prediction of august atlantic basin hurricane activity

1044 VOLUME 19W E A T H E R A N D F O R E C A S T I N G

q 2004 American Meteorological Society

Prediction of August Atlantic Basin Hurricane Activity

ERIC S. BLAKE AND WILLIAM M. GRAY

Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado

(Manuscript received 20 May 2003, in final form 28 April 2004)

ABSTRACT

Although skillful seasonal hurricane forecasts for the Atlantic basin are now a reality, large gaps remain inour understanding of observed variations in the distribution of activity within the hurricane season. The monthof August roughly spans the first third of the climatologically most active part of the season, but activity duringthe month is quite variable. This paper reports on an initial investigation into forecasting year-to-year variabilityof August tropical cyclone (TC) activity using the National Centers for Environmental Prediction–NationalCenter for Atmospheric Research reanalysis dataset. It is shown that 55%–75% of the variance of August TCactivity can be hindcast using a combination of 4–5 global predictors chosen from a 12-predictor pool witheach of the predictors showing precursor associations with TC activity. The most prominent predictive signalis the equatorial July 200-mb wind off the west coast of South America. When this wind is anomalously strongfrom the northeast during July, Atlantic TC activity in August is almost always enhanced. Other July conditionsassociated with active Augusts include a weak subtropical high in the North Atlantic, an enhanced subtropicalhigh in the northwest Pacific, and low pressure in the Bering Sea region.

The most important application of the August-only forecast is that predicted net tropical cyclone (NTC)activity in August has a significant relationship with the incidence of U.S. August TC landfall events. Betterunderstanding of August-only TC variability will allow for a more complete perspective of total seasonalvariability and, as such, assist in making better seasonal forecasts.

1. Introduction

During the past 20 years, large advances have oc-curred in our understanding of the seasonal variabilityof tropical cyclones (TCs). Seasonal hurricane forecastsfor the Atlantic basin have shown significant skill, es-pecially forecasts issued each year on 1 August (Grayet al. 2001). There is, however, appreciable intraseasonalvariability occurring on month-to-month time scaleswithin most seasons, which has not been successfullyforecast. If these shorter time-scale variations can beanticipated, it will provide useful insight into the natureof this variability and increase the utility of seasonalforecasts.

Intraseasonal variability can be illustrated using a pa-rameter termed net tropical cyclone activity (NTC).NTC is defined as the average of the aggregate seasonalpercentage of six indices of Atlantic hurricane activitywith a value of 100 representing an average season(Gray et al. 1994). Note that NTC is most sensitive tointense hurricane (IH) activity because of the small de-nominator of average intense hurricanes. The active1961 season observed an NTC of 210, more than twice

Corresponding author address: Eric S. Blake, NOAA/NationalWeather Service/Tropical Prediction Center, 11691 SW 17th St., Mi-ami, FL 33165-2149.E-mail: [email protected]

the annual mean. However, no TCs were observed dur-ing August of 1961; a rare occurrence that has happenedonly twice since 1944. A contrasting situation was notedduring 1976 wherein August activity was approximatelytwice the normal incidence; however, the season endedwith total seasonal activity that was only about 86% ofaverage. Figures 1 and 2 show a larger sample of AugustNTC variation.

a. Previous work

Prior to Gray (1984b), extended-range forecasting ofTC activity was limited and focused on issues relatedto potential predictability rather than actual attemptedyearly forecasts. Ballenzweig (1959) distinguished ac-tive versus inactive TC months by studying compositesof large-scale atmospheric fields. His report compiledand differenced active from inactive months during Au-gust–October and attempted to link the variable activityduring these periods to circulation anomalies across theNorthern Hemisphere (NH). He found that months ofmaximum TC activity were associated with a north-eastward shift of the Atlantic anticyclone, thereby ex-panding and weakening the area of easterlies equator-ward of the ridge. Whereas Ballenzweig’s study wasdiagnostic rather than prognostic, it introduced and il-lustrated the notion of observable mean differences be-

DECEMBER 2004 1045B L A K E A N D G R A Y

FIG. 1. Additional examples of Aug NTC variability (shown shadedon the right) during active seasons, which are shown on the left.Average (climatology) activity for both seasonal and Aug-only pe-riods are indicated by the dotted lines.

FIG. 2. As in Fig. 1 but for Aug NTC variability during inactiveseasons.

tween active and inactive periods occurring on month-long time scales.

Another study dealing with short-term differences ofTC activity was performed by Shapiro (1987) who test-ed the predictability of monthly Atlantic TC frequencyusing monthly mean wind and sea level pressure anom-aly (SLPA) data poleward of 208N. These parameterswere calculated over the Atlantic basin 2 months inadvance of the August–October active part of the hur-ricane season and examined for predictive associations.Shapiro found statistically significant correlations in TCfrequency encompassing approximately 45% of thehindcast variance. His results suggested that the phaseof the El Nino–Southern Oscillation (ENSO) was thestrongest modulator of monthly Atlantic TC activitythough he estimated that only one-sixth of his skill wasdirectly related to ENSO.

Recent findings by Maloney and Hartmann (2000)indicate that hurricane activity in the Gulf of Mexico(GOM) and western Caribbean Sea is significantly mod-ulated by the passage of the Madden–Julian oscillation(MJO). The MJO propagates eastward across the Trop-ics as a wavenumber 1 oscillation with a period of about30–45 days. The authors noted a fourfold increase ofwestern Caribbean and GOM genesis events when MJO-linked 850-mb u-wind anomalies were westerly acrossthe eastern Pacific south of Mexico. Maloney and Hart-mann state that more accurate predictions of week-to-week genesis events may be possible with forecasts ofthe MJO.

DeMaria et al. (2001) were likely the first to utilizedaily data for real-time forecasting of intraseasonal var-iations in Atlantic TC activity. They created a ‘‘genesisparameter’’ derived from the 5-day mean of verticalwind shear, midlevel moisture, and vertical stability forthe tropical Atlantic east of the Lesser Antilles. Thisgenesis parameter explained about 50% of the varianceof TC activity that formed in this area during 1995–99.

The authors suggest that monitoring the genesis param-eter on a daily basis provides an improved measure ofthe probability of a tropical wave developing into a TC.

b. Objective and outline

A forecast of August-only Atlantic basin TC activityis the focus of the remainder of this paper. A test ofGray and colleagues’ seasonal forecast scheme ex-plained only about 20% of the August NTC hindcastvariance, and therefore new forecast techniques neededto be developed for shorter-period predictions. Tropicalcyclone data and August climatology are the subjectsof section 2, and the forecast methodology along withthe global reanalysis data used are detailed in section3. Section 4 details the physical relationships betweenpredictors and TC activity, and section 5 examines hind-casting results of the August-only forecast scheme. Ap-plications of the August-only forecast including a fore-cast of U.S. landfalling TCs are the topic of section 6.Section 7 details the conclusions and ideas for futurework.

2. Tropical cyclone data and climatology ofAugust activity

a. Tropical cyclone data

August TC information was calculated from the ‘‘besttrack’’ database maintained by the National HurricaneCenter in Miami, Florida. All available track and in-tensity estimates were used to prepare a smoothed post-analysis composed of the most accurate storm infor-mation (Jarvinen et al. 1984; Neumann et al. 1999).Tropical cyclones whose life cycle spanned any portionof August were tallied and stratified by their maximumintensity during August. The basic activity parameterscalculated were the number of days during August thata TC existed as a named storm (NS), hurricane (H), andintense hurricane (IH; Saffir–Simpson category 3 orgreater). This method thus takes into account TCs that


TABLE 1. Comparison of year-by-year and long-term mean seasonalNTC vs Aug NTC. The column on the right shows the percent ofseasonal NTC that is observed in Aug. The maximum value in eachcolumn is shown in boldface while minima are in italic.

YearSeasonal

NTCAugNTC

Percent ofseasonal

NTC

19491950195119521953195419551956195719581959

115243121

97121127198

6986

14099

32.371.631.712.113.512.189.724.72.4

66.32.3

28.129.526.212.411.29.5

45.335.82.8

47.32.3

1960196119621963196419651966196719681969

101222

33116168

86140

9741

157

9.40.09.5

30.837.614.924.42.38.0

60.7

9.30.0

28.826.622.417.317.42.4

19.538.6

1970197119721973

65952852

22.118.010.89.3

33.918.938.718.0

197419751976197719781979

769285468696

29.322.451.46.1

16.236.3

38.524.460.413.418.837.8

1980198119821983198419851986198719881989

135114

373277

1103848

121140

60.010.62.9

21.66.5

21.87.2

15.66.5

31.1

44.59.37.9

67.68.4

19.819.032.45.4

22.21990199119921993199419951996199719981999

10459625537

237199

54170193

45.516.130.227.610.776.464.60.0

51.061.0

43.827.348.750.228.832.332.50.0

30.031.6

Avg 104.3 26.4 25.4

formed in July and persisted into August but consideredonly the portion of activity that occurred during August.

August TCs were also classified on the basis of theirorigins as suggested by Hess et al. (1995), which showedimprovements in seasonal forecasting skill obtained byseparately forecasting TC formations based on origin-linked classes. One such origin-linked class is termedtropical only (TO). The TO systems are defined as thosehurricanes that formed without any obvious midlatitudeinfluences, typically from African easterly waves equa-torward of 23.58N. An alternative class was termed bar-oclinically initiated (BI) hurricanes wherein nontropicaldisturbances were involved. The latter class may includestorms that develop along a stationary front, from adecaying midlatitude low, or from a mesoscale convec-tive system (MCS) emerging from North America (Els-ner and Kara 1999). Categorizing TCs at the time theyfirst reached tropical storm intensity as either TO or BIposed an additional problem since Hess et al. (1995)considered only hurricanes. It was decided that the keycriterion for classifying an NS as a TONS was that itdeveloped from a tropical wave without any contribu-tion from midlatitude influences. Note however that aTONS that subsequently became a hurricane due to non-tropical enhancement effects remained a TONS but wasnot considered a TOH. Regression techniques for bothTO and BI formations were tested in this study thoughno BI forecasting technique proved successful.

b. August climatology

This section presents a summary of the main cli-matological qualities of August TC variability. Augustis the secondmost active month on average, encom-passing about 26% of the total seasonal NTC. The lattervalue has ranged from a high of 67.6% (1983) to a lowof 0% (1961, 1997) as shown in Table 1. On average,August has approximately three TCs, two of which be-come hurricanes (H), and one of which becomes anintense hurricane. Table 2 presents a summary of sta-tistics for August activity. Note that 1955 and 1995 werethe two most active Augusts while 1961 and 1997 hadno August TC activity.

Table 3 contains the linear cross correlations betweenall August TC parameters. Though most of the indicesin the table are closely associated, the NS and IH valuesare not well correlated. A large number of NSs in Augustis only weakly indicative that an above average numberof IHs will form. An example is August 1995 wheneight NS occurred, but only one of these reached IHstatus. This dichotomy suggests that conditions favor-able for TC formation are not necessarily the same asconditions conducive for strengthening into powerfulstorms. It is not surprising that different sets of predic-tors are chosen to forecast the various TC indices, asdetailed in section 5.

c. Multidecadal variations

Gray (1990) and Gray et al. (1997b) proposed thatthe recent 25-yr (1970–94) downturn of TC activity,especially IHs, was primarily due to variations in thebroadscale Atlantic thermohaline circulation for whichNorth Atlantic (508–608N, 108–508W) SSTs are used as


TABLE 2. Summary of Aug Atlantic tropical cyclone data for the years 1949–99. The columns (from left to right) indicate the Aug totalsof named storms (NS), named storm days (NSD), hurricanes (H), hurricane days (HD), intense hurricanes (IH), intense hurricane days (IHD),net tropical cyclone activity (NTC), tropical-only named storms (TONS), and tropical-only hurricanes (TOH). Boldface type indicates themaximum value in each column for the period.

Year NS NSD H HD IH IHD NTC TONS TOH

19491950195119521953195419551956195719581959

34323251141

12.5031.0013.008.008.506.50

36.759.251.75

20.501.50

24111131030

6.7518.00

7.254.003.254.75

19.758.000.00

12.250.00

12100021020

1.254.001.750.000.000.008.501.000.005.500.00

32.371.631.712.113.512.189.724.72.4

66.32.3

34321251041

24110131030

1960196119621963196419651966196719681969

2022421116

5.000.005.75

12.5013.757.25

10.001.504.00

21.25

1012121013

1.750.001.509.756.504.258.500.002.75

12.75

0001101002

0.000.000.000.753.000.000.750.000.002.75

9.40.09.5

30.837.614.924.42.38.0

60.7

1002211104

0002101002

19701971197219731974

34223

8.258.005.506.757.50

12112

2.253.253.500.753.75

10001

0.500.000.000.001.50

22.118.010.89.3

29.3

22022

10011

19751976197719781979

26143

5.2524.751.756.508.50

24121

2.7514.50

1.251.504.75

11001

0.500.750.000.004.00

22.451.46.1

16.236.3

12013

11011

1980198119821983198419851986198719881989

3212331334

17.7510.753.256.503.25

11.755.25

17.253.25

21.00

3102031103

13.000.500.002.250.005.251.001.750.00

11.25

1001000000

6.500.000.000.250.000.000.000.000.000.00

60.010.62.9

21.66.5

21.87.2

15.66.5

31.1

2111220114

2001020002

1990199119921993199419951996199719981999

7114284054

22.754.009.75

17.008.00

46.0021.750.00

20.5026.75

3111153023

7.502.254.254.502.00

25.2511.25

0.0013.0014.25

1111011012

1.000.253.250.250.001.757.250.003.503.00

45.516.130.227.610.776.464.60.0

51.061.0

5013184044

2000153023

Avg 2.8 11.56 1.6 5.67 0.6 1.25 26.4 1.9 1.0

a proxy. Multidecadal variations of this circulation andassociated ocean SST patterns tend to occur on periodsof 30 yr or more. Broadscale North Atlantic SSTs werein a relatively ‘‘cool’’ phase from the early 1970s to theearly 1990s, which closely coincides with observed re-duced TC activity. However, North Atlantic SSTs havewarmed dramatically since mid-1995 with a return to

an SST configuration and level of TC activity moreclosely resembling that of the 1950s and 1960s.

Multidecadal signals in the August monthly data areconsidered by compositing aggregate activity during the1950–69 and 1995–99 time periods (both of these pe-riods judged to be during an active Atlantic Ocean ther-mohaline circulation) versus the inactive 1970–94 pe-


TABLE 3. Cross-correlation matrix showing associations between nine Aug TC indices for the 51 yr from 1949 to 1999.

NS NSD H HD IH IHD NTC TONS TOH

NSNSDHHDIHIHDNTCTONSTOH

—0.830.770.720.500.410.760.850.68

0.83—

0.850.930.650.610.910.880.85

0.770.85—

0.850.590.500.830.740.84

0.720.930.85—

0.710.670.930.820.88

0.500.650.590.71—

0.730.850.600.67

0.410.610.500.670.73—

0.830.560.63

0.760.910.830.930.850.83—

0.830.86

0.850.860.740.820.600.560.83—

0.86

0.680.850.840.880.670.630.860.86—

Avg 0.69 0.81 0.75 0.81 0.66 0.62 0.85 0.77 0.78

TABLE 4. Comparison of Aug activity during 25 ‘‘active thermohaline’’ years (1950–69, 1995–99) vs 25 ‘‘inactive thermohaline’’ years(1970–94). Thermohaline conditions are inferred from North Atlantic SSTA. Note the abundance of IHD and lack of significant BI activityduring the active years.

Avg NS NSD H HD IH IHD NTC TONS TOH BINS BIH

1950–69, 1995–991970–94Ratio active/inactive

2.722.80.97

13.319.771.36

1.61.521.05

7.553.752.01

0.720.441.64

1.740.752.32

31.0821.431.45

2.21.61.38

1.320.681.94

0.200.880.23

0.040.520.08

riod. Table 4 shows the difference between the two 25-yr periods expressed both as comparative averages andas ratios. The most pronounced changes are for the mostintense type of activity, which was much greater in theactive thermohaline period. Whereas August TO activitywas notably suppressed during the inactive thermoha-line years, the incidence of higher-latitude BI systemsincreased. Thirteen August BI hurricanes formed duringthe inactive era, while only one developed during theactive era.

The development of the August-only TC climatologyallowed the formulation of indices describing measuresof August TC activity. Some of these August indicesproved useful as precursor signs for the remainder ofthe season’s activity.

3. Global data and forecast methodology

a. Global data

Analyses performed in this study were based on themonthly National Centers for Environmental Predic-tion–National Center for Atmospheric Research(NCEP–NCAR) global reanalysis (Kalnay et al. 1996)data for 1949–99. The reanalysis is composed of nu-merous atmospheric and SST fields integrated onto a2.58 3 2.58 global grid. The analysis technique gives arealistic, consistent interpretation of large-scale weatherfeatures back to 1948 using standardized assimilationand interpolation methods. This unique dataset allowsfor comparison of recent climate data versus data fromearlier years in an environment free of artificial ‘‘climateshifts’’ due to inaccurate data or changes in most ob-servational platforms or analysis procedures (Kalnay etal. 1996). However this does not take into accountchanges in observational platforms such as geostation-ary satellite networks affecting the monitoring of TCs.

The online data analysis and plotting resources at theClimate Diagnostics Center (CDC) facilitated the iden-tification and evaluation of correlations between atmo-spheric data fields and August TC activity parameters.All TC statistics were tested for antecedent correlationwith reanalysis data fields for prior months on the CDCWeb page [http://www.cdc.noaa.gov/correlation; seeSmith and Brown (1999)]. An example given in Fig. 3shows the correlation field for July SLP associated with1958–99 August NTC. The areas with large correlationsin the North Pacific and North Atlantic Oceans in Fig.3 were used as predictors after undergoing additionaltesting as detailed in the following section.

b. Methodology

It was clear that a new set of predictive factors wouldbe required for skillful August-only forecasts as sea-sonal forecast factors failed to capture variability onmonthly time scales. These new predictors were iden-tified in several ways. First, the 10 most active Augustsin the data record were composited and differenced fromthe 10 least active Augusts. The CDC reanalysis com-positing site was similarly helpful for delineating broadareas with noticeable circulation differences for activeversus inactive Augusts on monthly time scales. Asshown in the flow chart in Fig. 4, the methodologyidentified various global-scale difference fields, whichwere then tested further as potential predictors in a sta-tistical sense, and later in a physical one. Another groupof provisional predictors was identified by correlatingTC activity indices with global reanalysis atmosphericfields for 1958–99. Additional predictors were foundlater by correlating forecast residuals with global re-analysis data.

Predictors thus identified were extracted from the full


FIG. 3. Analysis of correlation between Jul SLP and Aug NTC for the years 1958–99. (Figure courtesy Climate Diagnostics Center.) Darkand light shading indicates correlations of greater than 10.3 and 20.3, respectively. Note areas of high correlations associated with theAtlantic and Pacific subtropical ridges.

FIG. 4. Flowchart detailing how provisional predictors formultivariate regression were chosen. (See text for explanation.)

reanalysis dataset and condensed into time series of pre-dictive indices. Provisional predictors were sought infour monthly grid point data fields: 200-mb u and ywinds, 1000-mb geopotential height (correlates .10.98 with sea level pressure), and 500-mb geopotentialheight. All predictor types (composite, correlation, andresidual fields) were used. It should be noted that allpredictors selected were reanalysis ‘‘A’’ variables,which according to Kalnay et al. (1996) are primarilyobservation driven and therefore are mostly reliable.

These predictors can also be reliably estimated in real-time applications for a forecast. The spatial domains ofselected predictors were required to be of sufficient sizeso as to represent larger-scale phenomenon. The smallestarea used was 108 3 158. This size constraint reducedthe possibility of selecting areas with purely randomcorrelations as predictors.

A procedure used to further minimize areas of chancefavorable covariation being chosen for predictors wasto limit the predictor selection to the years 1958–1999.Selected predictors were tested against independent dataduring the first 9 yr (1949–57) of the database. Theearly years were thus utilized as an independent datasetto help determine if potential predictors were artifactsof chance correlation in the reanalysis data fields or werereal representations of actual lagged physical telecon-nection processes. If the possible predictor was similarlyeffective from 1949 to 1957 as 1958 to 1999, it waskept; otherwise, it was discarded.

All predictors that passed the initial test were can-didates in the development of forecast equations from51 yr of data. The predictors were selected using an all-subset technique allowing any predictor to combine withanother predictor and these predictors were included inthe forecast equations until the inclusion of any predic-tor explained less than 3% of additional total variance,as recommended by Gray et al. (1994), or until fivepredictors were obtained. No more than five predictors


FIG. 5. Global predictor map showing the locations of areas fromwhich each predictor was derived. Table 5 provides a description ofeach predictor. The numbers under each box indicate how many timesit was selected for a predictive equation, and hence its importancein the hindcast equations.

TABLE 5. Detailed listing of the area and utilization of all predictors for each individual hindcast parameter and the sign of each predictorcorrelation for an active TC Aug hindcast. See Fig. 5 for a map of these predictor locations.

Predictor and signof correlation Area Eqs. used in

1 Jul 200-mb y wind (2)2 Jul SLPA (2)3 Jul SLPA (2)4 Jul 200-mb u wind (2)5 Jul 500-mb height (2)6 Jul 200-mb u wind (1)7 Jul 200-mb u wind (2)8 Jun 200-mb u wind (1)9 Jun SLPA (1)

10 Apr SLPA (2)11 Feb SLPA (2)12 Jan SLPA (2)

48S–88N, 1058–798W478–628N, 1568E–1648W258–37.58N, 47.58–258W408–358S, 1108–858W42.58–27.58S, 72.58–958E17.58–7.58S, 1458E–180858S–58N, 1108–858W808–858N, 458W–108E188–308N, 1348–1548E108S–58N, 358W–158E52.58–758N, 58W–358E308–408N, 1108–958W

NSD, H, IH, IHD, NTC, TONS, TOHNSD, H, NTC, TOHNS, NSD, HD, IHHD, IHDIH, IHDNS, IHDNSNSD, H, HD, IH, NTC, TONS, TOHNS, HD, IHDNSD, H, HD, NTC, TONS, TOHNS, TONSIH, NTC

were selected to reduce the chance of overfitting. Thecorrelation technique was an ordinary least sum ofsquare deviations (OLS) regression scheme, unless thismethod produced negative forecast values. In these cas-es (H, IH) a Poisson regression was used as the cor-relation technique. The weights of the coefficients andpredictors were selected to fit each of the nine dependentvariables (NS, NSD, H, HD, IH, IHD, NTC, TONS,TOH) in maximizing hindcast skill.

After the equations were developed, the final step wasperforming a cross-validation (jackknife) procedure onthe forecast models. The test, which was conducted fol-lowing the guidelines suggested by Elsner and Schmert-mann (1994), considered the extent to which each yearcould be predicted using data that were independent ofthe predictor observations for that year. Thus, cross val-idation in this paper consisted of predicting each year’sTC activity for each of the 51 yr in the sample usingparameters derived from the other 50 yr. This procedurepartly emulates actual forecast conditions and providessome measure of the real predictive skill, generally re-garded as an approximate upper boundary.

4. August predictor–activity relationships

Twelve predictors were found to be related to AugustTC forecasting, and hindcast schemes were developedfor the 51-yr period between 1949 and 1999 utilizingthe most effective predictors for each of the nine de-pendent variables. The map in Fig. 5 shows areas fromwhich the predictor data were taken and Table 5 liststhe predictors used with each equation and the areas ofthe individual predictors. The remainder of the sectionelucidates physical relationships between the predictorsand August TC activity.

a. Galapagos and southeast Pacific predictors

The two most utilized predictors are the July 200-mbu and y winds over the equatorial Pacific just west ofSouth America. One or the other of these two predictorsis involved in all of the TC forecast equations (see Table5). When these areas experience winds that are anom-alously westerly and southerly, August TC activity isgenerally suppressed in the Atlantic basin. The lattercondition appears to be linked primarily to warm ENSOconditions, which produce anomalous westerlies in thedeep tropical Atlantic (see Gray 1984b for a detailedexplanation).

However, El Nino is not the only climate factor thataffects the two ‘‘Galapagos’’ predictor values. Thereappears to be an influence from the Southern Hemi-sphere (SH) winter long-wave pattern. It is observedthat SH troughs periodically extend to near the equator,creating upper-level westerly winds in the tropical east-ern Pacific. These intrusions from the SH midlatitudesappear to be independent of the state of ENSO and tendto be short lived (on the order of 5–10 days) but canstill radically change the prevailing 200-mb wind pat-terns and significantly alter u and y flow anomalies (Fig.6a).

During July prior to an active August, it is generallyobserved that SH upper-level westerly winds are at high-er latitudes, replaced by a strong anticyclone over north-ern South America. This persistent upper high is char-


FIG. 6. (a) The idealized July 200-mb pattern typically seen beforean inactive Aug with an SH trough. (b) The idealized Jul 200-mbpattern typically observed before active Augs.

FIG. 7. Simplified conceptual summary of the key features in anidealized summer pattern prior to increased Aug TC activity.

acterized by easterly wind anomalies near the equatorextending from the eastern coast of north Brazil west-ward to 1208W with northerly anomalies off the westcoast of Ecuador, as shown in Fig. 6b. Such a patternis favorable for Atlantic tropical-only TC genesis in aconcurrent and predictive sense as it is associated withweak zonal wind shear and easterly 200-mb wind anom-alies across the tropical Atlantic. It is hypothesized thatconditions promoting the appearance and maintenanceof this state of the atmosphere during July are a pre-cursor signal to an active August. If this pattern recursin August when the background climatology has becomemore favorable for TC formation, it often leads to avery active month.

Another predictor that is somewhat related to theequatorial winds is the SH midlatitude July 200-mbu wind west of Chile (408–358S, 1108–858W) (region4 in Fig. 5). When these upper-level westerly windsare weaker than normal, the duration of Atlantic TCsis typically increased, and consequently the NTC isincreased. Although the equatorial 200-mb wind isonly moderately correlated with the midlatitude flow(r 5 0.25), using this midlatitude flow in combinationwith the equatorial wind significantly increases theamount of variance explained in HDs, IHDs, andNTCs.

b. Atlantic SLPA predictors

Two Atlantic Ocean SLPA areas are used as predic-tors. The most significant of these involves the JulySLPA in the central Atlantic (258–37.58N, 47.58–258W;region 3 in Fig. 5). In general, low early season pressurein the Atlantic correlates well with increased Atlanticbasin activity, a fact that has been known since the mid-1930s (Brennan 1935; Ray 1935). Anomalous low pres-sure is indicative of weak midtropospheric subsidenceand less drying of the midatmosphere. Low pressure inthe Atlantic Ocean is also associated with reduced tradewinds, which is linked to warmer SSTs, partly due todecreased evaporation and upwelling.

April SLPAs in the equatorial Atlantic (108S–58N,358W–158E) also display a strong inverse relationshipwith August TC activity, especially activity in the deepTropics on both monthly (i.e., August) and seasonal timescales. Generally, if the April pressure in the equatorialAtlantic (region 10 in Fig. 5) is lower than normal, thenlower August SLPA and negative zonal wind anomaly(ZWA) values are typically observed in the tropical At-lantic, correlating at r 5 0.5. April SLPA may also belinked to ENSO wherein July SSTAs in Nino-4 correlateat r 5 0.3 with this April index.

c. Greenland predictor

Another predictor from the Atlantic region is the June200-mb zonal wind over the far northern region ofGreenland (808–858N, 458W–108E; region 8 in Fig. 5).When the June 200-mb wind in this area is anomalouslystrong from the west, August TC activity is enhanced.This connection possibly reflects midlatitude blockingconditions near Greenland and, hence, the negative


TABLE 6. Listing of predictors chosen for forecasting each TC activity parameter and the total hindcast variance explained for each Augactivity parameter. A more detailed description of the 12 predictors is given in Table 5.

Forecastparameter No. of predictors

Predictors chosenfrom table

Variability explainedby hindcast (R2)

(1949–99)

Likely independentforecast skill(jackknife)

NSNSDHHDIHIHDNTCTONSTOH

554555544

3, 6, 7, 9, 111, 2, 3, 8, 101, 2, 8, 103, 4, 8, 9, 101, 3, 5, 8, 121, 4, 5, 6, 91, 2, 8, 10, 121, 8, 10, 111, 2, 8, 10

0.550.710.570.690.680.780.740.680.64

0.410.610.470.590.590.720.660.600.56

1 Galapagos Jul 200-mb y; sign of correlation, 22 Bering Sea Jul SLP; sign of correlation, 23 Atlantic Ocean Jul SLP; sign of correlation, 24 SE Pacific Jul 200-mb u; sign of correlation, 25 S Indian Ocean Jul 500-mb height; sign of correlation, 26 Coral Sea Jul 200 mb u; sign of correlation, 17 Galapagos Jul 200-mb u; sign of correlation, 28 N Greenland Jun 200-mb u; sign of correlation, 19 NW Pacific Jun SLP; sign of correlation, 1

10 S Atlantic Ocean Apr SLP; sign of correlation, 211 Scandinavia Feb SLP; sign of correlation, 212 SW United States Jan SLP; sign of correlation, 2

phase of the North Atlantic Oscillation (NAO). En-hanced ridging attending a negative NAO leads to east-erly upper-level wind anomalies over northern midlat-itudes while westerly anomalies occur over the polarlatitudes. Van Loon and Rogers (1978) observed thatenhanced wintertime blocking was more prevalent overthe North Atlantic during the decades of the 1950s and1960s than during the 1970s and 1980s. As TC activityduring the 1950s and 1960s was much greater than dur-ing the 1970s and 1980s, it is hypothesized (Gray 2002a)in our 1 December seasonal forecast that the enhancedwintertime blocking patterns (implicit as westerlyanomalies for the Greenland 200-mb zonal wind) arethus a harbinger of active hurricane seasons. Wintercirculation patterns have become more similar to the1950s and 1960s and correlate well with the recent spurtof increased seasonal and August Atlantic TC activity.

d. Northwest Pacific predictors

The Pacific region also holds some key predictors forAugust TC genesis. One of these concerns June SLPAin the northwest Pacific Ocean (188–308N, 1348–1548E),southeast of Japan (region 9 in Fig. 5). When June pres-sure anomalies in this region are high, August TC ac-tivity tends to be increased in the Atlantic basin. LaNina events are typically associated with high pressurein this predictor region (Larkin and Harrison 2001). Thispredictor could be linked to an observed tendency forsignificant reduction of TC activity in the northwestPacific basin during June prior to active TC seasons inthe Atlantic. In addition, this predictor is most closely

linked to August IHD and HD explaining about 20%–25% of the variance of these two parameters.

Sea level pressure anomalies over the Bering Sea re-gion (478–628N, 1568E–1648W; region 2 in Fig. 5) dur-ing July are also strongly correlated with August At-lantic TC activity. There is more Atlantic TC activitywhen July pressure in this region is low. One of thesetwo Pacific Ocean SLPA indicators (i.e., either Juneregion 9 or July region 2) is selected by every forecastequation (except for TONS), emphasizing the linkagesbetween Atlantic TC activity and the global circulation.

These Pacific region predictors suggest that importantlarge-scale differences occur over the Pacific Ocean dur-ing summers prior to active versus inactive Augusts inthe Atlantic basin. An idealized look at global featuresassociated with an active pattern is shown in Fig. 7. Ingeneral, increased August Atlantic basin activity followsanomalous low pressure in the midlatitudes of the west-ern Pacific Ocean and high pressure in the western trop-ical Pacific during the early summer. An effect of thispressure pattern is that vertical shear is typically greaterin the tropical northwest Pacific Ocean during summersbefore active Atlantic years due to increased 200-mbwesterly winds and 850-mb easterly winds, similar towhat occurs during a La Nina event.

It is notable that a reverse (from the Pacific) synoptic-scale pattern is typically observed in the Atlantic Oceanin the early summer before an active August (Fig. 7).In particular, anomalous low pressure is noted in theAtlantic Tropics with a diminished Bermuda high andslightly increased pressures in higher latitudes. Theselower pressures in the Atlantic subtropical high region


TA

BL

E7.

Mat

rix

ofth

eli

near

cros

sco

rrel

atio

nsof

the

12pr

edic

tors

inF

ig.

5.T

heav

erag

esin

the

low

erro

war

eco

mpu

ted

from

abso

lute

valu

es,

henc

edi

sreg

ardi

ngsi

gn.

Var

i-ab

le1

23

45

67

89

1011

12

1 2 3 4 5 6 7 8 9 10 11 12 Avg

— 0.10

0.10

0.24

20.

290.

090.

822

0.29

20.

412

0.15

0.26

0.20

0.27

0.10 — 0.25

0.20

0.03

0.00

0.13

20.

142

0.21

20.

020.

280.

230.

14

0.10

0.25 — 0.13

0.13

0.22

0.17

20.

090.

030.

270.

180.

090.

15

0.24

0.20

0.13 —

20.

022

0.04

20.

102

0.13

0.10

0.06

0.10

0.09

0.11

20.

290.

030.

132

0.02 — 0.01

0.22

0.12

0.12

0.34

20.

030.

080.

13

0.09

0.00

0.22

20.

040.

01 — 0.19

0.06

20.

072

0.16

20.

062

0.09

0.09

0.82

0.13

0.17

0.10

20.

220.

19 —2

0.27

20.

240.

030.

200.

090.

22

20.

292

0.14

20.

092

0.13

0.12

0.06

20.

27 — 0.23

20.

062

0.21

20.

190.

16

20.

412

0.21

0.03

0.10

0.12

20.

072

0.24

0.23 — 0.06

20.

092

0.30

0.17

20.

150.

020.

230.

060.

342

0.16

0.03

20.

060.

06 — 0.11

0.05

0.12

0.26

0.28

0.18

0.10

20.

032

0.06

0.20

20.

212

0.09

0.11 — 0.22

0.16

0.20

0.23

0.09

0.09

0.08

20.

090.

092

0.19

20.

300.

050.

22 — 0.15

1G

alap

agos

Jul

200-

mb

y;si

gnof

corr

elat

ion,

22

Ber

ing

Sea

Jul

SL

P;

sign

ofco

rrel

atio

n,2

3A

tlan

tic

Oce

anJu

lS

LP

;si

gnof

corr

elat

ion,

24

SE

Pac

ific

Jul

200-

mb

u;si

gnof

corr

elat

ion,

25

SIn

dian

Oce

anJu

l50

0-m

bhe

ight

;si

gnof

corr

elat

ion,

26

Cor

alS

eaJu

l20

0m

bu;

sign

ofco

rrel

atio

n,1

7G

alap

agos

Jul

200-

mb

u;si

gnof

corr

elat

ion,

28

NG

reen

land

Jun

200-

mb

u;si

gnof

corr

elat

ion,

19

NW

Pac

ific

Jun

SL

P;

sign

ofco

rrel

atio

n,1

10S

Atl

anti

cO

cean

Apr

SL

P;

sign

ofco

rrel

atio

n,2

11S

cand

inav

iaF

ebS

LP

;si

gnof

corr

elat

ion,

212

SW

Uni

ted

Sta

tes

Jan

SL

P;

sign

ofco

rrel

atio

n,2


TABLE 8. Cross-correlation matrix for predictors vs predictants. The average values in the lower row are computed withoutrespect to sign.

Pre-dictor NS NSD H HD IH IHD NTC TONS TOH Avg

123456789

101112

20.4920.3120.3420.1920.08

0.1820.53

0.350.30

20.2720.3920.32

20.5420.4320.4420.35

0.010.06

20.570.430.35

20.3620.4420.35

20.3920.5020.4020.30

0.030.13

20.420.400.29

20.3420.3020.36

20.4420.4020.4520.3820.07

0.0220.44

0.460.41

20.4220.3520.42

20.4920.3820.4120.3420.24

0.0420.38

0.480.40

20.2520.2320.47

20.4820.3220.0820.4520.33

0.1320.33

0.260.51

20.2320.1720.40

20.5520.4420.3720.4120.15

0.1120.49

0.450.47

20.3520.3320.46

20.5620.2620.2520.25

0.010.14

20.580.420.31

20.3820.5320.37

20.4720.4020.3220.3220.10

0.0920.50

0.440.30

20.4220.4320.37

0.490.380.330.330.110.100.470.410.370.340.350.39

Avg 0.31 0.36 0.32 0.36 0.34 0.31 0.38 0.33 0.35



FIG. 8. Forecast Aug NTC (solid line) vs forecast residual NTC(dashed line).

are usually accompanied by low vertical wind shear inthe main TC development region along with reducedtrade winds and more easterly upper-level flow.

e. Southwest Pacific predictors

Another change related to August Atlantic TC activityoccurs in the SH winter during July. In this instance,an index of 500-mb heights in the Southern IndianOcean (SIO) (42.58–57.58S, 72.58–958E; region 5 in Fig.5) is used as a predictor for IH and IHD. A weaker thannormal ridge in the SIO is favorable for August TCactivity in the Atlantic basin. The physical mecha-

nism(s) for this relationship are not clear at present butmay be linked to processes altering August ZWA-linkedshear conditions across the eastern Atlantic. In this case,lower heights in the SIO are well correlated with easterly200-mb zonal wind anomalies across the main AtlanticTC development region.

A predictor for August TC activity found in the SHis the July 200-mb u wind in the general vicinity of theCoral Sea (17.58–7.58S, 1458E–1808; region 6 in Fig.5). When winds in this domain are enhanced from thewest, Atlantic TC formation tends to be increased duringthe subsequent August. This predictor is also well cor-related with ENSO, especially Nino-3.4 in the centralPacific as well as with increased zonal shear in the west-ern Pacific. Anomalous westerly Coral Sea winds areclosely tied to high pressure in the tropical central andeastern Pacific, as well as to cool water conditions inNino-3.4 during August (hence La Nina).

f. Early season predictors

Additional TC predictors were found based on NHwinter climate conditions. These include January sealevel pressure over the southwest United States (308–408N, 1108–958W; region 12 in Fig. 5) and Februarysea level pressure over Scandinavia (52.58–758N, 58W–358E; region 11 in Fig. 5), both of which are negativelycorrelated with subsequent TC activity. Low Februarysea level pressure in Scandinavia appears to be relatedto enhanced August TC activity through its association


TABLE 9. Ratios of tropical cyclone activity parameters during the 10 Augs with the largest values of each predictor to the values for the10 Augs associated with the lowest values for that predictor for 1949–99.

Predictor NS NSD H HD IH IHD NTC TOH TONS

123456789

101112

3.752.503.272.501.032.043.993.353.332.112.442.80

7.335.334.002.861.633.679.006.333.402.863.144.50

3.302.984.473.321.301.873.795.115.872.902.386.57

5.503.674.003.332.252.506.008.005.001.402.67

11.00

4.062.872.792.761.402.053.913.733.891.752.204.13

25.607.152.029.912.623.29

28.0031.507.781.492.16

114.00

4.751.832.201.831.062.274.433.862.312.232.923.38

2.362.632.641.921.072.142.882.202.751.751.772.30

2.691.451.821.480.831.942.602.001.951.482.051.67



TABLE 10. Summary statistics for Aug U.S. landfall events byintensity class (1949–99).

Namedstorms Hurricanes

Intensehurricanes

Landfalling TCsMax in AugAvg per year

3940.76

2320.45

1010.20

with stronger midlatitude blocking over Greenland. En-hanced blocking in the North Atlantic, and hence higherpressure near Greenland, are linked to a low pressuretrough in the vicinity of Scandinavia.

The relationship between lower January pressure overthe southwest United States and enhanced August TCactivity is difficult to explain but could occur as a com-bination of an ongoing negative Pacific decadal oscil-lation (PDO) pattern (correlates at approximately 0.4with southwest U.S. January pressure) and effects at-tending cold ENSO. Despite the lack of an obviousphysical linkage, this January southwest SLPA associ-ation is very robust (see section 5) and needs to beexplored further.

5. Hindcast test results

The prediction equations typically explain about55%–75% of the August TC variance. Table 6 showsthe predictors chosen for the forecast equations as wellas the amount of variance (R2) explained for each TCactivity parameter. See Fig. 5 for a map of the regions.The most effective predictions were made for NTC

where 86% of the variance was explained by the hind-cast equations. This requires some explanation as sta-tistical NTC (equation) hindcasts explained about 74%of the NTC variance (Table 6). A second method fordeveloping NTC hindcasts was obtained by calculating(forecast) NTC by summing the statistical forecasts forthe six main NTC dependent variables (NS, NSD, H,HD, IH, IHD). This approach is termed the indirecttechnique. The indirect method is considerably betterthan the direct statistical technique (variance explainedincreased 12%) and is therefore used as our primaryAugust NTC forecast. Results obtained with the indirecttechnique are always shown for the hindcast values intables and figures.

Considerable skill was also shown for other forecastparameters. Note that hindcast variance explained wassignificantly greater for TO cyclones than was obtainedfor all cyclones (see Table 6), which agrees well withHess et al. (1995). The most difficult parameter to fore-cast was NS. This result is likely due to a combinationof effects including a basic problem that the differencebetween a 35-kt storm and a 30-kt depression is rathersubjective. Hence, changes in the warning policy at theNational Hurricane Center as well as new observationtechniques have almost certainly biased the NS data. Inaddition, more TCs are likely being found in recentyears as compared to before the advent of daily satellitepictures (pre-1966).

Some statistical hindcasts (i.e., for August IH, HD)were combined by addition with other hindcasts to in-crease hindcast skill for one or the other of the param-eters involved. For example, the final IH hindcast was


TABLE 11. Comparison of numbers and ratios for observed vs hindcast Aug landfall events by intensity class during the 15 largest vs the15 smallest Aug values of NTC from 1949 to 1999.

Namedstorms Hurricanes

Intensehurricanes

Largest 15-yr observed NTCLargest 15-yr hindcast NTCSmallest 15-yr observed NTCSmallest 15-yr hindcast NTCRatio of largest–smallest observedRatio of largest–smallest hindcast

1715662.82.5

121012

12.05.0

5401`4.0

FIG. 9. (a) Aug U.S. hurricane landfalls during the 15 yr with the greatest observed Aug NTC(1949–2000). (b) As in (a) but for the 15 yr with the greatest hindcast Aug NTC. (c) As in (a)but for the 15 yr with the smallest observed Aug NTC. (d) As in (a) but for the 15 yr with thesmallest hindcast Aug NTC. Bold lines signify a major hurricane landfall.

obtained by combining the original IH hindcast with theHD hindcast, and similarly the HD hindcast was com-bined with the NSD hindcast to form the final HD hind-cast, obtained by OLS regression. An example of thisprocedure is illustrated in Eq. (1). These combinationsare possibly due to the relatively high correlations be-tween the parameters (see Table 3). The variance ex-plained increased about 5% for the hindcasts of HD andIH by using the hindcast combination approach:

IH 5 0.9(IH ) 1 0.1(HD ).final forecast forecast (1)

Table 7 contains a matrix showing cross correlationsbetween all pairs of predictors. The Galapagos 200-mbu and y parameters correlate very highly with each other(r . 0.8), which suggests that when the u wind is pos-itive (westerly), the y wind also tends to be positive(southerly) and vice versa. Table 8 shows the relation-ships between the 12 individual hindcast predictors and


TABLE 12. Linear correlation matrix for total seasonal activity indices vs Aug-only parameters for the period of 1949–99.

TotalNS

TotalNSD

TotalH

TotalHD

TotalIH

TotalIHD

TotalNTC

Totalavg

Aug NSAug NSDAug HAug HDAug IHAug IHDAug NTCAug TONSAug TOH

0.670.590.480.580.360.350.550.700.60

0.640.730.590.740.450.510.690.760.72

0.590.670.620.740.540.540.710.720.73

0.480.640.520.730.470.550.660.650.68

0.350.530.430.570.510.550.590.510.64

0.200.400.280.480.280.500.450.400.49

0.480.630.500.690.470.570.650.650.72

0.490.600.490.650.440.510.610.630.65

Avg 0.54 0.65 0.65 0.60 0.52 0.39 0.59 0.56

TABLE 13. Years wherein two or more TOHs occurred during Aug(1949–99) and the associated seasonal TC activity indices.

YearTOH in

AugAugNTC

Seasonaltotal of

hurricanesSeasonal

NTC

19491950195519581963196919801985198919901995199619981999

24332222225323

3272906631616022314676655161

711

977

129778

119

108

119243198140117157137111138102237199178193

Avg 2.6 55 8.7 162

nine forecast parameters. In general, the predictors cor-relate much better with the predictants than with eachother. Figure 8 displays the residual versus predictedvalues of NTC with the active years, in general, havingthe highest errors.

A method to measure the skill of the predictors is toexamine the August TC activity in the years associatedwith the top 10 values of each predictor in comparisonto August TC activity in the years with the bottom 10values of each predictor. The TCs were composited forthe 10 yr with the largest and smallest values for eachpredictor. These two composites of TC activity werecomputed and put into ratio form as shown in Table 9.Note the extremely large differences associated with theGalapagos predictors as well as the Greenland and thesouthwest U.S. January predictors.

The results in this section show strong hindcast skillfor making monthly forecasts of August-only TC activ-ity. In fact, August-only hindcast skill is significantlygreater than the seasonal forecast skill. The parametersdeveloped for the August forecast are also quite differ-ent from those used for the seasonal TC prediction il-lustrated in Gray et al. (1994).

6. Applications of the August forecastThe August forecast assists in the production of two

other forecasts, the August-only landfall probabilitiesand the rest-of-the-season activity. The section endswith a review of the results of independent test forecastsof August activity for 2000–03.

a. August landfall relationships

Other authors have attempted landfalling hurricaneforecasts, most notably Lehmiller et al. (1997) and Els-ner et al. (2000). These studies have shown some suc-cess in forecasting areas that could have a higher chanceof a hurricane landfall than average. However, theseforecasts were for the entire hurricane season and didnot deal with month-to-month changes.

August TC landfalls in the United States are a rela-tively common occurrence. Table 10 gives a brief sum-mary of statistics on U.S. August landfalls. Thirty-nineTCs have come ashore during the past 51 yr, an averageof about one TC every 1.3 yr with about one H landfallduring August every 2 yr and one IH every 5 yr. Mul-tiple August hurricane strikes are a relatively rare event,with only six instances during the past 100 yr. In anygiven August, there is a 41% chance of an H landfalland a 10% chance of an IH landfall event. These prob-abilities could be adjusted based on a strong positiverelationship between August NTC and the number ofAugust TC landfalls on the U.S. coastline. Table 11details this relationship for the years 1949–99 dividedinto two 15-yr subsections with vast differences in land-falling cyclones. Note that these relationships were verysimilar for observed and hindcast NTCs. Figure 9 com-pares the H and IH landfalls in the observed versushindcast values for 15-yr periods. These strong contrastsbetween high and low NTC values allow for landfallprobability forecasts based on forecast NTC. Monthlylandfall probabilities allow for a more specific predic-tion than a broad-brushed seasonal forecast.

b. How August activity relates to total season activity

Relationships between August activity parametersand the seasonal totals are also considered. Table 12


TABLE 14. Aug 2000 forecast and observed activity.

Forecastparameter

Experimental statisticalforecast

Final statisticalforecast

Final adjustedforecast Actual activity

NSNSDHHDIHIHDNTC

2.2914.211.778.271.131.04

32.2

4.0220.322.299.210.801.09

37.3

314.2528.2511.25

33.0

424.752

1411

42.6


Forecastparameter

Experimental statisticalforecast

Final statisticalforecast

Final adjustedforecast No. Actual activity

NSNSDHHDIHIHDNTC

1.6021.66

0.6221.09

0.770.59

12.6

1.799.241.611.040.230.63

15.8

3712.510.5

21.8

311.7500009.5

displays a correlation matrix for August-only TC activ-ity parameters versus total seasonal TC activity. Themost reliable August indicators for total seasonal activ-ity are HD and TOH. These parameters are influencedby generally long-lived hurricanes forming from trop-ical waves south of 208N. When two or more TOHsoccurred during August, seasonal NTC averaged morethan 50% above normal, as shown in Table 13. Thus,the incidence of TOH during August is an importantindicator for seasonal activity yet to come.

c. Test forecasts for August 2000–03

One of the few methods for diagnosing true forecastskill is to analyze the real-time performance of a tech-nique. It is acknowledged that 3 yr of August TC fore-casts are insufficient to determine forecast skill conclu-sively, but it may provide some useful insights into thefuture skill of the model.

Monthly forecasts for August of 2000–03 were issuedat the beginning of August of these three years anddiscussed in the August seasonal forecast update pro-duced by Gray et al. (2000, 2001, 2002b, 2003). TheAugust forecasts included statistical and analog fore-casts and a final adjusted forecast based upon input fromboth sources. The past 3 yr can be considered com-pletely independent from the 1949–99 developmentaldatabase and are thus a useful test of forecast skill. Itis important to note that the 2000 and 2001 forecastswere issued during the development of these statisticalforecast schemes. The statistical technique utilized forthe issuance of the 2000 and 2001 forecast (labeledexperimental statistical forecast) has been changed. Thefinal forecast scheme, which was developed with thesame 1949–99 database (no input from 2000 or 2001),

but which was not available at the time, is shown inTables 14 and 15 as ‘‘final statistical forecast.’’

Even during the development stage the forecasts ver-ified reasonably well. Tables 14–Tables 17 show the2000–03 statistical forecasts along with the qualitativelyadjusted forecast and observed data. Forecasts for 2000generally verified quite well and had skill well aboveclimatology alone. The main disagreement between ob-served activity and the August 2001 forecast concernedwhether an IH would occur during the month. The sta-tistical models suggested there would be an intense hur-ricane while the analog analysis indicated only a smallchance. This difference was the main cause of the errorin the final August 2001 forecast. The statistical forecastfor August 2002 was too high owing to favorable earlyseason predictors. However, the final qualitative forecasttook into account the rapidly changing climate condi-tions (early summer conditions very unfavorable due toENSO) and resulted in a better forecast. The August2003 forecast was excellent and was considered verysuccessful.

The overall results for the test forecasts are verypromising with indications of true forecasting skill. Thestatistical model is a useful guide for forecasting Augustmonthly TC activity with additional subjective refine-ments further increasing skill. Further understanding ofthe underlying physical processes will lead to betterforecasts, especially when extreme and unusual circum-stances present themselves.

7. Conclusions and future work

Extended-range seasonal TC forecasting began 20years ago with the discovery that two Atlantic basin TCmodulators—ENSO and quasi-biennial oscillation



Forecastparameter

Finalstatisticalforecast

Finaladjustedforecast

Actualactivity

NSNSDHHDIHIHDNTC

2.0918.62.67.110.681.06

31.8

410

1400

18

3500006.9


Forecastparameter

Finalstatisticalforecast

Finaladjustedforecast

Actualactivity

NSNSDHHDIHIHDNTC

2.517.30.73.80.630.42

17.9

381410.5

22

3622.511.25

26

(QBO)—in combination with SLPA over the CaribbeanSea could be used to make skillful forecasts (Gray1984a). Other factors have been added into the seasonalforecast schemes through the years as greater insightand longer, more detailed records of atmospheric con-ditions became available. The NCEP–NCAR reanalysisand similar datasets were absolutely crucial in devel-oping the current August forecast scheme and providedinsight into the global nature of TC forecasts.

Hindcast results using the reanalysis dataset make astrong case for true skill in future forecasts. As notedearlier, August-only hindcast skill exceeds current sea-sonal hindcast skill. This type of shorter-term predictionwas expected to be less reliable than the seasonal fore-cast (W. M. Gray 2002, unpublished manuscript) whereactive and inactive multiweek periods tend to be av-eraged. Conversely, a forecast that extends only 1 monthinto the future versus 31 months for the 1 August sea-sonal forecast is less vulnerable to effects of short-termclimate ‘‘drift’’ away from the conditions diagnosed atthe beginning of the target month. These results suggestthat TC activity in other months can also be forecast ina similar manner, and Klotzbach and Gray (2003) hasdeveloped a similar September-only forecast. It may bepossible to eventually forecast total seasonal activity asthe aggregate combination of 3-monthly forecasts ofAugust, September, and October, which typically com-pose about 90% of the total Atlantic basin seasonal ac-tivity.

Persistent variations in broadscale, global circulationfeatures create precursor signals for August TC activityin the Atlantic basin. Anomalous early summer highpressure over the northwestern Pacific Ocean, low pres-sure in the Bering Sea and the subtropical AtlanticOcean, and northeasterly 200-mb winds near the Ga-lapagos are all indicative of an active August in theAtlantic (Fig. 7). Additional signs that occur much ear-lier in the year include low pressure in the southwestUnited States in January, reduced pressure over Scan-dinavia in February, and low equatorial pressure in thecentral Atlantic in April. For some of these relation-ships, mechanisms linking TCs and the associated con-ditions are not independently obvious, and further re-search is needed to gain better insights into why thesephysical relationships act as precursor signals.

The study has also revealed some useful parameters

for forecasting the post–1 September TC activity. Thenumber of August TOHs is closely related to total sea-sonal activity. Generally, if two or more August hur-ricanes occur in the deep Tropics, then the entire seasonis likely to be more active than normal (see Table 13).The addition of just one August TOH above the mean(1) reveals considerable information about the likely TCactivity for the remainder of the season. A large numberof August TOHs is useful as a predictor for activityduring the remainder of the season.

Possible future comparison work might involve theNCEP–NCAR reanalysis and the European Centre forMedium-Range Weather Forecasts (ECMWF) reanaly-sis project. It would be useful to determine if the Augustforecast is as effective when utilizing the ECMWF da-tabase for predictor hindcast values. Additionally, if theECMWF results were similar, it would be additionalevidence of concrete relationships between predictorsand August TC activity.

It is hoped that the research and resulting August-only forecast will prove to be as useful as the seasonalTC predictions are for the public. Awareness of thethreat that TCs pose to the United States will increaseby including the August-only forecast with the seasonalprognostication. The associated landfall probabilitiesshould give the public a more reasonable idea of thevariable likelihood for a TC landfall and thus provideguidance on possible long-term emergency managementdecisions. It is planned that the August forecast will beissued by the Colorado State University research teamas long as the seasonal forecast is issued.

Acknowledgments. The first author would to thank hisadvisor and coauthor, William Gray, for extensive dis-cussions over three years and also Philip Klotzbach formuch discussion and manuscript assistance. He wouldalso like to thank past and present members of Dr. Gray’sresearch team for extensive and beneficial discussionand advice. Besides Drs. Gray and Klotzbach, this re-search team has included John Sheaffer, Todd Kimber-lain, John Knaff, and Matthew Eastin. Barbara Brumitand Amie Hedstrom provided invaluable manuscript as-sistance. This research was supported by the NationalScience Foundation with supplementary support fromthe Gertrude E. Skelly Charitable Foundation and theResearch Foundations of State Farm and USAA Insur-


ance Groups. The first author would also like to thankthe American Meteorological Society for their NASA/Mission to Planet Earth fellowship, which provided sup-port from 1998 to 1999.

REFERENCES

Ballenzweig, E. M., 1959: Relation of long-period circulation anom-alies to tropical cyclone formation and motion. J. Meteor., 16,121–138.

Brennan, J. F., 1935: Relation of May–June weather conditions inJamaica to the Caribbean tropical disturbances of the followingseason. Mon. Wea. Rev., 63, 13–14.

DeMaria, M., J. A. Knaff, and B. H. Connell, 2001: A tropical cyclonegenesis parameter for the tropical Atlantic. Wea. Forecasting,16, 219–233.

Elsner, J. B., and C. P. Schmertmann, 1994: Assessing forecast skillthrough cross validation. Wea. Forecasting, 9, 619–624.

——, and A. B. Kara, 1999: Hurricanes of the North Atlantic: Climateand Society. Oxford University Press, 488 pp.

——, K.-B. Liu, and B. Kocher, 2000: Spatial variations in majorU.S. hurricane activity: Statistics and a physical mechanism. J.Climate, 13, 2293–2305.

Gray, W. M., 1984a: Atlantic seasonal hurricane frequency. Part I:El Nino and 30 mb QBO influences. Mon. Wea. Rev., 112, 1649–1668.

——, 1984b: Atlantic seasonal hurricane frequency. Part II: Fore-casting its variability. Mon. Wea. Rev., 112, 1669–1683.

——, 1990: Strong association between West African rainfall and USlandfall of intense hurricanes. Science, 249, 1251–1256.

——, C. W. Landsea, P. W. Mielke, and K. J. Berry, 1994: PredictingAtlantic seasonal tropical cyclone activity by 1 June. Wea. Fore-casting, 9, 103–115.

——, ——, J. A. Knaff, P. W. Mielke, and K. J. Berry, 1997a: EarlyAugust forecast of Atlantic basin seasonal hurricane activity for1997. Dept. of Atmospheric Science Rep., Colorado State Uni-versity, Fort Collins, CO, 22 pp.

——, J. D. Sheaffer, and C. W. Landsea, 1997b: Climate trends as-sociated with multi-decadal variability of intense Atlantic hur-ricane activity. Hurricanes, Climatic Change and SocioeconomicImpacts: A Current Perspective, H. F. Diaz and R. S. Pulwarty,Eds., Springer Press, 15–53.

——, ——, P. W. Mielke, and K. J. Berry, 2000: Early August updatedforecast of Atlantic basin seasonal hurricane activity and landfallprobabilities for 2000. Dept. of Atmospheric Science Rep., Col-orado State University, Fort Collins, CO, 28 pp.

——, ——, ——, and ——, 2001: Updated early August forecast ofAtlantic seasonal hurricane activity and U.S. landfall strike prob-

abilities for 2001. Dept. of Atmospheric Science Rep., ColoradoState University, Fort Collins, CO, 30 pp.

——, P. J. Klotzbach, and C. W. Landsea, 2002a: Extended rangeforecast of Atlantic seasonal hurricane activity and US landfallstrike probabilities for 2003. Dept. of Atmospheric Science Rep.,Colorado State University, Fort Collins, CO, 17 pp.

——, C. W. Landsea, and P. J. Klotzbach, 2002b: Updated forecastof Atlantic seasonal hurricane activity and U.S. landfall strikeprobabilities for 2002. Dept. of Atmospheric Science Rep., Col-orado State University, Fort Collins, CO, 26 pp.

——, ——, and C. W. Landsea, 2003: Updated forecast of Atlanticseasonal hurricane activity and U.S. landfall strike probabilitiesfor 2003. Dept. of Atmospheric Science Rep., Colorado StateUniversity, Fort Collins, CO, 24 pp.

Hess, J. C., J. B. Elsner, and N. E. LaSeur, 1995: Improving seasonalhurricane predictions for the Atlantic basin. Wea. Forecasting,10, 425–433.

Jarvinen, B. R., C. J. Neumann, and M. A. S. Davis, 1984: A tropicalcyclone data tape for the North Atlantic basin, 1886–1983: Con-tents, limitations, and uses. NOAA Tech. Memo. NWS NHC 22,Miami, FL, 21 pp.

Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Re-analysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.

Klotzbach, P. J., and W. M. Gray, 2003: Forecasting September At-lantic basin tropical cyclone activity. Wea. Forecasting, 18,1109–1128.

Larkin, N. K., and D. E. Harrison, 2001: Tropical Pacific ENSO coldevents, 1946–95: SST, SLP, and surface wind composite anom-alies. J. Climate, 14, 3904–3931.

Lehmiller, G. S., T. B. Kimberlain, and J. B. Elsner, 1997: Seasonalprediction models for North Atlantic basin hurricane location.Mon. Wea. Rev., 125, 1780–1791.

Maloney, E. D., and D. L. Hartmann, 2000: Modulation of hurricaneactivity in the Gulf of Mexico by the Madden–Julian oscillation.Science, 287, 2002–2004.

Neumann, C. J., B. J. Jarvinen, C. J. McAdie, and G. R. Hammer,1999: Tropical cyclones of the North Atlantic Ocean, 1871–1998.National Climatic Data Center and National Hurricane Center,206 pp.

Ray, C. L., 1935: Relation of tropical cyclone frequency to summerpressures and ocean surface water temperatures. Mon. Wea. Rev.,63, 10–12.

Shapiro, L. J., 1987: Month-to-month variability of the Atlantic trop-ical circulation and its relationship to tropical storm formation.Mon. Wea. Rev., 115, 2598–2614.

Smith, C. A. and J. T. Brown, cited, 1999: A Web-based compositingprogram utilizing the NCEP reanalysis dataset. [Available onlineat http://www.cdc.noaa.gov/composites.]

van Loon, H., and J. C. Rogers, 1978: Seesaw in winter temperaturesbetween Greenland and northern Europe. Part I: General de-scription. Mon. Wea. Rev., 106, 296–310.

prediction of august atlantic basin hurricane activity

Documents