Seasonal Tropical Cyclone Forecasts by Suzana J

Home , Atmospheric science, Tropical cyclogenesis

Seasonal tropical cyclone forecasts by Suzana J. Camargo¹, Anthony G. Barnston1, Philip J. Klotzbach2 and Christopher W. Landsea3

Introduction Statistical and dynamical seasonal tropical cyclone

Seasonal forecasts of tropical activity forecasts are proposed to be made available cyclone activity in various regions have been developed since the fi rst on a public Website for forecasters and other users. attempts in the early 1980s by Neville Nicholls (1979) for the Australian region and William Gray (1984(a), (b)) for the North Atlantic region. Over et al., 2006). These quadrennial seasonal tropical cyclone forecasts time, forecasts for different regions, workshops, co-sponsored by the has increased tremendously since using differing methodologies, have WMO Commission for Atmospheric they were fi rst produced, especially been developed. Tourism in various Science Tropical Meteorology after 2004, when 10 tropical cyclones regions, such as the US Gulf and Research Programme and the struck Japan and four hurricanes East Coasts and the Caribbean, World Weather Watch Tropical impacted Florida, USA. is impacted by these seasonal Cyclone Programme, bring together forecasts. Insurance and re-insurance tropical cyclone forecasters and Although landfall forecasts are companies also make use of seasonal researchers to review progress and particularly important to users, forecasts in their policy decisions. plan for future activities in topics landfall forecast skill is still limited. It is fundamental to provide these such as seasonal forecasts. During As seasonal tropical cyclone forecasts users with information about the IWTC-VI, forecasters from various improve, more attention will be given accuracy of seasonal forecasts. countries shared information about to particular details such as regional Seasonal forecasts have limited use seasonal tropical cyclone forecasts landfall probabilities. The use of for emergency managers, because of currently being issued by their such specifi c forecasts will become the lack of skill in predicting impacts respective countries—which was more widespread and signifi cant to at the city or county level. often information not well known decision-makers and residents in by other scientists present. coastal areas. As has been the case in some of the previous WMO International Forecasters in National Meteorological With the popularization of these Workshops on Tropical Cyclones and Hydrological Services are forecasts, it is fundamental that (ITWC), a review of the progress interested in seasonal forecasts their documentation and verifi cation on seasonal forecasts of tropical because they are frequently asked become widely available. It is cyclone activity was presented at questions by the media and various recommended that WMO develop the IWTC-VI in San José, Costa decision-makers. Interest from the guidelines for the development Rica, in November 2006 (Camargo media and the general public in and validation of these forecasts, similar to the protocol that has been developed for global seasonal climate 1 International Research Institute for Climate and Society, The Earth Institute at Columbia (temperature and precipitation) University, Palisades, New York, USA forecasts (WMO, 2001). A summary 2 Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado, USA of grops that issue tropical cyclone 3 NOAA National Hurricane Center, Miami, Florida, USA seasonal forecasts is given in Table I.

WMO Bulletin 56 (4) - October 2007 | 297 Table I — Seasonal tropical cyclone forecasts: groups that issue the forecasts, regions in which the forecasts are issued, forecast type, Website where the forecast is available.

Group Basins Type Website

City University of Hong Kong, Western North Pacifi c Statistical http://aposf02.cityu.edu.hk China (CityU) Colorado State University, Atlantic Statistical http://hurricane.atmos.colostate.edu USA (CSU) Cuban Meteorological Atlantic Statistical http://www.met.inf.cu Institute (INSMET) European Centre for Medium- Atlantic Dynamical http://www.ecmwf.int Range Weather Forecasts Australian (collaborating agencies only) (ECMWF) Eastern North Pacifi c North Indian South Indian South Pacifi c Western North Pacifi c International Research Atlantic Dynamical http://iri.columbia.edu/forecast/tc_fcst/ Institute for Climate and Australia Society (IRI) Eastern North Pacifi c South Pacifi c Western North Pacifi c Macquarie University, Australia / southwest Statistical http://www.iges.org/ellfb/past.html Australia Pacifi c Meteorological Offi ce, United North Atlantic Dynamical http://www.metoffi ce.gov.uk/weather/ Kingdom (MetOffi ce) tropicalcyclone/northatlantic National Meteorological Eastern North Pacifi c Statistical http://smn.cna.gob.mx Service, Mexico (NSM) National Climate Centre, Western North Pacifi c Statistical http://bcc.cma.gov.cn China NOAA hurricane outlooks Atlantic Statistical http://www.cpc.noaa.gov Eastern North Pacifi c http://www.cpc.noaa.gov Central North Pacifi c http:// www.prh.noaa.gov/hnl/cphc Tropical Storm Risk (TSR) Atlantic Statistical http://tsr.mssl.ucl.ac.uk Western North Pacifi c Australian region

Statistical seasonal of ENSO, when the QBO was in its Atlantic seasonal tropical cyclone hurricane forecasts west phase and Caribbean basin sea- forecasts relative to climatology and level pressures were below normal. persistence. Their analysis indicated Colorado State University Statistical forecast techniques for that for the analysed period (1984– North Atlantic tropical cyclones have 2001), both the basic statistical Initial seasonal predictions for the evolved since these early forecasts. forecasts and an adjusted version North Atlantic basin (Gray, 1984(a), Additional predictors were added to demonstrated skill over climatology (b)) were issued by Colorado State the original forecast scheme, the QBO and persistence, with the adjusted University in early June and early is not used as a predictor anymore forecasts being more skilful than the August, beginning in 1984, using and the seasonal forecasts started basic forecasts. statistical relationships between being issued in early December of tropical cyclone activity and El Niño/ the previous year. Klotzbach and Gray Figure 1 shows the skill of the CSU Southern Oscillation (ENSO), the (2004) and Klotzbach (2007) explain forecasts for various leads, using linear Quasi-Biennial Oscillation (QBO) and the current forecast scheme. correlation as a skill measure. The skill Caribbean basin sea-level pressures. improves tremendously in June and Comparatively, more tropical cyclones Owens and Landsea (2003) examined August, probably because the ENSO were predicted in the cool phase the skill of Gray’s operational spring barrier is over. Since the ENSO

298 | WMO Bulletin 56 (4) - October 2007 the number of tropical cyclones in the 1 Central North Paciﬁ c region based on NS NSD the ENSO state and the Paciﬁ c decadal 0.8 H oscillation. HD IH 0.6 IHD Tropical Storm Risk (TSR) NTC Tropical Storm Risk issues statistical 0.4 forecasts for tropical cyclone activity

Correlation in the Atlantic, western North Paciﬁ c 0.2 and Australian regions. The seasonal prediction model uses ENSO forecasts 0 (Lloyd-Hughes et al., 2004) to predict the western North Paciﬁ c ACE index and is skilful in hindcast mode in that -0.2 region (Lea and Saunders, 2006).

-0.4 In a recent paper (Saunders and Lea, December April June August 2005), TSR describes its new forecast Month model, issued in early August, for Figure 1 — Correlations of the CSU seasonal forecasts for different leads: December seasonal predictions of hurricane (1992– 2006), April (1995-2006), June (1984-2006 or 1990-2006) and August (1984-2006 landfall activity for the US coastline. or 1990-2006). The correlations are given for: number of named storms (NS), number The model uses July wind patterns of named storm days (NSD), number of hurricanes (H), number of hurricane days (HD), to predict the seasonal US ACE index number of intense hurricanes (IH), number of intense hurricane days (IHD) and net (effectively, the cumulative wind tropical cyclone activity (NTC). Signifi cant correlations at the 95% signifi cance level are: energy from all tropical cyclones June – NS, NSD, H, HD, IHD, NTC, August – NS, NSD, H, HD, IH and NTC. None of the which strike the USA). The July height- correlations is signifi cant for the December and April leads. averaged winds in these regions are indicative of atmospheric circulation patterns that either favour or hinder state is usually defi ned by June, the as deterministic and probabilistic, hurricanes from reaching US shores. hurricane forecasts made in June or using terciles. They are based on The model correctly anticipates later become more skilful. Another the state of ENSO (Gray, 1984(a)) whether US hurricane losses are reason for a higher skill in June and and the tropical multi-decadal mode above- or below-median in 74 per cent August is that the season is about to (e.g. Chelliah and Bell, 2004), which of the hindcasts for the 1950–2003 start or has already started. incorporates the leading modes of period. The model also performed tropical convective rainfall variability well in “real-time” operation in 2004 CSU started issuing forecasts of occurring on multi-decadal time and 2005, while over-predicting in landfall probabilities in August 1998. scales. Important aspects of this signal 2006. The landfall probabilities are based that are related to an active Atlantic upon a forecast of net tropical cyclone hurricane season include a strong City University of activity. In general, when an active West African monsoon, reduced season is predicted (high net tropical vertical wind shear in the tropical Hong Kong, China cyclone activity), the probability of Atlantic, suppressed convection in landfall is increased (Klotzbach, the Amazon basin and high tropical Johnny Chan and colleagues have 2007). Atlantic sea-surface temperatures issued seasonal tropical cyclone (SSTs) (Goldenberg et al., 2001). The forecasts for the North-west Pacifi c National Oceanic NOAA forecasts and verifi cations for basin (number of tropical cyclones and named storms, hurricanes, major typhoons) since 1997. The statistical and Atmospheric hurricanes and accumulated cyclone predictions are based on various Administration (NOAA) energy (ACE) (Bell et al., 2000) over environmental conditions in the prior the period from 1998–2006 are given year, up to the northern hemisphere NOAA has been issuing seasonal in Figure 2. spring of the forecast season. The hurricane outlooks for the Atlantic and most prominent atmospheric and the eastern North Pacifi c regions since Since 1997, the Central Pacific oceanic conditions include ENSO, 1998 and 2003, respectively. These Hurricane Center issues in May the extent of the Pacifi c subtropical outlooks are provided to the public seasonal forecasts for the range of ridge and the intensity of the India-

WMO Bulletin 56 (4) - October 2007 | 299 parameters (see Table II) (Ballester NOAA tropical storms forecasts NOAA hurricanes forecasts 18 et al., 2004(a) and (b)). The Cuban Observations Observations May forecasts May forecasts Meteorological Institute also issues 26 August forecasts August forecasts Climatological mean 15 Climatological mean statistical landfall forecasts for Cuba 22 based on a discriminant function 12 methodology (Davis, 1986). 18 9 14 Florida State 6 10 Number of hurricanes University (FSU) Number of troprical storms 6 3 1998 2000 2002 2004 2006 1998 2000 2002 2004 2006 James Elsner and colleagues have Year Year been developing techniques for NOAA major hurricanes forecasts NOAA ACE forecasts modelling seasonal hurricane activity 8 360 Observations Observations and landfall. Although their forecasts May forecasts May forecasts August forecasts 300 August forecasts are not produced operationally, their Climatological mean Climatological mean 6 methodology is currently used to 240 issue region-specific forecasts for 4 180 various companies (James Elsner,

120 personal communication, 2006). The 2 FSU group pioneered various topics in 60 seasonal forecasting, such as the use Percentage of median ACE

Number of major hurricanes 0 0 of a Poisson distribution for hurricane 1998 2000 2002 2004 2006 1998 2000 2002 2004 2006 counts (Elsner and Schmertmann, Year Year 1993), the influence of the phase of the North Atlantic Oscillation on Figure 2 — NOAA forecasts (May and August leads) and observations for tropical Atlantic hurricane tracks and US cyclones with tropical storm intensity or higher, hurricanes, major hurricanes and ACE coastal hurricane activity (Elsner et (Accumulated Cyclone Energy, Bell et al., 2000) for the period 1998-2006 al., 2001), and the development of a skilful statistical model for seasonal forecasts of landfall probability over Burma trough (Chan et al., 1998). For a few years, forecasts of the number City University of Hong Kong June forecasts of tropical cyclones making landfall 40 were also issued (Liu and Chan, TS+TY forecasts 2003). Currently, the landfall forecast TS+TY observations TS+TY forecasts 35 scheme for the South China Sea is TS+TY forecasts being improved. The City University of Hong Kong, China, forecasts and the verifications are shown in 30 Figure 3. In most years, the observed number of tropical cyclones is within 25 the range of the forecast number of tropical cyclones, with the exception of 2006. 20 Number of tropical cyclones

Cuban Meteorological 15 Institute

1999 2000 2001 2002 20032004 2005 2006 2007

The Cuban Meteorological Institute Year has been issuing seasonal forecasts of Atlantic hurricane activity since Figure 3 — Veriﬁ cation of the City University of Hong Kong, China, forecasts issued 1996. Currently, the Cuban seasonal in early June: (top) the number of tropical storms and typhoons (TS+TY) observed and forecast is based on the solution of a forecast range; (bottom) the number of typhoons (TY) observed and forecast range. regression and an analogue method In green are the mean climatological number and the corresponding climatological and predicts various tropical cyclone standard deviation.

300 | WMO Bulletin 56 (4) - October 2007 Table II — Seasonal tropical cyclone forecasts: predictors and outputs used for each group. The group acronyms are deﬁ ned in Table I. Other acronyms: TCs (tropical cyclones), ENSO (El Niño-Southern Oscillation) , SST (sea-surface temperature), SLP (sea- level pressure), SOI (Southern Oscillation Index), OLR (outgoing long-wave radiation) and MDR (main development region).

Group Predictors Outputs

CityU 1. ENSO 1. Number of TCs 2. Extent of the Pacifc subtropical ridge 2. Number of named TCs 3. Intensity of India-Burma trough 3. Number of typhoons

CSU 1. SST North Atlantic 1. Number of named TCs 2. SST South Atlantic 2. Named of named TC days 3. SLP South Paciﬁ c 3. Number of hurricanes 4. ENSO 4. Number of hurricane days 5. Atlantic meriodinal Mode 5. Number of major hurricanes 6. Named of major hurricane days 7. Accumulated cyclone energy 8. Net tropical cyclone energy

INSMET 1. North Atlantic winds 1. Number of named TCs 2. ENSO 2. Number of hurricanes 3. Intensity of the Atlantic subtropical ridge 3. Number of named TCs in the Atlantic MDR, Caribbean and 4. SST North Atlantic Gulf of Mexico (separately) 5. Quasi Biennial Oscillation 4. First day with TC genesis in the season 5. Last day with a TC active in the season 6. Number of named TCs that form in the Atlantic MDR and impact the Caribbean

ECMWF 1. Coupled dynamical model 1. Number of named TCs 2. Model TCs identiﬁ ed and tracked 2. Mean location of TC genesis

IRI 1. Various SST forecast scenarios. 1. Number of named TCs 2. Atmospheric models 2. Accumulated cyclone energy (northern hemisphere only) 3. Model TCs identiﬁ ed and tracked. 3. Mean location of TCs (western North Paciﬁ c only)

Macquarie U. 1. SOI index 1. Number of TCs 2. Equivalent potential temperature gradient 2. Number of TCs in the Coral Sea

Met Ofﬁ ce 1. Coupled dynamical model 1. Number of named TCs 2. Model TCs identiﬁ ed and tracked

SMN 1. SST anomalies 1. Number of TCs 2. Equatorial wind anomalies 2. Number of tropical storms 3. Equatorial Paciﬁ c OLR 3. Number of hurricanes 4. Number of major hurricanes

NOAA 1. ENSO 1. Number of named TCs (Atlantic and 2. Tropical multi-decadal mode 2. Number of hurricanes Eastern Paciﬁ c) 3. Atlantic SST 3. Number of major hurricanes 4. Accumulated cyclone energy

NOAA 1. ENSO 1. Number of TCs (Central Paciﬁ c) 2. Paciﬁ c Decadal Oscillation

Tropical Storm 1. Trade winds 1. Number of named TCs Risk (TSR) 2. MDR SST 2. Number of hurricanes 3. ENSO 3. Number of major hurricanes 4. Sea-level pressure central Northern 4. Accumulated Cyclone Energy Paciﬁ c 5. ACE landfalling TCs 6. Number of landfalling named TCs 7. Number of landfalling hurricanes 8. Number of landfalling major hurricanes

WMO Bulletin 56 (4) - October 2007 | 301 the south-eastern USA (Lehmiller et western North Pacifi c since the early The International Research Institute al., 1997). More recently, Elsner and 1980s. Since 1995, when the National (IRI) for Climate and Society, the Jagger (2006) built a Bayesian model Climate Centre was established, a European Centre for Medium-range for seasonal landfall over the USA, nationwide workshop has been held Weather Forecasts (ECMWF) and using as predictors May-June values in April. Forecasts for landfalling more recently the UK Met Offi ce issue of the North Atlantic Oscillation; the typhoons in the South China Sea experimental seasonal forecasts of Southern Oscillation Index; and the and eastern China have also been tropical storm frequency based on Atlantic Multi-decadal Oscillation. developed. These seasonal forecasts dynamical models. The IRI and Met are being continuously improved by Offi ce forecasts are freely available National Meteorological the National Climate Centre and the on the Web. The ECMWF forecasts Shanghai Typhoon Institute. are available online to collaborating Service of Mexico agencies. The ECMWF and Met Offi ce The North Carolina State University forecasts are based on coupled ocean- The National Meteorological Service forecast group presented a new atmosphere models (Vitart and of Mexico has produced a seasonal seasonal forecast methodology Stockdale, 2001). The experimental tropical cyclone activity forecast for Atlantic hurricanes at the 27th IRI forecasts are obtained using a two- for the North-east Pacific basin Conference on Hurricanes and tier procedure. First, various possible since 2001. Their methodology uses Tropical Meteorology of the American scenarios for SSTs are predicted, analogue years and was originally Meteorological Society (T. Yan et al., using statistical or dynamical models. developed by Arthur Douglas at 2006) and gave their forecast for the Then, atmospheric models are forced Creighton University. The forecasts 2006 season. These forecasts for with those predicted SSTs. In both are first issued in January and number of hurricanes and number cases, the tropical cyclone-like updated in May, June and August. of landfalling hurricanes are based vortices are identifi ed and tracked Various predictors are used, including on ENSO, vertical wind shear, the in the atmospheric model outputs SSTs and atmospheric circulation Atlantic dipole mode and the North (e.g. Camargo and Zebiak, 2002). The patterns over the North Pacifi c and Atlantic Oscillation, as discussed in IRI also issues ACE forecasts based outgoing long-wave radiation over the Xie et al. (2004, 2005). on dynamical models for several equatorial Pacifi c. A cluster analysis is northern hemisphere regions. The IRI then used to identify the most similar It is likely that other statistical forecasts are probabilistic by tercile years in the historical record. forecasts are being issued by various category (above normal, normal, agencies around the world of which below normal), as in the example for Australia/South- we are not aware. the Atlantic in 2006 (Figure 4). The rank probability skill score for the West Pacifi c Dynamical tropical IRI July forecasts for the months of cyclone seasonal August to October in the Atlantic for Forecasts for the Australian/South- forecasts the period 2003-2006 is positive with West Pacific region are presented an approximate value of 0.12. annually in the December issue of Many studies have shown that low- the Experimental Long-Lead Forecast resolution climate models are able The skill of some of the best Bulletin since the 2004–2005 season. to simulate tropical cylone-like performing dynamical models in These forecasts are based on a disturbances (e.g. Manabe et al., predicting the frequency of tropical Poisson regression model and use 1970; Bengtsson et al., 1982). These storms is comparable to the skill of as predictors the September saturated disturbances have properties similar statistical models in some ocean equivalent potential temperature to those of observed tropical cyclones basins. Over the North and South gradient and the Southern Oscillation but are typically weaker and larger Indian Ocean, dynamical models Index (McDonnell and Holbrook, in scale. They are more realistic in usually perform poorly (Camargo 2004(a), (b)). They also developed higher-resolution simulations (e.g. et al., 2005). It is not clear to what forecasts for smaller subregions, Bengtsson et al., 1995). extent this is due to model errors or among which the highest hindcast to a lack of predictability. Similarly skill is in the Coral Sea, where ENSO While low-resolution simulations are to the seasonal climate forecasts, has its strongest infl uence. not adequate for forecasting individual combining different model forecasts cyclone tracks and intensities, some (multi-model ensemble forecasts) Other forecasts climate models have skill in forecasting appears to produce overall better levels of seasonal tropical cyclone forecasts than individual model The China Meteorological activity. They are able to reproduce ensemble forecasts (Vitart, 2006). The Administration has been issuing typical ENSO infl uences (e.g. Vitart hindcast skill of various dynamical forecasts of typhoon activity for the et al., 1997). climate models in predicting seasonal

302 | WMO Bulletin 56 (4) - October 2007 Probability forecasts for number of tropical cyclones An alternate approach for forecasting 80 Norht Atlantic tropical cyclones using climate models Below normal involves simulating the interannual 70 ASO 2006 Normal variability of environmental variables Above normal 60 that affect tropical cyclone activity (e.g. Ryan et al., 1992). A drawback of this 50 approach is that it requires a choice of which variables or combinations 40 of variables should be analysed. Climatological Recently, a few studies compared 30 probability both approaches using the same (33%)

Probability (percentage) climate models (e.g. McDonald et al., 20 2005; Camargo et al., 2007(b)). Both approaches may be used in the future, 10 since they are complementary. 0 April May June July August

Month forecast was issued The importance of

Figure 4 — IRI experimental dynamical forecast probabilities for the August-October ENSO prediction (ASO) 2006 period in the Atlantic for different lead times. The normal category is defi ned as six to nine named tropical cyclones, the below-normal category as fi ve or less named ENSO events shift the seasonal tropical cyclones and the above-normal category as 10 or more named tropical cyclones. temperature and precipitation In 2006 there were seven named tropical cyclones in the Atlantic during ASO, i.e. the patterns in a consistent manner in season was in the normal category. many parts of the world (Bradley et al., 1987; Ropelewski and Halpert, 1987). Depending on the time of the year, tropical cyclone activity is discussed Mozambique (Vitart et al., 2003). ENSO phenomena can be predicted in Camargo et al. (2005) and Vitart Another possible approach to predict with modest-to-moderate skill months (2006). The European multi-model the risk of tropical cyclone landfall in advance (Cane and Zebiak, 1985). (EUROSIP) dynamical forecasts of using dynamical models would ENSO forecasts are routinely used as tropical cyclone frequency skilfully incorporate statistical techniques a major component in probabilistic distinguished the very active Atlantic such as track clustering (Camargo seasonal climate forecasts at various hurricane season in 2005 from the et al., 2007(a)). centres (Goddard et al., 2001). below-average season in 2006 (Vitart Correlation = 0.78 (1.00) Forecast Observations 2 standard deviations et al., 2007). The EUROSIP forecasts are RMS error = 3.07 (4.56) 26 not currently available to the public. 25 24 The predicted number of tropical 23 22 storms in the EUROSIP hindcasts 21 20 (1993-2004) and real-time forecasts 19 18 17 (2005-2006) is shown in Figure 5 16 15 (Vitart et al., 2007, Figure 3). 14 13 12 11 Seasonal prediction of tropical cyclone 10 9 8 landfall represents a major challenge 7 6 for dynamical models. Tropical 5 4 cyclones take an unrealistically Number of tropical storms 3 2 poleward track in some of the models 1 used in seasonal forecasting systems, 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 due partly to the coarse horizontal Year model resolution, which leads to larger vortices than observed ones. Figure 5 — Number of tropical storms from July to November predicted by the EUROSIP These larger vortices would likely be (median) starting on 1 June (blue solid line) for the period 1993-2006. Hindcasts were more infl uenced by the beta effect. used for the period 1993-2004, and real-time forecasts in 2005-2006. The observations Finer-resolution climate models are are given in the dotted red line and the green vertical lines represent two standard able to reproduce landfall differences deviations within the multi-model ensemble distribution. (Figure originally from Vitart related to ENSO impacts, such as in et al., 2007)

WMO Bulletin 56 (4) - October 2007 | 303 Tropical cyclones are also affected by predicting its continuation for the next forecasts obtained with the Zebiak ENSO in various parts of the world. 9 to 12 months is a much easier task and Cane (1987) simple coupled The relationship between them was than predicting its initial appearance. model (Figure 6). While these skills fi rst documented in a series of papers Even a strong El Niño, such as that of are for a particular model, they by Neville Nicholls for the Australian 1997/1998, was not well anticipated roughly approximate the skills for region (Nicholls, 1979). During warm before signs of the initial onset were predictions of other dynamical as ENSO events, fewer cyclones occur observed in the northern hemisphere well as statistical models, because near Australia, while in cold events, an spring of 1997 (Barnston et al., 1999). they represent basic predictability enhanced risk of landfall in Australia Even after becoming apparent in the that is refl ected similarly across most exists with more cyclones affecting observations in late April and May of the present models. It is clear that Queensland. The impact of ENSO 1997, the strength of this extreme predictive skill for forecasts made in on North Atlantic cyclones was fi rst El Niño event was underpredicted March is high for only 2-3 months, discussed by William Gray (Gray, by most models, although a few while, for forecasts made in August, 1984(a)). The infl uence of ENSO on models did correctly anticipate the the skill extends to longer lead western North Pacifi c typhoon activity rapid weakening in the spring of 1998 times. Improvements in predictive was fi rst explored in Chan (1985). In all (Landsea and Knaff, 2000). skill using today’s more advanced cases, the relationship of ENSO and dynamical models have been small tropical cyclones was subsequently There is varying skill in ENSO and it remains to be seen whether developed into statistical forecasts forecasts as evidenced by the Nino3 or not substantial improvements predicting seasonal activity. Jan. The state of ENSO is of fundamental importance in the seasonal activity level and character of tropical cyclones Nov. in all ocean basins. This is the case not only because of the obvious relevance of the ENSO state to the SST anomaly pattern in the tropical ocean basin Sept. but also because of the infl uence of ENSO on fi elds of local atmospheric variables, such as the large-scale July horizontal pattern of anomalous circulation and geopotential height, upper-level divergence and vertical May wind shear. Thus, our ability to predict ENSO state several months in advance is critical to being able to predict tropical cyclone activity in the same Mar. timeframe, using either statistical or dynamical methodologies. Jan. ENSO predictability follows a well- 0 5 10 15 20 known seasonal cycle, in which the Lead time in months ENSO state for 4-6 months into the future is more accurately predicted from a starting time between July and November than between January and March. This is due to a “predictability 0.2 0.4 0.6 0.8 barrier” that exists between April and June, such that forecasts made just Figure 6 — Skill of the Zebiak and Cane ENSO forecast model for prediction of Nino3 before this period are hindered by sea-surface temperature anomalies for varying hindcast start months and hindcast the barrier. The seasonal timing of lead time. The colours indicate skill as a correlation between the hindcasts and the the predictability barrier is related to corresponding observations. The vertical axis indicates the month from which the the life cycle of ENSO episodes, which hindcast is made and the horizontal axis is the lead time. For example, a hindcast made often emerge between April and June from July with a lead time of two months would be a hindcast for September and with a and endure until the following March lead time of 24 months (right side of fi gure) would be a hindcast for the July two years to May. Once an episode has begun, after the hindcast was made.

304 | WMO Bulletin 56 (4) - October 2007 are possible, given the inherent IRI Model Forecasts of ENSO from May 2006 signal-to-noise characteristics of the 3 ocean-atmosphere system. The “slow Verification physics” relevant to ENSO dynamics 2.5 Observations by April 2006 may become better predicted by both Dynamical forecasts 2 Statistical forecasts statistical and dynamical models of the future. However, better prediction 1.5 of the shorter time-scale events that can also be important in triggering 1 El Niño onset, such as the Madden- Julian Oscillation, may prove to be 0.5 nearly impossible at multi-month lead times. The May forecasts by many Nino3.4 (°C) 0 statistical and dynamical models for the El Niño event of 2006/2007 are −0.5 shown in Figure 7. Very few models were able to forecast this event, which −1 had a late onset and was not very strong. −1.5 Obs. Forecasts The ENSO predictability barrier has −2 clear-cut implications for predictions FMA Apr AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM of tropical cyclone activity in the 2006Season 2007 northern hemisphere when compared to predictions of activity in the Figure 7 — May 2006 statistical and dynamical forecasts for the Nino 3.4 index southern hemisphere. Tropical cyclone (anomalies) using April observations. Observations for the Nino 3.4 region are also activity in the northern hemisphere shown. is considerably more challenging to predict because its peak seasons occur predictions for any ocean basin is the subject to legitimate verifi cation and shortly after the ENSO predictability far-from-perfect quality of today’s evaluation (Hastenrath, 1990; Owens barrier. When an ENSO event appears state-of-the-art ENSO forecasts. and Landsea, 2003). Verification somewhat later than usual (as was the Indeed, Landsea and Knaff (2000) measures help communicate the case in the late northern hemisphere showed that it is still very diffi cult to quality of future predictions to users summers of 1986 and 2006), the outperform a simple statistical model who need to know how to reasonably inhibiting effect on North Atlantic that uses as predictors only the recent apply them in their decision-making tropical cyclones is unanticipated evolution of SST anomalies in a few processes. Predictions of tropical until the peak season of August to tropical Pacifi c regions. This modest cyclone activity are expressed October is already beginning. This can skill level for detecting El Niño onset deterministically (e.g. a forecast of necessitate a sudden change as a fi nal still exists in 2007, as demonstrated either the exact number of tropical update to the seasonal prediction and by the poor predictions of the late- cyclones or a specific range of can potentially disrupt plans already starting 2006/2007 El Niño (Figure 7). If their numbers in a given ocean being followed in accordance with an the ENSO forecast challenge could be basin during the peak season) earlier seasonal prediction. overcome, the skill of predictions could or probabilistically (e.g. forecast improve signifi cantly—most notably probabilities for an underactive, The peak season for southern hemi- in the northern hemisphere. a near-normal or an over-active sphere activity occurs at least 6 months season). Suitable verification after the northern hemisphere spring measures for past predictions of the ENSO predictability barrier, which Verifi cation and same kind given in real-time and/or provides a safer cushion of lead time properly cross-validated hindcasts in which to become fairly certain about evaluation of over an extended past period during the ENSO state to be expected during seasonal tropical which real-time forecasts were not the peak season. Thus, last-minute cyclone forecasts issued, are an absolute necessity. surprises in seasonal outlooks for Even the best attempts at cross- basins south of the Equator are less validation may produce a somewhat likely to be impacted by inaccurate As for predictions of any aspect too optimistic skill estimate relative ENSO outlooks. Nonetheless, it is of seasonal climate, predictions to the skill expected for real-time clear that a major hurdle in improving of tropical cyclone activity are forecasts (Barnston et al. 1994).

WMO Bulletin 56 (4) - October 2007 | 305 Given the widespread dissemination actual forecasts attain skill equal to Epstein, 1969) for deterministic and of new seasonal tropical cyclone that of the reference forecasts and probabilistic forecasts, respectively. forecasts, it is fundamental that one (or 100 per cent) when they The conventional (i.e. Pearson) all forecast agencies should follow are perfectly accurate. The choice correlation coefficient or the standard guidelines for producing and of a reference forecast is vital to an Spearman rank correlation coeffi cient verifying these forecasts. A Website understanding of the meaning of are also informative when applied to is currently being developed that will the skill score. Climatology is often deterministic forecasts. When the include seasonal climate forecasts used as the reference forecast— correlation is applied to a small subset issued by various agencies. To a forecast for the long-term mean of a much longer climatological base participate, the agencies must follow number of tropical cyclones as a period, the uncentred correlation the WMO guidelines for seasonal deterministic forecast or a forecast coeffi cient, where the climatological forecasts. At the IWTC-VI meeting for climatological probabilities (e.g. mean is not removed in computing the in Costa Rica, it was suggested that a 33.3 per cent for each of the tercile- standard deviations and subsequently similar Website could be developed for based categories) as a probabilistic for the deviations of the cross- seasonal tropical cyclone forecasts. forecast. products, may be a more suitable Guidelines and standards must fi rst be verifi cation measure than the standard developed such as for other seasonal Outperforming a climatology reference correlation. This is particularly true climate forecasts. forecast is not usually considered when the mean values for the sub- diffi cult when some predictive skill is period differ noticeably from the One necessary step is to define assumed to exist. On the other hand, overall climatological means. common metrics for the seasonal setting the reference forecast to be the tropical cyclone forecast outputs, skill of a statistical model in verifying In the case of dynamical forecasts, such as number of named storms, a dynamical model could be setting the skill of hindcasts should be number of hurricanes, number of the reference forecast standard too provided (e.g. Camargo et al., 2005). major hurricanes, and accumulated high. Statistical models may be able For statistical prediction methods, cyclone energy). Dynamical forecasts to capture much of the available cross-validation (Michaelsen, 1987) is currently do not issue forecast number predictability inherent in the climate needed to help reduce artifi cial skill of hurricanes and major hurricanes, system through the observed historical that can exist in the training data mainly because of the low resolution data. Table III has a suggested list of sample but vanishes when the method of the models, but that could be verifi cation measures for deterministic is applied to a real-time forecast for the achieved in the future with higher and probabilistic forecasts. future. Statistical methods optimize resolution models. Some forecast results within the training sample and variables produced by individual Examples of verifi cation measures cannot fi lter out the component of groups, such as number of hurricane that have commonly been used with skill related to fi tting the adjustable days, would not be required for all climatology forecasts as the reference parameters (such as the weighting of agencies. are the mean square error skill each of the predictors in a multiple score (MSESS; WMO, 2002) and the linear regression technique) to the Another important consideration ranked probability skill score (RPSS; random variations in the sample being for setting up guidelines for tropical cyclone seasonal forecasts is the Table III — Verifi cation skill scores suggested for deterministic and probabilistic verifi cation measures. Table III has TC activity seasonal forecasts. Most of these scores are described in standard a list of skill measures that could statistical books, such as Wilkes (1995). The natural categories skill score is in be used. It is emphasized that a Owens and Landsea (2003), while the likelihood score is described in Harte and combination of various skill measures Vere-Jones (2005). gives a more complete evaluation of the skill of the forecasts. Type of Forecast Verifi cations : skill scores

Verification measures are often Deterministic 1. Root mean square error skill score formulated as a comparison with 2. Pearson correlation coeffi cient a set of reference forecasts made 3. Spearman rank correlation coeffi cient 4. Uncentred correlation coeffi cient (short periods) using a much simpler and uninformed 5. Bias compared to climatology method, such as perpetual climatology 6. Percentage improvement over trend forecasts or perpetual persistence 7. Normalized natural categories skill score of observations from the previous Probabilistic 1. Ranked probability skill score year or averaged over the previous 2. Relative operating characteristic (ROC) skill score n years. Such so-called “skill scores” 3. Likelihood skill score are often scaled to be zero when the

306 | WMO Bulletin 56 (4) - October 2007 used. In cross-validation, forecast with ENSO and the timing of the basin Currently, various agencies are issuing models are derived from all cases tropical cyclone season relative to tropical cyclone forecasts. It is of except for one (or more) that are the ENSO cycle. Further research fundamental importance to establish withheld, and these cases are then is necessary to gain a clear picture standards for the development and used as the target(s) of the prediction. of this potential predictability. verification of these forecasts, so This is repeated with all possible Unfortunately, the inhomogeneity that users are able to apply them cases, or sets of cases, withheld and of best-track datasets and possible appropriately. A single Website could used as the target(s). The anomaly biases in dynamical models prevent be developed by WMO to give easy values of the cases withheld must be an accurate estimate of the potential access to all forecasts that follow the expressed in terms of the climatology predictability. WMO guidelines. formed from the remaining years, which changes slightly each time a Summary new case(s) is withheld. Acknowledgments Statistical seasonal tropical cyclone Skill estimates resulting from cross- forecasting has come a long way since Some sections of this paper are based validated forecasts are nearly always it began in the early 1980s. Along on the sub-topic report of seasonal and sub-seasonal TC forecasts for the somewhat lower and more indicative with predictions of total seasonal IWTC-VI workshop held in November of skills to be expected in future cases. activity, several forecasts now 2006 in Costa Rica. We thank the When true skill is very high (e.g. the include individual monthly forecasts contributors to the seasonal forecasts in correlation between forecasts and and predictions of probability of that report: Maritza Ballester (Instituto corresponding observations > 0.6), landfall. As the availability of global de Meteorología de la República de Cuba), Mark. A. Saunders (University cross-validation results in only slightly datasets such as the various re- College London, United Kingdom) and lower skill than those in the training analysis products continues to be Frédéric Vitart (European Centre for sample. When true skills are low (e.g. improved, so will statistical forecasts Medium-range Weather Forecasts, 0.2 to 0.4), cross-validation results of tropical cyclones. An updated and United Kingdom). We also thank Michel are markedly lower and occasionally homogenous quality global best-track Rosengaus (Comisión Nacional del negative. dataset would also contribute to more Agua, Mexico), Fumin Ren (Chinese Academy of Meteorological Sciences, skilful forecasts (Landsea et al., 2004; China), Wes Browning and James Skill estimates produced using cross- Kossin et al., 2007). Weyman (Central Pacifi c Hurricane validation may be used to dampen Center, Hawaii, USA) for information on the amplitude of real-time statistical Dynamical seasonal tropical cyclone their forecasts. Suggestions of Russell forecasts that use an entire training forecasts are now currently issued Elsberry, Johnny Chan, Eric Blake and Richard Pasch signifi cantly improved sample for predicting a future case in for various regions. Increasing model this manuscript. We are grateful to real-time. Such damping helps hedge resolution should help improve the Frédéric Vitart and co-authors for giving against some artifi cial skill that may skill of these forecasts. To be able to permission to reproduce Figure 5. be present in the real-time forecast. forecast landfall probabilities using For example, if a real-time regression dynamical models where possible, forecast has an expected skill of 0.5 (as systematic biases in the tracks of References indicated by the variance explained by model tropical cyclones need to be the model using all available past years examined, explained and corrected. BALLESTER, M., C. GONZÁLEZ, and R. PERÉZ for training) but the cross-validated Some of the biases are probably due SUARÉZ, 2004(a): Modelo estadísitico correlation skill score for the same to factors other than low resolution para el pronóstico de la actividad model is 0.4, then the amplitude of the and more research is needed in ciclónica en el Oceáno Atlántico, el Golfo de México y el Mar de Caribe, anomaly of the real-time forecast for a understanding the atmospheric Revista Cubana de Meteorología, Vol,. future month would be decreased by model ability to forecast tropical 11, No 1, 9 pp (in Spanish), available the factor of 0.4/0.5 or 0.8 to account cyclones. from [email protected] . for the over-confi dence associated with over-fitting to the finite data Future improvement of seasonal BALLESTER, M., C. GONZÁLEZ, R. PERÉZ S UARÉZ, sample. forecasts is strongly dependent A. ORTEGA, and M. SARMIENTO, 2004(b): Pronóstico de la actividad ciclónica on improving ENSO forecasts, en la region del Atlántico Norte, con One important aspect to keep in including greater detail about the énfasis en el Caribe y Cuba, Informe mind is that different tropical cyclone characteristics of the ENSO event Científi co, Instituto de Meteorología regions have very different potential such as its magnitude, location and (in Spanish), available from biblio@ predictabilities. These differences are spatial pattern of SST anomalies. Both met.in.cu. due to various factors, such as the statistical and dynamical forecasts BARNSTON, A.G., M.H. GLANTZ, and Y.HE, basin climatological characteristics, are dependent on the quality of ENSO 1999. Predictive skill of statistical and the strength of the basin relationship forecasts for some of their skill. dynamical climate models in SST

WMO Bulletin 56 (4) - October 2007 | 307 forecasts during the 1997-98 El Niño 2007(a): Cluster analysis of typhoon EPSTEIN, E.S., 1969: A scoring system episode and the 1998 La Niña onset. tracks. Part I: General Properties, J. for probability forecasts of ranked Bull. Amer. Meteor. Soc., 80, 217-243. Climate, 20, 3635-3653. categories. J. Appl. Meteor., 8, 985-987. BARNSTON, A.G., H.M. VAN DEN DOOL, CAMARGO, S.J. , A.H. SOBEL, A.G. BARNSTON S.E. ZEBIAK, T.P. BARNETT, M. JI, D.R. and K.A. EMANUEL, 2007(b): Tropical GODDARD, L., S.J. MASON, S.E. ZEBIAK, RODENHUIS, M.A. CANE, A. LEETMAA, N.E. cyclone genesis index in climate C.F. ROPELEWSKI, R.E. BASHER and GRAHAM, C.F. ROPELEWSKI, V.E. KOUSKY, models, Tellus 59A, 428-443, doi: M.A. CANE, 2001: Current approaches E.A. O’LENIC and R. E. LIVEZEY, 1994: 10.1111/j.1600-0870.2007.00238.x to seasonal to interannual climate Long-lead forecasts – Where do we predictions, Int. J. Climatol., 21, stand? Bull. Amer. Meteor. Soc., 75, CANE, M.A. and S.E. ZEBIAK, 1985: A 1111-1152. 2097-2114. theory for El Niño and the Southern Oscillation, Science, 228, 1085-1087. GOLDENBERG, S.B., C.W. LandSEA, BELL, G.D. and M. S. HALPERT, R.C. SCHNELL, A.M. MESTAS-NUÑEZ and W.M. GRAY, R.W. HIGGINS, J. LAWRIMORE, V.E. KOUSKY, CHAN, J.C.L., 1985: Tropical cyclone 2001: The recent increase in Atlantic R. TINKER, W. THIAW, M. CHELLIAH and activity in the Northwest Pacifi c hurricane activity: causes and A. ARTUSA, 2000. Climate Assessment in relation to El Niño Southern implications. Science, 293, 474-479. for 1999. Bull. Amer. Meteor. Soc., 81, Oscillation phenomenon. Mon. Wea. S1-S50. Rev., 113, 599-606. GRAY, W.M., 1984(a): Atlantic seasonal hurricane frequency. Part I: El Niño BENGTSSON L., H. BÖTTGE and M. KANAMITSU, CHAN, J.C.L., J. SHI and C. LAM, 1998: and 30 mb quasi-biennial oscillation 1982: Simulation of hurricane-type Seasonal forecasting of tropical infl uences. Mon. Wea. Rev., 112, vortices in a general circulation cyclone activity over the western 1649–1668. model, Tellus, 34, 440-457. North Pacifi c and the South China Sea. Wea. Forecasting, 13, 997-1004. GRAY, W.M, 1984(b): Atlantic seasonal BENGTSSON L., M. BOTZET and M. ESCH, hurricane frequency. Part II: 1995: Hurricane-type vortices in a CHAN, J.C.L., J.E. SHI and C.M. LAM, Forecasting its variability. Mon. Wea. general circulation model, Tellus, 2001: Improvements in the seasonal Rev., 112, 1669–1683. 47A, 175-196. forecasting of tropical cyclone activity over the western North Pacifi c, Wea. HARTE, D. and D. VERE-JONES, 2005: BLAKE, E.S. and W.M. GRAY, 2004: Forecasting, 16, 997-1004. The entropy score ant its use in Prediction of August Atlantic basin earthquake forecasting. Pure Appl. hurricane activity. Wea. Forecasting, CHAUVIN, F., J.-F. ROYER and M. DÉQUÉ, 2006: Geophys., 162, 1229-1253. 19, 1044-1060. Response of hurricane-type vortices to global warming as simulated by HASTENRATH, S., 1990: Tropical climate BRADLEY, R.S., H.F. DIAZ, G.N. KILADIS ARPEGE-Climat at high-resolution. prediction: A progress report, 1985- and E.I. EISCHED, 1987: ENSO signal Clim. Dyn., 27, 377-399. 1990. Bull. Amer. Meteor. Soc., 71, in continental temperatures and 819-825. precipitation records, Nature, 327, CHELLIAH M. and G.D. BELL, 2004: Tropical 497-501. multidecadal and interannual KLOTZBACH, P.J., 2007: Recent climate variability in the NCEP/NCAR developments in statistical prediction CAMARGO, S.J, M. BALLESTER, A.G. BARNSTON, reanalysis. J. Climate, 17, 1777-1803. of seasonal Atlantic basin tropical P. K LOTZBACH, P. ROUNDY, M. A. SAUNDERS, cyclone activity, Tellus 59A, 511-518. F. VITART and M.C. WHEELER, 2006: DAVIS, J., 1986: Statistics and Data doi:10.1111/j.1600-0870.2007.00239.x. Short-term climate (seasonal and Analysis in Geology, John Willey and intra-seasonal) prediction of tropical Sons, Toronto, 646 pp. KLOTZBACH, P.J. and W.M. GRAY, 2004: cyclone activity and intensity. Topic Updated 6-11 month prediction of 4.3, Workshop Topic Reports, Sixth ELSNER, J.B. and T.H. JAGGER, 2004: A Atlantic basin seasonal hurricane International Workshop on Tropical hierarchical Bayesian approach to activity. Wea. Forecasting, 19, Cyclones (IWTC-VI), San José, Costa seasonal hurricane modeling. J. 917-934. Rica, November 2006, WMO, TMRP Climate, 17, 2813-2827. 72, 493-499. KOSSIN, J.P., K.R. KNAPP, D.J. VIMONT, ELSNER, J.B. and T.H. JAGGER, 2006: R.J. MURNANE and B.A. HARPER, 2007: CAMARGO, S.J. and S.E. ZEBIAK, 2002: Prediction models for annual US A globally consistent reanalysis of Improving the detection and tracking hurricane counts. J. Climate, 19, hurricane variability and trends. of tropical storms in atmospheric 2935-2952. Geophys. Res. Lett., 34, L04815, general circulation models. Wea. doi:10.1029/2006GL028836. Forecasting, 17, 1152-1162. ELSNER, J.B. and C.P. SCHMERTMANN, 1993: Improving extended-range seasonal LANDSEA, C. W. and J. A. KNAFF, 2000: How CAMARGO, S.J., A.G. BARNSTON and S.E. predictions of intense Atlantic much skill was there in forecasting the ZEBIAK, 2005: A statistical assessment hurricane activity. Wea. Forecasting, very strong 1997-1998 El Niño? Bull. of tropical cyclones in atmospheric 8, 345–351. Amer. Meteor. Soc., 81, 2107-2119. general circulation models. Tellus, 57A, 589-604. ELSNER, J.B. and B.H. BOSSAK and X.-F. NIU, LANDSEA, C. W., C. ANDERSON, N. CHARLES, 2001. Secular changes to the ENSO– G. CLARK, J. DUNION, J. FERNANDEZ- CAMARGO, S.J., A.W. ROBERTSON, US hurricane relationship. Geophys. PARTAGAS, P. HUNGERFORD, C. NEUMANN S.J. GAFFNEY, P. SMYTH and M. GHIL, Res. Lett., 28, 4123-4126. and M. ZIMMER, 2004: The Atlantic

308 | WMO Bulletin 56 (4) - October 2007 database re-analysis project: MCDONNELL, K.A. and N.J. Holbrook, VITART F.D., D. ANDERSON and T. STOCKDALE, Documentation for the 1851-1910 2004(b): A Poisson regression model 2003: Seasonal forecasting of tropical alterations and additions to the approach to predicting tropical cyclone landfall over Mozambique. HURDAT database. Hurricanes and cyclogenesis in the Australian/ J. Climate, 16, 3932-3945. Typhoons: Past, present and future. southwest Pacifi c Ocean region using R.J. Murnane and K.–B. Liu, Editors., the SOI and saturated equivalent VITART F.D., 2006: Seasonal forecasting Columbia University Press, 177-221. potential temperature gradient as of tropical storm frequency using predictors. Geophys. Res. Lett., 31, a multi-model ensemble. Q. J. R. LEA, A.S. and M.A. SAUNDERS, 2006: L20110, doi: 10.1029/2004GL020843. Meteorol. Soc. 132, 647-666. Seasonal prediction of typhoon activity in the Northwest Pacifi c Michaelsen, J., 1987. Cross-validation in VITART F., M.R. HUDDLESTON, M. DÉQUÉ, basin, 27th Conference on Hurricanes statistical climate forecast models. D. PEAKE, T.N. PALMER, T.N. STOCKDALE, and Tropical Meteorology, Monterey, J. Clim. Appl. Meteor., 26, 1589-1600. M.K. DAVEY, S. INENSON and 24-28 April, P. 5.23. A. WEISHEIMER, 2007: Dynamically- NICHOLLS, N., 1979: A possible method for based seasonal forecasts of Atlantic LEHMILLER, G.S., T.B. KIMBERLAIN and predicting seasonal tropical cyclone tropical storm activity, Geophys. Res. J.B. ELSNER, 1997: Seasonal prediction activity in the Australian region, Mon. Lett., doi:10.1029/2007GL030740, in models for the North Atlantic basin Wea. Rev., 107, 1221-1224. press. hurricane location. Mon. Wea. Rev., 125, 1780-1791. OWENS, B.F. and C.W. LANDSEA, 2003: WILKES, D.S., 1995: Statistical methods in Assessing the skill of operational the atmospheric sciences. Academic LIU, K.S. and J.C.L. CHAN, 2003: Atlantic seasonal tropical cyclone Press, 467 pp. Climatological characteristics and forecasts, Wea. Forecasting, 18, seasonal forecasting of tropical 45-54. WMO, 2001: Standardized Verifi cation cyclones making landfall along the System (SVS) for Long-Range South China coast. Mon. Wea. Rev., ROPELEWSKI, C.F. and M.S. HALPERT, Forecasts (LRF). Report 6.3B, April 131, 1650–1662. 1987: Global and regional scale 30 to May 3, 2001. Available at: http:// precipitation patterns associated www.wmo.int/web/wcp/clips_ictt/ LLOYD-HUGHES, B., M.A. SAUNDERS and with the El Niño/Southern Oscillation. SVS-for-LRF.html P. R OCKETT, 2004: A consolidated Mon. Wea. Rev., 115, 1606-1626. CLIPER model for improved August- XIE L., T. YAN, and L.J. PIETRAFESA, September ENSO prediction skill, RYAN, B.F. and I.G. WATTERSON, 1992: 2005(a): The effect of Atlantic sea Wea. Forecasting, 19, 1089-1105. Tropical cyclone frequencies inferred surface temperature dipole mode from Gray’s yearly genesis parameter: on hurricanes: Implications for the MANABE S., J.L. HOLLOWAY and H.M. STONE, Validation of GCM tropical climate. 2004 Atlantic hurricane season. 1970: Tropical circulation in a time- Geophys. Res. Lett., 19, 1831-1834. Geophys. Res. Lett., 32, L03701, doi: integration of a global model of 10.1029/2004GL021702. the atmosphere. J. Atmos. Sci., 27, SAUNDERS, M.A. and A.S. LEA, 2005: 580-613. Seasonal prediction of hurricane XIE L., T. YAN, L.J. PIETRAFESA, J.M. MORRISON activity reaching the coast of and T. KARL, 2005(b): Climatology MCDONALD, R.E., D.G. BLEAKEN, the United States. Nature, 434, and interannual variability of North D.R. CRESSWELL, V.D. POPE, and 1005-1008. Atlantic hurricane tracks. J. Climate, C.A. SENIOR, 2005: Tropical storms: 18, 5370-5381. representation and diagnosis in VITART F.D., J.L. ANDERSON and W.F. STERN, climate models and the impact 1997: Simulation of interannual YAN T., L. XIE, L.J. PIETRAFESA and T.R. KARL, of climate change. Clim. Dyn., 25, variability of tropical storm frequency 2006: Spatial and temporal variability 19-36. in an ensemble of GCM integrations. of North Atlantic hurricane tracks. J. Climate, 10, 745–760. Proc. 27th Conference on Hurricanes MCDONNELL, K.A. and N.J. HOLBROOK, and Tropical Meteorology, P5.20, 2004(a): A Poisson regression VITART F.D. and T.N. STOCKDALE, 2001: Monterey, CA, April, 2006. model of tropical cyclogenesis for Seasonal forecasting of tropical the Australian-southwest Pacifi c storms using coupled GCM ZEBIAK, S.E. and M.A. CANE, 1987: A Ocean region. Wea. Forecasting, 19, integrations. Mon. Wea. Rev., 129, model El-Niño Southern Oscillation. 440-455. 2521–2537. Mon. Wea. Rev., 321, 827-832.

WMO Bulletin 56 (4) - October 2007 | 309