Global Warming and Hurricanes
An Overview of Current Research Results
1. Has Global Warming Affected Atlantic Hurricane Activity?Geophysical Fluid Dynamics Laboratory/NOAA
Last Revised: June 15, 2015
A. Summary Statement
Two frequently asked questions on global warming and hurricanes are the following:
- Have humans already caused a detectable increase in Atlantic hurricane activity?
- What changes in hurricane activity are expected for the late 21st century, given the pronounced global warming scenarios from current IPCC models?
In this review, we address these questions in the context of published research findings. We will first present the main conclusions and then follow with some background discussion of the research that leads to these conclusions. The main conclusions are:
- It is premature to conclude that human activities--and particularly greenhouse gas emissions that cause global warming--have already had a detectable impact on Atlantic hurricane activity. That said, human activities may have already caused changes that are not yet detectable due to the small magnitude of the changes or observational limitations, or are model-estimated changes with considerable uncertainty (e.g., aerosol effects).
- Anthropogenic warming by the end of the 21st century will likely cause hurricanes globally to be more intense on average (by 2 to 11% according to model projections for an IPCC A1B scenario). This change would imply an even larger percentage increase in the destructive potential per storm, assuming no reduction in storm size.
- There are better than even odds that anthropogenic warming over the next century will lead to an increase in the numbers of very intense hurricanes in some basins—an increase that would be substantially larger in percentage terms than the 2-11% increase in the average storm intensity. This increase in intense storm numbers is projected despite a likely decrease (or little change) in the global numbers of all tropical storms.
- Anthropogenic warming by the end of the 21st century will likely cause hurricanes to have substantially higher rainfall rates than present-day hurricanes, with a model-projected increase of about 20% for rainfall rates averaged within about 100 km of the storm center.
B. Statistical relationships between SSTs and Atlantic hurricanes
Observed records of Atlantic hurricane activity show a correlation, on multi-year time-scales, between local tropical Atlantic sea surface temperatures (SSTs) and the Power Dissipation Index (PDI) --see for example Fig. 3 on this EPA Climate Indicators site. PDI is an aggregate measure of Atlantic hurricane activity, combining frequency, intensity, and duration of hurricanes in a single index. Both Atlantic SSTs and PDI have risen sharply since the 1970s, and there is some evidence that PDI levels in recent years are higher than in the previous active Atlantic hurricane era in the 1950s and 60s.
Model-based climate change detection/attribution studies have linked increasing
tropical Atlantic SSTs to increasing greenhouse gases, but the link between increasing
greenhouse gases and hurricane PDI or frequency has been based on statistical correlations.
The statistical linkage of Atlantic hurricane PDI to and Atlantic SST suggests at least the possibility of a large anthropogenic influence on Atlantic
hurricanes. If the correlation between tropical Atlantic SSTs and hurricane activity is used to infer future changes in Atlantic hurricane activity,
the implications are sobering: the large increases in tropical Atlantic SSTs projected
for the late 21st century would imply very substantial increases in hurricane destructive
potential--roughly a 300% increase in the PDI by 2100 (Figure 1 a).
On the other hand, Swanson (2008) and others have noted that Atlantic hurricane power dissipation is also well-correlated with other SST indices besides tropical Atlantic SST alone, and in particular with indices of Atlantic SST relative to tropical mean SST (e.g., Figure 1b from Vecchi et al. 2008). This is in fact a crucial distinction, because the statistical relationship between Atlantic hurricanes and local Atlantic SST shown in the upper panel of Figure 1 would imply a very large increases in Atlantic hurricane activity (PDI) due to 21st century greenhouse warming, while the statistical relationship between the PDI and the alternative relative SST measure shown in the lower panel of Figure 1 would imply only modest changes of Atlantic hurricane activity (PDI) with greenhouse warming. In the latter case, the alternative relative SST measure in the lower panel does not change very much over the 21st century in global warming projections from climate models, because the warming projected for the tropical Atlantic in the models is not very different from that projected for the tropics as a whole.
A key question then is: Which of the two future Atlantic hurricane scenarios inferred from the statistical relations in Figure 1 is more likely? To try to gain insight on this question, we have first attempted to go beyond the ~50 year historical record of Atlantic hurricanes and SST to examine even longer records of Atlantic tropical storm activity and second to examine dynamical models of Atlantic hurricane activity under global warming conditions. These separate approaches are discussed below.
C. Analysis of century-scale Atlantic tropical storm and hurricane records
To gain more insight on this problem, we have attempted to analyze much longer (> 100 yr) records of Atlantic hurricane activity. If greenhouse warming causes a substantial increase in Atlantic hurricane activity, then the century scale increase in tropical Atlantic SSTs since the late 1800s should have produced a long-term rise in measures of Atlantic hurricanes activity.
Existing records of past Atlantic tropical storm or hurricane numbers (1878 to present) in
fact do show a pronounced upward trend, which is also correlated with rising SSTs
(e.g., see blue curve in Fig. 4 or
Vecchi and Knutson 2008). However, the density of reporting ship traffic over
the Atlantic was relatively sparse during the early decades of this record, such
that if storms from the modern era (post 1965) had hypothetically occurred during
those earlier decades, a substantial number would likely not have been directly
observed by the ship-based "observing network of opportunity." We find that, after
adjusting for such an estimated number of missing storms, there is a small nominally
positive upward trend in tropical storm occurrence from 1878-2006. But statistical
tests reveal that this trend is so small, relative to the variability in the series,
that it is not significantly distinguishable from zero (Figure 2). In addition,
Landsea et al. (2010) note that the rising trend in Atlantic tropical storm
counts is almost entirely due to increases in short-duration (<2 day) storms alone.
Such short-lived storms were particularly likely to have been overlooked in the
earlier parts of the record, as they would have had less opportunity for chance
encounters with ship traffic.
If we instead consider Atlantic basin hurricanes, rather than all Atlantic tropical storms, the result is similar: the reported numbers of hurricanes were sufficiently high during the 1860s-1880s that again there is no significant positive trend in numbers beginning from that era (Figure 3, black curve, from CCSP 3.3 (2008) ). This is without any adjustment for "missing hurricanes".
The evidence for an upward trend is even weaker if we look at U.S. landfalling hurricanes, which even show a slight negative trend beginning from 1900 or from the late 1800s (Figure 3, blue curve). Hurricane landfalling frequency is much less common than basin-wide occurrence, meaning that the U.S. landfalling hurricane record, while more reliable than the basin-wide record, suffers from degraded signal-to-noise characteristics for assessing trends.
While major hurricanes (Figure 3, red curve) show more evidence of a rising trend from the late 1800s, the major hurricane data are considered even less reliable than the other two records in the early parts of the record. Category 4-5 hurricanes show a pronounced increase since the mid-1940s (Bender et al., 2010) but again, we consider that these data need to be carefully assessed for data inhomogeneity problems before such trends can be accepted as reliable.
The situation for Atlantic hurricane long-term records is summarized in Figure 4. While global mean temperature and tropical Atlantic SSTs show pronounced and statistically significant warming trends (green curves), the U.S. landfalling hurricane record (orange curve) shows no significant increase or decrease. The unadjusted hurricane count record (blue curve) shows a significant increase in Atlantic hurricanes since the early 1900s. However, when adjusted with an estimate of storms that stayed at sea and were likely “missed” in the pre-satellite era, there is no significant increase in Atlantic hurricanes since the late 1800s (red curve). While there have been increases in U.S. landfalling hurricanes and basin-wide hurricane counts since the since the early 1970s, Figure 5 shows that these increases are not representative of the behavior seen in the century long records. In short, the historical Atlantic hurricane record does not provide compelling evidence for a substantial greenhouse warming induced long-term increase.
D. Model simulations of greenhouse warming influence on Atlantic hurricanes
Direct model simulations of hurricane activity under climate change scenarios
offer another perspective on the problem. We have developed a regional dynamical
downscaling model for Atlantic hurricanes and tested it by comparing with observed
hurricane activity since 1980. This model, when forced with observed sea surface
temperatures and atmospheric conditions, can reproduce the observed rise in hurricane
counts between 1980 and 2012, along with much of the interannual variability (Figure 5). Animations showing the development and evolution of hurricane activity in the model are available here.
Turning to future climate projections, current climate models suggest that tropical Atlantic SSTs will warm dramatically during the 21st century, and that upper tropospheric temperatures will warm even more than SSTs. Furthermore, most of the models project increasing levels of vertical wind shear over parts of the western tropical Atlantic (see Vecchi and Soden 2007). Both the increased warming of the upper troposphere relative to the surface and the increased vertical wind shear are detrimental factors for hurricane development and intensification, while warmer SSTs favor development and intensitification. To explore which effect of these effects might "win out", we can run experiments with our regional downscaling model.
Our regional model projects that Atlantic hurricane and tropical storms are
substantially reduced in number, for the average 21st century climate
change projected by current models, but have higher rainfall rates,
particularly near the storm center. The average intensity of the storms that
do occur increases by a few percent (Figure 6), in general agreement with previous
studies using other relatively high resolution models, as well as with hurricane
intensity theory (Emanuel 1987).
Earlier, Knutson and Tuleya (2004) estimated the rough order of magnitude of
the hurricane sensitivity to be about 4% per deg C SST warming for maximum intensities
and about 12% per deg C for near-storm (100 km radius) rainfall rates (see also
Knutson and Tuleya (2008) abstract
here). These sensitivity estimates have
considerable uncertainty, as
CCSP 3.3 (2008) , gives
an estimated range of 1-8% per deg C SST warming for hurricane intensity, and 6-18%
per deg C for near-storm rainfall rates.
A review of existing studies, including the ones cited above, lead us to conclude that it is likely that greenhouse warming will cause hurricanes in the coming century to be more intense globally and have higher rainfall rates than present-day hurricanes.
Turning now to the important question of the frequency of very intense hurricanes, the regional model of Knutson et al. (2008) has an important limitation in that it does not simulate such very intense hurricanes. For example, the maximum surface wind in the simulated hurricanes from that model is less than 50 m/s (which is borderline category 3 hurricane intensity). Furthermore, the idealized study of Knutson and Tuleya (2004) assumed the existence of hurricanes and then simulated how intense they would become. Thus, that study could not address the important question of the frequency of intense hurricanes.
In our latest Atlantic basin dynamical downscaling studies (Bender et al. 2010; Knutson et al. 2013), we have tried to address both of these limitations by letting the Atlantic basin regional model of Knutson et al. (2008) provide the overall storm frequency information, and then downscaling each individual storm from the regional model study into the GFDL hurricane prediction system. The GFDL hurricane model (with a grid spacing as fine as 9 km) is able to simulate the frequency, intensity, and structure of the more intense hurricanes, such as category 3-5 storms, much more realistically than the regional (18 km grid) model.
Using this additional downscaling step, the new GFDL hurricane model study is able to reproduce some important historical characteristics of very intense Atlantic hurricanes, including the wind speed distribution and the change of this distribution between active and inactive decadal periods of hurricane activity (Fig. 1 of Bender et al. 2010). The model also supports the notion of a substantial decrease (~25%) in the overall number of Atlantic hurricanes and tropical storms with projected 21st century climate warming. However, using the CMIP3 and CMIP5 multi-model climate projections, the hurricane model also projects that the lifetime maximum intensity of Atlantic hurricanes will increase by about 5% during the 21st century in general agreement with previous studies. The hurricane model further projects a significant increase (+90%) in the frequency of very intense (category 4 and 5) hurricanes using the CMIP3/A1B 18-model average climate change projection (Fig. 7). Downscaled projections using CMIP5 multi-model scenarios (RCP4.5) as input (Knutson et al. 2013) still show increases in category 4 and 5 storm frequency, but these are only marginally significant for the early 21st century (+45%) or the late 21st century (+40%) CMIP5 scenarios. Downscaling individual CMIP3 model projections instead of the multi-model ensemble, we find that three of ten models produced a significant increase in category 4 and 5 storms, while the other seven produced no significant change. While multi-model ensemble results are probably more reliable than individual model results, each of the individual model results can be viewed as at least plausible at this time.
Returning to the issue of future projections of aggregate activity (PDI, as in Fig. 1), while there remains a lack of consensus among various studies on how Atlantic hurricane PDI will change, no model we have analyzed shows a sensitivity of Atlantic hurricane PDI to greenhouse warming as large as that implied by the observed Atlantic PDI/local SST relationship shown in Figures 1 (top panel). In other words, there is little evidence from current dynamical models that 21st century climate warming will lead to large (~300%) increases in tropical storm numbers, hurricane numbers, or PDI in the Atlantic. As noted above, there is some indication from high resolution models of substantial increases in the numbers of the most intense hurricanes even if the overall number of tropical storms or hurricanes decreases. In the Bender et al. 2010 study, we estimate that the effect of increasing category 4-5 storms outweighs the reduction in overall hurricane numbers such that we project (very roughly) a 30% increase in potential damage in the Atlantic basin by 2100. This estimate does not include the influence of future sea level rise or other important factors such as coastal development or changes in building practices.
Finally, one can ask whether the change in Category 4-5 hurricanes projected by our model is already detectable in the Atlantic hurricane records. Owing to the large interannual to decadal variability of SST and hurricane activity in the basin, Bender et al (2010) estimate that detection of this projected anthropogenic influence on hurricanes should not be expected for a number of decades. While there is a large rising trend since the mid 1940's in category 4-5 numbers in the Atlantic, our view is that these data are not reliable for trend calculations, until they have been further assessed for data homogeneity problems, such as those due to changing observing practices.
E. Other possible human influences on Atlantic hurricane climate
Apart from greenhouse warming, other human influences conceivably could have
contributed to recent observed increases in Atlantic hurricanes. For example, Mann
and Emanuel (2006) hypothesize that a reduction in aerosol-induced cooling over
the Atlantic in recent decades may have contributed to the enhanced warming of the
tropical North Atlantic, relative to global mean temperature. However, the cause
or causes of the recent enhanced warming of the Atlantic, relative to other tropical
basins, and its effect on Atlantic tropical cyclones, remains highly uncertain (e.g., Booth et al. 2012; Zhang et al. 2013; Dunstone et al. 2013; Villarini and Vecchi 2013). A number of anthropogenic and natural factors
(e.g., aerosols, greenhouse gases, volcanic activity, solar variability, and internal
climate variability) must be considered as potential contributors, and the science
remains highly uncertain in these areas. Finally recent work (Kossin et al. 2014; see GFDL Research Highlight) indicates that the latitude at which the maximum intensity of tropical cyclones occurs has expanded poleward globally in recent decades, although the causes for this have not been firmly established and a significant change was not seen in the Atlantic basin statistics.
Sea level rise must also be considered as a way in which human-caused climate change can impact Atlantic hurricane climate--or at least the impacts of the hurricanes at the coast. The vulnerability of coastal regions to storm-surge flooding is expected to increase with future sea-level rise and coastal development, although this vulnerability will also depend upon future storm characteristics, as discussed above. There are large ranges in the 21st Century projections for both Atlantic hurricane characteristics and for the magnitude of regional sea level rise along the U.S. coastlines. However, according to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, the average rate of global sea level rise over the 21st Century will very likely exceed that observed during 1961-2003 for a range of future emission scenarios.
F. Synthesis and Summary
In summary, neither our model projections for the 21st century nor our analyses of trends in Atlantic hurricane and tropical storm counts over the past 120+ yr support the notion that greenhouse gas-induced warming leads to large increases in either tropical storm or overall hurricane numbers in the Atlantic. A new modeling study projects a large (~100%) increase in Atlantic category 4-5 hurricanes over the 21st century, but we estimate that this increase may not be detectable until the latter half of the century.
Therefore, we conclude that despite statistical correlations between SST and Atlantic hurricane activity in recent decades, it is premature to conclude that human activity--and particularly greenhouse warming--has already caused a detectable change in Atlantic hurricane activity. ("Detectable" here means the change is large enough to be distinguishable from the variability due to natural causes.) However, human activity may have already caused some some changes that are not yet detectable due to the small magnitude of the changes or observation limitations, or are not yet properly modeled (e.g., aerosol effects on regional climate).
We also conclude that it is likely that climate warming will cause hurricanes
in the coming century to be more intense globally and to have higher rainfall rates
than present-day hurricanes. In our view, there are better than even odds that the
numbers of very intense (category 4 and 5) hurricanes will increase by a substantial
fraction in some basins, while it is likely that the annual number of tropical storms globally will either decrease or remain essentially unchanged. These assessment statements are intended to apply to climate warming of the type projected for the 21st century by IPCC AR4 scenarios, such as A1B.
The relatively conservative confidence levels attached to these projections, and the lack of a claim of detectable anthropogenic influence at this time contrasts with the situation for other climate metrics, such as global mean temperature. In the case of global mean surface temperature, the IPCC 5th Assessment Report (2013) presents a strong body of scientific evidence that most of the global warming observed over the past half century is very likely due to human-caused greenhouse gas emissions.
2. Global Warming and Tropical Cyclone Activity Around the Globe
The main focus of this web page is on Atlantic hurricane activity and global warming. However, an important question concerns whether global warming has or will substantially affect tropical cyclone activity in other basins.
In terms of historical tropical cyclone activity, a 2010 WMO assessment of tropical cyclones and climate change concluded that "it remains uncertain whether past changes in tropical cyclone activity have exceeded the variability expected from natural causes." This conclusion applied to all basins around the globe.
For future projections, GFDL atmospheric modelers have developed global models capable of simulating many aspects of the seasonal and year-to-year variability of tropical cyclone frequency in a number of basins, using only historical sea surface temperatures as input. Examples of the performance of this model on historical data are provided on this web page.
Figure 8 shows the distribution of late-21st-century projected changes in annual hurricane genesis frequency as obtained using a GFDL atmospheric model forced by sea surface temperature projections from several CMIP3 climate models. The top panel is the average frequency change based on individually downscaling the sea surface temperature changes from eight different CMIP3 climate models. Dots show where downscaled results from at least six of eight models agree on the sign of the storm frequency change. The bottom panel shows the downscaled result using the average sea surface temperature change from 18 CMIP3 models. The CO2 concentration in the downscaling model was doubled for all warm climate experiments. The results show a general preponderance of negative changes (i.e., reduced future hurricane genesis frequency) in most basins, with the exception being the Northeast Pacific basin. The tendency for models to project a reduction in global tropical cyclone frequency over the 21st century with climate warming was found across a range of studies as reviewed in the 2010 WMO assessment.
3. Recent GFDL Papers, Collaborative Papers, Commentary, & Animations on Global Warming and Hurricanes
- The poleward migration of the location of tropical cyclone maximum intensity, (Kossin, Emanuel, and Vecchi; Nature 2014). (see GFDL Research Highlight)
- Dynamical Downscaling Projections of Twenty-First-Century Atlantic Hurricane Activity: CMIP3 and CMIP5 Model-Based Scenarios, J. Climate 2013.
- TC-permitting GCM simulations of hurricane frequency response to sea surface temperature anomalies projected for the late 21st century. J. Climate 2012.
- Modeled Impact of Anthropogenic Warming on the Frequency of Intense Atlantic Hurricanes. Science 2010
- FAQ (Frequently Asked Questions) on our recent Science paper (updated Jan. 25, 2010)
- How well do we know past Atlantic hurricane activity? A web site on adjusting for "missing storms" in the past Atlantic hurricane data.
- Simulations of global hurricane climatology, interannual variability, and response to global warming using a 50km resolution GCM. J. Climate, 2009.
- Simulated reduction in Atlantic hurricane frequency under twenty-first-century warming conditions, Nature Geoscience, doi:10.1038/ngeo202 (Published online May 2008)
- FAQ (Frequency Asked Questions) on our recent Nature Geoscience study (Posted June 11, 2008)
- On Estimates of Historical North Atlantic Tropical Cyclone Activity -- J. Climate (July 15, 2008 issue)
- Simulation of the recent multidecadal increase of Atlantic hurricane activity -- BAMS (October 2007 issue)
- Simulated Hurricane Animations Web Page
4. WMO Expert Team 2010 Assessment of Tropical Cyclones and Climate Change
Cyclones and Climate Change", an assessment by a World
Meteorological Organization Expert Team on Climate Change Impacts on
Tropical Cyclones is now available. This assessment was published in Nature Geoscience (March 2010). For more information on the expert team, see this WMO
This report assesses published research on "Tropical Cyclones and Climate Change" from the international scientific literature.
5. An Overview of Earlier GFDL Research on Global Warming and Hurricanes
The strongest hurricanes in the present climate may be upstaged by even more intense hurricanes over the next century as the earth's climate is warmed by increasing levels of greenhouse gases in the atmosphere. Although we cannot say at present whether more or fewer hurricanes will occur in the future with global warming, the hurricanes that do occur near the end of the 21st century are expected to be stronger and have significantly more intense rainfall than under present day climate conditions. This expectation (Figure 1) is based on an anticipated enhancement of energy available to the storms due to higher tropical sea surface temperatures.
The results shown in Figure 1 are based on a simulation study carried out by Thomas R. Knutson and Robert E. Tuleya at NOAA's Geophysical Fluid Dynamics Laboratory (GFDL). In this study hurricanes were simulated for a climate warming as projected to occur with a substantial build-up of atmospheric CO2. An increase of intensity of about one-half category on the Saffir-Simpson scale was simulated for an 80 year build-up of atmospheric CO2 at 1%/yr (compounded). For hurricane wind speeds, our model shows a sensitivity of about 4% per degree Celsius increase in tropical sea surface temperatures, with a larger percentage increase in near-storm rainfall.
Early Studies on Global Warming and Hurricanes
An increase in the upper-limit intensity of hurricanes with global warming was suggested on theoretical grounds by M.I.T. Professor Kerry Emanuel in 1987. In the late 1990s, Knutson, Tuleya, and Kurihara at GFDL/NOAA began simulating samples of hurricanes from both the present-day climate and from a greenhouse-gas warmed climate. This was done by "telescoping-in" on coarsely resolved tropical storms in GFDL's global climate model using the high-resolution GFDL hurricane prediction model (Figure 2). A research report describing this work was published in Science (1998), with a more detailed paper in Climate Dynamics (1999, vol. 15). All of these studies, as well as our more recent ones, include the moderating effect of atmospheric stabilization aloft under high CO2 conditions, rather than simply increasing the sea surface temperature alone.
In a follow-up study, which appeared in the Journal of Climate (2001), NOAA scientists Knutson and Tuleya teamed up with Isaac Ginis and Weixing Shen of the University of Rhode Island to explore the climate warming/ hurricane intensity issue using hurricane model coupled to a full ocean model. The coupled model was used to simulate the "cool SST wake" generated by the hurricanes as they moved over the simulated ocean (Figure 3). The model simulations including this additional feedback still showed a similar percentage increase of hurricane intensity under warm climate conditions as the original model without ocean coupling.
A comprehensive idealized hurricane intensity modeling study by Knutson and Tuleya, published in Journal of Climate (2004), confirms the general conclusions of previous studies but makes them more robust by using future climate projections from nine different global climate models and four different versions of the GFDL hurricane model. The GFDL hurricane model used for the study is an enhanced resolution version of the model used to predict hurricanes operationally at NOAA's National Centers for Environmental Prediction. According to this latest study, an 80 year build-up of atmospheric CO2 at 1%/yr (compounded) leads to roughly a one-half category increase in potential hurricane intensity on the Saffir-Simpson scale and an 18% increase in precipitation near the hurricane core. A 1%/yr CO2 increase is an idealized scenario of future climate forcing. As noted by the Intergovernmental Panel on Climate Change (IPCC), there is considerable uncertainty in projections of future radiative forcing of earth's climate. A criticism of our paper by Michaels et al. is responded to here.
An implication of the GFDL studies is that if the frequency of tropical cyclones remains the same over the coming century, a greenhouse-gas induced warming may lead to an increasing risk globally in the occurrence of highly destructive category-5 storms.