Abstract

The 2000 tropical cyclone season over the South Indian Ocean (SIO) was exceptional in terms of tropical cyclone landfall over Mozambique. Observed data suggest that SIO tropical cyclones have a track significantly more zonal during a La Niña event and tend to be more frequent when local SSTs are warmer. The combination of both conditions happened during the 2000 SIO tropical cyclone season and may explain the exceptional number of tropical cyclone landfalls over Mozambique during that season. A set of experiments using an atmospheric model of fairly high resolution (TL159, with a Gaussian grid spacing of 1.125°) forced by prescribed SSTs confirms the role of La Niña conditions and warmer local SSTs on the frequency of tropical cyclone landfalls over Mozambique. This also suggests that a numerical model can simulate the mechanisms responsible for the exceptional 2000 tropical cyclone season, and therefore could be used to explicitly predict the risk of landfall over Mozambique.

A coupled model with a TL159 atmospheric component has been integrated for 3 months starting on 1 January of each year 1987–2001 to test this hypothesis. The hindcast produces significantly more tropical cyclone landfalls in 2000 than in any other year, and years with a predicted high risk of landfall generally coincide with years of observed tropical cyclone landfall.

1. Introduction

The 2000 tropical cyclone season over the South Indian Ocean (SIO) was exceptional in term of tropical cyclone landfall over Mozambique, with serious consequences. In February 2000, Mozambique was already devastated by severe flooding after two weeks of steady rain before the landfall of Tropical Cyclone Eline with hurricane intensity (maximum sustained wind larger than 32 m s–1) on 25 February 2000. Eline brought more than 20 cm of precipitation per day over the already flooded region, creating a catastrophic situation. An estimated 700 people died and 2 million people were affected by the flood. At the beginning of April, a second tropical cyclone with hurricane intensity (Hudah) made a landfall over the northern part of Mozambique, but fortunately without the devastating consequences of Tropical Cyclone Eline.

Although tropical cyclones cannot be blamed for all the flooding over Mozambique, they played a dramatic role in worsening an already serious situation. Such strong tropical cyclone activity over Mozambique (two landfalls) is rare. According to historical records (Neumann et al. 1993), Mozambique had been hit by more than one tropical cyclone only twice in the previous 60 yr (1962 and 1988), and never by more than one tropical cyclone with hurricane intensity. In most years, Mozambique does not suffer any tropical cyclone landfall, since tropical cyclones tend to have a southward recurvature well before reaching the Africa coast, thus sparing Mozambique. Therefore, the 2000 tropical cyclone season over the SIO can be viewed as exceptional and the seasonal forecast of tropical cyclone landfall over Mozambique could be a valuable tool for water resource management.

It would be excessively optimistic to think that the occurrence of one specific landfall could be predicted months in advance, since the synoptic conditions associated with that particular landfall have a predictability of the order of a few days at most. However, large-scale conditions, which might be predicted months in advance, can impact the frequency and tracks of tropical cyclones, and as a consequence the risk of landfall. The present paper will evaluate the predictability of the risk of landfall over Mozambique, using both observations and numerical model integrations. In section 2, a 51-yr record of observed tropical cyclone tracks from Neumann et al. (1993) and reanalyses data from the National Centers for Environmental Prediction (NCEP; Kalnay et al. 1996) will be used in order to identify predictors and potential physical mechanisms responsible for the year-to-year change in the risk of landfall over Mozambique. Observations are not sufficient to prove all the relations, but lead into model studies in the third section, which will describe sensitivity experiments using an atmospheric general circulation model (AGCM) forced by prescribed sea surface temperatures (SSTs). The goal of these experiments is to isolate the impact of the different predictors on the simulated tropical cyclones. This will demonstrate the validity of the physical mechanisms identified in the observational study and confirm that they can be reproduced in a numerical model. Section 4 will evaluate the skill of a coupled GCM to explicitly predict the risk of landfall over Mozambique. The last section will present a discussion of the main results of this paper.

2. Observations

In the present paper, tropical cyclones are defined as tropical cyclonic systems with a maximum sustained wind larger than 17 m s–1. The SIO is defined as the region west of 110°E. SIO tropical cyclones are defined as the tropical cyclones with at least a portion of their track within the SIO, though the genesis location does not necessarily need to be within SIO. Most SIO tropical cyclones occur during a specific period of the year—from September to April; the tropical cyclone season 2000, for example, refers to the period September 1999 to April 2000. All the statistics concerning observed tropical cyclones have been obtained from the Joint Typhoon Warning Center (JTWC, Guam), and Neumann et al. (1993). The tropical cyclone record over the SIO prior to the use of meteorological satellites in the 1960s may be less reliable, since they were based mostly on ship reports. On the other hand, these statistics do not display any obvious difference between the periods prior to and after the introduction of satellite data, as is the case over the eastern North Pacific. The statistics concerning tropical cyclone landfall over Mozambique since 1950 may be reliable, because of the large population living in the area.

A comparison between the tropical cyclone tracks in 1998, 1999, and 2000 (Fig. 1) indicates that SIO tropical cyclone tracks can display a significant variability from one season to another. In 2000, the tropical cyclone tracks were generally more zonal than in 1998, where all the tracks west of 80°E have a strong southward component. In 2000, the Tropical Cyclones Eline and Hudah had a genesis in the eastern edge of the basin with a landfall over Mozambique after weeks of a very unusual zonal track. In 1999, tropical cyclones’ tracks were also very zonal, but the number of intense tropical cyclones (maximum sustained wind larger than 34 m s–1) was significantly lower than climatology in 1999 and significantly higher in 2000. In 1998, the tropical storm tracks were more poleward than in 2000, and most tropical storms with a genesis location in the eastern part of the basin recurved east of Madagascar. None of these storms crossed Madagascar. However, a few tropical cyclone geneses took place over the Mozambique Channel, but these storms had a southeastward track most of the time. In this section, we will discuss what factors can explain the interannual variability in tropical cyclone frequency and tracks in the SIO, and how this impacts the risk of landfall over Mozambique using observed data.

Fig. 1.

Tropical storm tracks observed during the tropical cyclone seasons of (top) 2000, (middle) 1999, and (bottom) 1998. Squares represent the tropical cyclone’s first position. This figure suggests that the tracks of tropical cyclones over the South Indian Ocean can display a significant variability between two seasons

Fig. 1.

Tropical storm tracks observed during the tropical cyclone seasons of (top) 2000, (middle) 1999, and (bottom) 1998. Squares represent the tropical cyclone’s first position. This figure suggests that the tracks of tropical cyclones over the South Indian Ocean can display a significant variability between two seasons

The interannual variability of tropical cyclone statistics can often be related to the interannual variability of the large-scale circulation. The frequency of tropical cyclones is sensitive to the vertical shear of the horizontal wind, defined as the amplitude of the difference between the wind in the upper and the lower troposphere (see, e.g., Gray 1979; Frank 1987; Vitart and Anderson 2001) and SST anomalies (Saunders and Harris 1997; Goldenberg and Shapiro 1996; Vitart and Anderson 2001). Low vertical wind shear and warm SST anomalies are conducive to more tropical cyclone activity. Variability in tropical cyclone tracks has also been related to the large-scale circulation. To a first order, tropical cyclone motion is attributed to the environmental wind circulation. Dong and Neumann (1986) indicate the optimum level of 400 mb for hurricanes and 700 mb for tropical cyclones for the Atlantic basin. However, vertical means over various depths of the atmosphere are usually considered for modeling the tracks of tropical cyclones rather than using a single level (Holland 1983).

a. Impact of the zonal steering flow

The zonal wind from the NCEP reanalysis (Kalnay et al. 1996) has been vertically averaged from 200 to 850 mb over the SIO tropical cyclone main development region (10°–25°S, 40°–100°E) and has been averaged over the period from 1 January to 31 March (the peak period of the SIO tropical cyclone season). It will henceforth be referred to as the SIO zonal steering flow. Figure 2 displays the interannual variability of the SIO zonal steering flow from 1950 to 2000. Considering one single level instead of averaging over a vertical column would produce similar results. The SIO zonal steering flow displays a large interannual variability. It is not clear if the discontinuity around 1980 is due to a change in the quality of the atmospheric data or due to an interdecadal variability of the atmospheric circulation.

Fig. 2.

Interannual variability of the zonal steering flow (defined as the zonal mean flow averaged from 200 to 850 mb) over the SIO (10°–25°S, 40°–100°E) averaged from 1 Jan to 31 Mar. The mean is close to 0 m s–1. Crosses indicate years with landfall, squares years with La Niña conditions, and diamonds years with El Niño conditions. El Niño and La Niña seasons are defined as the seasons when the Niño-3.4 SSTs averaged over the period Jan–Mar are one std dev above and below the 1950–2000 climatology

Fig. 2.

Interannual variability of the zonal steering flow (defined as the zonal mean flow averaged from 200 to 850 mb) over the SIO (10°–25°S, 40°–100°E) averaged from 1 Jan to 31 Mar. The mean is close to 0 m s–1. Crosses indicate years with landfall, squares years with La Niña conditions, and diamonds years with El Niño conditions. El Niño and La Niña seasons are defined as the seasons when the Niño-3.4 SSTs averaged over the period Jan–Mar are one std dev above and below the 1950–2000 climatology

In order to evaluate if the interannual variability of the zonal steering flow is strong enough to generate an interannual variability of tropical cyclone motion, the mean tropical cyclone tracks have been evaluated over each SIO tropical cyclone season, when the zonal steering flow is at least one standard deviation above or below average (not shown). The direction of motion of each mean tropical cyclone track has been defined as the difference of longitude between day 5 and day 1 divided by the difference of latitude between day 5 and day 1. The Kolmogorov–Smirnov test (KS test; Knuth 1981) indicates that the direction of motion of the mean tropical cyclone tracks for the low steering-flow cases is significantly more zonal than in the high steering-flow cases with a confidence level higher than 97%.

This difference in tropical storm tracks is likely to impact significantly the spatial distribution of tropical cyclone days. Figure 3 displays a composite of the tropical cyclone days calculated from tropical cyclones with a genesis location between 50° and 120°E and over all the seasons when the zonal steering flow is at least one standard deviation below average (Fig. 3a) and at least one standard deviation above average (Fig. 3b). When the zonal steering flow is high, SIO tropical cyclones with a genesis location east of 50°E tend to recurve east of Madagascar and very few cross the island (Fig. 3b). However, when the zonal steering flow is low (Fig. 3a), the number of tropical cyclone days over the Mozambique Channel due to tropical cyclones crossing Madagascar (like Eline in 2000) is significantly higher than when the zonal steering flow is high (Fig. 3c). At this location, the amplitude of the difference represents 70% of the climatology, suggesting that the zonal steering flow has a significant impact on the tropical cyclone activity off the coast of Mozambique. Increased tropical cyclone activity to the west of Madagascar is likely to increase the risk of landfall over Mozambique.

Fig. 3.

Observed tropical cyclone days per season on a 4° × 4° latitude grid when the zonal steering flow is (a) 1 std dev or more below climatology (defined over the period 1950–2000) and (b) 1 std dev or more above climatology. (c) The difference between (a) and (b). The contour interval is 0.5 tropical cyclone day and the first contour is 0.5 tropical cyclone day. Shaded areas represent regions with more than 1 tropical cyclone day. Only tropical storms with a genesis location between 50° and 120°E have been considered

Fig. 3.

Observed tropical cyclone days per season on a 4° × 4° latitude grid when the zonal steering flow is (a) 1 std dev or more below climatology (defined over the period 1950–2000) and (b) 1 std dev or more above climatology. (c) The difference between (a) and (b). The contour interval is 0.5 tropical cyclone day and the first contour is 0.5 tropical cyclone day. Shaded areas represent regions with more than 1 tropical cyclone day. Only tropical storms with a genesis location between 50° and 120°E have been considered

Figure 3 does not include the tropical cyclones with a genesis location within the Mozambique Channel (west of 50°E) as these storms, which represent about 10% of the total number of SIO tropical cyclones, are usually short-lived and weak (see, e.g., Figs. 1b,c). Therefore, they are usually less dangerous for Mozambique than the hurricanes with a genesis location east of 50°E, like Eline. Over the past 60 yr, Mozambique has been hit only once (in 1988) by a hurricane with a genesis location in the Mozambique Channel. The zonal steering flow does not seem to significantly impact the statistics of these tropical cyclones. However, even by including them, the total number of tropical cyclone days over the Mozambique Channel is still significantly higher with low steering flow than with high steering flow.

The years with tropical cyclone landfall are marked with a cross in Fig. 2. According to the KS test, years with landfall display a zonal steering flow significantly more negative (95% confidence level) than years without landfall. During the past 50 yr, tropical cyclones have hit Mozambique during 25% of the total number of years. When the zonal steering flow is at least one standard deviation below average, the percentage increases to 58%, but is reduced to 11% when the zonal steering flow is at least one standard deviation above average. This suggests that the risk of landfall over Mozambique is indeed related to the zonal steering flow, with a probability about 5 times higher when the zonal steering flow is more than one standard deviation below average than when it is more than one standard deviation above average. If the zonal steering flow could be predicted a month or more in advance, it could be used as a predictor for the risk of landfall over Mozambique.

b. Impact of SIO SSTs

Several observational studies (Saunders and Harris 1997; Goldenberg and Shapiro 1996) and model experiments (Vitart and Anderson 2001) have demonstrated that warmer SSTs are conducive to a significant increase of tropical cyclone activity over the Atlantic. SSTs have also an effect on the frequency of tropical cyclones over the SIO. Jury et al. (1999) uses SSTs averaged over the region 8°–22°S and 50°–70°E as a predictor for the frequency of tropical cyclones over the SIO. Xie et al. (2002) demonstrate that warmer SSTs produced by the propagation of oceanic Rossby waves across the SIO lead to significantly more tropical cyclone days.

All the SIO tropical cyclone seasons have been classified according to the zonal steering flow and anomalous SST averaged over the region where most tropical cyclones originate (8°–22°S; 50°–110°E) and over the period January to March (Fig. 4a). The years with landfall are marked with a square. According to Fig. 4a, there is a strong correlation between SSTs and zonal steering flow. Years with more positive zonal steering flow coincide with years with high SIO SSTs. As discussed in the previous section, the percentage of years with landfall increases with more negative zonal steering flow. Overall, years with higher SSTs do not display more frequent landfalls. However, when the zonal steering flow is more negative, the percentage of years with landfall increases with warmer SSTs. If we consider only the years when the zonal steering flow is at least one standard deviation below average, then years with landfall display significantly warmer SSTs with a confidence level of 90% according to the KS test. Figure 4 has been divided into four regions. Region I is the region with low zonal steering flow (at least one standard deviation below average) and positive SST anomalies (calculated from the 1950–2000 climatology). During the past 51 yr, only four seasons are within this domain. Three of them had a landfall over Mozambique (75% of the cases). Over region II (steering flow at least one standard deviation below average and negative SST anomalies), the percentage of years with landfall decreases to 36%, which is about half that in region I. The difference of tropical cyclone landfall statistics between the two La Niña seasons 1999 and 2000 could be explained by the fact that 2000 is in region I, whereas the inactive tropical cyclone season 1999 is in region II. Over regions III and IV, no signal is evident. This result suggests that the impact of SSTs is particularly important when the steering flow is favorable for the possibility of landfall.

Fig. 4.

(a) Distribution of all the tropical cyclone seasons from 1950 to 2000 as a function of the zonal steering flow over the SIO (x axis) and SIO SST anomaly (y axis). Squares represent years with a landfall over Mozambique and circles years without a landfall. Four regions are defined. In region I, the SST anomaly is greater than 0 and the zonal steering flow is at least 1 std dev below climatology. This region displays the largest percentage of seasons with landfalls, and 2000 is within this region. Region II corresponds to zonal steering flow below at least 1 std dev and negative SST anomalies. The 1999 tropical cyclone season is in this region. Regions III and IV correspond to a zonal steering flow above about –0.9 m s–1. These regions display the lowest frequency of tropical cyclone landfall. (b) Distribution of all the El Niño (circles) and La Niña seasons (squares) from 1950 to 2000 as a function of SIO SST anomaly (y axis) and zonal steering flow over the SIO (x axis)

Fig. 4.

(a) Distribution of all the tropical cyclone seasons from 1950 to 2000 as a function of the zonal steering flow over the SIO (x axis) and SIO SST anomaly (y axis). Squares represent years with a landfall over Mozambique and circles years without a landfall. Four regions are defined. In region I, the SST anomaly is greater than 0 and the zonal steering flow is at least 1 std dev below climatology. This region displays the largest percentage of seasons with landfalls, and 2000 is within this region. Region II corresponds to zonal steering flow below at least 1 std dev and negative SST anomalies. The 1999 tropical cyclone season is in this region. Regions III and IV correspond to a zonal steering flow above about –0.9 m s–1. These regions display the lowest frequency of tropical cyclone landfall. (b) Distribution of all the El Niño (circles) and La Niña seasons (squares) from 1950 to 2000 as a function of SIO SST anomaly (y axis) and zonal steering flow over the SIO (x axis)

c. Impact of the vertical wind shear

The vertical wind shear is known to have a significant impact on Atlantic tropical cyclones (Gray 1984; Goldenberg and Shapiro 1996; Vitart and Anderson 2001) and explains mostly why the Atlantic tropical storm activity is significantly reduced during El Niño events. The impact of the vertical wind shear on the tropical cyclone frequency is due to the ventilation of the warm core above the center of the cyclone. A figure similar to Fig. 4 has been produced (not shown), but this time the tropical cyclone seasons have been classified according to vertical wind shear and steering wind. Unlike with SSTs, if we consider only the years with zonal steering flow one standard deviation below average, there is no significant difference of vertical wind shear between the years with tropical cyclone landfalls and those without. Therefore, only the steering wind and the SIO SSTs will be considered as predictors for the risk of landfall over Mozambique in the rest of this paper.

d. Role of El Niño–Southern Oscillation (ENSO)

ENSO is the most important source of interannual variability in the Tropics. It is often used as a predictor for tropical cyclone activity over the Atlantic (Gray et al. 1992), the western North Pacific (Chan et al. 1998) and the Australian basin (Nicholls 1992). The present section will investigate its impact on the zonal steering flow and SIO SSTs. In the present paper, El Niño seasons are defined as the seasons when the Niño-3.4 (5°N–5°S, 120°–170°W) SSTs averaged over the period January–March are at least one standard deviation above the 1950–2000 climatology, and La Niña seasons when Niño-3.4 SSTs are at least one standard deviation below climatology. The results presented in this paper are not sensitive to the choice of the ENSO index. The choice of Niño-3.4 was motivated by the fact that it is a widely used index for ENSO activity. The definition of ENSO years may not be consistent with other definitions in which an ENSO year is defined as the year where the ENSO event starts, as in Bove et al. (1998) for instance. Following our definition, there have been 10 El Niño seasons and 11 La Niña seasons over the period 1950–2000. In Fig. 2, El Niño years are marked with diamonds and La Niña years with squares. It appears clearly that El Niño seasons generally correspond to positive anomalies of the zonal steering flow, whereas La Niña seasons generally correspond to negative anomalies. This is confirmed by a KS test, which indicates that the 10 El Niño years display a zonal steering flow significantly more positive than La Niña years with a confidence level larger than 95%. The linear correlation between the interannual variability of zonal steering flow and the Niño-3.4 SSTs is 0.7 (99.99% level of confidence). The impact of ENSO on the zonal steering flow over the SIO suffers some exceptions, such as 1989, which was a La Niña season but with a positive zonal steering flow over the SIO, or 1990 which was not an El Niño season, but which displayed a very high and positive zonal steering flow. Therefore there is no perfect match between ENSO and the zonal steering flow over the SIO, but El Niño significantly increases the probability of a positive zonal steering flow, whereas La Niña increases the risk of a negative zonal steering flow. The next section will demonstrate that the strong link between ENSO and the zonal steering flow is due to a remote impact of Pacific equatorial SSTs on the wind circulation over the SIO. Because of its significant link with the zonal steering flow, ENSO is likely to impact the risk of landfall over Mozambique. Historical records of tropical cyclones over the SIO show that tropical cyclone landfalls occurred during 25% of all the tropical cyclone seasons. This percentage increases to 45% during La Niña seasons and reduces to 10% during El Niño seasons. Therefore, the probability of landfall over Mozambique seems to be 4.5 times higher during a La Niña year than during an El Niño year. Although the data is too short to prove its significance, this result suggests that ENSO impacts the risk of landfall over Mozambique. Because of its link with the zonal steering flow, ENSO could replace the zonal steering flow as a predictor for the risk of landfall over Mozambique.

ENSO also impacts local SSTs over the SIO. Xie et al. (2002) discuss the interannual variability of SIO SSTs, and found that ENSO is the dominant forcing for SIO thermocline variability. When an El Niño event takes place, anomalous easterlies appear in the equatorial Indian Ocean, forcing a westward-propagating downwelling oceanic Rossby wave in the SIO. This explains why the SSTs during the 10 El Niño seasons are significantly warmer than during the 11 La Niña seasons, with a level of confidence of 95% according to the KS test. The impact of ENSO on both the zonal steering flow and SIO SSTs explains the significant correlation (0.6) between these two quantities (e.g., Fig. 4a).

El Niño and La Niña seasons since 1950 have been classified according to SST anomalies and the zonal steering flow in a similar way to Fig. 4a (Fig. 4b). It appears clearly that El Niño years largely outnumber La Niña years in region IV, whereas La Niña seasons greatly outnumber El Niño seasons in regions I and II, where the probability of landfall is higher. However, if we consider only regions I and II, La Niña seasons do not display significantly lower SSTs than El Niño seasons of the same region. Xie et al. (2002) identified SST variability off Sumatra, which is not necessarily linked to ENSO, as having an impact on the SST variability over the SIO. Therefore, some of the SIO SST variability is not related to ENSO. This variability may explain why La Niña years like 1996 and 2000 displayed warmer SSTs than expected for a La Niña season.

In summary, La Niña conditions over the tropical Pacific and positive SST anomalies in the SIO seem to be conducive to an increase of tropical cyclone landfall over Mozambique. However, the data length (51 yr) is too short to prove this result, because only 2 yr (both with landfalls) fulfill both conditions. There is also some uncertainty in the quality of the NCEP reanalysis data. However, a recent 40-yr reanalysis from the European Centre for Medium-Range Weather Forecasts (ECMWF) produces a year-to-year variability in the SIO winds that is in good agreement with the NCEP reanalysis. In addition, the links between ENSO, the zonal steering flow, and tropical cyclone landfall seem to be stronger in the more recent part of the analysis, when the data are of better quality. The use of GCM integrations supports the overall results, as will be discussed in the next section.

3. Sensitivity studies using a high-resolution AGCM

A series of sensitivity studies using an AGCM forced by prescribed SSTs has been performed in order to evaluate the impact of SSTs on the tropical cyclones simulated by the AGCM. The model resolution is TL159L40, which represents a spectral resolution of T159 in the free atmosphere and a Gaussian grid spacing of 1.125° × 1.125° at the surface with 40 vertical levels. The AGCM (IFS cycle 23r4) has been integrated for 3 months forced by prescribed SSTs in order to evaluate the sensitivity of model tropical storms over the SIO to SSTs. Since the length of the model integrations exceeds the limits of deterministic predictability, a single model integration is certainly not enough to evaluate the risk of landfall over Mozambique. Therefore, the AGCM has been integrated with 20 different initial conditions, taken from the ECMWF operational analysis from 15 November to 21 December 1999, at 2-day intervals. In order to remove the effect of the initial state, the first 2 weeks of the forecasts have been discarded. Since the starting dates are not the same, the period covered by the integration varies from one member of the ensemble to another, but this is unlikely to generate significant differences in tropical cyclone numbers.

In order to clarify the role of SSTs in the Indian and Pacific Ocean basins, a series of experiments with different SSTs was conducted. For this purpose, a series of 20-member ensemble runs as described earlier was conducted. The initial conditions vary from one member of the ensemble to another, but are the same in all the sets of experiments. The key aspect of the design of the sets of experiments is that the only difference between the experiments is the SST forcing. Therefore all the significant differences in tropical storm statistics and large-scale circulation between the 20-member ensemble integrations can only be attributed to the difference in SSTs.

a. Impact of ENSO

To evaluate the impact of ENSO on model tropical cyclones, a 20-member integration of the AGCM forced by SSTs of 1999/2000 (EXP2000) has been compared to a 20-member ensemble forced by SSTs of 1997/98 (EXP1998). In EXP1998, the atmospheric model has been forced by observed SSTs during one of the strongest El Niño events on record. In EXP2000, the prescribed SSTs correspond to La Niña conditions. The present section will explore the differences in the large-scale circulation and the tropical storm statistics in the two sets of experiments. However, SSTs outside the tropical Pacific region could also have an impact on the statistics of tropical cyclones. Therefore, in order to isolate the impact of tropical Pacific SSTs, two additional 20-member ensembles have been generated. In EXP2000_Pac, the prescribed SSTs correspond to the 1997/98 season everywhere, except over the tropical Pacific (20°N–20°S, 120°–200°E) where SSTs of 1999/2000 have been imposed. Therefore, the only difference between EXP1998 and EXP2000_Pac is the SST forcing over the tropical Pacific. EXP1998_Pac has the opposite setting to EXP2000_Pac—that is, SSTs of 1999/2000 are used everywhere except over the tropical Pacific, where SSTs of 1997/98 are imposed. A summary of the experiment design is displayed in Table 1.

Table 1.

Description of the experiments EXP2000, EXP1998, EXP2000_Pac, and EXP1998_Pac

Description of the experiments EXP2000, EXP1998, EXP2000_Pac, and EXP1998_Pac
Description of the experiments EXP2000, EXP1998, EXP2000_Pac, and EXP1998_Pac

1) Zonal steering flow

Figure 5a displays the difference in the NCEP reanalysis between the zonal steering flow averaged over the period December–January–February 2000 and the zonal steering flow averaged over the period December–January–February 1998. Over the SIO, the zonal steering flow is weaker in 2000 than in 1998 over the main tropical cyclone development region (between 10° and 20°S). It is not clear at this point if this is a consequence of ENSO or it is due to other factors, such as local SSTs. The difference between the zonal steering flow simulated by EXP2000 and EXP1998 (Fig. 5b) displays a pattern that is very similar to the one from the NCEP reanalysis, although its amplitude over the eastern Pacific seems to be underestimated by the model. Over the SIO, EXP2000 simulates a zonal steering flow significantly lower than in EXP1998, with a confidence level larger than 95% according to the KS test.

Fig. 5.

Difference of zonal steering flow between (a) the zonal steering flow (m s–1) averaged over the period from 1 Dec 1999 to 1 Mar 2000 and the zonal steering flow averaged from 1 Dec 1997 to 1 Mar 1998 in the NCEP reanalysis, (b) EXP2000 minus EXP1998, (c) EXP1998_Pac minus EXP1998, and (d) EXP2000_Pac minus EXP1998. Dotted lines represent negative values. As shown, both EXP2000 and EXP2000_Pac simulate similar patterns with a significant reduction of zonal steering flow over the main SIO tropical storm development region. On the other hand, the difference between EXP1998_Pac and EXP1998 is much smaller in amplitude than between EXP2000 and EXP1998, suggesting that tropical Pacific SSTs are responsible for most of these anomalous patterns. In (b)–(d), regions where the KS test detects a confidence level in the difference lower than 95% are blanked out

Fig. 5.

Difference of zonal steering flow between (a) the zonal steering flow (m s–1) averaged over the period from 1 Dec 1999 to 1 Mar 2000 and the zonal steering flow averaged from 1 Dec 1997 to 1 Mar 1998 in the NCEP reanalysis, (b) EXP2000 minus EXP1998, (c) EXP1998_Pac minus EXP1998, and (d) EXP2000_Pac minus EXP1998. Dotted lines represent negative values. As shown, both EXP2000 and EXP2000_Pac simulate similar patterns with a significant reduction of zonal steering flow over the main SIO tropical storm development region. On the other hand, the difference between EXP1998_Pac and EXP1998 is much smaller in amplitude than between EXP2000 and EXP1998, suggesting that tropical Pacific SSTs are responsible for most of these anomalous patterns. In (b)–(d), regions where the KS test detects a confidence level in the difference lower than 95% are blanked out

In order to estimate if the difference in zonal steering flow is due to a remote impact of eastern Pacific SSTs, the zonal steering flows in EXP2000_Pac and EXP1998_Pac have been compared to the zonal steering flow in EXP1998 (Figs. 5c and 5d). The difference between the zonal steering flow of EXP2000_Pac and EXP1998 is similar to the difference between EXP2000 and EXP1998. In particular the zonal steering flow over the SIO is significantly lower in EXP2000_Pac than in EXP1998. Since the only difference between EXP2000_Pac and EXP1998 is the tropical Pacific SSTs, this suggests that the reduction of zonal steering flow simulated by the model between 1999/2000 and 1997/98 is due to a teleconnection from the Pacific. EXP1998_Pac and EXP1998 have the same SSTs over the tropical Pacific, but different SSTs elsewhere. EXP1998_Pac does not display a significant decrease of zonal steering flow compared to EXP1998 (Fig. 5c), indicating that SSTs outside the tropical Pacific have a small effect and supporting the result that the significant reduction of zonal steering flow between EXP2000 and EXP1998 is due to a remote impact of tropical Pacific SSTs, and, therefore, to ENSO.

2) Tropical storm statistics

The model tropical cyclones are tracked using an improved version of the objective procedure described in Vitart and Stockdale (2001). In the present paper, all the criteria need to be verified only twice along the whole trajectory, instead of all times, and the maximum permitted distance between two successive tropical cyclone positions has been increased. These changes improve the realism of the tropical cyclone tracks and help to capture portions of the tracks where the tropical cyclone has the intensity of a tropical depression. In order to evaluate its skill in detecting tropical cyclones, the objective procedure has been applied to 9 months of ECMWF operational analyses interpolated to the same resolution as the AGCM from 1 September 1999 to 1 June 2000 (the 2000 tropical cyclone season). Figure 6 displays the tropical cyclone tracks detected by the new objective procedure, and can be compared to the top panel of Fig. 1, which displays the tropical cyclone tracks from JTWC during the same period. The tropical cyclone tracking procedure detects 9 of the 10 observed tropical cyclones during the 2000 SIO tropical cyclone season. Tropical Cyclone Kirrily was not detected because it is too weak in the ECMWF operational analysis. On the other hand, the procedure detects two additional systems that were classified as tropical depressions by JTWC but have tropical storm intensity in the ECMWF operational analysis. In summary, the objective procedure seems to detect tropical cyclones in the analysis rather well. The two false alarms and the nondetection of Tropical Cyclone Kirrily are more likely due to inconsistencies between ECMWF operational analyses and JTWC’s tropical storm statistics, than due to deficiencies in the tropical cyclone detection algorithm. The tracks of the detected tropical cyclones look consistent with observed tracks. Most importantly for the present paper, the two landfalls of Leon-Eline and Hudah are well detected by the objective procedure (Fig. 6), and no nonobserved landfall was detected.

Fig. 6.

Tropical cyclone tracks over the SIO detected when applying the objective procedure for tracking tropical cyclones to the ECMWF operation analysis interpolated to a 1.125° × 1.125° horizontal grid from Sep 1999 to Jun 2000

Fig. 6.

Tropical cyclone tracks over the SIO detected when applying the objective procedure for tracking tropical cyclones to the ECMWF operation analysis interpolated to a 1.125° × 1.125° horizontal grid from Sep 1999 to Jun 2000

The tropical cyclones simulated by the AGCM have been tracked using the algorithm mentioned earlier. Table 2 displays the frequency of tropical cyclones simulated in the different experiments. Table 3 displays the observed frequency. By comparing Table 2 with Table 3, it appears that the model simulates less than half as many tropical cyclones as were observed. The low global frequency of model tropical storms in the ECMWF seasonal forecasting system has been discussed in Vitart and Stockdale (2001), and is most probably caused by the cumulus parameterization of the AGCM.

Table 2.

Tropical cyclone frequency, number of landfall, and ratio of number of landfall with the total number of tropical cyclones over the SIO region. All these numbers have been averaged over the 20-member ensemble. The numbers in parentheses delimit the 95% confidence interval, calculated by assuming a Poisson distribution of tropical storm frequency and landfalls and using a standard conservative method

Tropical cyclone frequency, number of landfall, and ratio of number of landfall with the total number of tropical cyclones over the SIO region. All these numbers have been averaged over the 20-member ensemble. The numbers in parentheses delimit the 95% confidence interval, calculated by assuming a Poisson distribution of tropical storm frequency and landfalls and using a standard conservative method
Tropical cyclone frequency, number of landfall, and ratio of number of landfall with the total number of tropical cyclones over the SIO region. All these numbers have been averaged over the 20-member ensemble. The numbers in parentheses delimit the 95% confidence interval, calculated by assuming a Poisson distribution of tropical storm frequency and landfalls and using a standard conservative method
Table 3.

Observed tropical statistics over the SIO region

Observed tropical statistics over the SIO region
Observed tropical statistics over the SIO region

The frequency of SIO tropical cyclones varies greatly from one experiment to the other (Table 2). In EXP2000, the frequency of tropical storms is about twice as large as in EXP1998, as observed. The KS test applied to the 20-member ensemble indicates that the difference is significant with a confidence level exceeding 99%. EXP2000_Pac simulates significantly more tropical storms than all other experiments (confidence level of 99%). Since the only difference between EXP2000_Pac and EXP1998 is the tropical Pacific SSTs, this suggests that tropical Pacific SSTs have a significant impact on the frequency of SIO tropical storms. This is probably due to the fact that La Niña conditions over the tropical Pacific significantly reduce the vertical wind shear over the SIO in EXP2000_Pac (not shown), which creates more favorable conditions for tropical cyclone genesis. EXP2000_Pac simulates about twice as many tropical storms as EXP2000, although both experiments have the same SST forcing over the tropical Pacific. This is likely due to the warmer SSTs over the SIO in EXP2000_Pac than in EXP2000. The difference in local SSTs between these two experiments exceeds 2 K over most of the SIO, which is large enough to impact the frequency of tropical cyclones (see section 2). This is also confirmed by the very low number of tropical cyclones in EXP1998_Pac, with negative SST anomalies over the SIO and El Niño conditions over the tropical Pacific. Both conditions are conducive to less tropical cyclone activity over the SIO. This result suggests a dual impact of ENSO on SIO tropical cyclone frequency. El Niño conditions are conducive to a significant increase of vertical wind shear over the SIO (not shown) and to warmer SSTs over the SIO. The increase of vertical wind shear is conducive to less SIO tropical cyclones, whereas the increase of SIO SSTs is conducive to more tropical cyclones. These impacts are of opposite sign, and this may explain why the interannual frequency of tropical cyclones over the SIO is poorly correlated with ENSO (correlation of 0.013 over the period 1950–2000).

The tropical cyclone tracks vary significantly from EXP2000 to EXP1998 (Fig. 7). Tropical cyclones in EXP1998 tend to recurve or die east of Madagascar, whereas tropical cyclones in EXP2000 tend to cross Madagascar and sometimes make a landfall over Mozambique. Figures 8a,b display the distribution of tropical cyclone days in EXP1998 and EXP2000 after normalization by the total number of tropical cyclone days over the entire basin simulated by each experiment. Figure 8c displays the difference between the two experiments. EXP2000 displays more tropical cyclone activity west of Madagascar, but less activity east of Madagascar. The KS test applied to the ensemble distribution of the number of tropical cyclone days averaged over the region west of Madagascar (25°–15°S, 30°–50°E) indicates that the difference between EXP1998 and EXP2000 is significant with a confidence level exceeding 95%. The same test applied to the region east of Madagascar (15°–25°S, 50°–60°E) indicates that the difference in that region is also significant with a confidence level exceeding 95%. This result demonstrates that in the model, the SST forcing has a significant impact on the tropical cyclone tracks.

Fig. 7.

Tropical storm tracks simulated by the 20-member ensemble of (a) EXP2000, (b) EXP1998, (c) EXP2000_Pac, and (d) EXP1998_Pac. This figure shows a significant impact of the SST forcing on the simulated tropical cyclone distribution. The tropical cyclone activity west of Madagascar is significantly larger in EXP2000 and EXP2000_Pac than in EXP1998 and EXP1998_Pac

Fig. 7.

Tropical storm tracks simulated by the 20-member ensemble of (a) EXP2000, (b) EXP1998, (c) EXP2000_Pac, and (d) EXP1998_Pac. This figure shows a significant impact of the SST forcing on the simulated tropical cyclone distribution. The tropical cyclone activity west of Madagascar is significantly larger in EXP2000 and EXP2000_Pac than in EXP1998 and EXP1998_Pac

Fig. 8.

Distribution of tropical storm days simulated by (a) EXP2000 and (b) EXP1998, and (c) the difference of EXP2000 and EXP1998. Over each grid point, the number of tropical cyclone days has been normalized by the total number of tropical storm days simulated by each experiment over the entire SIO. The first contour is 0.02, and areas where the value exceeds 0.04 are shaded. This figure shows that tropical cyclones have significantly different tracks in EXP2000 than in EXP1998, consistent with observations in Fig. 3 

Fig. 8.

Distribution of tropical storm days simulated by (a) EXP2000 and (b) EXP1998, and (c) the difference of EXP2000 and EXP1998. Over each grid point, the number of tropical cyclone days has been normalized by the total number of tropical storm days simulated by each experiment over the entire SIO. The first contour is 0.02, and areas where the value exceeds 0.04 are shaded. This figure shows that tropical cyclones have significantly different tracks in EXP2000 than in EXP1998, consistent with observations in Fig. 3 

EXP2000_Pac exhibits similar tracks to EXP2000, whereas EXP1998_Pac exhibits tracks consistent with the tracks in EXP1998. Figures 9a,b display the distribution of tropical cyclone days in EXP2000_Pac and EXP1998. The difference between EXP2000_Pac and EXP1998 (Fig. 9c) is consistent with the difference in tropical cyclone days between EXP2000 and EXP1998 (Fig. 8c). The KS test also shows that this difference is significant. Since the only difference between EXP1998 and EXP2000_Pac is the SST forcing over the tropical Pacific, this result demonstrates that the difference in tracks between EXP1998 and EXP2000 is essentially due to the remote impact of tropical Pacific SSTs.

Fig. 9.

Same as in Fig. 8 but for EXP2000_Pac instead of EXP2000. (c) The difference between EXP2000_Pac and EXP1998 is consistent with the difference between EXP2000 and EXP1998 in Fig. 8(c)

Fig. 9.

Same as in Fig. 8 but for EXP2000_Pac instead of EXP2000. (c) The difference between EXP2000_Pac and EXP1998 is consistent with the difference between EXP2000 and EXP1998 in Fig. 8(c)

According to the previous paragraphs, the tropical Pacific SSTs have a significant impact on the frequency and tracks of tropical cyclones over the SIO. One would expect that these differences have significant consequences on the risk of landfall over Mozambique. The number of tropical cyclone landfalls over Mozambique has been counted for the four sets of experiments and the results are displayed in Table 2. When forced by SSTs of 2000 (EXP2000), 14 of the 20 AGCM integrations simulate a landfall over Mozambique, whereas only one member of the ensemble simulates a landfall when forced by SSTs of 1998 (EXP1998). The difference is significant with a confidence level of 99.9% according to a simple binomial test. In EXP2000 the proportion of landfalling storms is 19%, which is close to the observed one (Table 3). In EXP1998, the percentage of tropical cyclones with a landfall over Mozambique is just 2%. The difference in the probability of landfall between both experiments is significant, and this demonstrates that the risk of landfall over Mozambique in the model is affected by SSTs. Thus the results obtained with the model look consistent with observations, supporting the conclusions of section 2.

The difference between EXP2000_Pac and EXP1998 is even more striking, with 26 landfalls in EXP2000_Pac and just 1 in EXP1998. The only difference between the two experiments is the tropical Pacific SSTs. This demonstrates that the tropical Pacific SSTs have a significant impact on the risk of landfall over Mozambique. This is also found to be true when comparing EXP2000 and EXP1998_Pac, experiments that differ only in the tropical Pacific SSTs.

b. Impact of local SSTs

EXP2000_Pac simulates significantly more tropical cyclone landfalls over Mozambique than EXP2000, although both experiments were forced by the same SSTs over the tropical Pacific. Interestingly, the fractions of tropical cyclone landfall in EXP2000 and EXP2000_Pac are very close, near 20% (Table 2), but EXP2000_Pac simulates about twice as many tropical cyclones as EXP2000. This could be explained by warmer SSTs over the SIO in EXP2000_Pac than in EXP2000. The present section will investigate this hypothesis.

In order to evaluate the impact of local SSTs, the SIO SSTs have been warmed and cooled by 0.5 K over the main tropical cyclone development region in SIO (10°–25°S, 30°–110°E). The choice of 0.5 K was made because it is close to the amplitude of the standard deviation of SSTs over the SIO in La Niña years. Twenty-member integrations of the AGCM forced by the modified SSTs (EXP2000_SIN+0.5 and EXP2000_SIN–0.5) have been compared to EXP2000. Table 4 describes the setting of the two experiments. EXP2000_SIN+0.5 simulates significantly more tropical cyclones (103 events instead of 73) than the control experiment EXP2000 with a confidence level of 95%. The number of landfalls is slightly higher (19 instead of 14), but the percentage of tropical cyclones with landfall is about the same in both experiments (Table 5). On the other hand, cooling SIO SSTs by 0.5 K reduces the number of landfalls to 7 (instead of 14), which is significantly less than in EXP2000_SIN+0.5 (within the 95% level of confidence), although the total number of tropical cyclones is not significantly reduced compared to EXP2000. The tracks are not significantly different between the three ensemble integrations (not shown). However, tropical cyclones are significantly less intense in EXP2000_SIN–0.5, and as a consequence have a shorter duration. This may explain why the percentage of tropical cyclones with a landfall over Mozambique in EXP2000_SIN–0.5 is only half that in EXP2000.

Table 4.

Description of the settings of EXP2000, EXP2000_SIN+0.5, EXP2000_SIN–0.5

Description of the settings of EXP2000, EXP2000_SIN+0.5, EXP2000_SIN–0.5
Description of the settings of EXP2000, EXP2000_SIN+0.5, EXP2000_SIN–0.5
Table 5.

Same as in Table 2 but for the experiments described in Table 4 

Same as in Table 2 but for the experiments described in Table 4
Same as in Table 2 but for the experiments described in Table 4

c. Conclusion of the sensitivity experiments

The sensitivity experiments described in this section suggest that tropical Pacific SSTs have a significant impact on the risk of landfall over Mozambique, through their impact on the large-scale circulation. Model tropical storms tend to have a more zonal track during the 2000 La Niña season than during the 1998 El Niño season. The experiments also suggest that SIO SSTs impact the risk of landfall through the frequency and intensity of tropical cyclones. These results are consistent with the observational analysis presented in section 2. From these sensitivity experiments, it can be concluded that an AGCM can simulate the impact of SSTs on the risk of landfall over Mozambique.

4. Forecasting the risk of landfall over Mozambique

In order to evaluate the skill of a coupled GCM to predict explicitly the risk of landfall over Mozambique, a coupled ocean–atmosphere model has been integrated for 3 months starting on 1 January for each year from 1987 to 2001 (15 yr). The atmospheric component is the same as that used in section 3. In order to remove the most deterministic part of the forecast, the first 10 days of integrations have been removed. The atmospheric initial conditions have been perturbed using singular vectors (Palmer et al. 1998) and stochastic perturbations are applied to the physics tendencies (Palmer 2001) during the integrations. Oceanic initial conditions are perturbed in two ways: random perturbations are applied to the wind stress during the data assimilation in order to produce five different realizations of the ocean state; random perturbations are also applied to the SSTs in order to produce a 10-member ensemble. The SST perturbations have been created by taking the difference between two different weekly mean SST analyses from 1985 to 1999, and also by taking the differences between a weekly mean SST analysis and its 1-week persistence. In order to have a 3D structure, the SST perturbations are linearly interpolated to zero at an oceanic depth of 40 m. The SST perturbations are added to the SSTs produced by the operational ocean analyses with a plus (+) and minus (–) sign.

The objective procedure for tracking model tropical cyclones described in section 3 has been applied to each member of the ensemble from 1987 to 2001. The mean number of simulated tropical cyclones in the period from 11 January to 31 March is 4.6, which is lower than the observed mean number of tropical cyclones over the same period of time (7.5). This is likely to originate from the atmospheric component of the coupled GCM, since the AGCM used for the sensitivity experiments described in section 3 displays a similar bias in the frequency of SIO tropical cyclones (Tables 2 and 3).

The frequency of predicted SIO tropical cyclones displays an interannual variability that is positively and significantly correlated with the observed interannual variability (correlation of 0.55 between the mean of the ensemble and observations), although the model fails to predict the exceptionally strong number of SIO tropical cyclones in 1994. The interannual variability predicted by the higher-resolution forecast discussed here is more realistic than the one predicted by the operational seasonal forecasting system (Vitart and Stockdale 2001).

The coupled model predicts a significant interannual variability in tropical cyclone tracks. Figure 10 shows an example of such strong interannual variability. In Fig. 10a, the tracks of all the tropical cyclones generated by 10-member ensemble integrations of the coupled GCM starting on 1 January 2000 are displayed. They look consistent with the observations (Fig. 1a). A couple of model tropical cyclones have a westward trajectory and landfall over Mozambique as did Leon-Eline and Hudah, although their genesis location is in the central part of the SIO instead of the eastern edge of the basin. Fig. 10b shows the predicted trajectories of the ensemble forecasts starting on 1 January 1998. As in the observations, most of the tropical cyclone activity is concentrated just east of Madagascar with few tropical cyclones crossing the island.

Fig. 10.

Tropical cyclone tracks predicted by the 10-member ensemble during the 3 months of integration when starting from (a) 1 Jan 2000 and (b) from 1 Jan 1998. This figure suggests a significant interannual variability in the tracks of the tropical cyclones predicted by the coupled GCM

Fig. 10.

Tropical cyclone tracks predicted by the 10-member ensemble during the 3 months of integration when starting from (a) 1 Jan 2000 and (b) from 1 Jan 1998. This figure suggests a significant interannual variability in the tracks of the tropical cyclones predicted by the coupled GCM

The number of landfalls over Mozambique has been counted for each member of the ensemble. The mean number of landfalls (averaged over the 10 members of the ensemble) has been calculated for each year from 1987 to 2001. As the model simulates fewer tropical cyclones than observed by a factor of 1.5, the number of landfalls per year and per ensemble member has been multiplied by 1.5. The interannual variability of the predicted number of landfalls is significantly correlated with the predicted interannual variability of Niño-3.4 SSTs (correlation of 0.7). The four El Niño years (1987, 1992, 1995, and 1998) correspond to four of the years where the model predicts the lowest risk of landfall over Mozambique (Fig. 11). The coupled model predicts a risk of landfall over Mozambique to be above average during 3 of the 4 La Niña years (1996, 1999, and 2000), but not in 1989 (Fig. 11). Therefore, the impact of ENSO on the risk of landfall over Mozambique, discussed in sections 1 and 2, seems to be present in the forecast.

Fig. 11.

Interannual variability of the observed and predicted number of tropical storm landfalls over Mozambique. The predicted number of tropical storm landfalls has been calculated by taking the mean of the ensemble distribution, and multiplying it by 1.5, since the model simulates 1.5 less tropical cyclones than observed

Fig. 11.

Interannual variability of the observed and predicted number of tropical storm landfalls over Mozambique. The predicted number of tropical storm landfalls has been calculated by taking the mean of the ensemble distribution, and multiplying it by 1.5, since the model simulates 1.5 less tropical cyclones than observed

During the 15 yr of the hindcast, Mozambique has been hit by tropical cyclones in 4 yr (1988, 1994, 1996, and 2000). Interestingly, the model predicts a risk of landfall during these 4 yr as greater than average. All the years where the model predicts a reduced risk of landfall coincide with years with nonobserved landfall over Mozambique. There are 3 yr when the model predicts an above average risk of landfall but no tropical cyclone landfall was observed (1990, 1991, and 1999). However, the seasonal forecasts are probabilistic forecasts, not deterministic. An increased risk of landfall does not necessarily mean that there will be a landfall over Mozambique, especially when the predicted number of landfalls is less than 1. The model successfully predicts a high risk of landfall over Mozambique for the 2000 season, higher than in any other year. Interestingly, it also predicts a high risk of landfall in 1988 and 1994, which are not La Niña years. Both 1988 and 1994 seasons displayed a tropical cyclone landfall over Mozambique. Therefore, it seems that although ENSO has a significant impact on the predicted risk of landfall, other factors, such as local SST patterns, may play a significant role. The predicted high risk of landfall over Mozambique in 1988 and 1994 suggests that the model is able to predict the impact of these additional factors on the risk of landfall over Mozambique. The predicted interannual variability of tropical cyclone landfall is positively and significantly correlated with the observed interannual variability (correlation of 0.81 with a level of confidence of 99%). In summary, this hindcast experiment strongly suggests that the high-resolution coupled system has skill in predicting the number of landfalls over Mozambique.

5. Conclusions

Observational studies of section 2 suggest, but do not entirely prove, that two factors significantly impact the risk of landfall over Mozambique: ENSO and SIO SSTs. Sensitivity tests in section 3 using an AGCM reproduce this impact and therefore confirm the importance of ENSO and SIO SSTs on the risk of landfall over Mozambique. This demonstrates that an AGCM can be a very useful tool for quantifying how the variability in the large-scale circulation impacts the statistics of tropical cyclones. The observational study and the sensitivity experiments demonstrate that the skill of a dynamical model in predicting the risk of landfall over Mozambique depends strongly on its skill in predicting SSTs over the SIO and tropical Pacific.

The present paper shows that a coupled GCM with sufficiently high horizontal resolution has skill in predicting the risk of landfall over Mozambique. More especially, the coupled model predicts a number of landfalls for the 2000 tropical cyclone season that is higher than in any of the other 14 yr of the hindcast period. This result gives a strong indication that the risk of landfall over Mozambique can be predicted, and that coupled GCM integrations could be a useful tool for predicting such risk. The TL159 horizontal resolution that was used in the present paper seems fine enough to allow the explicit seasonal probabilistic prediction of tropical cyclone landfalling. Such a high atmospheric resolution is likely to be used for operational seasonal forecasting in the coming years. Therefore, there is hope that in the near future, operational dynamical seasonal forecasting systems will be able to explicitly predict the seasonal risk of landfall over Mozambique.

Current operational seasonal forecasting systems have a horizontal resolution much coarser than the one used in the experiments described in the present paper. At such low resolution, tropical storm tracks tend to be too short and poleward (Vitart et al. 1997). Therefore, it is unlikely that these operational dynamical systems have skill in predicting explicitly the risk of landfall over Mozambique. However, a study not reported in the present paper indicates that the ECMWF seasonal forecasting system with a T63 horizontal resolution displays strong skill in predicting the large-scale predictors (ENSO and SIO SSTs) a few months in advance. This suggests that current seasonal forecasting systems could be useful for the indirect seasonal prediction of the risk of landfall over Mozambique.

The present study focuses on the SIO and the risk of landfall over Mozambique prompted by the exceptional 2000 tropical cyclone season and its catastrophic consequences for Mozambique. Future plans include investigating if a coupled model could also be useful in predicting the risk of landfall over other areas. It may be possible to predict the risk of landfall over several basins such as the North Indian Ocean, the western North Pacific, and the eastern North Pacific, and North Atlantic where there is significant interannual variability of tropical cyclone statistics. For that purpose, additional GCM integrations will be needed, since the tropical cyclones seasons over these basins differ from the SIO tropical cyclone season.

Acknowledgments

The authors would like to thank the anonymous reviewers whose comments proved invaluable in improving the presentation of the material.

REFERENCES

REFERENCES
Bove
,
M. C.
,
J. B.
Elsner
,
C. W.
Landsea
,
X.
Niu
, and
J. J.
O’Brien
,
1998
:
Effect of El Niño on U.S. landfalling hurricanes, revisited.
Bull. Amer. Meteor. Soc.
,
79
,
2477
2482
.
Chan
,
J. C. L.
,
J. E.
Shi
, and
C. M.
Lam
,
1998
:
Seasonal forecasting of tropical cyclone activity over the western North Pacific and the South China Sea.
Wea. Forecasting
,
13
,
997
1004
.
Dong
,
K.
, and
C. J.
Neumann
,
1986
:
The relationship between tropical cyclone motion and environmental geostrophic flows.
Mon. Wea. Rev.
,
114
,
115
222
.
Frank
,
W. M.
,
1987
:
Tropical cyclone formation.
A Global View of Tropical Cyclones, R. L. Elsberry, Ed., Office of Naval Research, 53–90
.
Goldenberg
,
S. G.
, and
L. J.
Shapiro
,
1996
:
Physical mechanism for the association of El Niño and west African rainfall with Atlantic major hurricane activity.
J. Climate
,
9
,
1169
1187
.
Gray
,
W. M.
,
1979
:
Hurricanes: Their formation, structure and likely role in the tropical circulation.
Meteorology over the Tropical Oceans, D. B. Shaw, Ed., Royal Meteorological Society, 155–218
.
Gray
,
W. M.
,
1984
:
Atlantic seasonal hurricane frequency. Part I: El Niño and 30 mb quasi-biennal oscillation influences.
Mon. Wea. Rev.
,
112
,
1649
1668
.
Gray
,
W. M.
,
C. W.
Landsea
,
P. W.
Mielke
Jr.
, and
K. J.
Berry
,
1992
:
Predicting Atlantic seasonal hurricane activity 6–11 months in advance.
Wea. Forecasting
,
7
,
440
455
.
Holland
,
G.
,
1983
:
Tropical cyclone motion: Environmental interaction plus beta effect.
J. Atmos. Sci.
,
40
,
328
342
.
Jury
,
M. R.
,
N.
Pathack
, and
B.
Parker
,
1999
:
Climatic determinants and statistical prediction of tropical cyclone days in the southwest Indian Ocean.
J. Climate
,
12
,
3369
3372
.
Kalnay
,
E.
, and
Coauthors
.
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
431
471
.
Knuth
,
D. E.
,
1981
:
Seminumerical Algorithms. Vol. 2.
The Art of Computer Programming. Addison-Wesley, 688 pp
.
Neumann
,
C. J.
,
B. R.
Jarvinen
,
C. J.
McAdie
, and
J. D.
Elms
,
1993
:
Tropical Cyclones of the North Atlantic Ocean, 1871–1992.
Historical Climatology Series, Vol. 6-2, NOAA, 193 pp
.
Nicholls
,
N.
,
1992
:
Recent performance of a method for forecasting Australian seasonal tropical cyclone activity.
Aust. Meteor. Mag.
,
21
,
105
110
.
Palmer
,
T. N.
,
2001
:
A nonlinear dynamic perspective of model error: A proposal for non-local stochastic–dynamic parameterization in weather and climate prediction models.
Quart. J. Roy. Meteor. Soc.
,
127
,
279
304
.
Palmer
,
T. N.
,
R.
Gelaro
,
J.
Barkmeijer
, and
R.
Buizza
,
1998
:
Singular vectors, metrics, and adaptative observations.
J. Atmos. Sci.
,
55
,
633
653
.
Saunders
,
M. A.
, and
A. R.
Harris
,
1997
:
Sea warming as a dominant factor behind near-record numbers of Atlantic hurricanes.
Geophys. Res. Lett.
,
24
,
1255
1258
.
Vitart
,
F.
, and
J. L.
Anderson
,
2001
:
Sensitivity of Atlantic tropical storm frequency to ENSO and interdecadal variability of SSTs in an ensemble of AGCM integrations.
J. Climate
,
14
,
533
545
.
Vitart
,
F.
, and
T. N.
Stockdale
,
2001
:
Seasonal forecasting of tropical storms using coupled GCM integrations.
Mon. Wea. Rev.
,
129
,
2521
2537
.
Vitart
,
F.
,
J. L.
Anderson
, and
W. F.
Stern
,
1997
:
Simulation of the interannual variability of tropical storm frequency in an ensemble of GCM integrations.
J. Climate
,
10
,
745
760
.
Xie
,
S-P.
,
H.
Annamalai
,
F. A.
Schott
, and
J. P.
McCreary
Jr.
,
2002
:
Structure and mechanisms of South Indian Ocean climate variability.
J. Climate
,
15
,
864
878
.

Footnotes

Corresponding author address: Dr. Frédéric Vitart, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading, RG2 9AX, United Kingdom. Email: nec@ecmwf.int