Projected changes in tropical cyclone activity under future warming scenarios using a high-resolution climate model


This study examines how characteristics of tropical cyclones (TCs) that are explicitly resolved in a global atmospheric model with horizontal resolution of approximately 28 km are projected to change in a warmer climate using bias-corrected sea-surface temperatures (SSTs). The impact of mitigating from RCP8.5 to RCP4.5 is explicitly considered and is compared with uncertainties arising from SST projections. We find a reduction in overall global TC activity as climate warms. This reduction is somewhat less pronounced under RCP4.5 than under RCP8.5. By contrast, the frequency of very intense TCs is projected to increase dramatically in a warmer climate, with most of the increase concentrated in the NW Pacific basin. Extremes of storm related precipitation are also projected to become more common. Reduction in the frequency of extreme precipitation events is possible through mitigation from RCP8.5 to RCP4.5. In general more detailed basin-scale projections of future TC activity are subject to large uncertainties due to uncertainties in future SSTs. In most cases these uncertainties are larger than the effects of mitigating from RCP8.5 to RCP4.5.


Tropical cyclones (TCs) have been explicitly simulated in a number of global climate models with resolutions finer than 50 km. Despite some sensitivities to model physics and dynamical core design, climate models running at high resolution are capable of capturing many aspects of the observed TC climatology with reasonable fidelity, including geographical, seasonal and even inter-annual variations (e.g.; Zhao et al. 2009; Manganello et al. 2012; Zhao et al. 2012; Murakami et al. 2012a, 2012b, Knutson et al. 2015). The Community Atmosphere Model (CAM) has also produced reasonable simulations of TC climatology (Bacmeister et al. 2014; Reed et al. 2015) when run at horizontal resolutions of around 28 km.

Determining possible changes to the global TC climatology in a warming climate is a critical but vexing problem for models (e.g.; Murakami et al. 2012a, 2012b; Walsh et al. 2015). Indices of TC activity derived from large scale atmospheric quantities in low-resolution simulations, e.g., Genesis Potential Index (GPI; Emanuel and Nolan 2004) do not always agree with explicit simulations of TC activity in high-resolution simulations using the same climate model (e.g.,Wehner et al. 2015). Explicit counts of storms in low-resolution simulation with a given model are also poor predictors of TC counts in the same model at higher resolution (Wehner et al. 2015). Thus, explicit projections of TC activity from high resolution models may provide new information not available from more economical approaches.

In addition to the obstacle posed by the resolution sensitivity of climate change projections in TC activity, a second major obstacle is posed by the scenario design itself. A straightforward approach to assessing future TC activity would be to simply conduct high-resolution experiments using the fully-coupled Community Earth System Model (CESM) into the future using Representative Concentration Pathways (RCP) scenario forcing. However, CESM at high horizontal resolution retains substantial sea surface temperature (SST) biases in the tropical Atlantic and Pacific oceans (Small et al. 2014). These biases are similar to those common in lower-resolution CMIP5 models (e.g; Wang et al. 2014) and are seen to have a noticeable negative impact on the climatology of TCs at high resolution, particularly in the N Atlantic and NE Pacific basins (Small et al. 2014). Idealized prescribed scenarios such as uniform 2 K SST warming with a forced doubling of CO2 (e.g., Walsh et al. 2015; Wehner et al. 2015) avoid these biases, but do not capture geographic variation in warming.

In this study we have chosen to use a hybrid approach to force high-resolution atmospheric simulations. We force the atmospheric model using SSTs from fully-coupled RCP4.5 and RCP8.5 CESM1 simulations at a lower standard resolution, but apply a bias correction to the SSTs based on present day biases in CESM (e.g.; Murakami et al. 2012a). We perform a total of 11 simulations (Table 1, Section 2). A total of 4 future SSTs are used, of which two are derived from the 30-member Large Ensemble (LE, Kay et al. 2015). This study complements that of Done et al. (2015) which uses large-scale model fields from lower-resolution ensembles to make projections of hurricane damage in the future. Simulated storm tracks from the simulations described here, are also used by Gettelman et al. (2016) as inputs for a statistical model of storm damage used in the re-insurance industry.

Table 1 Experiment names and configurations used in this study. Asterisks following “BRACE set” experiments are intended as a reminder that 1o air/sea coupling was used in these runs

This paper is part of a larger project on the Benefits of Reducing Anthropogenic Climate changE (BRACE; O’Neill and Gettelman in preparation). It seeks to understand the impacts of mitigation from RCP8.5 to RCP4.5 – the “BRACE signal” – on TC activity, and to compare it with differences related to uncertainties in SST projections derived from coupled climate models.

Model, experiments, and analysis methods

Model configuration and experimental setup

This study uses the Community Atmosphere Model (CAM) - the atmospheric component of CESM with version 5 atmospheric physics as described by Neale et al. (2012). We utilize the atmospheric spectral element dynamical core (SE; Dennis et al. 2011) at a horizontal resolution of around 28 km. CAM5 is able to capture TCs at these high horizontal resolutions (e.g., Bacmeister et al. 2014). Reed et al. (2015) discuss the sensitivity of tropical cyclone activity in decadal simulations with CAM5 to the dynamical core used. Our study uses prescribed SSTs in both present day and future runs. Present day simulations use observed monthly-mean SSTs (Hurrell et al. 2008). Our technique for generating bias corrected future SSTs from fully-coupled simulations is described in the Supplementary Material and in Section 2.2. All forcing other than SST is historical through 2005 followed by RCP4.5 or RCP8.5 forcing (VanVuuren et al. 2011) for 2006 onwards.

The high-resolution runs discussed here represent the “bleeding edge” of what is feasible for extended (multi-decadal) runs with CAM. They are demanding in terms of both human and computational resources. Simulation costs for the 28 km configuration of CAM-SE are approximately 100 times that of the LE configuration used by several of the studies in this issue. In addition, performing 28-km simulations requires significant investment of human resources to port and optimize model codes on numerous high-performance computing (HPC) platforms, and to manage the large volumes of high-resolution output generated. As a consequence the number and length of high-resolution simulations we can perform and analyze is small compared to the lower resolution CESM runs discussed elsewhere in this issue.

Some aspects of the CESM high-resolution configuration are still evolving. Initially, we used a coarse 1o ocean model grid to couple between ocean and atmosphere. This was done in part to minimize the number of separate boundary forcing data sets in CESM. However, we discovered that this coupling had detectable impacts on the climatology of tropical cyclone winds (Supplementary Material). Better physical consistency is obtained by coupling on the atmosphere’s higher resolution 28 km grid using SST data sets that have been pre-interpolated to this resolution. The discussion in the Supplementary Material shows that TC statistics for both configurations can be largely harmonized using a rescaling procedure based on pressure-wind relationships although some statistically significant differences remain.

We performed an initial set of 3 experiments (present day, RCP4.5 and RCP8.5) with coupling on the 1o degree grid. These were used to characterize the basic model response to a warming climate and to explore differences between RCP4.5 and RCP8.5 scenarios. We then performed two ensembles (3 members) in the present and in the future under RCP8.5, to characterize internal atmospheric variability. Ensemble members were generated by statistically perturbing initial temperatures at levels below 0.01 K, and used coupling on the 28 km atmospheric grid. Finally, in order to probe uncertainties introduced by future SST variability, two more runs were done using two different future SST data sets constructed from LE members. These additional SSTs were chosen based on a rough analysis of features thought to control N Atlantic hurricane activity.

Table 1 summarizes the runs examined in this study. Asterisks in the experiment names are to emphasize that the 1o coupler grid was used in these simulations.

Prescribed, bias-corrected SSTs

Fully coupled CESM1 retains substantial SST biases (Fig. S1) in the tropical Atlantic and Pacific oceans, which have a noticeable impact on the climatology of tropical cyclones (Small et al. 2014). Although coupled simulations would be preferable in many ways – more frequent air/sea coupling, realistic representation of TC cold wakes and consequent reduction of TC intensities – we decided that the large errors in TC climatology that result from coupled SST biases are unacceptable in a study that seeks to understand impact of climate change on TC statistics. In this study we utilize simulations with prescribed SSTs. Future SSTs are obtained from fully-coupled CESM1 simulations at standard 1o atmospheric and 1o ocean resolution. However, before applying these SSTs as forcing for our future simulations they are corrected by subtracting present day CESM1 biases with respect to observations. Our procedure for generating future SSTs is described in the Supplementary Material. A key assumption in our approach is that CESM biases will not change significantly with time.

Figure 1 illustrates the 4 bias corrected SSTs used in our simulations. SSTs for 2070–2090 in the single RCP4.5 scenario run (Fig. 1a) show 1 to 3 K warming over most of the world ocean. The RCP8.5 SST1 (Fig. 1b) shows warming of 2 to 3 K over most tropical ocean areas with even larger values (>3.5 K) in the eastern Pacific. The RCP4.5 SST is typically 0.5-1 K cooler over most of the tropics compared to RCP8.5 SST1. Both SSTs show a distinctive El Niño-like pattern of warming with the strongest tropical warming in the eastern equatorial Pacific. The off-equatorial tropical N. Atlantic in both SST sets is typically 0.5 to 1 K cooler than the tropical eastern Pacific. This relatively cool region is roughly coincident with the main development region (MDR) for Atlantic hurricanes and would therefore be expected to affect the formation of N. Atlantic storms (e.g. Zhao and Held 2012).

Fig. 1

a Annual mean (2070–2090) surface temperature difference from the present day (1985–2005) in RCP4.5 scenario. b Same as a except for RCP8.5 SST1. c Mean 2070–2090 difference in SST’s between RCP8.5 SST1 and SST2 (warmer Atlantic MDR). d Same as c except for RCP8.5 SST3 (warmer Atlantic colder tropics)

Figures 1c, d show the mean differences between the additional LE-based datasets and SST1 during the northern hemisphere warm season (July–November). The differences are small compared to the overall warming in Fig. 1b. However even small differences (0.1–0.5 K) in SSTs within main basin development regions relative to the tropical mean may affect cyclogenesis (Zhao and Held 2012). SST2 (Fig. 1c) has a broad tongue of slightly warmer SSTs in the eastern tropical Atlantic and a somewhat cooler central equatorial Pacific than SST1. SST3 (Fig. 1d) is sharply colder (0.2–0.4 K) over much of the tropics than SST1. In the tropical Atlantic SST3 and SST2 have similar east-west dipoles of cooling/warming with respect to SST1, but SST3 is slightly cooler overall in the tropical Atlantic than SST2. SST3 also possesses the least El Niño-like pattern of warming of the 3 SSTs used.

Cyclone tracker

The TC detection algorithm and tracker utilized for this analysis is that used and described in Zhao et al. (2009) with 3-hly model output. Following the same approach as in Bacmeister et al. (2014), the surface winds (commonly taken to be at a height of 10 m) used for the TC tracker are estimated using the lowermost model level winds at around 60 m and the power-wind law. The basic output of the tracker includes storm center longitude λ c ; n (t), latitude ϕ c ; n (t) as well as maximum wind V max ; n (t) and minimum central pressure p min ; n (t). The subscript n in these expressions designates the n-th storm identified by the tracker. The time variable t is discretized in 3-h intervals.

Track density

The storm location output from the tracker is used to calculate tropical cyclone track densities on a regular 4ox4o lat-lon grid. Occurrences of storms exceeding a given threshold within each 4ox4o gridbox are counted to give a “density” in units of hrs yr.−1 (4o)−2.

Storm precipitation

In addition to the standard tracker fields described above we also examine precipitation and storm size statistics for simulated TCs. We calculate average precipitation falling within 500 km of the storm center diagnosed by the TC tracking algorithm (e.g.; Jiang and Zipser 2010). This quantity \( {\mathcal{P}}_{500;n}(t) \) is calculated using instantaneous 3-hrly precipitation fields from the model and is analogous to other track quantities like V max ; n (t) and p min ; n (t). Storm size is discussed in the Supplementary Material.

Bootstrap analysis of significance

To test the significance of our results, we perform a simple bootstrap analysis (e.g.; Efron and Tibshirani 1998) using the period 1985–2005 for the present day runs and 2070–2090 for the future runs. We generate 2000 synthetic 20-year TC track files from each of our runs. Sampling is with replacement and individual storms are assigned to years according to their genesis time. PDFs of a statistic over the 2000 member synthetic ensembles are compared to evaluate significance.


BRACE signal

Figure 2 shows global maps of 20-year annual mean tropical cyclone track densities obtained from the BRACE set PD-1*, RCP4*, and RCP8-SST1–1* along with present day biases. The top row of Fig. 2 shows 20-year mean track densities for all storms (TS-Cat5). Detailed discussions of TC distributions in CAM5 versus observations can be found in Bacmeister et al. (2014), Wehner Small et al. (2014) and Reed et al. (2015). Biases with respect to IBTrACS (Knapp et al. 2010) are shown in the bottom row of Fig. 2. Here we will focus on projected changes to TC distributions in the future. At first glance there is little change in overall TC activity projected for 2070–2090, but closer inspection reveals that TC activity in the N. Atlantic decreases significantly under both RCP4* and RCP8-SST1–1*. In particular, in RCP8-SST-1* the largest TC densities decrease to 8–10 h yr.−1 4o-2 in the N Atlantic from values near 18 h yr.−1 4o-2 in PD-1*. The area of high track densities in the N. Atlantic is also dramatically smaller in RCP8.5 2070–2090 than in the present day. Under RCP4.5 N. Atlantic TC activity is reduced as well, but by a smaller amount. Elsewhere, small reductions in TC activity are also evident, e.g. central Pacific. Murakami et al. 2014 have analyzed the relationship between present day model biases and projections of future TC activity. This is discussed further in the Supplementary Material (S3).

Fig. 2

Track densities for the BRACE simulation set PD-1*, RCP4*, and RCP8-SST1–1*. The top row shows the track densities for all tropical cyclones (TS-Cat 5), while the middle row shows the track densities for all extreme storms (Cat 4 and 5). Units are average hours per year in which a storm is found within a 4o x 4o gridbox. Bottom panels show present day biases in PD-1* with respect to IBTrACS best-track estimates

The middle row of Fig. 2 shows track densities for extreme storms (Cat 4 and 5). There are pronounced increases in extreme storm activity in both RCP4* and RCP8-SST1–1*. These increases are most pronounced in the NW Pacific, but are also apparent in the Southern Indian Ocean near Madagascar and in the South Pacific east of Australia. Little or no increase in extreme storm frequency is projected for the N Atlantic or NE Pacific. Increases in extreme TC frequency under warming scenarios have been found in other studies (e.g; Murakami et al. 2012a, 2012b; Knutson et al. 2015). We note that our simulations underpredict present day Cat 4 and 5 storms in all basins (Fig. 2, bottom row) This may have implications for future projections.

Bootstrap results for global and basin average track densities are given in Table 2. Bootstrap ensemble PDFs for the BRACE set are shown in the top row of Fig. 3. There is a significant BRACE signal evident in global mean and N. Atlantic TS-Cat5 track densities. The global mean track density of TS-Cat5 shows significant decreases in the future; from a present day value of around 5.8 h yr.−1 4o-2 in PD-1* to a value of around 4.7 h yr.−1 4o-2 for 2070–2090 in RCP8-SST1–1*. Experiment RCP4* has an intermediate value of around 5.4 h yr.−1 4o-2. For Cat4–5 storms in the NW Pacific track densities for RCP4* and RCP8-SST1–1* overlap almost completely, but are clearly distinct from the PD-1* result. Both global mean and NW Pacific densities of Cat4–5 storms double in both future runs with respect to their value in PD-1*.

Table 2 Results from bootstrap analysis of tropical cyclone track densities (see text). Ensemble means from 2000 bootstrapped 20-year samples are shown along with 1st %-ile and 99th %-ile values in italics. Results are for 1985–2005 in the present day scenario and for 2070–2090 under RCP4.5 and RCP8.5. The RCP8.5 results include 4 experiments with SST1 as well as two additional experiments with SST2 and SST3. Results are shown for areal averages over the global tropics (42S-42N), N. Atlantic (80 W–20 W,10 N–38 N), NW. Pacific (120E-180,10 N–38 N), NE. Pacific (120 W–96 W,10 N–26 N), S. Pacific (140E-180, 30S–10S) and S. Indian Ocean (28E-68E, 34S–10S). Results from runs using the 1o are highlighted. Blank cells indicate that no storms were detected
Fig. 3

First column: Frequency distribution of 20-year mean global (42S-42N) averaged track densities for TS-Cat5. Horizontal axis is units of hrs yr.−1 4o-2. Distributions are accumulated over 2000 randomly sampled 20-year subsets of data for each experiment. The top panel shows results for PD-1* (black), RCP4* (green) and RCP8-SST1–1* (red). Middle panel shows results for PD-2,3, and 4. Bottom panel shows results for RCP8-SST1–2,3,4 (red) as well as RCP8-SST2 (light blue) and RCP8-SST3 (magenta). Second column: Same as first expect for N Atlantic means (80 W–20 W,10 N–38 N). Third column: Same as first except for for NW Pacific (120E-180, 10 N–38 N) Cat4–5 averaged track densities

As suggested by Fig. 2, the increased frequency of extreme storms is not uniform over all basins. It is dominated by strong increases in the NW Pacific basin with secondary centers of action in the western Southern Indian Ocean and the S Pacific. This is confirmed in Table 2. Extreme storm activity in the NW Pacific in the future is projected to more than double, and dramatic increases occur in the S Pacific and S Indian oceans as well. Importantly for the purposes of the BRACE study, the differences in extreme storm activity between RCP4* and RCP8-SST1–1* are generally small except in the S Pacific storm basin. It is also noteworthy that CAM5 projects no increased Cat4–5 frequency in the N. Atlantic.

Uncertainty and variability

Results from our variability set are shown in Table 2 as well as in the second and third rows of Fig. 3. First, we note that there remain significant differences between calculations using the 1o degree coupler grid and those using the 28-km coupler grid despite the rescaling described in the Supplementary Material. Runs with the 1o coupler grid exhibit smaller mean track densities for TS-Cat5 than those using the 28 km coupler grid for present day and future conditions both globally and in the N Atlantic basin and NW Pacific means for Cat4–5 are larger.

Despite differences in the means, the spread in the PDFs of mean track densities is similar using both coupler grids. This justifies the use of 28-km coupler runs to evaluate the significance of the BRACE signal obtained from PD-1*, RCP4* and RCP8-SST1–1*. The PDFs for PD-2,3, and 4 as well as RCP8-SST-2,3,4 overlap in the global means and for most major basins. This is evident in Fig. 3 as well as in the tabulated results in Table 2. This gives us confidence that 20-year means of track densities with given SST fields are sufficiently stable for use in detecting climate signals under both present day and future conditions.

From these results we argue that the difference between PD-1*, RCP4*, and RCP8-SST1–1* seen in global mean and N Atlantic mean track densities for TS-Cat5, i.e. the BRACE signal, is significant. The global BRACE signal is brought about by strong signals in NE Pacific (not shown) and N Atlantic. Other basins show projected decreases TS-Cat5 densities, but not significant differences between RCP4* and RCP8*. The same is true for the strong future increase in Cat4–5 densities projected in the NW Pacific in both RCP4* and RCP8-SST1–1*. It is worth noting here that Cat4–5 activity in CAM5 is overwhelmingly dominated by activity in the NW Pacific, and while 20-year means seem sufficient to characterize this activity it is not clear this is true of other basins. Table 2 shows large differences in 20-year Cat4–5 mean track densities among members of the variability set for both the S. Indian Ocean and S. Pacific basins.

Large uncertainties exist in future SSTs even under a single forcing scenario such as RCP8.5 (Kay et al. 2015). These must be considered when projecting future TC activity. As described in Section 2 we performed 2 additional simulations with SSTs derived from the LE selected because they possessed warmer tropical Atlantic SSTs (Fig. 1b,c). The light blue and magenta curves in the bottom row of Fig. 3 show bootstrap results for RCP8-SST2 and RCP8-SST3. The global mean track densities for TS-Cat5 appear relatively insensitive to SST choice (Fig. 3, lower left). This increases our confidence in the signal obtained from the BRACE set, which suggests that mitigating from RCP8.5 to RCP4.5 would result in greater overall TC activity globally.

At the basin scale and for Cat4–5 storms, SST differences are more significant. Track densities for TS-Cat5 in the N Atlantic are substantially higher in RCP8-SST2 and RCP8-SST3, with SST3 producing around 50 % more N. Atlantic activity than SST1,. In the NW Pacific both SST2 and SST3 produce a noticeable and significant decrease in the density of Cat 4 and 5 storms relative to SST1. This is reflected in global mean densities for Cat4–5, although in RCP8-SST2 the substantial decrease in the NW Pacific is offset by somewhat enhanced high track densities in the S. Indian Ocean, S. Pacific and NE Pacific. This offset does not appear in RCP8-SST3 perhaps related to the generally cooler tropics in SST3 (Fig. 1).

The use of different couplers for the BRACE set and for the variability set (Table 1) adds an unfortunate complication to the analysis, but it seems clear that SST uncertainties can have an impact comparable to mitigation from RCP8.5 to RCP4.5. The TS-Cat5 track densities in the N Atlantic in RCP8-SST3 for 2070–2090 are halfway between those for the present day and those for 2070–2090 using SST1. This is similar to the position of RCP4* results with respect to PD-1* and RCP8-SST1–1*. Even though different couplers are used in these comparisons we would argue that the size of the SST effect in the N Atlantic is similar to the BRACE signal in our experiments. A more complete exploration of SST variability under both RCP4.5 and RCP8.5 could reveal a robust BRACE signal, but the small number of SST realizations explored here is not sufficient to do so.

The relationship of basin-scale TC activity to SSTs is not yet completely understood (Bell and Chelliah 2006; Zhao and Held 2012). N Atlantic activity is believed to respond to both Atlantic and Pacific SSTs, with cool conditions in the E Pacific, as during La Niña events, enhancing Atlantic hurricanes, in contrast to suppression during warm E. Pacific/El Niño conditions (Gray 1984; Pielke Jr and Landsea 1999). Warm SSTs in the main development region (MDR) of any basin, relative to the tropical mean, may enhance TC activity in that basin (Zhao and Held 2012). Camargo and Sobel (2005) contend that El Niño conditions enhance TC activity in the NW Pacific. Our results are roughly consistent with all of these proposals. The generally El Niño-like character of the RCP4.5 and RCP8.5 warming signals (Fig. 1a, b) could explain both the enhancements in intense NW Pacific storms and the overall suppression of Atlantic activity. As the El Niño-like character of the warming is reduced, or alternatively the relative warmth of the tropical Atlantic is increased, as in SST3, N Atlantic TC activity increases and NW Pacific activity decreases.

Tropical cyclone precipitation

Extreme precipitation is expected to become more frequent under global warming (e.g.; O’Gorman and Schneider 2009). Knutson et al. (2010) also discuss precipitation increases in simulated TCs under global warming scenarios. Figure 4 (left) presents annual frequency histograms of\( {\mathcal{P}}_{500;n}(t) \), the areally-averaged precipitation within 500 km of diagnosed storm centers (Section 2.5). These histograms are compiled over each 3-hourly sample in the lifetime of each storm diagnosed by the tracker. The histograms on the left were compiled using the raw 3-hourly precipitation fields in each simulation. Note that these histograms are in terms of absolute annual frequency so that increased frequency represents an increased annual frequency of intense TC rain. It is immediately clear that strong precipitation (>40 mm d−1) near TCs is projected to become more common in warmer climates. The frequency of TC precipitation over 80 mm d−1 is projected to increase by a factor of nearly 10 in 2070–2090 under RCP8.5 over its present day value, and by a factor of 3 to 5 over present day frequency under RCP4.5. These increases with warming are generally consistent with the work of Villarini et al. (2014) and Wehner et al. (2015) which found that idealized warming scenarios produced substantial increases in mean daily and maximum instantaneous TC precipitation rates.

Fig. 4

a: Intensity distributions of mean 500 km storm-precipitation along storm tracks: black) 1985–2005; green) RCP4.5 2070–2090; and red) RCP8.5 2070–2090 b Same as a except for storm precipitation scaled by saturation specific humidity (see text)

Since saturation specific humidity q sat (T, p) is a rapidly increasing function of temperature for constant pressure we should expect a warmer climate to possess higher absolute specific humidities in the lower atmosphere, which will translate to higher overall rain rates if other aspects of meteorological forcing remain reasonably constant (Held and Soden 2006). To isolate the role of increased q sat in generating more frequent intense rainfall we scaled \( {\mathcal{P}}_{500;n} \) by the local saturation specific humidity before compiling frequency histograms. We used the relationship;

$$ {P^{*}}_{500;n}(t)=\frac{q_{sat}\left({T}_0,{p}_0\right)}{q_{sat}\left({T}_{s;n}(t),{p}_0\right)}{P}_{500;n}(t) $$

where T 0 is set to 300 K, T s ; n (t) is the average surface temperature within a 500 km radius following the n-th storm track, and p 0 is a reference pressure set to 1000 hPa. Assuming the moisture content of air entering TCs is controlled by low-level temperature near the storm, this scaling should remove the thermodynamic effect responsible for increased precipitation intensities.

Histograms of \( {{\mathcal{P}}^{*}}_{500} \)are shown in Figure 4 (right). We see that a large part, but not all, of the increased frequency of intense rain events has been removed by the q sat scaling. This suggests that most of the projected change in the intensity histograms of \( {\mathcal{P}}_{500} \) is a simple consequence of increased atmospheric humidity, but that some increase in the dynamical forcing of TC precipitation is also taking place. Wehner et al. (2015) also found increases in maximum TC precipitation in their idealized future SST simulations (i.e., uniformly increased SSTs of +2 K) that exceed what would be expected from increases in q sat alone. Storm size also shows a modest gain in the future (Fig. S5). Similar results are found when this analysis is performed for individual basins. The exception is the N Atlantic where decreases in projected TC frequency may compensate for warmer SSTs leading to lower absolute frequencies of TC precipitation for all intensities.

Discussion and conclusions

Frequency/track density

As in several other models (e.g. Knutson et al. 2010, etc.) CESM projects a modest (~20 %) decrease in global TS-Cat 5 frequency in warmer climates. According to our results, this potential “benefit” of a warmer climate could be reduced under RCP4.5 compared to the RCP8.5 scenarios (Figure 3, Table 2). Projections of TC activity for individual basins are difficult to make with confidence because of significant differences in projected SSTs at the basin scale, as well as remaining uncertainty about the factors that control TC activity in individual basins. Figure 3 suggests that a decrease in future TC activity in the N Atlantic is a robust feature of a warmer climate. However, the magnitude of the reduction depends on details of the future SSTs. In particular, our simulations suggest that uncertainties in future SSTs are as important in projecting future N Atlantic hurricane activity as the effects of mitigating from RCP8.5 to RCP4.5.

Other changes in future TC climatology projected by CESM under global warming are not so benign. CESM projects a significant increase (200–300 %) in the global frequency of very intense (Cat4–5) storms under RCP8.5. This increase is concentrated in several hotspots around the world, with by far the largest contribution coming in the NW Pacific basin. This feature appears in all of the future runs we have performed, and our experiments show no benefit from mitigating to RCP4.5 (Table 2, Fig. 3). We should emphasize that the projected increase in Cat4–5 storms in the NW Pacific is significantly affected by the choice of future SST (Fig. 3). Since only one RCP4.5 SST was used, it is possible we have missed a potential benefit of mitigation from RCP8.5 to RCP4.5.

Storm precipitation

The risk of intense TC precipitation (>50 mm d−1 average within 500 km of center) increases dramatically as the climate warms (Fig. 4). This increased risk is largely due to increased humidity in the atmosphere rather than intensified dynamical forcing and would be reduced by about a half by mitigating to RCP4.5 from RCP8.5. In the case of extreme TC precipitation the benefits of mitigation are clearly larger than the effects of SST uncertainties (Fig. S5).

Our experiments suggest that much of the increased impact from TCs in a warmer climate may be felt away from the United States and in quantities not commonly assessed in current impact models (e.g.; Gettelman et al. 2016) such as rainfall and storm size (Fig. S5). A complete picture of future societal and economic impacts of TCs will require extending impact models both geographically and in terms of physical variables considered.

The BRACE signal we were able to detect in this study is comparable to that arising from different future SSTs in many of the features we examined. A complete evaluation of the benefits of mitigation from RCP8.5 to RCP4.5 would require the use of a large number of SST projections under both scenarios. It is not clear whether projects such as the large ensemble (Kay et al. 2015) or even coupled model intercomparisons have spanned the range of possible future SSTs. Global mean track densities for TS-Cat5 appear to have a clear BRACE signal although it is not necessarily a beneficial one. Our projection for the global frequency of intense TC precipitation also shows a clear signal from mitigation to RCP4.5. However, for basin scale projections of TC frequency, or for projections of intense tropical cyclone activity, our study suggests that SST uncertainty is a key factor. This uncertainty is important not only in the context of the BRACE project, but also for any detailed projections of TC activity in a warming climate.


  1. Bacmeister JT et al. (2014) Exploratory high-resolution climate simulations using the community atmosphere model (CAM). J Clim 27(9):3073–3099

    Article  Google Scholar 

  2. Bell GD, Chelliah M (2006) Leading tropical modes associated with interannual and multidecadal fluctuations in North Atlantic hurricane activity. J Clim 19(4):590–612

    Article  Google Scholar 

  3. Camargo SJ, Sobel AH (2005) Western North Pacific tropical cyclone intensity and ENSO. J Clim 18(15):2996–3006

    Article  Google Scholar 

  4. Dennis J et al. (2011) CAM-SE: A scalable spectral element dynamical core for the Community Atmosphere Model. International Journal of High Performance Computing Applications 26:74–89

    Article  Google Scholar 

  5. Done JM, PaiMazumder D, Towler E et al (2015) Estimating Impacts of North Atlantic Tropical Cyclones Using an Index of Damage Potential. Clim Change. doi:10.1007/s10584-015-1513-0

  6. Efron B, Tibshirani RJ (1998) An Introduction to the bootstrap. Monographs on Statistics and Applied Probablity 57. Chapman and Hall/CRC Press, Boca Raton FL. 435 pp.

  7. Emanuel KA, D Nolan, 2004: Tropical cyclone activity and the global climate system. Preprints, 26th Conf. on Hurricanes and Tropical Meteorology, Miami, FL, Amer. Meteor. Soc., 10 A.2. [Available online at]

  8. Gettelman, A., D. Bresch, C. C. Chen, et al. (2016) Projections of future tropical cyclone damage with a high resolution global climate model. Climatic Change. (in review)

  9. Gray WM (1984) Atlantic seasonal hurricane frequency. Part I: El Nino and 30 mb quasi-biennial oscillation influences. Mon Weather Rev 112(9):1649–1668

    Article  Google Scholar 

  10. Held IM, Soden BJ (2006) Robust responses of the hydrological cycle to global warming. J Clim 19(21):5686–5699

    Article  Google Scholar 

  11. Hurrell JW, Hack JJ, Shea D, Caron JM, Rosinski J (2008) A new sea surface temperature and sea ice boundary dataset for the Community Atmosphere Model. J Clim 21:5145–5153. doi:10.1175/2008JCLI2292.1.

    Article  Google Scholar 

  12. Jiang H, Zipser E (2010) Contribution of tropical cyclones to the global precipitation from eight seasons of TRMM data: Regional, seasonal, and interannual variations. J Clim 23:1526–1543

    Article  Google Scholar 

  13. Kay J et al. (2015) The Community Earth System Model (CESM) Large Ensemble Project: A Community Resource for Studying Climate Change in the Presence of Internal Climate Variability. Bull. Amer. Meteorol. Soc. doi:10.1175/BAMS-D-13-00255.1

    Google Scholar 

  14. Knapp KM et al. (2010) The International Best Track Archive for Climate Stewardship (IBTrACS): Unifying tropical cyclone data. BAMS 91:363–376

    Article  Google Scholar 

  15. Knutson TR, McBride JL, Chan J, Emanuel K, Holland G, Landsea C, Held I, Kossin JP, Srivastava AK, Sugi M (2010) Tropical cyclones and climate change. Nat Geosci 3(3):157–163

    Article  Google Scholar 

  16. Knutson TR et al. (2015) Global Projections of Intense Tropical Cyclone Activity for the Late Twenty-First Century from Dynamical Downscaling of CMIP5/RCP4.5 Scenarios. J Clim 28(18). doi:10.1175/JCLI-D-15-0129.1

  17. Manganello JV et al. (2012) Tropical cyclone climatology in a 10-km global atmospheric GCM: toward weather-resolving climate modeling. J Clim 25(11):3867–3893

    Article  Google Scholar 

  18. Murakami H, Mizuta R, Shindo E (2012a) Future changes in tropical cyclone activity projected by multi-physics and multi-SST ensemble experiments using the 60-km-mesh MRI-AGCM. Clim Dyn 39(9–10):2569–2584

    Article  Google Scholar 

  19. Murakami H et al. (2012b) Future changes in tropical cyclone activity projected by the new high-resolution MRI-AGCM. J Clim 25:3237–3260

    Article  Google Scholar 

  20. Murakami H, Hsu P-C, Arakawa O, Li T (2014) Influence of model biases on projected future changes in tropical cyclone frequency of occurrence. J Clim 27:2159–2181

    Article  Google Scholar 

  21. Neale, R. J. et al. (2012) Description of the NCAR Community Atmosphere Model (CAM 5.0). NCAR Tech. Note NCARTN-4861STR, 274 pp

  22. O’Gorman PA, Schneider T (2009) The physical basis for increases in precipitation extremes in simulations of 21st-century climate change. Proc Natl Acad Sci 106(35):14773–14777

    Article  Google Scholar 

  23. Pielke Jr RA, Landsea CN (1999) La nina, el nino and atlantic hurricane damages in the united states. Bull Am Meteorol Soc 80(10):2027–2033

    Article  Google Scholar 

  24. Reed KA, Bacmeister JT, Rosenbloom N et al. (2015) Impact of the dynamical core on the direct simulation of tropical cyclones in a high-resolution global model. Geophys Res Lett 42(9):3603–3608

  25. Small RJ et al. (2014) A new synoptic scale resolving global climate simulation using the Community Earth System Model. Journal of Advances in Modeling Earth Systems 6(4):1065–1094

    Article  Google Scholar 

  26. VanVuuren DP et al. (2011) Representative concentration pathways: an overview. Clim Chang 109:1–2 5–31

    Article  Google Scholar 

  27. Villarini G et al. (2014) Sensitivity of tropical cyclone rainfall to idealized global-scale forcings. J Clim 20(12):2307–2314

    Google Scholar 

  28. Walsh KJ et al. (2015) Hurricanes and climate: the US CLIVAR working group on hurricanes. Bull Am Meteorol Soc 96:997–1017. doi:10.1175/BAMS-D-13-00242.1

    Article  Google Scholar 

  29. Wang C, Zhang L, Lee SK, Wu L, Mechoso CR (2014) A global perspective on CMIP5 climate model biases. Nat Clim Chang 4(3):201–205

    Article  Google Scholar 

  30. Wehner M et al. (2015) Resolution dependence of future tropical cyclone projections of CAM5. 1 in the US CLIVAR Hurricane Working Group idealized configurations. J Clim 28(10):3905–3925

    Article  Google Scholar 

  31. Zhao M, Held IM (2012) TC-permitting GCM simulations of hurricane frequency response to sea surface temperature anomalies projected for the late-twenty-first century. J Clim 25(8):2995–3009

    Article  Google Scholar 

  32. Zhao M, Held IM, Lin SJ, Vecchi GA (2009) Simulations of global hurricane climatology, interannual variability, and response to global warming using a 50-km resolution GCM. J Clim 22(24):6653–6678

    Article  Google Scholar 

  33. Zhao M, Held IM, Lin S-J (2012) Some counter-intuitive dependencies of tropical cyclone frequency on parameters in a GCM. J Atmos Sci 69(7). doi:10.1175/JAS-D-11-0238.

Download references


Computing resources for this work were provided by; The Argonne Leadership Computing Facility at Argonne National Laboratory (Office of Science of the US Department of Energy) through the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program, and the Climate Simulation Laboratory at NCAR’s Computational and Information Systems Laboratory, sponsored by the National Science Foundation and other agencies. This work also utilized part of the “Using Petascale Computing Capabilities to Address Climate Change Uncertainties” PRAC allocation support by the National Science Foundation (NSF), and the Blue Waters sustained-petascale computing project supported by the NSF and the state of Illinois.

The authors would also like to acknowledge support from the Regional and Global Climate Modeling Program (RGCM) of the US Department of Energy, Office of Science (BER), Cooperative Agreement DE-FC02-97ER62402 and from NSF’s EaSM program.

Author information



Corresponding author

Correspondence to Julio T. Bacmeister.

Additional information

This article is part of a Special Issue on “Benefits of Reduced Anthropogenic Climate ChangE (BRACE)” edited by Brian O’Neill and Andrew Gettelman.

Electronic supplementary material


(PDF 7455 kb)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Bacmeister, J.T., Reed, K.A., Hannay, C. et al. Projected changes in tropical cyclone activity under future warming scenarios using a high-resolution climate model. Climatic Change 146, 547–560 (2018).

Download citation


  • Tropical cyclones
  • Climate change
  • High-resolution