using Monthly Surface Pressure Data those that crossed the 75°W longitude prior to reaching 35°N latitude (Fig . 2) . We created two additional categories for the 116 named storms that made landfall in the United States . A storm that made landfall along any portion of the Gulf of Mexico coastline at any stage in its life cycle was placed in the category “Gulf .” Other storms that made landfall in the United States, but did not affect the Gulf Coast, were placed in the category “non-Gulf .” Tropical Storm Dottie made landfall along the south coast of Florida in 1976 and was not included in this study because we could not clearly determine whether Tropical Storm Dottie significantly affected the portion of Florida that borders the Gulf of Mexico . We use six variables to describe annual variations in tropical cyclone tracks . All of these variables are meant to separate storms that affect southeastern states from those that have more northerly tracks and impact states farther to the north . Each variable is analyzed for all named storms occurring that year, but analyses are also performed for more specific categories (hurricanes only, tropical storms only) . This allows us to identify any relationships that might exist solely for stronger or weaker systems . The six variables are: • Percentage of total tropical cyclones with a “south” track • Percentage of total hurricanes with a “south” track Figure 2 . Triangles represent the average longitude where east-moving or west-moving storms first crossed 35º N latitude . Circles represent the average latitude where west-moving storms first crossed 75º W longitude . Both groups are subdivided based on the average May–June SLP at Nassau, Bahamas: 1015 = 1015 .0–1015 .9 hPa; 1016 = 1016 .0–1016 .9 hPa; 1017 = 1017 .0–1017 .9 hPa; 1018 = 1018 .0–1018 .9 hPa . 81 P. Grady Dixon et al. • Percentage of total tropical storms with a “south” track • Percentage of total tropical cyclones making landfall in Gulf • Percentage of total hurricanes making landfall in Gulf • Percentage of total tropical storms making landfall in Gulf We compared each of the pressure-based variables to each of the storm-based variables using bivariate correlation and discriminant analysis . The bivariate correlations measure the linear relationship between the two variables, with the pressure-based data acting as the independent variables . The discriminant analysis used in this study requires that the dependent variable be a binary value . We classified each year’s hurricane activity as “above average” and “below average,” based solely on the data used during the study . For example, approximately 20% of all cyclones that occurred during the study period eventually made landfall somewhere along the U .S . Gulf Coast . Years with less than 20% of the cyclones making landfall along the Gulf Coast would be considered “below average .” Discriminant tests are applied in order to determine whether above average years can be distinguished from below average years based just on the independent variable, which makes this analysis method more useful than the commonly used bivariate correlation . An average value typically represents the mid-point of extreme values rather than a value that is most often expected . With that in mind, there are rarely “average” years . Hence, all years of this study were easily included within one of the two categories used for discriminant analysis . For the discriminant analyses, the dependent and independent variables can also be reversed in order to further test the relationships . For example, storm-based data can also become the independent variables while the pressure-based data are reduced to binary “above average” and “below average” values . This test determines whether years with anomalous pressure values can be successfully identified using only storm data . Table 2 . Median p values, based on the six statistical tests conducted for each relationship between pressure-based and storm-based variables . Variable Median p value All Tropical Cyclones 0.625 0.632 0.526 0.452 0.632 0.399 0.519 0.583 0.250 0.611 0.702 0.405 0.528 Median p value Hurricanes 0.815 0.347 0.507 0.207 0.924 0.854 0.676 0.583 0.155 0.706 0.786 0.405 0.580 Median p value Tropical Storms 0.625 0.476 0.860 0.563 0.482 0.327 0.559 0.596 0.221 0.611 0.586 0.495 0.533 2-month NAOI 6-month NAOI 2-month Reykjavik 6-month Reykjavik 2-month Hatteras 6-month Hatteras 2-month Hatteras PG 6-month Hatteras PG 2-month Nassau 6-month Nassau 2-month Nassau PG 6-month Nassau PG TOTAL 82 Predicting Paths of Atlantic Tropical Cyclones Using Monthly Surface Pressure Data We tested a total of six relationships for each of the twelve pressure-based variables, and we recorded two-tailed statistical significance values for each test, along with the respective statistical score (Pearson Correlation Coefficient and Wilks’ Lambda) . We computed average statistical significance values for each pressure-based variable in order to determine which measures show the strongest relationships to the various stormbased variables . RESULTS We calculated the median statistical significance values for all tests (bivariate and discriminant analysis) for each of the pressure-based variables for hurricanes, tropical storms, and all tropical cyclones . By finding the median significance of all tests used, an overall usefulness of the pressure pattern for forecasting the paths of tropical systems (north vs . south and Gulf landfall vs . non-Gulf landfall) is discerned (Table 2) . The most interesting results include the much stronger correlation values (i .e ., lower p values) displayed by the 2-month SLP at Nassau . In addition, NAO index values (both two- and six-month) display some of the weakest relationships with the storm-based variables (Tables 3 and 4) . Based on these Table 3 . Results of bivariate correlations between 2-month NAO index values and the six storm-based variables . ALL = all named storms; H = hurricanes only; TS = tropical storms only Dependent Variable R2 p 0.898 0.911 0.573 0.974 0.626 0.340 % gulf landfalls (ALL) <0.001 % gulf landfalls (H) % gulf landfalls (TS) % south (ALL) % south (H) % south (TS) <0.001 0.009 <0.001 0.007 0.027 results, we chose to focus on Nassau SLP data for the remainder of the analysis . The two-month Nassau SLP shows the lowest median significance values, and is the only pressure-based variable that illustrates consistent notable relationships with the storm-based variables . In fact, this pressure variable correlated quite well with the percent of southern-track hurricanes and percent of Gulf-landfall hurricanes with significance values of p = 0 .076 and 0 .044, respectively (Table 5) . These linear correlation analyses suggest that between 11 .6% and 14 .7% of the hurricane track variance can be explained by using the 2-month Nassau SLP . With the Table 4 . Results of bivariate correlations between 6-month NAO index values and the six storm-based variables . ALL = all named storms; H = hurricanes only; TS = tropical storms only Dependent Variable % gulf landfalls (ALL) % gulf landfalls (H) % gulf landfalls (TS) % south (ALL) % south (H) % south (TS) R2 0.002 0.027 0.001 0.035 0.026 0.008 p 0.789 0.338 0.872 0.271 0.349 0.603 Table 5 . Results of bivariate correlations between 2-month Nassau SLP and the six storm-based variables . ALL = all named storms; H = hurricanes only; TS = tropical storms only Dependent Variable % gulf landfalls (ALL) % gulf landfalls (H) % gulf landfalls (TS) % south (ALL) % south (H) % south (TS) R2 0.001 0.116 0.075 0.014 0.147 0.022 p 0.899 0.076 0.159 0.548 0.044 0.455 83 P. Grady Dixon et al. Table 6 . Results of bivariate correlations between 6-month Nassau SLP and the six storm-based variables . ALL = all named storms; H = hurricanes only; TS = tropical storms only Dependent Variable % gulf landfalls (ALL) % gulf landfalls (H) % gulf landfalls (TS) % south (ALL) % south (H) % south (TS) R2 <0.001 0.005 0.019 0.019 0.123 0.006 p 0.949 0.703 0.477 0.471 0.062 0.685 exception of the percent of Gulf-landfall hurricanes (Table 6), the six-month Nassau pressure variable did not perform as well as the two-month index . This is likely due to the temporal proximity of the 2-month mean (May–June) to the summer-season strengthening of the Bermuda-Azores high . While tropical storms do not appear to correlate with Nassau SLP data as well as the hurricanes do, it should be pointed out that the median significance value for the relationship between the two-month Nassau SLP and tropical storms was among the lowest values (i .e ., strongest relationship) . We made use of discriminant analyses to classify cases into groups, based on relationships, and then to test whether cases actually occur as predicted . Similar to the bivariate correlation, the discriminant analyses revealed the potential usefulness of the 2-month Nassau SLP for predicting the percent of southern-track hurricanes and percent of Gulf-landfall hurricanes, with both yielding statistically significant relationships (p = 0 .019 and 0 .013, respectively)(Table 7) . Furthermore, the three highest prediction success rates are associated with the 2-month Nassau SLP (Table 7) . All other variables that yielded statistically significant relationships, based on discriminant analyses, are listed in Table 7, but the 2-month Nassau SLP still stands out as the most reliable predictor . However, even this variable seems to be reliable only Table 7 . Results of statistically significant (α = 0 .10) discriminant analyses . Dependent variables are categorized as “above average” or “below average .” Success rates explain the percentage of years that were accurately identified as “above average” or “below average” based on the independent variable . Binary Dependent Variable % gulf landfalls (H) % south (H) % gulf landfalls (ALL) % gulf landfalls (H) 2-month Nassau 2-month Nassau 6-month Nassau PG 6-month Hatteras PG 2-month Iceland 6-month Iceland 6-month Iceland 84 Independent Variable 2-month Nassau 2-month Nassau 6-month Nassau 6-month Nassau % gulf landfalls (H) % south (H) % south (TS) % south (TS) % gulf landfalls (ALL) % south (H) % south (TS) Wilks’ Lambda 0.805 0.889 0.890 0.889 0.808 0.786 0.861 0.861 0.918 0.912 0.870 p 0.019 0.083 0.079 0.077 0.020 0.013 0.047 0.033 0.090 0.079 0.031 Success Rate (%) 78.6 57.1 62.1 48.3 75.0 67.9 55.2 63.6 63.9 63.9 66.7 Predicting Paths of Atlantic Tropical Cyclones Using Monthly Surface Pressure Data for the prediction of hurricane tracks, but not tropical storm tracks . In a broad sense, these results suggest that the variation of the 2-month Nassau SLP is helpful in determining the preferred track of tropical systems, especially hurricanes, for a given year . CONCLUSIONS Results of this research show that mean SLP for the months of May and June act as better predictors of storm tracks during the following hurricane season than do any of the other pressure-based variables (NAO index, Iceland SLP, Hatteras SLP) . However, it should be noted that none of the researched variables are able to account for large portions of the variance in storm tracks . Nevertheless, the 2-month Nassau SLP was able to distinguish between years with above average and below average percentages of landfalls in the Gulf of Mexico nearly 80% of the time . The results of this study show that higher SLP values in Nassau during May and June should typically lead to a larger percentage of hurricanes approaching the southeastern United States . This is supported by the fact that the number of “south” tracks increases with increasing SLP in Nassau . Conversely, the study also shows that there is a negative correlation between Nassau SLP and hurricanes that make landfall in the Gulf of Mexico . In other words, higher SLP values in Nassau during May and June usually result in a smaller percentage of hurricanes making landfall in the Gulf . Perhaps storms affected by very strong subtropical high pressure tend to remain south until they are just off the east coast of the United States before traveling northward around the western edge of the high pressure . Weaker high-pressure systems likely allow storms to drift northward while still well out in the Atlantic Ocean, or conversely, the pressure systems may not be strong enough to alter storm paths . Higher pressure at Nassau tends to cause west-moving storms to stay farther to the south in the vicinity of 75ºW longitude, but it also tends to cause storms to move northward earlier than during lower pressure conditions (Fig . 2) . Again, this appears to be due to storms following the circulation along the northwest edge of the high-pressure cell . More research is warranted in this area in order to gain a better understanding of how to use pressure data from the western Atlantic Ocean for predicting tracks of tropical systems . This study has shown that SLP at Nassau provides some utility in this process, and future research will likely combine these results with variables that can be used to predict the annual number and intensity of tropical systems . REFERENCES Barnston, A . G ., and R . E . Livezey . 1987 . Classification, Seasonality and Persistence of Low-Frequency Atmospheric Circulation Patterns . Monthly Weather Review 115 (6): 1083-1126 . Bell, G ., E . Blake, C . Landsea, M . Chelliah, R . Pasch, K . Mo, and S . Goldenberg . 2007 . The 2006 North Atlantic Hurricane Season: A Climate Perspective, 10: Climate Prediction Center . Elsner, J . B . 2003 . Tracking Hurricanes . Bulletin of the American Meteorological Society 84: 353-356 . Elsner, J . B ., and T . H . Jagger . 2004 . A Heirarchical Bayesian Approach to Seasonal Hurricane Modeling . Journal of Climate 17: 2813-2827 . Elsner, J . B ., K .-b . Liu, and B . Kocher . 2000 . Spatial Variations in Major U .S . Hurricane Activity: Statistics and a Physical Mechanism . Journal of Climate 13: 2293-2305 . Elsner, J . B ., X . Nu, and T . H . Jagger . 2004 . Detecting Shifts in Hurricane Rates Using a Markov Chain Monte Carlo Approach . Journal of Climate 17: 2652-2666 . Goldenberg, S . B ., C . W . Landsea, A . M . Mestas-Nuñez, and W . M . Gray . 2001 . The Recent Increase in Atlantic Hurricane Activity: Causes and Implications . Science 293: 474-479 . Hurrell, J . W . 1995 . Decadal Trends in the North Atlantic Oscillation: Regional 85 P. Grady Dixon et al. Temperatures and Precipitation . Science 269(5224): 676-679 . Hurrell, J . W ., and H . V . Loon . 1997 . Decadal Variations in Climate Associated with the North Atlantic Oscillation . Climatic Change 36: 301-326 . Jones, P . D ., T . Jonsson, and D . Wheeler . 1997 . Extension to the North Atlantic Oscillation Using Early Instrumental Pressure Observations from Gibraltar and South-west Iceland . International Journal of Climatology 17: 1433-1450 . Klotzbach, P . J ., and W . M . Gray . 2004 . Updated 6-11-Month Prediction of Atlantic Basin Seasonal Hurricane Activity . Weather and Forecasting 19: 917-934 . Lamb, P . J ., and R . A . Peppler . 1987 . North Atlantic Oscillation: Concept and an Application . Bulletin of the American Meteorological Society 68: 1218-1225 . Rogers, J . C . 1984 . The Association between the North Atlantic Oscillation and the Southern Oscillation in the Northern Hemisphere . Monthly Weather Review 112 (10): 1999-2015 . Walker, G . T . 1924 . Correlations in Seasonal Variations of Weather, IX . In Memoirs of the Indian Meteorology Department, 275332 . Walker, S . G . T ., and E . W . Bliss . 1932 . World Weather V . Memoirs of the Royal Meteorological Society 4(36): 53-84 . Wallace, J . M ., and D . S . Gutzler . 1981 . Teleconnections in the Geopotential Height Field during the Northern Hemisphere Winter . Monthly Weather Review 109(4): 784-812 . Xie, L ., T . Yan, L . J . Pietrafesa, J . M . Morrison, and T . Karl . 2005 . Climatology and Interannual Variability of North Atlantic Hurricane Tracks . Journal of Climate 18: 5370-5381 . 86