A 66-year tropical cyclone record for south-east Africa: temporal trends in a global context

INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. (2014) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/joc.3932 A 66-year tropical cyclone record for south-east Africa: temporal trends in a global context Jennifer M. Fitchett and Stefan W. Grab* School of Geography, Archaeology and Environmental Studies, University of the Witwatersrand, Johannesburg, Gauteng, South Africa ABSTRACT: This study investigates changes in the frequency and timing of tropical cyclone landfalls over the south- west Indian Ocean during the last 66 years. Little is known about the spatial and temporal trends of such storm landfalls during recent historical times, specifically the last ca. 100 years. By analysing three storm track records spanning periods of 66–161 years, we establish that much of the perceived change in storm numbers can be attributed to improvements in storm detection methods over the past century. Furthermore, we find no statistically significant trends in the frequency of tropical cyclone landfalls over Madagascar and Mozambique over the past 6 decades, despite more comprehensive records during the most recent period. There is, however, considerable interannual variability in the number of storms making landfall over the countries investigated; most probably driven by cyclical atmospheric forcing, including El Ni˜no-Southern Oscillation (ENSO) and the Quasi-Biennial Oscillation (QBO). Recent trends indicate an increasing number of tropical cyclones tracking to the south of Madagascar, potentially associated with the southward shift of the 26 ◦ C isotherm, combined with a decrease in the steering flow during La Ni˜na years. KEY WORDS tropical cyclones; frequency; tracking direction; south-west Indian Ocean; climate change Received 26 July 2013; Revised 3 December 2013; Accepted 27 December 2013 1. Introduction of these develop in the northern hemisphere, and twice as many in the eastern than western hemisphere. An There has been considerable focus on the impact that estimated average nine tropical cyclones develop in the climate change, and in particular the increase in sea south-west Indian Ocean each year (11 if tropical storms surface temperatures, has had on the frequency of are included), of which only 5% make landfall over the tropical cyclones making landfall over coastal regions (Goldenberg et al., 2001; Mann and Emanuel, 2006). African continent (i.e. 0.45 landfalls per annum) (Reason, However, some propose that it is rather the extent 2007; Mavume et al., 2009). By contrast, an average 2.3 to which inhibiting factors control storm formation, and 3.7 extreme tropical cyclones make landfall over the which determines tropical cyclone frequency (Henderson- south-east USA and Guangdong province (China) each Sellers et al., 1998; Chan, 2006; Knutson et al., 2010). year, respectively (Goldenberg et al., 2001; Elsner and Particular focus explores the inhibiting role of vertical Liu, 2003). Clearly, coastal countries of south-east Africa shear (the difference in magnitude and direction between are not as regularly affected by tropical cyclones as those lower and upper tropospheric winds), as it is likely to in the Gulf of Mexico, the south-eastern United States, increase with rising sea surface temperatures and the sta- China, Japan and Australia. However, the consequences bilization of the atmospheric regions 5–20◦ north and of recent southern African tropical cyclones have been south of the equator, associated with increased tropical devastating (Table 1). As a developing region with rel- precipitation (Goldenberg et al., 2001; Sugi et al., 2002; atively poor disaster warning, preparedness and coping Mavume et al., 2009). Furthermore, the relative impor- strategies, countries such as Madagascar and Mozam- tance of factors influencing tropical cyclone formation bique are in some respects more vulnerable to tropi- may change in response to atmospheric and sea surface cal cyclone disasters than regions with robust disaster temperature changes over time, thus changing the hydro- risk reduction and coping strategy initiatives (Ash and logical cycle (Webster et al., 2005; Engelbrecht et al., Matyas, 2012). The landfall of an occasional severe tropi- 2008). cal cyclone may have substantial consequences including According to Gray (1975), 87% of all tropical cyclones loss of human life, and devastation to agriculture and develop between latitudes 20◦ north and south; two thirds infrastructure (Jury et al., 1993; Shanko and Camberlin, 1998; Vitart et al., 2003; Table 1). Given the 0.3 ◦ C increase in mean sea surface tem- * Correspondence to: Jennifer M. Fitchett, School of Geography, Archaeology and Environmental Studies, University of the Witwater- peratures over the south-west Indian Ocean since 1960 srand, P/Bag 3, WITS 2050, Johannesburg, South Africa. (Reason and Keibel, 2004; Gouretski et al., 2012), and E-mail: stefan.grab@wits.ac.za projected air temperature increases of as much as 3–5 ◦ C  2014 Royal Meteorological Society  J. M. FITCHETT AND S. W. GRAB Table 1. Impacts of south-west Indian Ocean tropical cyclone landfalls. Year Month Name Effects Madagascar 1951 January unknown 500 dead, 2 coastal trading vessels sunk 1959 April unknown 300–350 dead, 83 000 lost shelter 1994 February Geralda 70 dead, 90% city of Toamasina destroyed, rice stock severely damaged 2000 February Eline 64 dead, 10 000 homeless, 80% destruction of Mahanoro 2000 February Gloria 150 dead, over 100 homes destroyed, outbreak of cholera 2002 May Kesiny 33 dead, 1200 injured, 5000 homeless, 50% of cropland destroyed, 17 bridges destroyed 2004 March Gafilo 172 dead, offshore ferry capsized killing further 111, 200 000 affected, $250 million damage 2004 January Elita 33 dead, 129 injured, 55 983 homeless, 80% of rice crop flooded 2005 January Ernest 19 dead, 32 000 homeless, food shortages and price increases 2006 December Bondo 11 dead, 20 000 homeless 2007 March Indlala 88 dead, 3600 houses destroyed, 19 000 acres of rice crop lost 2008 February Ivan 93 dead, 80 000 homeless, 400 acres of crops destroyed 2008 March Jokwe 400 homeless, 44 buildings destroyed 2009 April Jade 15 dead, 22 900 homeless 2009 January Fanele 10 dead, 4 102 homeless, damage to bridges and roads leaving regions isolated 2010 March Hubert 85 dead, 132 injured, 66 000 homeless, 7 regions cut off due to damage to roads and bridges 2011 February Bingiza 22 dead, 13 injured, 22 845 homeless, 500 km2 destroyed crops 2012 February Giovanna 35 dead, 60% of homes across Madagascar destroyed 2012 February Haruna 24 dead, 92 injured, 10 000 homeless 2012 February Irina 77 dead, 911 homeless, landslides blocked entry to villages Mozambique 1956 April unknown 107 dead, Memba fishing port in ruins 1994 March Nadia 204 dead, 1.5 million homeless, $240 million damage 1996 January Bonita 11 dead 2000 February Eline 150 dead from storm, total 1000 casualties from flooding, 300 000 displaced, 4 ships sunk 2001 March Dera 100 deaths, 250 000 displaced, severe flooding 2003 January Delfina 47 deaths, 19 deaths from flooding, 22 000 displaced, several days power outage in Nampula, $3.5 million in damage 2003 March Japhet 17 dead, 23 000 homeless, 237 000 ha cropland destroyed, livestock losses 2007 February Favio 10 dead, 100 injured, 33 000 homeless, $71 million in damage 2008 March Jokwe 16 dead, 55 000 homeless, 75% of power lines in Nampula destroyed 2012 January Funso 15 dead from ship sinking, 56 000 homeless, 70 000 with no access to clean drinking water over southern Africa by the end of the 21st century 2. Data and methodology (Davis, 2010), it is anticipated that the frequency of trop- 2.1. Data selection ical cyclone landfalls over the south-east African region may have changed (Mavume et al., 2009). To this end, Multi-decadal records of tropical cyclone activity over there is value in examining the spatio-temporal patterns of the south-west Indian Ocean were scrutinized, and for tropical cyclone occurrences over the region, as this may the sake of consistency and brevity, data were limited to assist ongoing climate modelling and projection work. tropical cyclone storm track records. For the purpose of Our objective is to analyse the annual counts of south- this study, these are named according to the companies west Indian Ocean tropical cyclones making landfall responsible for the data administration; namely NOAA, over south-east Africa (Madagascar and Mozambique), Severe Weather and Unisys. The storm track records based on one 161-year and two 66-year storm track were selected based on availability, but cover a range of records. The specific aim is to determine whether there spatial scales from global (NOAA: http://www.ncdc.noaa. have been any changes in the frequency and seasonal gov/oa/ibtracs/), to the south Indian (Severe Weather: timing of such storms in this sub-tropical region. More http://australiasevereweather.com/cyclones/index.html) specifically, we investigate the timing of any changes and south-west Indian Oceans (Unisys: http://weather. in the number of recorded tropical cyclones making unisys.com/hurricane). landfall, and compare these to published findings from The continuous 161 year National Oceanographic and other global regions where tropical cyclones regularly Atmospheric Association of the United States (NOAA) have an impact on coastlines. record is a best track composite record for all ocean  2014 Royal Meteorological Society Int. J. Climatol. (2014)  A TEMPORAL TROPICAL CYCLONE RECORD FOR SOUTH-EAST AFRICA basins. Ship logs and accounts from the populated 2.2.1. Analysis of trends coastlines provided data from the 1850s to early 1900s Patterns of south-west Indian Ocean tropical cyclone fre- (Landsea, 1995). Subsequently, aircraft observations and quency are determined, either using the long-term dataset atmospheric readings, with routine aircraft reconnais- (should it be sufficiently accurate), or the most robust sance from 1944, provided the majority of information shorter-term dataset. Trend-lines through the annual trop- for the mid 20th Century. From the late 1960s onwards, ical cyclone counts, together with Pearson correlation information was derived solely from satellite observa- analysis of these time trends, permits the determination of tions. The dataset uses the Saffir–Simpson Scale to clas- the nature, extent and statistical strength of any tropical sify storms as tropical cyclones, with requirements of cyclone frequency changes. The 5-year running means lower tropospheric air pressure of below 980 mb and wind for the Mozambique and Madagascar data are analysed speeds in excess of 64 knots (Landsea, 1995). against the overall mean trends, thus allowing for the The Severe Weather dataset, compiled by the Aus- relative timing of storm frequency changes to be iden- tralian Bureau of Meteorology (2012), contains storm tified. Any significant cycles in tropical cyclone land- tracks for the south Indian Ocean over a 66-year period fall counts are determined through spectral analysis. The from 1944 to present. Using the Saffir–Simpson scale South African data were omitted from these analyses, requirements for the identification of tropical cyclones, given the infrequent tropical cyclone landfalls over the this dataset uses aircraft reconnaissance data from 1944 southern-most sub-continent. to 1970, together with satellite imagery and infrared sea surface temperature data for the period 1970 to present; obtained from the Joint Typhoon Warning Centre. The 3. Results Unisys records for the south-west Indian Ocean cover the period 1944–2011, and derive all their data from 3.1. Consistency of datasets the Joint Typhoon Warning Centre. These datasets simi- The storm track records display similar information for larly use the Saffir–Simpson scale classification of trop- the period of overlap (1964–2011), with no particular ical cyclones. The storm tracks are analysed for tropical bias toward over- or under-representation (see Table 2). cyclones making landfall over Madagascar and Mozam- The Unisys record shows the highest correlation with bique, with monthly and annual storm counts made the other two records, and thus appears the most robust for each country over the periods for which data are dataset. However, with no correlation coefficient below available. 0.6, the long term NOAA record appears sufficiently robust over the 66-year period, hence facilitating a critical 2.2. Data analysis data quality analysis for the full 161-year storm track The assumption is made that with a number of datasets record. recording the same phenomena, and compiled similarly, there should be little discrepancy in their outputs. Con- 3.2. The long-term NOAA dataset sequently, a comparison using correlation tables would We compare three broadly defined periods (Figure 1) reveal any discrepancies in the datasets. The record cor- for recording cyclone landfalls over the south-west relating best with the other two datasets for both countries Indian Ocean: 1850–1899 based primarily on ship logs; is thus probably the most robust. 1900–1943 based on ship logs, land-based records and To determine whether the 161-year NOAA dataset is early air-based observations; and 1944–2011 based pri- sufficiently robust to indicate any changes in regional cli- marily on aircraft reconnaissance and satellite imagery mate anomaly, we follow the methodology of Emanuel (Landsea, 2007). The mean number of recorded tropi- (2006) and Landsea (2007) who analyse tropical cyclone cal cyclones which made landfall over Madagascar and numbers from datasets covering periods exceeding a Mozambique from 1850 to 1899 is 0.5 (σ = 0.7) and 0.1 century. A 5-year running mean is produced for the (σ = 0.7) per annum respectively. Whilst the number of full period, which together with the annual counts recorded cyclones increases substantially over this period is measured against the 161-year mean for the com- for Madagascar (by 0.01/annum; r = 0.27, p = 0.06), plete dataset. Following the methodology of Holland the recorded frequency for Mozambique decreases (by and Webster (2007), a 5-year average scatter plot is 0.003/annum; p = 0.05, p = 0.73) (Figure 1(A and B)). divided into three periods, consistent with methodolog- Both the inter-annual variability and increasing num- ical changes for storm detection. This would reveal bers of cyclones recorded for Madagascar between the effect of technological data capturing techniques of 1850 and 1899 are a likely product of changing ship- tropical cyclones. If these periods coincide with large ping trends. A better indication of absolute numbers technological changes, it suggests an improvement in of tropical cyclone landfalls for this period would measuring technique, rather than an abrupt oceanic or thus depend on more widespread historical documentary climatic change. If, however, these data groups do not evidence. exhibit patterns particularly different from trends for the The number of recorded tropical cyclones making whole dataset, changes in methods of measurement can landfall from 1900 to 1943 averages 2.4 (σ = 1.41) be disregarded. and 0.32 (σ = 0) per annum for Madagascar and  2014 Royal Meteorological Society Int. J. Climatol. (2014)  J. M. FITCHETT AND S. W. GRAB A B Figure 1. Time-trends through the mean five year cyclone count from the NOAA records for (A) Mozambique and (B) Madagascar for the periods 1850–1899, 1900–1943 and 1944–2011. Table 2. Correlation table for the Madagascar and Mozambique identified for Madagascar from 1925 onwards, with datasets. a mean of 3.5 (σ = 0.7) per annum for the period 1925–1943, and is likely due to the increased recording Madagascar capacity of aerial reconnaissance. NOAA Severe weather Unisys The data from NOAA indicate negligible change in the NOAA 1 0.7607 0.8098 average number of cyclones making landfall over Mada- Severe weather 1 0.9171 gascar (3.1/annum, σ = 1) and Mozambique (1.1/annum Unisys 1 σ = 0.5) during the last 68 years (Figure 1(A and B)). For Synoptic maps both Madagascar and Mozambique, the NOAA record Mozambique shows a decrease in the number of tropical cyclone NOAA Severe weather Unisys landfalls during this period, by −0.01/annum (r = 0.11, NOAA 1 0.7163 0.829 p = 0.38) and −0.004/annum (r = 0.07, p = 0.60) respec- Severe weather 1 0.3208 tively (Figure 1A, B); this is consistent with an earlier Unisys 1 finding, suggesting reduced tropical cyclone formation Synoptic maps over the region (Mavume et al., 2009). However, as for the 1925–1943 period, a significantly larger number of cyclones have made landfall over Madagascar than Mozambique respectively, with increases for both Mozambique (∼300% difference). Madagascar (0.06/annum, r = 0.49, p = 0.0007) and Only 5% of tropical cyclones making landfall over Mozambique (0.004/annum, r = 0.13, p = 0.42) (Figure Madagascar subsequently reach Mozambique; and of 1(A and B)). A pronounced increase in records is those making landfall over Mozambique, 34.5% develop  2014 Royal Meteorological Society Int. J. Climatol. (2014)  A TEMPORAL TROPICAL CYCLONE RECORD FOR SOUTH-EAST AFRICA Figure 2. Map indicating the predominant tropical cyclone trajectories before making landfall on Madagascar and Mozambique. within the Mozambique Channel (Figure 2). The remain- The mean isotherm shift (40 km/decade or 0.6◦ /decade) ing 65.4% develop within the greater south Indian Ocean over the last 70 years (Figure 3) is consistent with the basin, with 44.1% passing to the north of Madagascar 0.5◦ pole-ward shift of the Indian Ocean subtropical and subsequently moving in a south-westerly direction gyre, observed for the period 1960–1999 (Alory et al., to make landfall over Mozambique; 20.5% pass to the 2007). Global increases in sea-surface and near surface south of Madagascar and move in a north-westerly direc- temperatures also account for the poleward shift of trop- tion toward Mozambique; the remaining 35.3% (23.2% of ical cyclones (Mavume et al., 2009; Gouretski et al., all Mozambique tropical cyclones) comprise the 5% of 2012). A notable deviation from this southward isotherm Madagascan tropical cyclones which continue over the shift occurred between 1961 and 1970; this coincided island and through the Mozambique Channel to make with global atmospheric cooling between ca. 1950 and landfall over Mozambique (Figure 2). 1970 (Mann et al., 2008). Should the isotherm shift con- Over the last few decades, there has been no sig- tinue southwards due to on-going global warming, it nificant change in the relative proportions of tropi- would have important implications for near-future trop- cal cyclones developing in the Mozambique Channel, ical cyclone tracks and consequent landfalls over both in comparison to those forming in the greater south Madagascar and South Africa (see also Malherbe et al., Indian Ocean. There has, however, been an increase in 2012). the number of tropical cyclones tracking to the south of Madagascar, before making landfall over Mozam- 3.3. Trends in tropical cyclone frequency bique. Seven tropical cyclones which made landfall over Analyses are henceforth made on the most robust 66- Mozambique during the period 1944–2011 followed year Unisys record. This record indicates an average of this south tracking path (20.5% of Mozambican tropi- 2.9 (σ = 1.7) and 0.8 (σ = 0.8) cyclones per annum mak- cal cyclones, Figure 2), whilst four took such a path ing landfall over Madagascar and Mozambique, respec- during the last 20 years (11.7% of Mozambican tropi- tively. A statistically insignificant decreasing number cal cyclones). Tropical cyclone Favio (in 2007) was one of tropical cyclones have made landfall over Mada- such storm which tracked south of Madagascar and made gascar (−0.2/decade; r = 0.18, p = 0.15) and insignifi- landfall over Mozambique, which has been attributed cant increasing number over Mozambique (0.04/decade; to a warm phase ENSO and negative southern Indian r = 0.08, p = 0.50) during the last 66 years. Despite a Ocean dipole (SIOD) (Klinman and Reason, 2008; Ash 0.3 ◦ C sea surface temperature increase over the south and Matyas, 2012). However, the increasing number Indian Ocean since 1960 (Reason and Keibel, 2004; of storms following this southward track may also be Mavume et al., 2009), the frequency of tropical cyclone driven by a southerly shift in the 26◦ C and 27◦ C sea landfalls over south-eastern Africa has not increased. surface temperature isotherm, which is demonstrated Should changes in tropical cyclone frequency have from NOAA decadal mean isotherm positions (Figure 3). been driven by global temperature increases associated  2014 Royal Meteorological Society Int. J. Climatol. (2014)  J. M. FITCHETT AND S. W. GRAB Figure 3. Southward shift in the decadal mean 26 ◦ C and 27 ◦ C isotherms for the south-west Indian Ocean. with either long-term (multi-decadal scale) increases in (Henderson-Sellers et al., 1998). ENSO and the Quasi- atmospheric CO2 concentrations, or short term cycles Biennial Oscillation (QBO) globally influence tropical associated with other factors (e.g. sunspots), then cyclone numbers and their geographic locality, whilst changes in tropical cyclone numbers of a similar magni- the Indian Ocean Dipole influences tropical cyclone fre- tude and direction would be expected across the world’s quency in the Indian Ocean (Jury et al., 1999; Ash ocean basins. We thus analyse similar tropical cyclone and Matyas, 2012). Functions of these variables, ver- counts from Unisys storm tracks for other ocean basins tical shear and the steering flow, all have an apparent in which tropical cyclones make landfall. South Pacific influence on such storm frequency and track paths (Vitart and north Indian Ocean Unisys storm track records are et al., 2003). limited to only a few decades and thus no comparison is Variability and cyclicity in the number of tropical possible. The west Pacific and north Atlantic regions are cyclones making landfall over both Madagascar and on average impacted by 16.9 (σ = 3.8) and 5.1 (σ = 2.1) Mozambique becomes more apparent when examining tropical cyclones per annum respectively (Figure 4). the 5 year running mean, and particularly so when com- As with the south-west Indian Ocean, there have been pared with the 66-year mean (Figure 5). Whilst there is insignificant changes in the annual occurrence of tropical considerable variability within this record, a long term cyclones over the north Atlantic (0.03/decade; r = 0.03, periodicity of 18–20 years is observed (consistent with p = 0.84) and west Pacific (−0.3/decade; r = 0.14, the findings of Malherbe et al., 2012), with the largest p = 0.25) (Figure 4). However, a statistically significant cyclic amplitudes of ca. two cyclones per annum for increase in annual cyclone landfalls is evident for the Madagascar and one per year for Mozambique. Smaller east Pacific (0.9/decade; r = 0.44, p = 0.0003) over the amplitudinal ranges with periods of 7–15 years are nested 66-year period (Figure 4). within this record (Figure 5), which, based on the tempo- In addition to the absolute sea surface temperature, ral pattern of their periodicity, are associated with peak shifts in the latitudinal position of the 26 ◦ C isotherm El Ni˜no and La Ni˜na events. As demonstrated in the could change the location of tropical cyclone land- overlaid ENSO and tropical cyclone count time series in falls and their frequency in higher latitude regions Figure 6, and confirming the findings of Reason et al.  2014 Royal Meteorological Society Int. J. Climatol. (2014)  A TEMPORAL TROPICAL CYCLONE RECORD FOR SOUTH-EAST AFRICA Figure 4. Frequency of tropical cyclone development in six oceanic regions. (2000), minima in tropical cyclone numbers frequently and for the La Ni˜na events of 1955–1956, 1970–1971, coincided with low ENSO index values (strong La Ni˜na and 1998–2000, all of which coincided with tropical events), whilst maxima in tropical cyclone events were cyclone minima (Figure 6). This is consistent with find- often concurrent with high ENSO values (strong El Ni˜no ings indicating a higher frequency of tropical cyclones events). This is particularly apparent for the El Ni˜no over the south-west Indian Ocean during El Ni˜no years, events of 1963–1964, 1972–1973 and 2002–2003, all of and over the south-east Indian Ocean during La Ni˜na which coincided with peaks in tropical cyclone landfalls; years (Ash and Matyas, 2012).  2014 Royal Meteorological Society Int. J. Climatol. (2014)  J. M. FITCHETT AND S. W. GRAB Table 3. Results of spectral analysis performed on annual counts tropical cyclone landfall over one sub-region whilst of tropical cyclone landfall over Madagascar and Mozambique inhibiting landfall over another. Using multivariate mul- for the period 1944–2010. tiple regression analysis from the NOAA Climate Indices Location of Significant cyclic p-Value (http://www.esrl.noaa.gov/psd/data/climateindices/list/), tropical cyclone period (years) (95% confidence) combinations of the Indian Ocean Dipole, QBO and El landfall Ni˜no/La Ni˜na events account for statistically significant (R 2 = 0.18, p = 0.0230) attributions of these shifts, with: Madagascar


 2.8 0.0002 Madagascar 3.498 0.294 7.6 0.0004 y = − ENSO Mozambique 1.212 0.289 12.2 <0.0001

 16.1 0.0002 −0.093 0.021 20.5 <0.0001 + IOD − QBO 0.245 0.037 Mozambique 3.5 <0.0001 However, the explanatory strength of this model is 8.3 <0.0001 limited by the poor temporal resolution of such a dataset. 11.1 <0.0001 14.3 0.0001 22.0 0.0001 4. Discussion 4.1. Scrutinizing the datasets At an even smaller scale, low amplitude 2–4 year A long-term analysis of tropical cyclone counts for the cycles are notable and consistent with south Indian Ocean northern hemisphere found three distinctive periods of QBO patterns for the south-west Indian Ocean (Jury increased storm events, each of which corresponded with et al., 1999). These visually observed cycles in tropical a period during which the method of storm observa- cyclone counts were confirmed through spectral analysis, tion had changed (Holland and Webster, 2007). Satel- with statistically significant cycles aligning with QBO, lite imagery during the past four decades provides twice ENSO and the 18–20 year cycle (Table 3). However, daily coverage of the earth, which has greatly improved inherent to any mathematical analysis of natural cycles tropical cyclone detection since the days of aerial recon- which do not have a fixed temporal period, the output naissance (1930s–1970s) or reliance on ship and coastal of this analysis cannot capture the variable lengths records (1850–1930s). As each new technology became and timing of these events. Given the high variability increasingly entrenched, so the ability to detect tropical and multiple cyclical patterns of different amplitudes, any climate change impacts on the number of tropical cyclones improved (Mann and Emanuel, 2006; Landsea, cyclones making landfall over the south-east African sub- 2007). The timing of abrupt increases in tropical cyclone continent are likely to be obscured. Furthermore, given frequency is consistent with our results for storms mak- these cyclic patterns, it is anticipated that any changes in ing landfall over Madagascar and Mozambique (Figures 2 sea surface temperature and vertical shear, in response to and 3). recent climate change, may yet take considerable time to induce changes in tropical cyclone numbers. 4.2. Climate change and tropical cyclones Of further interest are periods during which storm Despite global increases in atmospheric and sea surface frequency counts either exceed or fall below the 66-year temperatures over the past century, the slight decrease in mean (Figure 5). As would be expected for countries tropical cyclone occurrences over the south-west Indian adjacent to the same ocean basin, in the majority of cases Ocean and the west Pacific (Figure 4) is consistent with (87% of years), the periods during which the counts of previous findings (Sugi et al., 2002; Walther et al., 2002; tropical cyclone landfalls either exceed or fall below Walsh, 2004; Webster et al., 2005). In contrast, increasing the 66-year mean are common to both Madagascar and numbers of tropical cyclones over the east Pacific, and a Mozambique. This suggests that the frequency of tropical less substantial increase for the north Atlantic are noted cyclones is primarily driven by regional systems over the (Figure 4). Differences in the magnitude and direction south-west Indian Ocean basin, rather than local controls of these trends across various ocean basins indicate that over the Mozambique Channel. However, in a few cases, regional climate drivers, rather than sea surface and above average counts of tropical cyclones over Mada- air temperatures alone, are controlling tropical cyclone gascar occur contemporaneously with below average formation. However, long-term trends of tropical cyclone counts for Mozambique, or vice versa (most notably numbers in either direction are obscured by patterns of from 1994–2004). During this period, storm counts for climate variability where records span less than a century. the west Pacific and east Pacific Oceans were initially There are already notable cyclic patterns of inter-annual below their study period averages, whilst those for the tropical cyclone counts, with periodicities exceeding a north Atlantic and the north Indian oceans were above decade (Walsh, 2004; Chan, 2006; Knutson et al., 2010). their study period averages, followed by a period during Furthermore, these patterns of variability and cyclicity which the inverse occurred. This indicates periods during not only obscure statistical trends, but also act as drivers which prevailing ocean-atmospheric conditions favour to decrease the impacts of global warming on tropical  2014 Royal Meteorological Society Int. J. Climatol. (2014)  A TEMPORAL TROPICAL CYCLONE RECORD FOR SOUTH-EAST AFRICA Figure 5. Tropical cyclone counts for Madagascar and Mozambique from the Unisys record, overlaid with the 66-year mean and a 5-year running mean for each country. cyclone formation through hindering the development greater storm occurrences during recorded El Ni˜no years, of tropical storms (Singh et al., 2001; Walsh, 2004). as reported from the south-west Indian Ocean and else- The impact of global-scale climate change on tropical where (Henderson-Sellers et al., 1998; Jury et al., 1999; cyclones may thus take longer to detect than for other Reason and Keibel, 2004). This cyclic tropical cyclone weather systems (c.f. Singh et al., 2001; Goldenberg pattern has been attributed to a switch in the behaviour et al., 2001; Walsh, 2004; Knutson et al., 2010). of vertical shear (Goldenberg et al., 2001). Vertical shear Although GCMs project fewer tropical cyclones should increases with decreasing sea surface temperatures in the CO2 be doubled (Sugi et al., 2002), more intense storms south Indian Ocean during El Ni˜no events, additionally (defined by the maximum potential wind speed) are pro- inhibiting tropical cyclone formation (Henderson-Sellers jected to continue as the global climate continues to warm et al., 1998; Goldenberg et al., 2001). (Walsh and Ryan, 2000; Emanuel, 2006; Webster et al., Cycles of more frequent tropical cyclone activity have 2005, Elsner et al., 2008; Knutson et al., 2010). Whilst also been attributed to La Ni˜na events when warm sea Elsner et al. (2008) present an increase in tropical cyclone surface anomalies occur in semi-closed regions (such strength over the Atlantic Ocean basin, but less so else- as the Mozambique Channel), together with reduced where, Webster et al. (2005) suggest a large increase in steering flow due to decreased vertical shear (Vitart the number and proportion of tropical cyclones reaching et al., 2003). Conditions of low zonal steering facili- categories four and five for the north Pacific, north Indian tate a greater occurrence of tropical cyclones crossing and south-west Pacific Oceans. No consensus for the Madagascar and forming within the Mozambique chan- strength of future tropical cyclones has yet been reached; nel, whilst under high zonal steering flow storms deflect Sugi et al. (2002) predict no likely change in maximum to the southeast (Vitart et al., 2003). Such circumstances tropical cyclone intensity if CO2 levels were doubled, account for most cyclones developing in the Mozambique yet Walsh and Ryan (2000) propose a slight statistically Channel, which then move west and make landfall over insignificant increase in tropical cyclone strength for the Mozambique (Reason and Keibel, 2004). However, as the Australian region, but argue that such an increase in development of tropical cyclones within the Mozambique strength would be mitigated by an increase in vertical Channel has only recently been noted, the associations of wind shear associated with sea surface warming. such storms to La Ni˜na events cannot be statistically con- firmed. Given no apparent changes in the strength or fre- 4.3. South-west Indian Ocean tropical cyclones: quency of El Ni˜no events associated with climate change, cyclicity and atmospheric drivers such events are unlikely to inhibit tropical cyclone for- The cyclic temporal occurrence of tropical cyclones mation in future (van Oldenborgh et al., 2005). over various regions of the world is largely consis- With decreased steering flow, tropical cyclones origi- tent with those observed over the south-east Indian nating in the south-west Indian Ocean are more likely Ocean. They are likely directly connected to (and most to move north of Madagascar, making landfall over probably caused by) the same atmospheric forcing mech- northern Mozambique, rather than turning southwards to anisms (Sugi et al., 2002; Mann and Emanuel, 2006). make landfall over Madagascar during La Ni˜na events The 8–15 year tropical cyclone cycle observed for Mada- (Vitart et al., 2003; Reason, 2007). Given no clear trends gascar and Mozambique is noticeably associated with towards increasing frequency or strength of El Ni˜no/La  2014 Royal Meteorological Society Int. J. Climatol. (2014)  J. M. FITCHETT AND S. W. GRAB A B Figure 6. Time series of tropical cyclone numbers and NOAA bivariate ENSO index. Ni˜na events, changes in storm steering flow are unlikely (Henderson-Sellers et al., 1998; Singh et al., 2001; Mann in the near future (Reason and Keibel, 2004; Collins, and Emanuel, 2006). Further regional drivers which may 2005). However, recent trends indicate an increasing induce cyclical changes in tropical cyclone numbers over number of tropical cyclones tracking to the south of the south-west Indian Ocean include the Indian Ocean Madagascar; probably associated with the southward shift Dipole, and the strength and frequency of Rossby waves. of the 26 ◦ C isotherm, together with coincidental overlap However, such drivers can only be verified when a in IOD and ENSO phases, as was the case for tropical more robust higher resolution tropical cyclone database cyclone Favio in 2007 (Ash and Matyas, 2012). Should becomes available. such a trend continue, it could increase the number of tropical cyclones making landfall over South Africa in future decades (Reason and Keibel, 2004). 5. Conclusion Three further cycles are identified for tropical cyclone Despite some conflicting views concerning the impact landfall over the south-west Indian Ocean. A small of climate change on tropical cyclones, consensus has amplitude 2–4 year cycle is identified over both Mada- been reached, and confirmed through this study, on the gascar and Mozambique, and confirmed through spec- following: tral analysis (Table 3). This is consistent with findings from other northern and southern hemisphere regions, (i) Despite mean global atmospheric temperature hav- where such cycles have been attributed to the QBO ing increased over the past century, consequently of the lower troposphere (Goldenberg et al., 2001; raising mean sea surface temperature, the trends Holland and Webster, 2007). The considerably larger and rates of change are not consistent worldwide amplitude 20–25 year cycle could be associated with or within individual ocean basins (Sugi et al., 2002; the 29 year cycle observed over the north-west Indian Webster et al., 2005; Xie et al., 2010). Furthermore, Ocean regions of Bangladesh, Myanmar and India (Singh atmospheric conditions are not changing uniformly et al., 2001). These cycles are attributed to multi-decadal across all regions. Consequently, long-term changes cycles of thermohaline circulation strength, together in tropical cyclone frequency depend significantly with the consequential impact on monsoonal strength on the region and time period studied.  2014 Royal Meteorological Society Int. J. Climatol. (2014)  A TEMPORAL TROPICAL CYCLONE RECORD FOR SOUTH-EAST AFRICA (ii) Local sea surface temperatures >26 ◦ C are required Climate Change Handbook for North-Eastern South Africa. CSIR: Pretoria. for tropical cyclone formation, but temperature Elsner JB, Liu KB. 2003. Examining the ENSO-typhoon hypothesis. increases above the 26 ◦ C threshold do not directly Clim. Res. 25: 43–54. yield a greater number of tropical cyclones. A pole- Elsner JB, Kossin JP, Jagger TH. 2008. The increasing intensity of ward shift of the 26 ◦ C isotherm, or an increasing the strongest tropical cyclones. Nature 455: 92–95. Emanuel K. 2005. Increasing destructiveness of tropical cyclones number of days during which the 26 ◦ C isotherm over the past 30 years. Nature 436: 686–688. extends beyond 5◦ of the equator (required for Cori- Engelbrecht FA, McGregor JL, Engelbrecht CJ. 2008. Dynam- ics of the Conformal Cubic Atmospheric Model projected olis rotation), seems already to have initiated a climate-change signal over southern Africa. Int. J. Climatol. 29: pole-ward shift of tropical cyclone landfall. How- 1013–1033. ever, whilst sea surface temperature determines the Goldenberg SB, Landsea CW, Mestas-Nunez AM, Gray WM. 2001. maximum potential intensity of a tropical cyclone, The recent increase in Atlantic tropical cyclone activity: causes and implications. Science 293: 474–479. a strong vertical shear (difference in magnitude Gouretski V, Kennedy J, Boyer T, K¨ohl T. 2012. Consistent near and direction between lower and upper tropospheric surface ocean warming since 1900 in two largely independent winds) prevents the formation of tropical cyclones, observing networks. Geophys. Res. Lett. 39: 1–8. Gray WM. 1975. Tropical cyclone genesis. Colorado State Univer- as it inhibits the development of strong cyclonic cir- sity Department of Atmospheric Science Paper No. 234. Colorado culation (Hubbert and McInnes, 1999). Yet, warmer State University: Fort Collins. 1–121. sea surface temperatures are associated with even Henderson-Sellers A, Zhang H, Berz G, Emanuel K, Gray W, Landsea C, Holland G, Lighthill J, Shieh SL, Webster P, McGuffie greater vertical shear, thus moderating changes in K. 1998. Tropical cyclones and global climate change: a post- the number of tropical cyclones, and the spatial IPCC assessment. Bull. Am. Meteorol. Soc. 79(1): 19–39. extent of their regional formation (Hubbert and Holland GJ, Webster PJ. 2007. Heightened tropical cyclone activity in the North Atlantic: natural variability or climate trend? Philos. McInnes, 1999; Jury et al., 1999; Goldenberg et al., Trans. Math. Phys. Eng. Sci. 365(1860): 2695–2716. 2001). Hubbert GD, McInnes KL. 1999. A storm surge inundation model for (iii) The trend towards decreasing numbers of tropi- coastal planning and impact studies. J. Coastal Res. 15: 168–185. Jury MR, Pathack B, Wang B, Powell M, Raholijao N. 1993. A cal cyclone landfalls over Madagascar (although destructive Tropical cyclone season in the SW Indian Ocean: not statistically significant) is consistent with his- January-February 1984. S. Afr. Geogr. J. 75(2): 53–59. toric records for the Australian region (i.e. south- Jury MR, Pathack B, Parker B. 1999. Climatic determinants and statistical prediction of tropical cyclone days in the Southwest west Pacific and south-east Indian Oceans), and Indian Ocean. J. Climate 12: 1738–1746. with projected decreases across the entire south Klinman MG, Reason CJC. 2008. On the peculiar storm track of Indian Ocean (Nicholls et al., 1998; Sugi et al., TC Favio during the 2006–2007 Southwest Indian Ocean tropical cyclone season and relationships to ENSO. Meteorol. Atmos. Phys. 2002). However, future tropical cyclone storm tracks 100: 233–242. across Mozambique should be closely monitored, as Knutson TR, McBride JL, Chan J, Emanuel K, Holland G, Landsea the increasing trend of such storms in this south- C, Held I, Kossin JP, Srivastava AK, Sugi M. 2010. Tropical cyclones and climate change’. Nat. Geosci. 3(3): 157–163. east Indian Ocean sector is consistent with increas- Landsea CW. 1995. Best Track Data [online]. Retrieved October 13, ing numbers of tropical cyclones making landfall 2011. http://www.weather.unisys.com/hurricane. over countries of similar longitude, adjacent to the Landsea CW. 2007. Counting Atlantic tropical cyclones back to north-west Indian Ocean (Singh et al., 2001). Con- 1900. EOS Trans. Am. Geophy. Union 88(18): 197–208. Malherbe J, Engelbrecht FA, Landman WA, Engelbrecht CJ. 2012. tinued work to improve the understanding of spatio- Tropical systems from the Southwest Indian Ocean making temporal tropical cyclone landfalls is essential for landfall over the Limpopo River Basin, Southern Africa: a robust model developments; ultimately to improve historical perspective. Int. J. Climatol. 32: 1018–1032. Mann ME, Emanuel KA. 2006. Atlantic tropical cyclone trends storm projections and early warning systems. linked to climate change. EOS Trans. Am. Geophy. Union 87(24): 233–244. Mann ME, Zhang Z, Hughes MK, Bradley RS, Miller SK, Ruther- Acknowledgements ford S, Ni F. 2008. Proxy-based reconstructions of hemispheric and global surface temperature variations over the past two mil- We much appreciate the constructive input provided by lennia. Proc. Natl. Acad. Sci. U. S. A. 105(36): 13252–13257. anonymous referees who assisted us in refining an earlier Mavume AF, Rydberg L, Rouault M, Lutjeharms JRE. 2009. Climatology and landfall of tropical cyclones in the South-West version of this paper. Indian Ocean. West. Indian J. Mar. Sci. 8(1): 15–36. Nicholls N, Landsea C, Gill J. 1998. Recent trends in Australian region tropical cyclone activity. Meteorol. Atmos. Phys. 45: References 167–205. Reason CJC, Allan RJ, Lindesay JA, Ansell TJ. 2000. ENSO and Alory G, Wijffels S, Meyers G. 2007. Observed temperature trends climatic signals across the Indian Ocean Basin in the global in the Indian Ocean over 1960–1999 and associated mechanisms. context: part I, interannual composite patterns. Int. J. Climatol. Geophys. Res. Lett. 34: 1–6. 20: 1285–1327. Ash KD, Matyas CJ. 2012. The influences of ENSO and the Reason CJC, Keibel A. 2004. Tropical cyclone Eline and its unusual subtropical Indian Ocean dipole on tropical cyclone trajectories penetration and impacts over the Southern African Mainland. Wea. in the Southwestern Indian Ocean. Int. J. Climatol. 32(1): 41–56. Forecasting 19: 789–805. Chan JCL. 2006. Comment on “Changes in Tropical cyclone Reason CJC. 2007. Tropical cyclone Dera, the unusual 2000/01 Number, Duration and Intensity in a Warming Environment”. tropical cyclone season in the Southwest Indian Ocean and Science 311: 1713–1714. associated rainfall anomalies over Southern Africa. Meteorol. Collins M. 2005. El Ni˜no- or La Ni˜na-like climate change. Climate Atmos. Phys. 97: 181–188. Dynam. 24: 89–104. Shanko D, Camberlin P. 1998. The effects of the Southwest Indian Davis C, Archer E, Engelbrecht F, Landman W, Stevens N, Sinden Ocean tropical cyclones on Ethiopian drought. Int. J. Climatol. L, Nkambule C, Maserumule R, van der Merwe M. 2010. A 18(1373): 1388.  2014 Royal Meteorological Society Int. J. Climatol. (2014)  J. M. FITCHETT AND S. W. GRAB Singh OP, Khan TMA, Rahman MS. 2001. Has the frequency of Xie SP, Deser C, Vecchi GA, Ma J, Teng H, Wittenberg AT. 2010. intense tropical cyclones increased in the North Indian Ocean? Global warming pattern formation: sea surface temperature and Curr. Sci. 80(4): 575–580. rainfall. J. Climate 23: 966–986. Sugi N, Noda A, Sato N. 2002. Influence of the global warming on tropical cyclone climatology: an experiment with the JMA Global Model. J. Meteor. Soc. Japan 80(2): 249–272. Van Oldenborgh GJ, Philip SY, Collins M. 2005. El Ni˜no in a Data sources changing climate: a multi-model study. Ocean Sci. Discuss. 2: 267–298. Australian Bureau of Meteorology. 2012. Severe Weather Vitart F, Anderson D, Stockdale T. 2003. Seasonal forecasting Historical Tropical Cyclone Tracks [online]. Retrieved of tropical cyclone landfall over Mozambique. J. Climate 16: from 12 August 2011 to 02 June 2013 http://australia 3932–3945. severeweather.com/cyclones/index.html. Walsh KJE, Ryan BF. 2000. Tropical cyclone intensity increase near National Oceanic and Atmospheric Association. 2012. Historical Australia as a result of climate change. J. Climate 13: 3029–3036. Best Track Records [online]. Retrieved from 12 August 2011 to Walsh K. 2004. Tropical cyclones and climate change: unresolved 02 June 2013 http://www.ncdc.noaa.gov/oa/ibtracs/. issues. Climate Res. 27: 77–83. National Oceanic and Atmospheric Association. 2013. Cli- Walther GR, Post E, Convey P, Menzel A, Parmesan C, Beebee TJC, mate Indices: Monthly Atmospheric and Ocean Time Series Fromentin JM, Hoegh-Guldberg O, Bairlein F. 2002. Ecological [online]. Retrieved on 30 September 2013 http://www.esrl. responses to recent climate change. Nature 416: 389–395. noaa.gov/psd/data/climateindices/list/. Webster PJ, Holland GJ, Curry JA, Chang HR. 2005. Changes in Unisys. 2012. Historical Best Track Records [online]. Retrieved Tropical cyclone number, duration, and intensity in a warming from 12 August 2011 to 02 June 2013 http://weather. environment. Science 309: 1844–1846. unisys.com/hurricane.  2014 Royal Meteorological Society Int. J. Climatol. (2014)