Information obtained through July 2004 indicates that the 2004 Atlantic hurricane season will have above-average activity. We estimate that 2004 will have about 7 hurricanes (average is 5.9), 13 named storms (average is 9.6), 55 named storm days (average is 49), 30 hurricane days (average is 24.5), 3 intense (category 3-4-5) hurricanes (average is 2.3), 6 intense hurricane days (average is 5.0) and a Hurricane Destruction Potential (HDP) of 85 (average is 71). We expect Atlantic basin Net Tropical Cyclone (NTC) activity in 2004 to be about 125 percent of the long-term average. The probability of U.S. major hurricane landfall is estimated to be about 130 percent of the long-period average. This early August forecast is based on a newly-devised extended range statistical forecast procedure which utilizes 52 years of past global reanalysis data. Analog predictors are also utilized. The influence of El Niño conditions in our hurricane forecast are implicit in our predictor fields, and therefore we do not utilize a specific ENSO forecast. As of early August, it appears that warm neutral or weak warm ENSO conditions are likely for the remainder of the Atlantic hurricane season. These conditions are expected to bring about a small diminution of Atlantic hurricane activity from that predicted by our early April and late May forecasts.

1  Introduction

      This is the 21st year in which the first author has made forecasts of the coming season's Atlantic basin hurricane activity. Our Colorado State University research project has shown that a sizable portion of the year-to-year variability of Atlantic tropical cyclone (TC) activity can be hindcast with skill significantly exceeding climatology. These forecasts are based on a statistical methodology derived from 52 years of past global reanalysis data and a separate study of prior analog years which have had similar global atmosphere and ocean precursor circulation features to this year. Qualitative adjustments are added to accommodate additional processes which may not be explicitly represented by our statistical analyses. These evolving forecast techniques are based on a variety of climate-related global and regional predictors previously shown to be related to the forthcoming seasonal Atlantic tropical cyclone activity and landfall probability.

2  Forecast Methodology

       We believe that seasonal forecasts must be based on methods showing significant hindcast skill in application to long periods of prior years. Most atmosphere-ocean circulation and energy exchange processes are too complicated to allow for skillful deterministic initial value seasonal and yearly prediction. It is only through hindcast studies that one can demonstrate that seasonal forecast skill is possible. This is a valid methodology provided the atmosphere continues to behave in the future as it has in the past. Unlike initial value deterministic prediction, it is not necessary to fully understand all relevant atmospheric and oceanic processes to issue a skillful statistical forecast. One can use prior empirical associations without understanding all of the physical linkages and processes which are involved.

The last few years have seen tremendous growth in the accessibility of global atmospheric data on the Internet. An example of this accessibility is the NOAA/NCEP reanalysis which archives historical atmospheric and ocean surface data and makes this data easily available. Other cooperative research groups are developing similar reanalysis products. Many of these reanalysis data sets are available from the late 1940s and offer exciting and unique opportunities for the development of new and skillful extended range empirical climate forecasts.

There is a great curiosity as to how active the coming hurricane season will be. Any forecast with reasonable physical linkages which has demonstrated hindcast skill above that of climatology is of interest to most residents of the southeast U.S. and the Caribbean. It is important that the public be informed as to what the current atmospheric and oceanic conditions suggest may occur in the coming hurricane season.

3  Earlier 1 August Statistical Hurricane Forecast Schemes

       Our original early August seasonal hurricane forecast scheme was developed in the early 1990s (Gray et al. 1993) and demonstrated significant hindcast skill for the period of 1950-1991 (Gray et al. 1994). This scheme included measurements of West African rainfall as an important forecast input. A revised 1 August forecast scheme was developed in the mid-1990s. Both schemes did not have the advantage of the NOAA reanalysis products of the last 3-4 years.

Since the observed shift of Atlantic Ocean SST patterns in 1995 [and implied increase in the strength of the Atlantic Thermohaline Circulation (THC)], our earlier 1 August forecast schemes have consistently underpredicted Atlantic basin hurricane activity. The previously observed (1950-1994) strong association between West African rainfall and Atlantic hurricanes has not been reliable since 1994.

3.1  New 1 August Seasonal Forecast Scheme

       Most reanalysis data sets are available from the late 1940s and offer exciting and unique opportunities for the development of new and more skillful extended range empirical climate forecasts. For example:

1. Our new scheme has been developed on 11 more years of hindcast data (1950-2001).

2. Our new scheme has been able to use the recently developed NOAA/NCEP reanalysis data that were not available to us at the time we developed our earlier scheme. The reanalysis has allowed us to more readily search for new forecast parameters.

Through extensive analyses of the recently available NOAA/NCEP reanalysis products, Phil Klotzbach of our forecast team has developed a new set of 1 August extended range predictors which shows superior hindcast skill over our previous 1 August forecast schemes. This new 1 August forecast scheme does not use West African rain as a predictor. No significant improvement in hindcast skill was achieved by adding June-July fields to the 1 June forecast, and therefore, the 1 August seasonal forecast uses the same predictors as the 1 June forecast.

The pool of seven predictors for this new extended range forecast is given and defined in Table 1. The location of each of these new predictors is shown in Fig. 1. Strong statistical relationships can be extracted via combinations of these predictive parameters (which are available by the end of May), and quite skillful Atlantic basin hurricane forecasts for the following summer and fall can be made if the atmosphere and ocean continue to behave in the future as they have during the hindcast period of 1950-2001. Full documentation of the skill of this forecast can be found in our 28 May 2004 seasonal forecast.

Table 2 shows our statistical forecast for the 2004 hurricane season and the comparison of this forecast with climatology (average season between 1950-2000). All our forecast parameters are expected to be above average. Brief descriptions of our new early August seasonal predictors are contained in our 28 May 2004 seasonal forecast, which is listed on our Colorado State University hurricane forecast web site.

4  Recently Developed Predictions of Individual Monthly Atlantic TC Activity for August, September, and October

       A new aspect of our climate research is the development of TC activity predictions for individual months. There are often monthly periods within active and inactive Atlantic basin hurricane seasons which do not conform to the overall season. For example, 1961 was an active hurricane season (NTC of 222), but there was no TC activity during August; 1995 had 19 named storms, but only one named storm developed during a 30-day period during the peak of the hurricane season between 29 August and 27 September. By contrast, the inactive season of 1941 had only six named storms (average 9.3), but four of them developed during September. During the inactive 1968 hurricane season, three of the eight named storms formed in June (June average is 0.5).

We have started new research to see how well various sub-season or individual monthly trends of TC activity can be forecast. This effort has recently been documented in papers by Blake and Gray (2004) for August and Klotzbach and Gray (2003) for September. These reports show that it is possible to develop a skillful prediction scheme for August-only and September-only Atlantic basin tropical cyclone activity. We are issuing our second October-only forecast this year. On average, August, September, and October have about 26%, 48%, and 17% of total NTC, respectively. Initial August-only forecasts have now been made by Blake for 2000-2003, and their verification looks promising.

It has been generally thought that it is more difficult to predict hurricane activity during shorter periods than to predict activity for the entire season. Despite the presumed inherent difficulties with these shorter period forecasts, Blake and Klotzbach have devised quite skillful August-only, September-only, and October-only prediction schemes based on 51 years (1950-2000) of hindcast testing using a statistically independent jackknife approach. Predictors are largely derived from June and July NCEP global reanalysis data but also include information from earlier in the year.

4.1  Independent August-Only Statistical Forecast

       Figure 2 and Table 3 list the predictors used in the August-only hindcast (Blake and Gray 2004) for each of the nine different forecast parameters. The table also shows hindcast skill for the 51-year period 1950-2000, as well as the independent jackknife hindcast skill over this period. Table 4 gives the predictor values for August 2004. Table 5 gives our independent statistical prediction for August 2004. These predictors indicate slightly below-average activity for August 2004. However, due to the considerable amount of tropical cyclone activity that has already occurred during this August, we are forecasting a slightly above-average August.

4.2  Independent September-Only Statistical Forecast

       Figure 3 and Table 6 portray and list our 1 August predictors for September-only activity for this year. The number of predictors used and individual parameter hindcast skill are given in Table 7. Table 8 gives the predictor values for September 2004. Table 9 gives our independent September statistical forecast and our adjusted final forecast.

Data available through the end of July indicates that September 2004 will have about average activity. The most significant predictor, 1000 mb trade winds in the Atlantic (5-15^°N, 30-50^°W), is running slightly below normal. The other six predictors are mixed: three call for above-average activity while three call for below-average activity. We believe that September will have slightly more activity than forecast by our statistical forecast based on the continuation of positive sea surface temperature anomalies in the tropical Atlantic and the continuation of an active thermohaline circulation. An updated September-only statistical forecast will be issued on 3 September. This early September forecast will have the advantage of August data.

4.3  Independent October-only Statistical Forecast

       Through examination of the NCEP/NCAR reanalysis, we have discovered four predictors that in combination explain about 50 percent of the October cross-validated variance in Net Tropical Cyclone activity for the hindcast period of 1950-2001. We are currently unable to find combinations of predictors that explain large amounts of variance for the individual tropical cyclone parameters (i.e., named storms, hurricane days, etc.). Therefore, our October forecast consists of predicting NTC and consequently increasing or decreasing October's values for the other parameters accordingly. For example, if October NTC was 150 percent of normal and a typical October had two named storms, we would forecast three named storms for October. The predictors utilized in our initial October prediction are displayed graphically in Figure 4, and their 2004 values are displayed in Table 10. All four predictors are negative for storms. Therefore, we are calling for a below-average October with an NTC of about 50 percent of the climatological average. In round numbers, we are forecasting 2 named storms, 1 hurricane, 0 intense hurricanes and an NTC of 10 for October. This low forecast is primarily a consequence of the moderately positive sea surface temperature anomalies in the Nino 4 and Nino 3.4 regions. Warmth in these regions has been shown to reduce late-season tropical cyclone activity in the Atlantic. Table 11 displays our initial statistical forecast and our adjusted forecast for October tropical cyclone activity. Additional updates for the October-only forecast will be issued in early September and in early October. Physical links between the predictors and October tropical cyclone activity follow:

4.4  Physical Association of October Forecast Parameters

1. June-July Sea Level Pressure (10-25^°N, 10-40^°W) (-)

   Low sea level pressure in June-July in this part of the subtropical Atlantic is the most important predictor for October tropical cyclone activity. Low pressure indicates that a weak subtropical ridge is present, trade winds are weaker, and consequently, due to an evaporation decrease, the tropical Atlantic is warmer than normal. On a climatological average, tropospheric vertical wind shear and sea level pressure are directly related. Lower than normal sea level pressure indicates that late-season tropical cyclones are more likely to occur due to a combination of reduced wind shear and a warm tropical Atlantic.

2. July 200 mb Geopotential Height (20-35^°N, 5-45^°W) (+)

   High heights in the northern subtropical Atlantic indicate that there is an increased height gradient between the tropical and subtropical Atlantic which decreases the area and strength of upper-level westerly winds. Stronger easterlies tend to persist throughout the remainder of the hurricane season thereby reducing vertical wind shear and providing more favorable conditions for October tropical cyclone development.

3. July 200 mb U (35-47.5^°S, 160^°E-160^°W) (+)

   Increased upper-level westerlies near New Zealand indicate increased Southern Hemisphere winter baroclinicity which is typically associated with favorable conditions for tropical cyclones in the Atlantic. These conditions tend to persist through October, increasing the likelihood of late-season tropical cyclones.

4. Previous November SLP (45-65^°N, 115-145^°W) (-)

   Low sea level pressure in this area during November of the previous year implies a deeper and eastward-shifted Aleutian Low which is typical of a positive Pacific North American Pattern (PNA). A positive PNA is frequently associated with the final year of warm ENSO conditions (Horel and Wallace 1981) and therefore, a return to cooler conditions in the eastern tropical Pacific during the following year. Cool ENSO conditions provide a favorable environment for the development of October tropical cyclones.

4.5  Monthly Prediction Summary

       Tables 12 summarizes our individual monthly predictions and our monthly adjustments to these predictions. Based on jackknifed hindcast data from 1950-2000, the sum of the August, September, and October forecasts explains 79% of the variance in seasonal TC activity. Note that the sum of these three monthly predictions gives lower values of tropical cyclone activity than our full seasonal statistical forecast.

4.6  Seasonal Analogs for 2004

       Table 13 lists our best seasonal analogs for 2004. We selected years with warm anomalies in the tropical and north Atlantic Ocean. In addition, the years selected had warm anomalies in the central Pacific and cool anomalies in the east Pacific. Our seasonal analogs continue to point toward an active season.

5  Forecast Adjustments

       Table 14 provides a comparison of all of our forecast techniques along with the final full season adjusted forecast. Given the current (July) global conditions and other information we have, we anticipate somewhat greater activity than indicated by the sum of our individual monthly predictions. Although we give great credence to the sum of our individual monthly predictors and consider this our most skillful forecast, we feel we cannot totally disregard our prior seasonal forecasts of 6 December, 4 April, and 30 May which indicate a more active season. In addition, our full-season analogs also call for an active year. We are thus making only a slight downward adjustment of our earlier seasonal forecasts.

6  Discussion

       This is a difficult forecast. There are currently mixed climate signals which cause our full-season 2004 forecast to indicate above-average activity, while the sum of our three monthly forecasts indicates below-average activity. The Atlantic Ocean signals look very favorable for an active year; whereas, the Pacific Ocean signals indicate a suppressed season.

June and July changes in a number of climate signals and their influence on our individual monthly forecasts have led us to make a downward revision of our 2 April and 28 May forecasts by one storm number for named storms and hurricanes. We have also reduced our forecasts of the number of days that tropical cyclones will exist in the Atlantic, and we have consequently reduced our NTC value 20 points from 145 to 125. Several of the factors currently observed that lead us to believe that there will be enhanced hurricane activity in 2004 are as follows:

• Warm Atlantic SSTAs

• Below average SLPA in the Tropical Atlantic

• West-phase QBO

• Low equatorial 200 mb Geopotential Heights

• Persistence of favorable conditions associated with the active Atlantic Multidecadal Mode and a strong Atlantic thermohaline circulation

Factors currently observed that lead us to believe that Atlantic hurricane activity in 2004 will be reduced are as follows:

• Enhancement of warm ENSO conditions - especially in the Nino 4 and Nino 3.4 regions

• Stronger-than-average trade winds across the equatorial Atlantic - resulting in generally increased wind shear across the main development region

• Strong 200 mb easterly anomalies in the eastern equatorial Pacific - typically associated with warm ENSO conditions and reduced Atlantic seasonal activity

Consequently, based on these mixed climate signals, we are reducing our forecast somewhat from our early April and late May updates; however, we are not reducing the forecasts as much as suggested by the sum of our monthly forecasts.

7  Landfall Probabilities for 2004

       A significant focus of our recent research involves efforts to develop forecasts of the probability of hurricane landfall along the U.S. coastline. Whereas individual hurricane landfall events cannot be accurately forecast months in advance, the total seasonal probability of landfall can be forecast with statistical skill. With the observation that, statistically, landfall is a function of varying climate conditions, a probability specification has been developed through statistical analyses of all U.S. hurricane and named storm landfall events during the last 100 years (1900-1999). Specific landfall probabilities can be given for all cyclone intensity classes for a set of distinct U.S. coastal regions.

Figure 5 provides a flow diagram showing how these forecasts are made. Net landfall probability is shown linked to the overall Atlantic basin Net Tropical Cyclone activity (NTC; see Table 16) and to climate trends linked to multi-decadal variations of the Atlantic Ocean thermohaline circulation as inferred from recent past years of North Atlantic SSTA* in the region of 50-60^°N, 10-50^°W.

Higher values of SSTA* generally indicate greater Atlantic hurricane activity, especially for intense or major hurricanes. Atlantic basin NTC can be skillfully hindcast, and the strength of the Atlantic Ocean thermohaline circulation can be inferred as SSTA* from North Atlantic SST anomalies in the current and prior years. These relationships are then utilized to make probability estimates for U.S. landfall. The current (July 2004) value of SSTA* is 44. Hence, in combination with a prediction of NTC of 125 for 2004, a combination of NTC + SSTA* of (125 + 44) yields a value of 169.

As shown in Table 15, NTC is a combined measure of the year-to-year mean of six indices of hurricane activity, each expressed as a percentage difference from the long-term average. Whereas many active Atlantic hurricane seasons feature no landfalling hurricanes, some inactive years have experienced one or more landfalling hurricanes. Long-term statistics show that, on average, the more active the overall Atlantic basin hurricane season is, the greater the probability of U.S. hurricane landfall. For example, landfall observations during the last 100 years show that a greater number of intense (Saffir-Simpson category 3-4-5) hurricanes strike the Florida and U.S. East Coast during years of (1) increased NTC and (2) above-average North Atlantic SSTA* conditions.

Table 16 lists strike probabilities for different TC categories for the entire U.S. coastline, the Gulf Coast and Florida, and the East Coast for 2004. The mean annual probability of one or more landfalling systems is given in parentheses. Note that Atlantic basin NTC activity in 2004 is expected to be greater than the long-term average (125), and North Atlantic SSTA* values are measured to be above average (44 units). U.S. hurricane landfall probability is thus expected to be above average owing to both a higher NTC and above-average North Atlantic SSTAs. During periods of positive North Atlantic SSTA*, a higher percentage of Atlantic basin major hurricanes cross the Florida and eastern U.S. coastline for a given level of NTC.

8  United States Landfalling Hurricane Webpage Application

       Over the past four years, we have been compiling and synthesizing our landfalling hurricane data and have been developing a webpage application with extensive landfalling probabilities for the Gulf and East Coasts of the United States. In partnership with the GeoGraphics Laboratory at Bridgewater State College, a web application has been created that displays landfall probabilities for eleven regions, 55 subregions and 205 individual counties of the United States coast extending from Brownsville, Texas to Eastport, Maine. Individual probabilities of sustained winds of tropical storm force (40-75 mph), hurricane force ( > 75 mph) and intense or major hurricane force ( > 115 mph) are also given. These probabilities are based on the current forecast of NTC activity and on current values of SSTA*.

Figures 6 and 7 display example screens of data that is available on this website. The user can select tracks of all intense hurricanes that have made landfall in a given area over the last 100 years. This webpage is currently available at http://www.e-transit.org/hurricane . One can also reach this webpage from a link off the CSU Tropical Meteorology Project homepage: http://tropical.atmos.colostate.edu.

9  Increased Level of Atlantic Basin Hurricane Activity During Seven of the Last Nine Years - But Decreased U.S. Major Hurricane Landfall

       A major reconfiguration of the distribution of Atlantic SST anomalies began in mid-1995 and has largely persisted through the present. North Atlantic SSTs have become about 0.4 to 0.6^°C warmer than normal since 1995 and tropical Atlantic August-October upper tropospheric 200 mb winds have increased from the east, bringing about a significant decrease in tropospheric vertical wind shear. There have been large changes in SSTA and 200 mb zonal wind anomaly (ZWA) during 1995-2003 as compared to conditions from the prior 25-year period of 1970-1994. These changes are well associated with the large increase in major hurricane activity in the Atlantic basin during seven of the last nine years. As noted several times before, we hypothesize that these strong broadscale SST changes are associated with basic changes in the strength of the Atlantic Ocean thermohaline circulation (THC). This interpretation is consistent with changes in a long list of global atmospheric circulation features during the last nine years which conform to a prominent shift into hurricane-enhancing Atlantic circulation patterns, particularly the enhancement of major hurricane activity. Historical and geographic evidence going back thousands of years indicates that shifts in the Atlantic multi-decadal thermohaline circulation tend to occur on periods of 25-50 years. If the recent nine-year shift follows prior occurrences, it is likely that enhanced intense Atlantic basin hurricane activity will persist through the early decades of the 21st century in contrast with the diminished activity that persisted from 1970-1994.

Despite El Niño-linked reductions of Atlantic basin hurricane activity during 1997 and 2002, the last nine years (1995-2003) constitute the most active nine consecutive years on record. Table 17 provides a summary of the total number of named storms (103), named storm days (567), hurricanes (62), hurricane days (288), major hurricanes (29), major hurricane days (75.25) and Net Tropical Cyclone activity (1230) that have occurred during the recent active seven of the last nine-year period of 1995-1996, 1998-2001, and 2003. The seven-year annual average of NS, NSD, H, HD, IH, IHD and NTC during these years has been 143, 206, 151, 254, 261, 362 and 234 percent, respectively above the averages of the prior 25-year period of 1970-1994. These trends toward increased hurricane activity give strong support to the suggestion that we have indeed entered a new era of enhanced major hurricane activity.

10  The 1995-2003 Upswing in Atlantic Hurricanes and Global Warming

       Various groups and individuals have suggested that the recent large upswing in Atlantic hurricane activity (since 1995) may be in some way related to the effects of increased man-made greenhouse gases such as carbon dioxide (CO₂). There is no reasonable scientific way that such an interpretation of this recent upward shift in Atlantic hurricane activity can be made. Please see our 21 November 2003 verification report for more discussion on this subject: http://tropical.atmos.colostate.edu/Forecasts/2003/nov2003

11  Forecast Theory and Cautionary Note

      Our forecasts are based on the premise that those global oceanic and atmospheric conditions which precede comparatively active or inactive hurricane seasons in the past provide meaningful information about similar trends in future seasons. It is important that the reader appreciate that these seasonal forecasts are based on statistical schemes which, owing to their intrinsically probabilistic nature, will fail in some years. Moreover, these forecasts do not specifically predict where within the Atlantic basin these storms will strike. The probability of landfall for any one location along the coast is very low and reflects the fact that, in any one season, most US coastal areas will not feel the effects of a hurricane no matter how active the individual season is. However, it must also be emphasized that a low landfall probability does not insure that hurricanes will not come ashore. Regardless of how active the 2004 hurricane season is, a finite probability always exists that one or more hurricanes may strike along the US coastline or the Caribbean Basin and do much damage.

12  Forthcoming Update Forecasts of 2004 Hurricane Activity

      We will be issuing seasonal updates of our 2004 Atlantic basin hurricane activity forecast on Friday 3 September and Friday 1 October 2004. These updates will include separate forecasts for 2004 September-only activity and October-only Atlantic basin TC activity. A verification and discussion of all 2004 hurricane activity and forecasts will be issued in late November 2004. All of these forecasts will be available at our web address given on the front cover: http://tropical.atmos.colostate.edu/forecasts/index.html.

13  Acknowledgments

       The first author gratefully acknowledges past valuable input to our project research by former graduate students and colleagues Chris Landsea, John Knaff, Eric Blake, Paul Mielke and Ken Berry. A number of other meteorologists have furnished us with data and given valuable assessments of the current state of global atmospheric and oceanic conditions. This includes Arthur Douglas, Richard Larsen, Todd Kimberlain, Ray Zehr and Mark DeMaria. In addition, Barbara Brumit and Amie Hedstrom have provided excellent manuscript, graphical, and data analysis assistance over a number of years. We have profited over the years from many in-depth discussions with most of the current NHC hurricane forecasters. The first author would further like to acknowledge the encouragement he has received for this type of forecasting research application from Neil Frank, Robert Sheets, Robert Burpee, Jerry Jarrell, former directors of the National Hurricane Center (NHC), and from the current director, Max Mayfield. We also thank Bill Bailey of the Insurance Information Institute for his sage advice and encouragement.

The financial backing for the issuing and verification of these forecasts has in part been supported by the National Science Foundation, but this NSF support is insufficient. We appreciate the financial support of the Research Foundation of AIG - Lexington Insurance Company (a member of the American International Group) for the last two years. We are also grateful to the Research Foundations of the United Services Automobile Association (USAA) and to State Farm Insurance for prior support.

14  Citations and Additional Reading

15  Verification of Previous Forecasts

Table 18: Summary verifications of the authors' five previous years of seasonal forecasts of Atlantic TC activity between 1999-2003.