Tropical Cyclones and Climate Change: Review Article
Tropical Cyclones and Climate Change: Review Article
Tropical Cyclones and Climate Change: Review Article
Whether the characteristics of tropical cyclones have changed or will change in a warming climate — and if so, how — has
been the subject of considerable investigation, often with conflicting results. Large amplitude fluctuations in the frequency and
intensity of tropical cyclones greatly complicate both the detection of long-term trends and their attribution to rising levels of
atmospheric greenhouse gases. Trend detection is further impeded by substantial limitations in the availability and quality of
global historical records of tropical cyclones. Therefore, it remains uncertain whether past changes in tropical cyclone activity
have exceeded the variability expected from natural causes. However, future projections based on theory and high-resolution
dynamical models consistently indicate that greenhouse warming will cause the globally averaged intensity of tropical cyclones
to shift towards stronger storms, with intensity increases of 2–11% by 2100. Existing modelling studies also consistently project
decreases in the globally averaged frequency of tropical cyclones, by 6–34%. Balanced against this, higher resolution modelling
studies typically project substantial increases in the frequency of the most intense cyclones, and increases of the order of 20%
in the precipitation rate within 100 km of the storm centre. For all cyclone parameters, projected changes for individual basins
show large variations between different modelling studies.
T
he challenge for climate change detection and attribution coastal areas. In developing countries, in particular, the movement
research with regard to tropical cyclones is to determine of the population to the coast is the result of social factors that are not
whether an observed change in tropical cyclone activity easily countered. Climate change is hence one of several factors likely
exceeds the variability expected through natural causes, and to to affect the future evolution of damage from tropical cyclones.
attribute significant changes to specific climate forcings, such as We discuss here issues related to detection and attribution, and
greenhouse gases or aerosols. For future projections of tropical to future projections for tropical cyclones. The future projection
cyclone activity, the challenge is to develop both a reliable projec- statements in this report are intended to apply roughly to the IPCC
tion of changes in the various factors influencing tropical cyclones, A1B scenario2 as of the late twenty-first century. All likelihood
both local and remote, and a means of simulating the effect of statements follow conventions used by the IPCC2 (Supplementary
these climate changes on tropical cyclone metrics, such as storm Information S4).
frequency, intensity and track distribution. This two-step process We consider new developments in the field since the 2006 World
is required because the coupled atmosphere–ocean models used Meteorological Organization expert-team statement6, including: new
to project climate on a multidecadal to centennial timescale do not satellite-based intensity analyses; improved hindcast performance of
themselves simulate tropical cyclones adequately. downscaling techniques; substantial new analysis of data homoge-
Sea surface temperatures (SSTs) in most regions of tropical neity issues; new simulations with higher resolution global models;
cyclone formation have increased by several tenths of a degree and analyses of the sensitivity of tropical cyclone projections to the
Celsius during the past several decades1. The Intergovernmental choice of climate model being downscaled. A discussion of limita-
Panel on Climate Change (IPCC) fourth assessment report2 con- tions of tropical cyclone historical data is given in Supplementary
cluded that most of the global surface temperature increase over Information S5. For detection and attribution, the emphasis here is
the past half century is very likely due to the observed increase in on the Atlantic Ocean basin because the data records for this region
anthropogenic greenhouse-gas concentrations, and the US Climate are longer and relatively more reliable, though our assessment state-
Change Science Program 3.3 report3 extended this by concluding ments (summarized in Box 1) include consideration of all basins as
that human-induced greenhouse-gas increases have very likely appropriate. Comparisons with previous assessments and recom-
contributed to the increase in sea surface temperatures (SSTs) in mendations for future progress are contained in Supplementary
hurricane formation regions4. These results have raised the question Information S6 and S7.
of how substantial further warming, coupled with other changes in
the tropical environment, would affect tropical cyclone activity. tropical cyclone activity versus sea surface temperature
Recent decades have seen very large increases in the economic Over the past 50 years, a significant statistical correlation has existed
damage and disruption caused by tropical cyclones. Historical between Atlantic tropical cyclone power dissipation (definitions
analyses5 indicate that this has been caused primarily by rising in Supplementary Information S4) and SST on timescales of a
coastal populations and the increasing value of infrastructure in few years or more7. A comparable large correlation exists, on all
1
Geophysical Fluid Dynamics Laboratory/NOAA, 201 Forrestal Road, Princeton, New Jersey 08542, USA, 2Centre for Australian Weather and Climate
Research, Melbourne 3001, Australia, 3Guy Carpenter Asia-Pacific Climate Impact Centre, City University of Hong Kong, Kowloon, Hong Kong, China,
4
Program in Atmospheres, Oceans, and Climate, Massachusetts Institute of Technology, Room 54-1620 MIT, 77 Massachusetts Avenue, Cambridge,
Massachusetts 02139, USA, 5National Center for Atmospheric Research, Boulder, Colorado, USA, 6National Hurricane Center/NWS/NOAA, 11691 SW
17th Street, Miami, Florida 33165, USA, 7National Climatic Data Center/NOAA, 1225 W Dayton Street, Madison, Wisconsin 53706, USA, 8India
Meteorological Department, Shivajinagar, Pune 411005, India, 9Research Institute for Global Change, JAMSTEC, 3173-25 Showa-machi, Kanazawa-ku,
Yokohama, 236-0001 Kanagawa, Japan. *e-mail: Tom.Knutson@noaa.gov
detection and attribution twenty-first-century warming, although increases may not occur
It remains uncertain whether past changes in any tropical cyclone in all tropical regions. The frequency of the most intense (rare/
activity (frequency, intensity, rainfall, and so on) exceed the high-impact) storms will more likely than not increase by a
variability expected through natural causes, after accounting for substantially larger percentage in some basins.
changes over time in observing capabilities.
Rainfall. Rainfall rates are likely to increase. The projected mag-
tropical cyclone projections nitude is on the order of +20% within 100 km of the tropical
Frequency. It is likely that the global frequency of tropical cyclones cyclone centre.
will either decrease or remain essentially unchanged owing to
greenhouse warming. We have very low confidence in projected Genesis, tracks, duration and surge flooding. We have low con-
changes in individual basins. Current models project changes fidence in projected changes in tropical cyclone genesis-location,
ranging from −6 to −34% globally, and up to ±50% or more in tracks, duration and areas of impact. Existing model projections do
individual basins by the late twenty-first century. not show dramatic large-scale changes in these features. The vul-
nerability of coastal regions to storm-surge flooding is expected to
Intensity. Some increase in the mean maximum wind speed of increase with future sea-level rise and coastal development, although
tropical cyclones is likely (+2 to +11% globally) with projected this vulnerability will also depend on future storm characteristics.
timescales down to a year, between the power dissipation and the Some observational studies15–17 report substantial century-scale
tropical Atlantic Ocean SST relative to mean tropical SST8. Taken increases in Atlantic tropical cyclone frequency, that can be mod-
at face value, these two statistical relations lead to dramatically dif- elled statistically by the century-scale SST increases (Fig. 2, first
ferent inferences about late-twenty-first-century Atlantic tropical blue series versus second green series), and some of this increase
cyclone activity9, ranging from a dramatic increase of about 300% has been attributed to anthropogenic forcing15,16. However, it has
in the first case to little change in the second (Fig. 1). been found20 that the statistical significance of the trends in the
Tropical Atlantic Ocean SST has increased more rapidly than original storm frequency data is greatly reduced after adjustments
tropical mean SST over the past 30 years, coincident with the posi- are made20,21 for an estimated number of missing tropical cyclones
tive trend in the Atlantic power dissipation index over this period owing to a lower reporting-ship track density and other observa-
(Fig. 1b). This differential warming of the Atlantic can be affected tional limitations in pre-satellite (pre-1966) years (Fig. 2, first red
by natural multidecadal variability, as well as by aerosol forcing, series). Furthermore, the trend in storm count in the original data
but climate models9–13 indicate that it is not strongly influenced has been shown to be almost entirely due to an increase in short
by greenhouse-gas forcing (Fig. 1b). If the relationship between duration (<2-day) tropical storms22 — a phenomenon that has been
Atlantic power dissipation and this differential warming in Fig. 1b interpreted as likely being attributable to changes in observing capa-
is causal, then a substantial part of the increase in Atlantic power bilities22. There is a much smaller increasing trend in storms lasting
dissipation since 1950 is likely due to factors other than greenhouse- more than two days (Fig. 2, second red series) and after an estimated
gas-induced warming. adjustment for missing storms20,22, the resulting long-term trend is
On the other hand, the case for the importance of local SSTs not significant (p > 0.05).
would be strengthened by observations of an increase in power dis- Hurricane counts (with no adjustments for possible missing
sipation in other basins, in which local warming in recent decades cases) show a significant increase from the late 1800s to present, but
does not exceed the tropical mean warming. A study for the north- do not have a significant trend from the 1850s or 1860s to present 3.
west Pacific Ocean basin7 shows a statistical correlation between Other studies23 infer a substantial low-bias in early Atlantic tropical
low-frequency variability of power dissipation and local SSTs. But cyclone intensities (1851–1920), which, if corrected, would further
this correlation is considerably weaker than for the Atlantic Ocean, reduce or possibly eliminate long-term increasing trends in basin-
and other key measures of storm activity in the northwest Pacific, wide hurricane counts. Landfalling tropical storm and hurricane
such as the number of Category 4 and 5 typhoons, do not show a activity in the US shows no long-term increase (Fig. 2, orange
significant correlation with SST14. series)20. Basin-wide major hurricane counts show a significant
rising trend, but we judge these basin-wide data as unreliable for
tropical cyclone frequency climate-trend estimation before aircraft reconnaissance in 1944.
Detection and attribution. We first consider tropical Atlantic A study of a 1,500-year record of sediment overwash from a
Ocean SST variability, which has been used statistically to model number of sites along the US coast and one near Puerto Rico24 finds
Atlantic changes in the frequency of tropical storms15–17. Substantial evidence for relatively high numbers of strong Atlantic hurricane
multidecadal SST variability is evident in the North Atlantic basin landfalls at these sites during several periods (from around
(Fig. 2, second green series). The cause of this variability remains ad 1000–1200; the early 1400s; the early 1800s; the 1950s; and in
uncertain, with possible contributions from both internal climate recent decades). This record is not subject to the same data errors
variability and radiative-forcing changes16,18. Evidence from tropical that have made direct assessment of strong-hurricane frequency
African lake sediments19 indicates that rainfall variability before the from the observational record difficult, but is subject to uncertainties
twentieth century was at least as large as that seen in the twentieth in interpretation of storm characteristics from geological evidence
century, increasing the plausibility of substantial natural climatic and limited spatial coverage. Comparisons of this data set with
variability in the tropical Atlantic Ocean. The multidecadal SST other measures of strong landfalling tropical cyclones in the period
variability (evident in Fig. 2, second green series) complicates trend of direct records have yet to be documented.
detection in the tropical Atlantic, but model simulations indicate In terms of global tropical cyclone frequency, it was concluded25
that substantial proportions of the observed tropical Atlantic and that there was no significant change in global tropical storm or hur-
northwest tropical Pacific SST increases over the past half century ricane numbers from 1970 to 2004, nor any significant change in
arise from greenhouse warming 4. hurricane numbers for any individual basin over that period, except
Temperature
Five–year PDI based on observed absolute SST; r = 0.79
300 Statistical five-year PDI downscaling of
global climate models (1946–2100)
100
Unadjusted counts
Tropical storms: unadjusted
0
High-resolution
model projections
Past Future
–100
Tropical storms: >2–day:
1960 1980 2000 2020 2040 2060 2080 2100
unadjusted
Year
b
Adjusted counts
Tropical storms: adjusted
Change in power dissipation index (%)
Landfall counts
(unadjusted)
100
0 US landfalling hurricanes
(unadjusted)
Past Future
–100
1960 1980 2000 2020 2040 2060 2080 2100 1880 1900 1920 1940 1960 1980 2000
Year Year
Figure 1 | Past and extrapolated changes in atlantic hurricane power Figure 2 | Tropical atlantic indices. Green-shaded curves depict global
dissipation index (PDi). Anomalies are regressed onto a, tropical mean temperature (HadCRUT3 data set) and August–October main
Atlantic SST or b, tropical Atlantic SST relative to tropical mean SST development region (MDR; 10° N–20° N, 80° W–20° W) SST anomalies
(1946–2007), and these regression models are used to statistically (HadISST data set). Blue-shaded curves represent unadjusted tropical
estimate the PDI from several climate models. Anomalies are per-cent storm counts. Red-shaded curves include time-dependent adjustments
change relative to 1981–2000 average (2.13 x 1011 m3 s–2). The green bar for missing storms based on ship-track density20,22. The curve labelled
denotes approximate range of the PDI anomaly predicted by the statistical/ ‘>2-day’ depicts storms with a duration greater than 2.0 days22. Orange-
dynamical calculations of ref. 11. The green circle, star and diamond denote shaded curves depict US landfalling tropical storms and hurricanes (no
approximate values suggested by high-resolution dynamical models (refs adjustments). Solid black lines are five-year means (1878–2008); dashed
12, 10 and 13 respectively). SST region is 20° W–70° W, 7.5° N–22.5° N. black lines are linear trends. Vertical axis ticks represent one standard
Figure reproduced with permission from ref. 9 (© 2008 AAAS). deviation. Series normalized to unit standard deviation. Only the top three
series have significant linear trends (p = 0.05). Figure reproduced with
for the Atlantic (discussed above). Landfall in various regions of permission from refs 20 (© 2008 AMS) and 22 (© 2009 AMS).
East Asia26 during the past 60 years, and those in the Philippines27
during the past century, also do not show significant trends. coupled climate models to adequately simulate the changes in
Thus, considering available observational studies, and after large-scale conditions that affect cyclone development. Care must
accounting for potential errors arising from past changes in observ- be taken in interpreting results from regional models, as the use
ing capabilities, it remains uncertain whether past changes in of small domains or spectral nudging across the regional domain
tropical cyclone frequency have exceeded the variability expected constrains the model to follow the conditions imposed from the
through natural causes. driving large-scale model.
The general convergence of frequency projections from differ-
Projection. Progress has been made in developing dynamical ent approaches (Supplementary Table S1), in conjunction with the
and statistical/dynamical models for seasonal tropical cyclone hindcasting tests illustrated in Fig. 3, is beginning to provide some
frequency. Such models include: global coupled climate confidence in global and hemispheric projections of tropical cyclone
models13,28; relatively high-resolution atmospheric models frequencies. However, confidence in these projections remains very
running over observed or projected SST distributions10,29,30; low for individual basins (Supplementary Table S1), owing to uncer-
regional climate models used to downscale solutions from glo- tainties in the large-scale patterns of future tropical climate change,
bal coupled models12,31; and new statistical/dynamical techniques as evident in the lack of agreement between the model projections
aimed at avoiding the limitations on intensity simulations in of patterns of tropical SST changes29 as well as remaining limitations
dynamical models11. Many of these models reproduce key aspects in the downscaling strategies.
of observed past tropical cyclone variability when forced with Based on existing modelling studies (Supplementary Table S1)
historical variations in boundary conditions, such as SSTs or, in and limited existing observations, we judge that it is likely that
the case of regional models, by the SSTs and large-scale atmos- global mean tropical-cyclone-frequency will either decrease or
pheric winds, moisture and temperature distributions from remain essentially unchanged owing to greenhouse warming. Late-
atmospheric reanalyses (Fig. 3). But such tropical-cyclone-fre- twenty-first-century model projections indicate decreases ranging
quency simulations are highly dependent on the ability of global from −6 to −34% globally, with a comparatively more robust
12 Model ensemble
increase in both the mean intensities (Supplementary Table S2) and
10 Observed
in the frequency of cyclones at higher intensity levels13,38. Such shifts
8
are observed in the best track records of global tropical cyclone
6 intensities, but these records are known to have substantial hetero-
4 geneities, which can also manifest themselves as a shift towards
2 stronger storms. A substantial global increase (nearly doubling) in
0 the number of the most severe tropical cyclones (Category 4 and 5 on
1980 1985 1990 1995 2000 2005 the Saffir–Simpson scale) has been reported25 from 1975 to 2004.
c Other studies contested this finding, based on concerns about data
28
quality39,40 and the short record-length relative to multidecadal vari-
Number of tropical storms
Model ensemble
24
Observed ability41 in the northwest Pacific. Analyses of globally consistent
20 satellite-based intensity estimates since 1981 indicate that trends in
16 the best track-data are indeed inflated42, but do support an increase
12 globally in the intensities of the strongest tropical cyclones43.
8 The new satellite-based intensity data42,43 were designed to be
4 more homogeneous than the existing global data, but still carry
0
uncertainties, particularly in the Indian Ocean where the satel-
1986 1990 1995 2000 2005 lite record is less consistent43. The short time period of the data
d 18 does not allow any definitive statements regarding separation of
16 Model ensemble anthropogenic changes from natural decadal variability or the
Number of hurricanes
100*log(A1B/CTR) (%)
MIROC
tional increase in the occurrence rate of the strongest (rarest) trop- 20
MRI
ical cyclones12,38. For example, a recent downscaling study38 using
10
an operational (9-km grid) hurricane prediction model shows a
tendency towards increased frequency of Atlantic Category 4 and 0
5 hurricanes over the twenty-first century. We judge that a sub-
–10
stantial increase in the frequency of the most intense storms is
more likely than not globally, although this may not occur in all –20
tropical regions. Our confidence in this finding is limited, since –30
the model-projected change results from a competition between
the influence of increasing storm intensity and decreasing overall –40 10.3% –2.2% 19.1% -5.8% –13.8%
storm frequency. –50
Although such changes were not noted in several relatively low- Atlantic East Pacific West Pacific North Indian Southern
resolution simulations, these models are less reliable for investigat- Hemisphere
ing the most intense cyclones. As an example, it was found that, for
one series of models, a resolution of ~60 km was needed before a Figure 4 | sensitivity of projected tropical cyclone activity to different
warming-related intensification was simulated13. climate models providing downscaling conditions. a, Projected fractional
Further studies are needed to evaluate model projections of change in North Atlantic hurricanes (late twenty-first century) using
intensity changes, for example, by comparing model simulations of a global atmospheric model to downscale SST projections from three
the interannual variability of intensities to observations38. As there individual climate models or from an 18-model ensemble. The two
is a suggestion in existing studies that climate-warming-induced projections for each case (red and blue) used different controls based on
increases of intensity are larger in higher resolution models than different observed SST data. The vertical bars denote 90% confidence
in coarse-grid models13, it is plausible that existing models may intervals. Reproduced with permission from ref. 29 (© 2009 AMS).
systematically underestimate future intensity trends. The future b, Approximate percentage change in tropical cyclone power dissipation in
characteristics of intense tropical cyclones (Category 3–5) deserve various tropical storm basins projected for the late twenty-second century
particular attention, as these storms historically have accounted for using a statistical/dynamical downscaling framework forced with climate
an estimated 85% of US hurricane damage, despite representing change statistics from seven global models. The change here is given as
only 24% of US landfalling tropical cyclones5. Further studies with 100 multiplied by the logarithm of the ratio of the twenty-second- (A1B)
finer resolution models hopefully will increase our confidence in and twentieth-century power dissipation (see ref. 11). Reproduced with
future projections of tropical cyclone intensity and the frequency of permission from ref. 11 (© 2008 AMS).
very intense cyclones.
related rainfall has not been established by existing studies.
tropical cyclone rainfall Satellite-based studies report an increase in the occurrence of
Detection and attribution. Atmospheric moisture content has heavy-rain events, generally in the tropics during 1979–2003
increased in recent decades in many regions46, and climate models (ref. 47), and also an increase during warm periods of interan-
are unanimous that the integrated water column in the tropics nual variability48. A number of studies of land-based precipitation
will increase, on average, as the atmosphere warms. The expecta- data have identified increasing trends in the frequency of very
tion is that as the water-vapour content of the tropical atmosphere heavy precipitation events2,3. None of these studies isolate tropical
increases, the moisture convergence for a given amount of mass cyclone precipitation rates.
convergence is enhanced. This should increase rainfall rates in sys-
tems (such as tropical cyclones) where moisture convergence is an Projection. Tropical-cyclone-related rainfall rates are likely to
important component of the water-vapour budget. An increase in increase with greenhouse warming. This is a robust projection in
storm-wind intensities would add to this moisture convergence. model simulations of tropical cyclones in a warmer climate: all seven
Despite this expectation, a detectable change in tropical-cyclone- available studies report substantial increases in storm-centred rainfall