We use a high-resolution regional climate model to investigate the changes in Atlantic tropical cyclone (TC) activity during the period of the mid-Holocene (MH: 6000 years BP) with a larger amplitude of the seasonal cycle relative to today. This period was characterized by increased boreal summer insolation over the Northern Hemisphere, a vegetated Sahara and reduced airborne dust concentrations. A set of sensitivity experiments was conducted in which solar insolation, vegetation and dust concentrations were changed in turn to disentangle their impacts on TC activity in the Atlantic Ocean. Results show that the greening of the Sahara and reduced dust loadings (MH
Tropical cyclones (TCs) are one of the most powerful atmospheric phenomena on Earth. With increasing damages and costs due to natural disasters and recent upswing in Atlantic TCs, it becomes more and more important to understand how TC activity may change in the future. As TC development is strongly influenced by, among others, vertical wind shear, sea surface temperature (SST) and humidity, changes in these environmental parameters due to climate change may result in large variability in TC activity. The ongoing global warming can affect those environmental variables both directly by increasing the SST and indirectly through changes in the atmospheric stability and circulation. A recent study (Evan et al., 2016) has shown that changes in atmospheric circulation at the end of the century could potentially reduce dust loadings over the tropical North Atlantic by around 10 %. Evan et al. (2006) showed that Saharan dust layer is strongly linked to changes in North Atlantic TC activity, acting as an inhibiting factor for TC formation, as also previously suggested by Dunion and Velden (2004). These studies suggest that reducing the Saharan dust layer could lead to an increase in TC genesis occurrence, as well as more intense TCs by changes in the midlevel jet, directly impacting the vertical wind shear, and by increasing incoming solar radiation at the surface, directly warming the ocean surface. Local changes in the energy fluxes could also affect the atmospheric circulation through changes in the position of the Intertropical Convergence Zone (ITCZ) or the West African monsoon (WAM) affecting TC activity (Schneider et al., 2014; Seth et al., 2019). For these reasons, a better understanding of the role of WAM intensity and dust loading in altering hurricane activity is of paramount importance.
Dramatic intensifications of the WAM have occurred in the past (Shanahan et al., 2009), the most recent during the early and middle Holocene (MH, 12 000–5000 years BP), when the WAM was much stronger and extended further inland than today. The northward penetration of the WAM led to an expansion of the northern African lakes and wetlands, as well as to an extension of Sahelian vegetation into areas that are now desert, giving origin to the so-called “green Sahara” (e.g., Holmes, 2008; Kowalski et al., 1989; Rohling et al., 2004). Therefore, the MH climate represents a good test case to investigate the TC response to changes in orbital forcing and also investigate how radiative forcing caused by a greener Northern Hemisphere can impact their genesis.
Paleotempestology records are, however, sparse and most of them only span a few millennia, making it difficult to evaluate TC variability further back than the observational period. Nevertheless, records from the western North Atlantic suggest large variations in the frequency of hurricane landfalls during the late Holocene, together with strong positive anomalies in the WAM (Donnelly and Woodruff, 2007; Greer and Swart, 2006; Liu and Fearn, 2000; Toomey et al., 2013).
Only a handful of modeling studies investigating TC changes during the MH are currently available (Korty et al., 2012; Koh and Brierley, 2015; Pausata et al., 2017). Both Korty et al. (2012) and Koh and Brierley (2015) have focused on simulations of the Paleoclimate Modelling Intercomparison Project (PMIP), which only account for the change in orbital forcing and the greenhouse gas (GHG) concentrations during the MH, assuming pre-industrial vegetation cover and dust concentrations. These studies do not explicitly simulate the changes in TCs but rather investigate how key environmental variables affect TC genesis due to the insolation forcing. Both studies came to similar conclusions that considering changes in the orbital forcing makes the environment less prone to develop TCs in Northern Hemisphere summer and more prone in the Southern Hemisphere summer. More recently, Pausata et al. (2017) used a statistical thermodynamical downscaling approach (Emanuel, 2006; Emanuel et al., 2008) to generate a large number of synthetic TCs and assess their changes during the MH with an enhanced vegetation cover over the Sahara and reduced airborne dust concentrations. Their results suggest a large increase in TC activity worldwide and in particular in the Atlantic Ocean in the MH climate. However, this kind of downscaling approach does not consider how the TC genesis may have been affected by changes in atmospheric dynamics, such as those associated with the African easterly waves (AEWs; Gaetani et al., 2017) that are known to seed TC genesis (Caron and Jones, 2012; Frank and Roundy, 2006; Landsea, 1993; Thorncroft and Hodges, 2001; Patricola et al., 2018). Here, we use the same modeling simulations as in Pausata et al. (2017) to drive a high-resolution regional climate model to investigate the impact of the atmospheric dynamics changes induced by Saharan vegetation and dust reduction on TC activity during the MH compared to the PI climate. This study will compare the dynamical downscaling results to those obtained with the statistical thermodynamical downscaling approach used by Pausata et al. (2017) and how they compare with the findings of Koh and Brierley (2015) and Korty et al. (2012). It will also provide insights into how a potential warmer and greener Northern Hemisphere could alter future Atlantic TC activity.
The paper is structured as follows. The model description, experimental design and the analytical tools used in the study are presented in Sect. 2. Section 3 focuses on (1) the model's response to the changes in climate conditions on TC activity, (2) the seasonal distribution of TCs and (3) their intensity. Discussion and conclusions are presented in Sect. 4.
The simulations carried out by Pausata et al. (2016) and Gaetani et al. (2017) and performed with an Earth system model (EC-Earth version 3.1) at
horizontal resolution of
CRCM6 simulation domain (red box). The black/green shaded box shows the approximate present-day tropical cyclone main development region (MDR). Note that the data are projected over an equidistant cylindrical projection.
We performed three distinct 30-year experiments with CRCM6 (see Table 1). The first experiment, the control or reference case, is a pre-industrial (PI; performed at 0.11 and 0.22
Boundary conditions for each modeling experiment.
In this study, a storm-tracking algorithm was developed using a three-step procedure (storm identification, storm tracking and storm lifetime) to detect tropical cyclones, following previous studies (Gualdi et al., 2008; Scoccimarro et al., 2011; Walsh et al., 2007). In comparison to most routines, our algorithm performs a double filtering approach similar to that applied in Caron and Jones (2012) to ensure that the genesis and dissipation phases of TCs are well represented and that TCs are not counted twice in the case of a temporary decrease of intensity followed by a restrengthening. Looser detection criteria (with lower values than the standard thresholds values) were first used in order to detect all storm centers; then criteria were enforced to standard values following the literature (strict criteria). Centers that satisfy the strict criteria are then classified as being “strong” centers (storm identification), while the others are classified as “weak” centers. To correctly represent each track (storm tracking), the strong and weak centers are then paired following two different methods: the storm history using a similar approach to that of Sinclair (1997) and the nearest-neighbor method as in Blender et al. (1997), Blender and Schubert (2000) and Schubert et al. (1998). Once the storm tracks are defined, the algorithm determines the core of each track as the centers sitting between the first and the last strong centers found in the track, thus neglecting the genesis and dissipation phases. This subsection of the track (representing the main TC lifetime) has to satisfy a third set of criteria that reject TCs that do not live long enough, that do not travel a long-enough distance or that do not reach the strength of a tropical storm. If the core of the storm track satisfies all these criteria, the genesis and dissipation phases (represented by the weak centers that occurred before the first and after the last strong centers) are added to form the complete storm track. A detailed description of the storm identification and tracking can be found in Appendix A.
Many environmental proxies have been used to link the changes in the
dynamical and thermodynamical fields to TC activity. Here, two well-known
environmental proxies were adopted – the potential intensity (
To evaluate the CRCM6 performance in simulating tropical cyclones, an
additional 0.22
In this study, we use a two-tailed Student
To evaluate the statistical significance in the difference in TC counts between different climate states, a bootstrap method was used to create 100 randomly selected 30-year samples out of the 40-year (1979–2018) distributions of the annual number of observed TCs and ERA5 TCs.
In this section, the TCs in the MH and PI climate conditions are studied to evaluate how changes in orbital forcing, dust and vegetation feedbacks impact TC activity in the Atlantic Ocean, by focusing on TC trajectories and annual frequency (Sect. 3.1), seasonality (Sect. 3.2) and intensity (Sect. 3.3). We also highlight the impacts of such changes on the different variables known to affect TC genesis.
The PI climate simulation has a spatial distribution of Atlantic hurricanes
that is similar to present climate, where most of the TCs form in the main
development region (MDR) and move west–northwestward towards the North
American east coast (Fig. 2b). However, there are fewer TCs in the simulated
PI climate than in the present-day climate simulation driven by ERA-Interim
(cf. Fig. 2a and A2); this is due to a large extent to the SST cold bias in the
EC-Earth simulation (
June to November (JJASON) climatology of
The northward shift of the ITCZ in the MH is due to energetic constraints
associated with the changes in orbital forcing causing a warming of the NH
and a cooling of SH during boreal summer relative to PI (Merlis et al., 2013; Schneider et al., 2014; Seth et al., 2019). The ITCZ is associated with more favorable conditions for cyclogenesis by increasing the ambient vorticity and
therefore the TC activity (Merlis et al., 2013). Our analysis shows that the absolute vorticity maximum undergoes a northward shift relative to the control experiment, following the ITCZ displacement (Fig. A5), supporting the northward shift of the TC tracks (Fig. 2c). Higher
absolute vorticity values are also found over the Greater Antilles and the
western part of the Gulf of Mexico where there is a higher TC occurrence in
the MH
The northward shift and the increase of TC activity in the MH experiments are also related to the strengthening of the WAM, which amplifies the westerly winds, and the SST anomaly (Fig. A6). Such changes lead to the development of a wind shear pattern anomaly in the MDR, with lower values of wind shear in the central–western region of the MDR and higher values on the eastern side of the MDR relative to PI. Thus, while the area more favorable for TC development is reduced (Fig. A7), the more favorable conditions present on the western side more than compensate the decrease in the east, allowing more cyclones to develop in the MH experiment. In addition to the zonal atmospheric circulation changes, the enhanced northward penetration of the WAM together with the displacement of the ITCZ leads to a northward shift of the maximum in AEW activity in the MH experiments relative to PI (see Fig. A8). The poleward displacement of the AEWs may also contribute to the changes in TC genesis location as they influence the region where TCs develop (e.g., Caron and Jones, 2012).
The vegetation changes and the associated reduction in dust concentrations
further strengthen the WAM in the MH
The GPI anomalies of both MH experiments relative to PI closely follow the changes in the atmospheric and oceanic environmental factors that can affect TCs (cf. Figs. 3, A5–A9). The GPI shows more favorable conditions with higher values of vorticity and SST and lower wind shear values. Similarly to the absolute vorticity field (Fig. A5), the GPI shows a small northward shift relative to the control experiment, thus contributing to the poleward displacement of the TC genesis locations and therefore the the TC tracks (cf. Figs. 3 and 4).
Changes in seasonal GPI (JJASON) for
TC seasonal (JJASON) cyclogenesis density anomaly for
The largest changes in GPI are seen in the MH
In terms of frequency, an average of 5.5 TCs per year is simulated in the PI
experiment (Fig. 2b). This is
Bootstrap distributions based on
To analyze changes in TC seasonal cycle, we consider changes in the monthly
number of TCs rather than change of the length of the TC season. The PI
climate has a TC seasonal cycle that is similar to the present climate, with
a peak in TC in September (Fig. 6). The MH experiments show a distinct
pattern: a decrease in TC activity at the beginning of the hurricane season
for both MH experiments (statistically significant for the MH
TC climatological distribution throughout the extended TC season (JJASON) for each experiment. Error bars (whiskers) indicate the standard error of the mean.
Gaetani et al. (2017), using the same global model experiments performed
with EC-Earth, showed a large decrease in the AEWs in the MH
Monthly AEW anomalies represented
through the variance of the meridional wind at 700 hPa, filtered in the 2.5 to 5 d band, for
The reduction of the AEWs in the MH
AEJ represented through a vertical cross
section of zonal mean (0–40
Other factors could be playing a role in modifying the TC seasonal cycle. In particular, the shift in TC seasonal cycle could be related to changes in the orbital forcing, most importantly the precession of the equinoxes: during the MH, the perihelion was in September instead of January, as today, with the stronger insolation anomalies peaking in late summer at NH low latitudes. Furthermore, while higher potential intensity (due mostly to warmer SSTs; see Figs. A6 and A11) develops on the western part of the MDR and most of the North Atlantic Ocean from June to September relative to the PI experiment, the strengthening of the WAM causes a cold anomaly response over the eastern part of the MDR, together with stronger vertical wind shear and weaker absolute vorticity values. The withdrawal of the WAM in late September then causes the decrease in wind shear and positive anomalies in both absolute vorticity and SSTs to extend eastward. These environmental anomalies are likely the reason for the TC seasonality changes during the MH experiments (Figs. A11a, A12a, A13a). The cyclogenesis anomalies and the GPI changes are consistent with these assumptions (cf. Figs. 9a and A14a). Korty et al. (2012), who studied the response to orbital forcing in PMIP2 models during the MH, also found that the TC season in the Northern Hemisphere was less favorable during summer, while it became more favorable during fall (October and November) relative to pre-industrial climate. The authors pointed out that these findings were due to the difference between the warming rate of the atmosphere (which warms faster during the summer months) and that of the ocean surface, which led to a negative potential intensity anomaly during the first half of their TC season (June to September) and a positive anomaly during the second half (October to November). Using PMIP3 models, Koh and Brierley (2015) drew similar conclusions; however, the changes in fall were not a robust signal across models.
Changes in climatological monthly GPI
for
Accounting for the Saharan greening and reduced airborne dust concentrations
(MH
Seasonal variation (JJASON) of the GPI summed over the experimental domain for the three experiments.
Changes in climatological monthly GPI
for MH
To assess TC intensity, we considered the 10 m maximum wind speed in 3 h
intervals and then classified them using the Saffir–Simpson scale
categories. For the three experiments, most tropical cyclones reach only
tropical storm or hurricane Category 1 (93 %, 88 % and 91 % for PI, MH
Climatological number of TCs per year in various categories for the three experiment during the TC extended season (JJASON).
To better understand the cause of these changes, we turn to the seasonal
Changes in climatological seasonal
In this study, we use the CRCM6 regional climate model with a high horizontal resolution (0.11
Gaetani et al. (2017) showed a strong decrease in AEW in the MH
Our study suggests that the different orbital parameters together with the
changes in the WAM intensity may have been the main causes of the changes in
TC seasonality, offering better conditions for cyclogenesis towards the end
of the hurricane season. WAM intensity affects the wind shear on the eastern
side of the MDR. The WAM withdrawal towards the end of the summer extended
the more favorable conditions from the central–western portion towards the
eastern portion of the MDR, causing an increase in TC activity during the
second half of the season in the MH simulations. These results are
consistent with the findings of Korty et al. (2012), who also showed higher
cyclogenesis potential towards the end of the PI hurricane season in their
MH experiment, with likely increase in TC activity during October when the
GPI is at its maximum. However, their results are based on the entire
Northern Hemisphere, while here we only focus on the North Atlantic Ocean. In
addition, our results compare well to those of Koh and Brierly (2015), who
found less favorable environmental conditions for TC development during
Northern Hemisphere summer in the MH relative to pre-industrial when
analyzing PMIP3 model simulations. Hence, the impact on TCs on changes in
orbital forcing is consistent across different models and highlights an
interesting point where there may be a repression of the modeled
environmental conditions that negatively affects proxies associated with TC
(i.e., the
Finally, the simulated impact of dust changes needs further investigation,
as rainfall in the north of Africa can be strongly affected by the dust
optical properties (e.g., “heat pump” effect where the atmospheric dust
layer warms the atmosphere, enhancing deep convection and intensifying the
WAM; see Lau et al., 2009). In particular, in EC-Earth, dust particles are
moderately to highly absorbing particles (single scattering albedo
In conclusion, our study highlights the importance of vegetation and dust
changes in altering TC activity and calls for additional modeling efforts to
better assess their role on climate. For example, employing regional model
simulations with atmosphere and ocean coupling will be important to better
represent the interactions between TC activity and TC–ocean feedbacks as
a large amount of energy is transferred through TC activity between the
atmosphere and the ocean (Scoccimarro et al., 2017). Furthermore, to
validate the model results, additional new paleotempestology records across
the Gulf of Mexico and Caribbean Sea will be of paramount importance. While
our study shows an increase in TC frequency and intensity during a climate
state with warmer summers and a stronger WAM, it is difficult to draw a
direct conclusion for the future, as environmental proxies associated with
TCs (i.e., the
In this study, we developed a tracking algorithm that makes use of a three-step procedure to detect cyclones, following previous studies (Gualdi et al., 2008; Scoccimarro et al., 2011; Walsh et al., 2007).
The storms are identified with the following criteria:
The surface pressure at the center must be lower than 1013 hPa and lower than its surrounding grid boxes within a radius of 24 km ( The center must be a closed pressure center so that the minimum pressure difference between the center and a circle of grid points in a small and a large radius around the center (200 and 400 km radii) must be greater than 1 and 2 hPa, respectively. There must be a maximum relative vorticity at 850 hPa around the center (200 km radius) higher than The maximum surface wind speed around the center (100 km radius) must be stronger than 8 m s To account for the warm core, temperature anomalies at 250, 500 and 700 hPa are calculated, where each anomaly is defined as the deviation from a spatial mean over a defined region. The sum of the temperature anomalies between the three levels must then be larger than 0.5 If there are two centers nearby, they must be at least 250 km apart from each other; otherwise, the stronger one is taken.
To identify the genesis and dissipative phases of the TCs, a double
filtering approach was used, similar to that applied by Caron
and Jones (2012). The aforementioned threshold values were used to first
detect all the potential centers that could belong to a storm for each time
step. Then, these criteria were enforced to the standard values (defined
below) following the literature (Gualdi
et al., 2008; Scoccimarro et al., 2011; Walsh, 1997; Walsh et al., 2007),
and these new threshold values were applied to each center to identify the
ones that satisfied these enforced criteria among the potential weak ones
that were predefined. The centers that satisfied the standard criteria were labeled
by the algorithm as being strong centers (or real TC centers), while those
that only satisfied the first set of criteria were identified as being weak
centers (with standard value properties defined below).
The enforced criteria are the following:
surface pressure at the center deeper than 995 hPa; minimum pressure difference between the center and a 200 and 400 km radius greater than 4 and 6 hPa, respectively; relative vorticity maximum larger than wind speed maximum above 17 m s warm core temperature anomaly above 2 The maximum wind velocity at 850 hPa must be larger than the maximum wind velocity at 300 hPa.
Another condition was added that only the strong centers needed to satisfy:
In doing so, we avoided double counting cyclones that may decreased in intensity before re-intensifying. Conditions (e) and (f) are the main conditions that filtered the TCs centers from other low-pressure systems and extratropical cyclones, as TCs have a warm core in their upper part and stronger low-level wind speed than other storms.
Storms were then tracked as follows: for each potential center found, the algorithm used the nearest-neighbor method, which was also applied in many other studies (Blender et al., 1997; Blender and Schubert, 2000; Schubert et al., 1998), to find a corresponding center in the following 3 h time interval within a 250 km radius around the storm center. Once two centers were paired, they formed a storm track. The potential position of the next center to be a continuation of the storm was then calculated using the storm history, based on the position of the previous two centers, which allowed us to establish a possible speed and direction for the predicted center. A similar procedure was applied in Sinclair (1997) and was derived from Murray and Simmonds (1991). The algorithm then searches around the last storm center using the nearest-neighbor method and around this potential position at the next time step to find a matching center. The nearest center was always chosen first.
Once a track was completed, it had to satisfy the following final conditions:
The TC had to exist for at least 36 h (with a minimum of 12 centers at 3 h intervals). The TC needed to have at least 12 strong centers along its entire track, so that the shortest TC had only strong centers (36 h). The TC had to travel at least 10 The number of strong storm centers needed to represent at least 77 % of a subpart of the complete storm track delimited by the first and last strong center found by the algorithm. This way, we ensured that the storm was classified as a TC during most of its time.
JJASON climatology (1980–2009) of
Climatological track density (JJASON) for the
Climatological changes in SST between simulated EC-Earth pre-industrial climate and ERA-Interim reanalysis from 1979 to 2008. The black box shows the approximate present-day MDR. The contour lines follow the color-bar scale (dashed, negative anomalies; solid, positive anomalies); the 0 line is omitted for clarity.
Monthly mean climatological changes in precipitation for
Climatological (JJASON) changes in 850 hPa absolute vorticity relative to PI for the
Climatological (JJASON) changes in SST for the
Changes in vertical wind shear (300–850 hPa) for the
AEWs represented through the variance
of the meridional wind at 700 hPa, filtered in the 2.5 to 5 d band, for the
Climatological (JJASON) changes in 200 hPa wind speed
for the
Comparison between TC seasonal cyclogenesis density
anomalies for the
Monthly mean changes in climatological SSTs for the
Monthly mean changes in climatological wind shear (200–850 hPa) for the
Monthly mean changes in 850 hPa absolute vorticity for the
TC seasonal (JJASON) cyclogenesis density anomaly for the
Changes in climatological seasonal CAPE between a saturated boundary layer air parcel and an air parcel that has been isothermally lowered to a reference level (JJASON) for the
The code used to track the tropical cyclones is available from the corresponding author upon request.
The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.
FSRP conceived the study and designed the experiments with contributions from SD and RL. KW carried out the model simulations and SD analyzed the model output. SD and KW developed the tracking algorithm. SJC and KE provided the codes to compute the indexes. All authors contributed to the interpretation of the results. SD wrote the manuscript with contributions from all co-authors.
The authors declare that they have no conflict of interest.
The authors would like to thank Georges Huard and Frédérik Toupin
for the technical support; the Recherche en Prévision Numérique
(RPN), the Meteorological Research Branch (MRB) and the Canadian
Meteorological Centre (CMC) for the permission to use the GEM model as basis
for our CRCM6 regional climate model; and Qiong Zhang for sharing the global
model outputs. We acknowledge use of the ERA5 data by ECMWF provided through the Copernicus Climate Change Services (
This research has been supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) (grant nos. RGPIN-2018-04981 and RGPIN-2018-04208) and the Fond de recherche du Québec – Nature et Technologies (FRQNT) (grant no. 2020-NC-268559).
This paper was edited by Martin Claussen and reviewed by Robert Korty and one anonymous referee.