Decreasing global tropical cyclone frequency in CMIP6 historical simulations

Abstract

The impact of anthropogenic global warming on tropical cyclone (TC) frequency remains a challenging issue, partly due to a relatively short period of reliable observational TC records and inconsistencies in climate model simulations. Using TC detection from 20 CMIP6 historical simulations, we show that the majority (75%) of these models show a decrease in global-scale TC frequency from 1850 to 2014. We demonstrated that this result is largely explained by weakened mid-tropospheric upward motion in CMIP6 models over the Pacific and Atlantic main development regions. The reduced upward motion is due to a zonal circulation adjustment and shifts in Intertropical Convergence Zone in response to global warming. In the South Indian Ocean, reduced TC frequency is mainly due to the decreased survival rate of TC seeds because of an increased saturation deficit in a warming climate. Our analysis highlights global warming’s potential impact on the historical decrease in global TC frequency.

Most CMIP6 simulations show a decline in global-scale historical tropical cyclone frequency.

INTRODUCTION

Tropical cyclones (TCs) are one of the most destructive extreme weather events, causing substantial societal impacts. To date, human activities are estimated to have caused ~1.0°C of global warming above preindustrial levels (1). Therefore, understanding how TC activity responds to anthropogenic warming is an important topic from a climate change impact assessment perspective. Detection and attribution of anthropogenic influences on TC activity remain challenging due to a limited period of reliable TC observations, particularly in the presence of multidecadal variability (2–4). Several recent studies have indicated, with relatively high confidence, that global warming will increase average TC intensity, the proportion of TCs reaching category 4 to 5 intensity (e.g., 1-min maximum sustained winds ≥ 113 kt) and average TC rain rates (3, 5–9). However, there remains less certainty regarding the impact of global warming on TC frequency (TCF) (10–12). On the basis of several decades of reliable best-track data, there is no clear evidence of an observed trend in global TCF that can be attributed to anthropogenic warming (13).

Previous studies have also suggested that anthropogenic aerosol forcing could affect regional TCF (14–17). The increased TCF over the North Atlantic (NA) since the 1980s has been partly attributed to the reduction in anthropogenic aerosol emissions over Europe and the United States. Some studies have also revealed a hemispherically asymmetric response of TCF to anthropogenic aerosol forcing (18, 19). Anthropogenic aerosol forcing could partly cancel the impact of greenhouse gas forcing on TCF changes (20), making it difficult to detect the long-term trend of TCF under global warming. To improve our understanding of the connection between global-scale TCF and climate change, as well as the related physical mechanisms and drivers, we perform exploratory analyses of global-scale TCF changes under global warming.

The ability of climate models to simulate TCs has been greatly improved in the past few decades, thus becoming one of the main tools used to explore the effect of anthropogenic warming on TCF (9–11, 21). A large majority of general circulation model simulations project a decrease in global TCF under future warming climate conditions (3, 5, 9). The most recent Intergovernmental Panel on Climate Change Report, the Sixth Assessment Report (IPCC AR6), states that global TCF is likely to decrease or remain unchanged (22). However, some models project an increase in global TCF (23, 24), diminishing our confidence in TCF changes in a warming climate (11). Using a statistical-dynamical TC downscaling framework that assumes a consistent number of TC seeds with warming (23, 25, 26), several results have shown an increase in global TCF with global warming, with other results showing a decrease. The differences in these results depend on the performance of various TC genesis indices (26). By directly detecting TCs in Coupled Model Intercomparison Project – Phase 5 (CMIP5) models under a warmer climate, there have been varying results obtained on the statistical characteristics of TCs. Some models have indicated an increase in global TCF, other models have indicated little change in global TCF, while yet other models have indicated a decrease in global TCF (27). The relatively low horizontal resolution of the CMIP5 models was regarded to be one of the main causes of the uncertainty in the assessment of TC activity in a warming climate (28). When excluding models with an extremely poor TC climatology in the present climate, most models projected a relatively robust reduction of global TCF with warming.

In recent years, high-resolution simulations of TC activity in climate models have been able to capture some characteristics of TCs more realistically (29–33), particularly improving the quality of TC intensity simulations. Most high-resolution global simulations indicate a substantial reduction of TCF in a future warming climate, with the decrease mainly occurring for relatively weak TCs (weaker than or equal to category 1 on the Saffir-Simpson Hurricane Wind Scale; 1-min maximum sustained winds ≤ 82 kt) (12, 29, 30, 32–34). However, a few studies based on high-resolution simulations have reported an increase in global TCF in a warmer climate (24, 35).

An alternative way to identify TCs is the Okubo-Weiss-Zeta parameter (OWZP) TC detector, which was developed to identify regions of enhanced vorticity with weak deformation. TCs were then diagnosed from relatively large-scale dynamic and thermodynamic conditions (36–39), which means that this scheme can be used to directly detect TCs in climate models that have a lower resolution. Using the OWZP TC detection method, some studies have shown a robust reduction in future global TCF based on CMIP5 models (38).

One of the fundamental challenges remains uncovering the underlying physical mechanisms behind the reduced global TCF observed in models (3, 5, 11). Upward mass flux reductions and saturation deficit increases have been proposed to explain TCF changes in response to climate change (5, 40). Several prior studies have suggested that the reduction of global TCF is closely related to reductions in upward mass flux (41–43), for instance, via decreased precipitation in 2× CO₂-only [fixed sea surface temperature (SST)] experiments and increased atmospheric stability in SST-only (2K uniform SST increase with fixed CO₂) experiments. Other studies have argued that increased saturation deficit appears to cause the reduction of global TCF in a warming climate (5, 11, 26, 44, 45), suggesting a greater importance for saturation deficit than relative humidity in TC formation. Increased saturation deficit prolongs the timescale for the free troposphere in a TC to reach saturation through surface evaporation, thereby inhibiting genesis (46, 47). When using saturation deficit as the moisture variable in empirical TC genesis indices, there is a decrease in global TCF with climate warming (44). Statistical-dynamical downscaled TCF changes in a warming climate were found to depend on the moisture variable used in the TC genesis indices, with an increasing trend using column relative humidity and a decreasing trend using saturation deficit (26). Under the presence of vertical wind shear, increased saturation deficit allows increased ventilation, entraining drier air into the upward mass flux of convection which results in less net heating, suppressing TC development (48–50).

Recent studies have increasingly emphasized the role of TC seeds as regulators of TCF (11, 12, 35, 51). The expected number of TCs has been considered as the product of the number of seeds and the survival rate for these seeds (11, 35). Some studies have suggested that the seeds simulated in high-resolution models could influence projected future TCF changes (12, 35, 51). An increase in TC seeds that exceeds the less efficient development of these seeds into TCs are thought to be responsible for increased TCF under a warmer climate (35). In contrast, some high-resolution TC simulations project a decrease in TC seeds, consistent with the decrease in TCF (12, 51). The TC seeds are sensitive to the simultaneous occurrence of large-scale ascent and background vorticity, while TC genesis probability depends on the ventilation effect (52). However, some numerical experiments have indicated that the TC climatology is insensitive to the characteristics of dynamically independent synoptic-scale disturbances (53, 54). The climatological TCF can be maintained in favorable large-scale environments, even when the external perturbations, such as typical TC precursors like African easterly waves in the NA, have been blocked (53). A recent study argued that the TC climatology is mostly controlled by the large-scale environment (55).

In summary, most studies have focused on future projections of TCF in a warming climate. Whether we can already detect the impact of anthropogenic global warming on TCF in the present climate remains unknown. As new generations of climate models become available, such as those from CMIP6, it is imperative to assess TCs in those models to have a better insight of their performance and potentially suggest improvements. In this work, we detected TCs in historical simulations using 20 CMIP6 models and found that most simulations showed a decrease in global TCF in the present climatology compared to the preindustrial climatology. We find that global warming results in anomalous SST gradients and Intertropical Convergence Zone (ITCZ; https://www.bbc.co.uk/bitesize/guides/z9yssbk/revision/1) shifts, weakening ascending motion in the main development region (MDR) for TCs in the Pacific and Atlantic basins, thus decreasing TCF. Changes in large-scale conditions cannot explain the decrease in South Indian Ocean (SIO) TCF. The decreased survival rate of TC seeds appears responsible for the reduced SIO TCF.

RESULTS

Declining global TCF in response to a warming climate

Historical simulations from 20 CMIP6 models (table S1) were used to investigate the impact of climate change on global TCF. To compare with observations, we examined the simulated mean annual TC number in the current climate for each model with a focus on the period characterized by reliable observed TC records (1985–2014) (table S2). The simulated climatological TC spatial distribution (fig. S1) and seasonal cycle (fig. S2) in CMIP6 models closely resemble observations. However, there are some exceptions, such as an overestimation of TCF over the CP, an underestimation of TCF over the NA, a 1-month lag in the season cycle of Eastern North Pacific (ENP) TCs, and an inconsistent seasonal cycle over the North Indian Ocean (NIO) as compared with observations. The models overestimated TCs during the peak monsoon season and could not capture the bimodal distribution over the NIO, which has also been reported in a previous study based on CMIP5 simulations (56). The bias in the seasonal cycle over the NIO between CMIP6 models and observations is likely due to the tracking algorithm incorrectly identifying monsoon depressions as TCs, especially in low-resolution models (57). In this study, NIO TCs have been excluded, as this basin only accounts for a small percentage of the global TC climatology (~5%).

Over the NA, there is a large dispersion in the simulation of TCs, with some models capturing TCF frequency in the basin and others underestimating TCF. Some models cannot simulate TC-like vortices, likely due to their relatively coarse spatial resolution, dynamic cores, or parameterization schemes that cannot resolve all TC-related processes (29, 32). Although there are fewer TCs simulated in CMIP6 models than in observations, the climatological features of NA TCs are similar to observations, thus allowing us to investigate the response of TCs to global warming over the NA. Also note that there is a 1-month lag in the season cycle of ENP TCs that appears in all models. This 1-month lag in the ENP warrants further investigation. However, the CMIP6 models generally can reproduce the observed spatial distributions and peak season for ENP TCs. Over the CP, there are more TCs identified in models than observations. Considering that this region is not a MDR for observed TCs, the changes in this region will be discussed separately.

The multimodel ensemble mean (MME) of the TCF anomaly shows a significant declining trend in TCF at the global scale and hemisphere scale (Fig. 1, A to C) during the peak TC seasons of July to October in the Northern Hemisphere (NH) and January to April in the Southern Hemisphere (SH). All trends are tested using the Mann-Kendall test and Sen’s slope analysis. To further explore the impact of global warming for each model, we compared the annual mean TCF between the period of 1850–1899 representing the preindustrial climatology and the period of 1965–2014 representing the recent historical period which includes global warming (Fig. 1D). Most models (15 of 20) show a reduction in both global and hemispheric TCF in the present period when compared to the preindustrial era, with a significant decrease in the MME global TCF.

Fig. 1. Annual global and hemispheric annual TCF and changes in the mean annual TCF between the present and preindustrial periods.

Time series of the MME of the (A) global, (B) NH, and (C) SH TCF anomaly during 1850–2014 (blue), along with a 5-year running mean (red) and linear trend (purple). The shading indicates the 95% confidence interval, and the asterisk (*) indicates that the trend is significant at the 95% confidence level. (D) Difference in annual TCF between the present period (1965–2014) and the preindustrial period (1850–1899) for each model and the MME. Shading indicates the 95% confidence interval.

On a regional scale, we observe an overall decrease in TC formation over the MDRs (Fig. 2A), with an increase over the Central Pacific (CP) and the subtropical NA. Note that the CMIP6 models used in this study tend to overestimate TCF over the CP region and underestimate TCF over the subtropical NA basin compared with observations (fig. S1), consistent with earlier studies using CMIP5 models (27, 38, 58). The regional increase in TCF largely offsets the MDR decrease, due to a model bias in simulating TCs over the CP and subtropical NA, thus resulting in no significant changes in basin TCF. Given the regional differences in TCF changes, key areas in each basin are selected to better reflect global TCF changes, as shown in Fig. 2. These key areas compare well with the observed MDRs (fig. S1). All MDRs show a decreasing trend in TCF with an increasing TCF over the CP region (Fig. 3). This CP region is termed a secondary MDR for simulations, as it has fewer observed TCs climatologically than the other MDRs. Using high-resolution model simulations, some studies have suggested that there may be an increase in TC activity in the CP under anthropogenic climate change (59, 60). All changes in TCF in the MME for all MDRs are statistically significant at the 95% confidence level. More than 70% of models also show a decrease in TCF in the MDRs and an increase in the CP in the current period compared with the preindustrial period (fig. S3).

Fig. 2. Changes in TCF, DGPI, and ENGPI between the present and preindustrial periods based on CMIP6 model simulations.

Difference in the spatial distribution of multimodel mean (A) TCF, (B) DGPI, and (C) ENGPI between 1965–2014 and 1850–1899. The NH season is defined to extend from July to October, and the SH is defined to extend from January to April. Gray dots denote that the differences are significant at the 95% confidence level. The boxes indicate the MDR for TC formation in each basin, defined as: the western North Pacific (WNP), 10° to 30°N, 120° to 150°E; the CP, 10° to 25°N, 150° to 200°E; the ENP, 5° to 20°N, 240° to 270°E; the NA, 5° to 20°N, 310° to 350°E; the SIO, 20° to 5°S, 60° to 110°E; and the SP, 25° to 10°S, 150° to 190°E.

Fig. 3. Annual TCF in the individual TC MDRs.

Time series of MME MDR TCF anomaly for the (A) WNP, (B) CP, (C) ENP, (D) NA, (E) SIO, and (F) SP during 1850–2014 (blue), along with a 5-year running mean (red) and linear trend (purple). The shading indicates the 95% confidence interval, and the asterisk (*) indicates that the trend is significant at the 95% confidence level.

Because of the limited temporal record, decadal or multidecadal internal variability (e.g., the Pacific decadal oscillation or Atlantic multidecadal oscillation) might have substantially obscured the impact of global warming (22). Although we used the MME of 20 CMIP6 models to filter out the influence of natural internal variability, it is still uncertain whether the slight decreasing trend in global TCF (Figs. 1 and 3) over the past century is a result of external forcing or internal variability. We have applied singular value decomposition (SVD) analysis to the MME SST and TCF from 20 CMIP6 models to extract the response mode of global TCF to the SST pattern associated with global warming (4). The first SVD mode explains ~78% of the total covariance between global SST and TCF. The first mode for the SST spatial pattern represents a global warming mode with the corresponding expansion coefficient (EC1_SST) consistent with the global mean SST (fig. S4). This mode represents a nonlinear long-term warming trend, with more rapid warming occurring after the 1960s. The spatial pattern of the first SVD TCF mode matches relatively well with TCF changes between the preindustrial and the present period. The expansion coefficient of the first SVD TCF mode (EC1_TCF) exhibits a nonlinear trend consistent with that of the global warming mode. The EC1_TCF is highly correlated with EC1_SST (r = 0.83, P = 0.01). A notable change in global TCF occurs after the 1960s, consistent with the global warming trend. The results of the SVD analysis support the impact of global warming on the reduced global TCF as revealed by the MME.

Primary factors driving the decreased global TCF

Two TC genesis potential indices, the Emanuel-Nolan genesis potential index (ENGPI) and the dynamical genesis potential index (DGPI), are used here to measure TC background conditions (44, 61). These indices have been widely used to explain the projected changes in tropical cyclogenesis frequency. Changes in GPIs were determined by the MME difference between the present period (1965–2014) and the preindustrial period (1850–1899) to show the impact of global warming on the TC environment. As shown in Fig. 2 (B and C), changes in the DGPI spatial distribution are similar to that of the TCF changes, with some differences observed in the SIO. By contrast, the ENGPI increases for all MDRs except the NA (Fig. 2, B and C). These results are substantially different from the changes in TCF from the MME, indicating that the DGPI better captures changes in the long-term trend of the TCF than the ENGPI (61).

This difference is due to the ENGPI emphasizing the influence of thermodynamic factors, including the maximum potential intensity (MPI) and 600-hPa relative humidity. The MPI is defined as the maximum intensity that a TC can attain given favorable oceanic and atmospheric thermodynamic conditions. We found that MPI remained constant or weakly increased in most regions during the historical period except for a noticeable decrease over the NA. The 600-hPa relative humidity significantly increased over the SIO and CP, contributing to the observed increase in ENGPI (fig. S5). These factors are generally increasing in a warming climate, primarily due to increased MPI associated with increased SST (44). Although the increase in TC-favorable thermodynamic conditions is not evident except for the SIO and CP in our results, these two thermodynamic factors cannot explain the decrease of global TCs observed in CMIP6 models. Previous studies have also examined other empirical genesis indices (62) using different combinations of thermodynamic variables (44) and have emphasized that the saturation deficit is an important predictor for capturing a global reduction in TCF in warmer climates (26, 44). These studies highlight the impact of thermodynamic parameters on the reduction of TCF in a warming climate. In this study, we also examine the changes of saturation deficit. Saturation deficit shows a global increase during the historical period and contributes to the global reduction in TCF. However, the GPIs used in this study highlight the role of dynamic factors. Here, we have mainly analyzed the contribution of dynamic factors through the DGPI, with the influence of saturation deficit discussed in the following section.

Overall, the DGPI shows a more consistent spatial distribution with the TCF changes in a warming climate, highlighting the importance of dynamic factors in driving the decline in global TCF. Further analysis shows that ω500 has a relatively consistent increase for each basin, except in the SIO and CP (fig. S5), suggesting that vertical motion is a key factor in global TCF change. The importance of upward motion has been highlighted in prior studies (5, 40), with reduced global TCF under global warming projected by climate models mainly due to decreased upward mass flux (as measured by the magnitude of the upward p-velocity at 500 hPa). However, neither the DGPI nor the ENGPI can explain the reduction of SIO TCF in the present period. There is also considerable model disagreement among the 20 CMIP6 models regarding the impact of environmental factors on SIO TCF changes.

Given the inconsistencies in the spatial distribution of changes in large-scale environmental fields, each factor’s contribution in each basin’s MDR is now analyzed using a DGPI budget analysis. The relative contribution of an individual factor is determined by the DGPI difference caused by that specific factor between the present period and the preindustrial period while keeping all other factors at their preindustrial values (18, 61). As shown in Fig. 4, the MDR DGPI changes significantly in all basins except for the western North Pacific (WNP). By examining the relative contribution of each factor, we find that vertical motion is the main contributor to the total DGPI change in each basin (significant at a 95% confidence level). Other factors do not have a statistically significant contribution and have considerable model-to-model disagreement. In the WNP, there are relatively large model biases in vertical wind shear and 850-hPa relative vorticity as well as in 500-hPa meridional shear vorticity. These biases may be associated with uncertainty in the simulated monsoon trough (63, 64).

Fig. 4. Contributions of individual factors to the dynamical genesis index (DGPI) in the individual TC MDRs.

Changes of average DGPI and the four individual factors’ contribution in the MDR between 1965–2014 and 1850–1899, for the (A) WNP, (B) CP, (C) ENP, (D) NA, (E) SIO, and (F) SP. SUM means the sum of the four individual factors’ contribution, with the four factors being vertical wind shear (VWS), 850-hPa absolute vorticity (VOR), 500-hPa meridional shear vorticity (MZW), and 500-hPa vertical pressure velocity (Omega). The dotted line indicates the range of values of the 20 models. The top and bottom bounds of the box indicate the 95% confidence interval, and the middle of the box indicates the multimodel mean.

Changes in the Walker Circulation associated with the global warming pattern

As mentioned above, weakened upward motion is the primary driver of the reduction of MDR TCF under global warming. The vertical motion in the MDR is largely modulated by the atmospheric circulation. Previous studies have related the weakened upward motion to the adjustment of the atmospheric circulation under global warming, especially the Walker and Hadley circulations (34, 65). Because of the differential warming rates in response to global warming across different regions, the resulting anomalous SST gradients could further influence changes in the atmospheric circulation. We examine differences in SST between the present period and the preindustrial period. During the NH summer [July–October (JASO)], the tropical CP and western NIO experience stronger warming compared to other basins. The positive SST gradient anomaly from the WNP to the CP results in anomalous low-level westerly winds over the WNP, while the negative SST gradient from the CP to the ENP leads to anomalous low-level easterly winds over the ENP (Fig. 5). The stronger warming and the convergence of anomalous zonal winds strengthen upward motion over the CP, contributing to an increase in TCF in this region. Meanwhile, the anomalous low-level zonal winds and enhanced local upward motion over the CP further induce an anomalous counterclockwise zonal circulation over the WNP and an anomalous clockwise zonal circulation over the ENP (fig. S6), thereby weakening the upward motion over these two basins. Similarly, the anomalous zonal SST gradient in the Indo-Pacific region also induces an anomalous clockwise vertical circulation, further weakening upward motion over the WNP. Corresponding to this CP-like warming pattern, the SLP over the Indo-Pacific region increases more than over the ENP, thus contributing to these anomalous zonal circulations over the Pacific.

Fig. 5. Changes in SST, SLP, and wind between the present and preindustrial periods.

Changes in MME (A) SST and 850-hPa wind, (B) SLP and 200-hPa wind during JASO, between 1965–2014 and 1850–1899. (C and D) As in (A) and (B) but for JFMA. All differences at every point displayed are significant at the 95% confidence level.

During the SH summer [January–April (JFMA)], the eastern South Pacific (SP) experiences stronger warming. This positive SST gradient anomaly induces anomalous westerlies over the SP (Fig. 5), thus resulting in an anomalous counterclockwise zonal circulation and weakening of upward motion over the western SP (fig. S6). The decreased Indo-Pacific SLP gradient also supports the weakening of the Walker Circulation (Fig. 5, B and D). We use an index of the large-scale tropical Indo-Pacific SLP gradient (ΔSLP) as a proxy for the Walker Circulation. We calculate ΔSLP as the difference in SLP averaged over the central/eastern Pacific (5°S to 5°N, 160°W to 80°W) and over the Indian Ocean/western Pacific (5°S to 5°N, 80°E to 160°E) (66). We find that changes in the MME ΔSLP exhibit a significant declining trend (P = 0.01) during 1850–2014 (fig. S7), indicating a weakening of the Walker Circulation due to global warming. Overall, the anomalous SST gradient corresponds to a CP-like warming pattern modulating changes in the zonal circulation, leading to a weakening of upward motion over the western WNP, eastern ENP, and western SP. This weakened upward motion inhibits TC formation in these regions, while the stronger warming over the CP enhances upward motion, thus favoring TC formation in this region.

ITCZ shift in response to the warming pattern

The adjustment of the zonal circulation associated with global warming cannot explain the reduction of SIO and NA TCF. The impact of the zonal SST gradient associated with the CP-like warming pattern is mainly confined to the Pacific. During the NH summer, there is a significant decrease in precipitation in the ENP and NA MDR, suggesting a weakening of deep convective activity. Precipitation increases over the equatorial ENP-NA, implying a southward shift of the ITCZ over the ENP-NA (Fig. 6). We note that there is a stronger increase in SST occurring in the tropical SP in the CMIP6 historical simulations, decreasing the cross-equatorial meridional SST gradient and weakening the cross-equatorial flow into the ENP (Fig. 5). The weakened cross-equatorial flow in response to global warming drives a southward shift of the ITCZ (67, 68). In the NA, the stronger warming near the equator enhances convergence and hinders northward transport of energy across the equator, leading to a displacement of the ITCZ toward the equator (69). This southward shift of the ITCZ causes the ENP-NA MDRs to become less favorable for TC formation due to less deep convective activity. In the SH, there is a significant increase in precipitation in the SIO, in contrast to the decrease in TCF. There is no apparent ITCZ shift across the entire tropics due to warming. Our analyses suggest that a weakening of upward mass flux is critical for TCF reduction.

Fig. 6. Changes in mid-tropospheric saturation deficit and precipitation between the present and preindustrial periods.

Changes in the MME (A) mid-tropospheric saturation deficit, with the NH displayed during JASO and the SH during JFMA, (B) NH precipitation during JASO, and (C) SH precipitation during JFMA, between 1965–2014 and 1850–1899. White dots denote differences that are significant at the 95% confidence level. The boxes indicate the MDRs for TCF, as defined in Fig. 2.

Changes in TC seed frequency and survival rate

Changes in mid-tropospheric saturation deficit mostly explain the decreased TCF, especially over the SIO. With global warming, the saturation deficit shows an overall increase (Fig. 6). This increase enhances the ventilation effect and increases the potential for dry air entrainment into convective clouds, thus creating a more hostile environment for TC formation (11, 65). Given the increase in convective activity associated with the increase in precipitation in the SIO, we speculate that the decrease in TCF may be related to a reduced TC seed survival rate (i.e., the efficiency of seeds developing into TCs) (12, 35, 51). In this study, TC seeds, namely, TC precursor disturbances or tropical depressions, are defined as TC-like vortices that persists for a duration of 24 hours rather than the 48-hour threshold used for developed TCs (70).

In the NH, there is a robust decrease in TC seed frequency in the WNP, ENP, and NA and an increase in the CP, consistent with corresponding changes in TCF. However, there is no robust change in TC seed frequency in the SH, with substantial disagreement among the CMIP6 models (Fig. 7). Although changes in TC seed frequency in response to global warming in the SH are uncertain, there is a robust decrease in TC seed survival rate over the SIO across models (18 of 20 models) (Fig. 7). Therefore, although there is no consistent response of changes in environmental factors and TC seed frequency in the SIO, the robust reduction of the survival rate associated with the increased saturation deficit could lead to a reduction of TCF in the SIO. Given these results, the reduction of SIO TCF is likely due to an increase in saturation deficit.

Fig. 7. Changes in the mean TC seed frequency and survival rate between the present and preindustrial periods for the individual MDRs.

(A) Difference in annual TC seed frequency between 1965–2014 and 1850–1899 for each model and the MME in the MDR, including the WNP, CP, ENP, NA, SIO, and SP. Shading indicates the 95% confidence interval. (B) As in (A) but for the survival rate (the ratio of TCF to TC seed frequency).

DISCUSSION

Our aim is to determine whether there has already been an impact of global warming on global TCF during the historical period (1850–2014). We attempt to answer this question via detection of TCs in the latest generation of CMIP6 multimodel historical climate simulations. The simulated TCs have a similar spatial distribution and seasonal cycle compared with observations. There are relatively small but significant declining trends in global MME TCF, as well as in both hemispheres separately. There is also a significant decrease in TCF in the MDR of various basins (including the western WNP, the eastern ENP, the southern NA, SIO, and western SP), with an increase over the CP (including the southeastern WNP). The first SVD mode of MME SST and the global TCF shows a distinct global warming SST pattern, with TCF modes that are highly similar to the difference in TCF between the present period (1965–2014) and the preindustrial period (1850–1899). The time series of expansion coefficients for the first SVD SST and TCF fields are highly correlated (r = 0.83), with a nonlinear upward trend that is consistent with global mean SST. These results highlight the potential impact of global warming on the reduction of TCF during the historical period.

Changes in upward motion appear to be the dominant driver of TCF, with a weakening of upward motion consistent with regional-scale reductions in TCF in most regions. There is an increase in CP TCF associated with enhanced upward motion. These changes in vertical motion are found to be closely associated with the global warming SST pattern. During the NH summer, stronger warming over the CP and NIO causes an anomalous zonal SST gradient. The negative SST gradient over the Indo-Pacific and ENP leads to anomalous low-level easterly winds, while the positive gradient over the WNP results in anomalous low-level westerly winds. These zonal wind anomalies enhance low-level convergence over the CP. The resulting anomalous circulation weakens upward motion over the western WNP and eastern ENP, thus inhibiting TCs. Meanwhile the enhanced zonal divergence over the Indo-Pacific further weakens upward motion over the western WNP. A similar process is found for the SH summer. The positive zonal SST gradient weakens the Walker Circulation, thereby weakening upward motion over the western SP.

In summary, the adjustment of the zonal circulation associated with the global SST warming pattern weakens upward motion over the MDR in various basins, thereby reducing the number of TCs. In the NA, stronger warming near the equator increases equatorial precipitation, resulting in an equatorward shift of the ITCZ. The southward shift of the ITCZ associated with global warming weakens upward motion and deep convective activity in the NA, creating an unfavorable environment for TC formation. Meanwhile, the meridional SST gradient associated with global warming weakens the cross-equatorial flow, also weakening upward motion and deep convection over the ENP. The reduction of TCF over the SIO is mainly due to the robust reduction in the survival rate of TC seeds associated with an increase in saturation deficit. Whether the TC seeds can independently control TCF changes is still a controversial topic; therefore, this is only a hypothesis on how the saturation deficit contributes to the reduction of TCF over the SIO.

Our work highlights the TC activity response to global warming characterized by the reduction of global TCF in the MME of 20 CMIP6 historical simulations, as well as providing a plausible physical mechanism for this reduction. We find a significant decrease in global TCF in a warming environment. Our study therefore increases the confidence in future projections of TC activity. Although we only analyzed TCF changes during the historical period (1850–2014), the stronger trend emerging after the 1960s agrees well with the SST warming trend.

Note that this study emphasizes that the CP–El Niño–like trend of SST in response to global warming alters the zonal SST gradient and the Walker Circulation, which weakens upward motion and reduces TCs during historical period. However, observations show a more La Niña–like trend in SST over the past 50 years, which is opposite to that simulated in CMIP6 models. Previous studies have also reported that most current climate models incorrectly simulate the equatorial Pacific response to the increase in greenhouse gases (71, 72). In a recent study that explored the impact of the tropical Pacific historical SST trends on TCF changes, only models forced with observed SST showed a pattern of increased Atlantic TC activity and decreased Pacific TC activity that was similar to observations. In contrast, coupled models with an El Niño–like SST response showed an increasing TC activity trend over the North Pacific (73). Our results show an increase in TCF over the eastern part of the WNP and CP, which is similar to the impact of El Niño, but a decrease of TCF over western part of the WNP which is consistent with the observational trend (15, 19, 74, 75). The CMIP6 models appear to only incorrectly simulate an increase of TC activity over the CP. As shown in fig. S1, CMIP6 models simulated continuous activity of TCs over the CP with an excessive number of TCs. The warm SST bias over the CP in CMIP6 models may contribute to a more favorable environment for TCF, thus leading to an overestimation of TCF over the CP (73). The SST bias is amplified with global warming in CMIP6 models, thus leading to an increasing TCF trend over the CP. Further analysis is required to ascertain to what extent that bias in SST trends in models could affect the simulated TCF trend, especially at regional scales.

To corroborate our findings, we performed an analysis using highresSST-present runs from high-resolution AMIP experiment from the CMIP6-HighResMIP (i.e., historical forced atmospheric experiments) for 1950–2014. These experiments are forced by daily SST and sea ice datasets from HadISST with the initial state derived from the ECMWF's atmospheric reanalysis of the 20th century (ERA-20C). Anthropogenic aerosol and greenhouse gas (GHG) concentrations forcing and other settings are the same as the historical simulations in CMIP6 models. The AMIP runs from five HighResMIP models including ECMWF, HadGEM, NICAM, EC-Earth, and MRI have been analyzed. Among these five models, the ECMWF contains five ensemble members, while the other four models have a single ensemble member. The MME from these five high-resolution models also shows a significant decrease in global-scale TCF during 1950–2014 (table S4). The trends in regional TCF from the MME of these AMIP runs are not statistically significant, differing from the results using the CMIP6 models. This lack of statistical significance may be due to the relatively small model sample size and the stronger influence of internal variability, given the relatively short duration of the simulations. Given that AMIP runs are forced by observed SST from 1950 to 2014, it makes sense that there is stronger internal variability in the MME of these five AMIP runs than in the MME of 20 low-resolution CMIP6 models. However, these AMIP runs also show a consistent decrease in MDR TCF (fig. S8), similar to the results from the CMIP6 models, with significant trends for TCF in most MDRs. In general, high-resolution AMIP experiments show declining TCF trends, similar to the results from the low-resolution CMIP6 models.

We only considered the impact of global increases in SST in the historical period, which includes the response to both greenhouse gases and aerosol forcing. Previous studies have suggested that, apart from the influence of multidecadal variability (e.g., the Atlantic Multidecadal Oscillation and the Pacific Decadal Oscillation), decreased anthropogenic aerosol emissions likely also contributed to increased TCF over the NA since the 1980s (14, 15). We also found an increase in TCF over the NA, with a significant increasing trend of 0.03/year since the 1980s. To determine whether the anthropogenic aerosol forcing affects the trend of NA TCF in our results, we would need to consider additional experiments which are forced with a single external forcing to assess the roles of anthropogenic greenhouse gases and anthropogenic aerosols independently.

There remains a fundamental question. Will changes in TC activity in the future, assuming global warming continues unabated, be similar to the changes that have occurred in the past? Will these changes be related to the same physical mechanisms? We intend to analyze simulations of TC activity in future scenarios in follow-up studies. We also note that the horizontal resolution of these CMIP6 models is relatively low, and high-resolution models will likely improve the representation of the statistical features of TCs (32).

MATERIALS AND METHODS

CMIP6 models

Historical experiments from 20 CMIP6 models (obtained from https://esgf-node.llnl.gov/search/cmip6/) are used to explore the response of global TCF to global warming. Only one member (“r1i1p1fi”) from each model has been used. To retain as many samples as possible, several versions of the same model are also selected, although these models from the same family typically have similar biases. All CMIP6 models that included the necessary variables at daily frequency required for TC detection have been evaluated. To investigate the possible impact of global warming on the long-term trend of TCF, we excluded the models with historical run simulations with the required output only available from 1950 to 2014 and retained the remaining models that had historical run simulations at high-frequency output from 1850 to 2014. In addition, several models that failed to capture global-scale and basin-scale TCs and even cyclonic disturbance have also been excluded (e.g., AWI-ESM-1-1-LR, IITM-ESM, MPI-ESM-1-2-HAM, MPI-ESM1-2-HR, and MPI-ESM1–2-LR). Model resolutions range from 100 to 200 km. More detailed information regarding model resolutions can be found in table S1. In this work, the period from 1850 to 1899 is used as the preindustrial climatology, in agreement with the IPCC (2018) definition, with the period from 1965 to 2014 representing the present climatology. The use of a 50-year climatology reduces uncertainties arising from internal variability. The impact of global warming on global TCF is assessed by comparing the two periods. The daily outputs of these models are used to detect TCs in CMIP6 simulations, and monthly outputs are used to analyze the large-scale environment.

TC and seed detection and tracking in models

The OWZP TC detector (36–39) has been used in this study to detect TCs in CMIP6 historical simulations. The OWZ variable is used to identify regions of near solid-body rotation in the lower to middle troposphere, which has been argued to be necessary for TC formation (39). This detection technique is more suitable for TC formation diagnostics and detection in coarse-resolution models since it only examines large-scale variables, enabling use of a uniform threshold. The OWZ detection approach takes the place of relative vorticity used in other detection schemes. The low-deformation vorticity is determined from a modified OWZ equation, defined as

$$\text{OWZ}=\text{max}({\text{OW}}_{\text{norm}},0)\times (\mathrm{\zeta }+f)\times \text{sign}(f)$$

where OW_norm is a normalized Okubo-Weiss parameter, ζ is the vertical component of relative vorticity, f is the Coriolis parameter, and ζ + f is the absolute vorticity. The normalized Okubo-Weiss parameter is defined as

$${\text{OW}}_{\text{norm}}=\frac{{\mathrm{\zeta }}^2-(E^2+F^2)}{{\mathrm{\zeta }}^2}$$

where the numerator is the Okubo-Weiss parameter, with E and F representing the stretching and shearing deformation, respectively

$$E=[\frac{\partial u}{\partial x}-\frac{\partial v}{\partial y}]$$

$$F=[\frac{\partial v}{\partial x}+\frac{\partial u}{\partial y}]$$

$$\mathrm{\zeta }=[\frac{\partial v}{\partial x}-\frac{\partial u}{\partial y}]$$

where u and v are the zonal and meridional horizontal wind components, respectively.

The OWZP TC detection method identifies and tracks circulations with the dynamic potential to support TC formation in climate model data. The identification is based on threshold values of OWZ on the 850- and 500-hPa pressure levels and other relatively large-scale dynamic and thermodynamic conditions (table S3). The detection procedure is summarized here in the following steps:

1) All model data used are interpolated to a 1° × 1° grid for consistency and to minimize grid-dependent problems.

2) Each grid point is assessed to see whether it is likely to be part of a circulation that would support TC formation, using the thresholds listed in table S3. The “likely” grid points are grouped into individual storms at a particular time.

3) The instantaneous storms from step 2 are linked progressively in time to produce a set of storm tracks.

4) Each storm track is assessed to see whether it satisfies a set of conditions for a minimum consecutive time period of 48 hours. In addition, storms lasting at least 24 hours are identified as TC seeds. Storms forming near or poleward of the subtropical jet (defined as the 200-hPa wind speed > 25 ms⁻¹ and the zonal wind component > 15 ms⁻¹) are deemed to be subtropical and are therefore removed from this analysis.

The spatial distribution of TCF from the MME is similar to observations (fig. S1). The NHC and JTWC best-track data are taken from the International Best Track Archive for Climate Stewardship version 4 (76, 77). The main areas where TCs form in each individual basin in the CMIP6 simulations are also consistent with observations. We also examine the seasonal cycle of TCs simulated by CMIP6 historical experiments during the period with reliable observed TC records (1985–2014) in each basin. In this study, we mainly consider TCs during the peak season, defined as July to October in the NH and January to April in the SH. As shown in fig. S2, except for the NIO, the CMIP6 historical experiments can reproduce the historical TC climatology. Consequently, the NIO is not considered in this study, as it only accounts for a small percentage of the global TC climatology (~5%). We do note that there are fewer NA TCs in CMIP6 models than in observations. Although there are fewer TCs simulated in CMIP6 models, the climatological features of NA TCs are similar to observations, thus allowing for the investigation of the response of NA TCs to global warming.

TC genesis potential indices

In this work, two genesis potential indices are used to measure the contribution of environmental factors to tropical cyclogenesis, including the Emanuel-Nolan GPI (ENGPI) (78) and the dynamic GPI (DGPI) (79). The ENGPI is defined as

$$\text{GPI}={\mid {10}^5\mathrm{\eta }\mid }^{\frac{3}{2}}{(1+0.1V_{\text{shear}})}^{-2}{\left(\frac{V_{\text{pot}}}{70}\right)}^3{\left(\frac{H}{50}\right)}^3$$

where η is the absolute vorticity at 850 hPa (s⁻¹), V_shear is the magnitude of the vertical wind shear between 850 and 200 hPa (m s⁻¹), V_pot is the potential intensity (PI, m s⁻¹), and H is the relative humidity at 600 hPa (%). The PI is a theoretical thermodynamic limit for the maximum intensity of a TC and is used to measure the favorability of ocean-atmosphere thermodynamic conditions for TC growth (80), defined as

$$V_{\text{pot}}^2=\frac{T_{\mathrm{s}}}{T_{\mathrm{o}}}\frac{C_{\mathrm{k}}}{C_{\mathrm{D}}}({\text{CAPE}}^{*}-{\text{CAPE}}^{\mathrm{b}})$$ (4)

where T_s is SST, T_o is the mean outflow temperature at the level of neutral buoyancy, C_k is the exchange coefficient for enthalpy, C_D is the drag coefficient, CAPE^* is the convective available potential energy (CAPE) for an air parcel at the radius of maximum winds, and CAPE^b is the CAPE of the boundary layer air (80). The code for the calculation of PI can be downloaded from ftp://texmex.mit.edu/pub/emanuel/TCMAX/.

The DGPI consists of dynamic factors only (79), defined as

$$\begin{array}{c}\text{DGPI}={(2.0+0.1\times V_{\mathrm{s}})}^{-1.7}{(5.5-\frac{\mathrm{d}{u}_{500}}{\text{dy}}\times {10}^5)}^{2.3}\\ {(5.0-20\times {\mathrm{\omega }}_{500})}^{3.4} {(5.5+\mid {\mathrm{\zeta }}_{\mathrm{a}850}\times {10}^5\mid )}^{2.4}{e}^{-11.8}-1.0\end{array}$$

where V_s is the vertical wind shear, u₅₀₀ is the 500 hPa zonal wind, ω₅₀₀ is the 500-hPa vertical pressure velocity, ζ_a850 is the 850-hPa absolute vorticity, and $\frac{\mathrm{d}{u}_{500}}{\text{dy}}$ is the meridional gradient of the zonal wind at 500 hPa, representing meridional shear vorticity.

To assess the relative role of each factor in the GPI, one of the variables in the GPI is taken as the climatological mean during the present-day (1965–2014), while the other three variables are calculated using their average values during the preindustrial period (1850–1899). The difference between the recalculated GPI during the present period and that during the preindustrial period is estimated as the possible impact of the individual factor in response to global warming (18, 61).

Bootstrap test

A bootstrap test was used to examine whether the multimodel mean changes were statistically significant. The values of differences from the 20 CMIP6 models were resampled randomly to form 1000 realizations of the 20-sample sets. The MME differences from the 1000 realizations were used to estimate the 95% confidence interval. The sign of the MME of the difference was considered significant at the 95% confidence level if the confidence interval did not include 0.

Supplementary Materials

This PDF file includes:

Tables S1 to S4

Figs. S1 to S8