Geophysical Research Letters

Is cumulus convection the concertmaster of tropical cyclone activity in the Atlantic?

Authors

  • Cristiana Stan

    Corresponding author
    1. Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, Virginia, USA
    2. Center for Ocean-Land-Atmosphere Studies, Calverton, Maryland, USA
      Corresponding author: C. Stan, Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, VA 22030, USA. (cstan@gmu.edu)
    Search for more papers by this author

Corresponding author: C. Stan, Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, VA 22030, USA. (cstan@gmu.edu)

Abstract

[1] The influence of the cloud representation in global climate models on the accuracy of the North Atlantic tropical cyclone simulations is investigated. The North Atlantic tropical cyclone activity is simulated with a standard climate model, CCSM. The conventional parameterization of cloud processes in CCSM is replaced by the “super-parameterization” and the simulation is run again. The comparison of tropical cyclone statistics reveals that the model with explicit representation of cloud processes produces a larger number of events, with stronger intensity and longer life-cycle. The results show that clouds have a significant impact on the mechanisms associated with tropical cyclone activity such as surface temperature and ocean subsurface processes, vertical wind shear, and the transport of moisture in the lower troposphere.

1. Introduction

[2] Tropical cyclone (TC) activity in the North Atlantic Ocean is the most challenging manifestation of low-latitude synoptic variability (less than 10 days) and has one of the largest socio-economic impacts among the nature's caprices. In this basin, TCs are observed to form beyond 5°N during the boreal summer and early fall. When conditions in the ocean and the atmosphere's large-scale circulation are favorable, cyclones can move over the land. As they make landfall, the coastal shores endure severe winds and rain, which sometimes have devastating effects. Simulation and prediction of TCs depend on the climate model accuracy. Both model resolution and representation of physical processes can have an impact on the model ability to capture the observed characteristics of the TC activity [Reed and Jablonowski, 2011].

[3] The onset and development of TCs involve a concerted effort among the favorable conditions in the ocean mean-state, large-scale atmospheric circulation, and cloud-scale processes associated with enhanced cumulus convection [Gray, 1968]. Warm sea-surface temperature (SST) anomalies fuel the feedback processes at the air-sea interface, alter the horizontal- and vertical-shear of the trade winds [Shapiro and Goldenberg, 1998], and the buoyancy of cumulus updrafts [Tompkins, 2001]. Strong rotational winds, which define the TCs, tend to develop in regions of existing asymmetric disturbances of cyclonic vorticity embedded in the prevalent tropical easterly flow. A region of organized convective activity is also necessary for the TC activity to occur and the mesoscale organization of cumulus convection acts as a thermodynamic engine of the TCs.

[4] These conditions are often encountered over the North Atlantic tropical ocean but seldom do they evolve into intense storms. Only when the cluster of clouds acts simultaneously to enhance the existing warming in the lower troposphere and to transport horizontal momentum in the vertical do the pre-existing conditions evolve into a strong TC. Numerical simulations and seasonal forecasts of TCs with either atmosphere-only models or coupled ocean-atmosphere climate models show poor results with respect to the observed frequency and intensity [Camargo et al., 2005; Emanuel et al., 2008]. Numerous studies [Randall et al., 2007; Emanuel et al., 2010] argue that horizontal resolution of climate models is not enough to simulate the sustained winds concentrated on very small areas, which are associated with TCs. While increasing the horizontal resolution shows a better simulation of TC statistics [Chauvin et al., 2006; Bengtsson et al., 2007; Zhao et al., 2009; Caron et al., 2011; Putman and Suarez, 2011; Manganello et al., 2012], these studies do not tackle the fundamental problem of the role of representing clouds in climate models. There is also evidence [Vitart et al., 2001] that conventional parameterizations of cloud processes have an impact on the simulation of TC statistics and Zhao et al. [2012] showed that global TC frequency and storm intensity are sensitive to the cumulus mixing rate.

[5] In this study, cloud processes, including the entrainment, are explicitly represented and the fluxes at air-sea interface are driven by the ocean-atmosphere feedbacks. The explicit representation is a non-hydrostatic formulation of cloud dynamics, includes prognostic equations for water vapor, cloud condensates (water droplets and ice particles) and precipitating particles (rain, snow, and graupel), and couples the cloud microphysics to the radiative transfer processes at their native scale.

2. Climate Simulations

[6] In this study, we compare the simulation of the North Atlantic TC activity in two versions of the Community Climate System Model, version 3.0 (CCSM3) [Collins et al., 2006]. In the first simulation, which is referred to as CCSM the representation of cloud processes is through conventional parameterizations based on the statistical effects of clouds on the large-scale variables. In the second simulation, cloud processes are explicitly resolved using the multi-scale modeling framework, also known as the super-parameterization [Grabowski, 2001; Khairoutdinov and Randall, 2001]. This version of the model is referred to as SP-CCSM [Stan et al., 2010] and it has the convection processes represented by a 2D cloud-process resolving model (CRM) embedded in each grid column of the atmospheric component. CCSM and SP-CCSM use the same horizontal resolution in their components, spectral T42 (∼300 km) for the atmosphere and land models, and 3 degrees (∼300 km) for the ocean and ice models. The horizontal resolution of the individual 2D CRM is 4 km.

[7] Both simulations have the same initial conditions. The atmospheric states are taken from an AGCM simulation with observed SST climatology as boundary condition and the ocean states are climatological values of temperature and salinity and assuming no ocean currents at start. The numerical integrations are 25 years long and the results presented in this paper are based on the last 20 years. SP-CCSM simulates many aspects of the observed tropical climate and its variability and with smaller biases than peer models [e.g.,Stan et al., 2010; DeMott et al., 2011, 2012].

[8] The TC activity simulated by the two climate models is investigated using a downscaling analysis technique [Emanuel, 2006], which generates statistics of synthetic storm tracks consistent with the large-scale circulation of the climate model (same means, variances and covariances). A hurricane model is run along each of the synthetic tracks to estimate the intensity of TCs.

3. Results

[9] Figure 1 compares the spatial distribution of TC activity blue obtained when downscaling the models using Emanuel's [2006] technique. The events included in TC statistics analysis are filtered to retain only those whose maximum wind speed exceeds 35 kt and genesis is defined as the first point of each event with maximum winds exceeding this threshold. The scatter plot of genesis points (Figures 1a and 1b) shows that throughout the basin, SP-CCSM simulates a uniform distribution of centers of action whereas in CCSM areas with generation of vortices are narrowly confined near the 10°N and along the eastern shore. Furthermore, the genesis density (Figures 1c and 1d) maps indicate that in CCSM within these areas the potential for development of TC activity is reduced. In addition, Figures 1e and 1fshow higher track density in SP-CCSM than in CCSM. The comparison of maps of the power dissipation indexEmanuel [2005], which takes into account the intensity, frequency, and duration of TCs reveals that SP-CCSM simulates events with increased intensity and lifetime. CCSM's results blue presented here are consistent with those from previous studies using this model. For example,Emanuel et al. [2008] have shown that CCSM3 simulates very little of the observed TC activity, including frequency, intensity, and duration in the North Atlantic basin.

Figure 1.

Comparison between the TC activity simulated by CCSM (left) and SP-CCSM (right). (a, b) Genesis points, (c, d) genesis density accumulated in a 5° × 5° area, (e, f) track density accumulated in a 5° × 5° area, and (g, h) power dissipation.

[10] To further investigate the models' TC activity statistics, Figure 2compares the annual number of events in the North Atlantic basin as function of maximum wind speed that each event reaches during its lifetime and the number of TCs in each month simulated by the two models. Both models simulate TCs with intensities in the same range and the SP-CCSM tends to produce more events for almost the whole spectrum of maximum wind speeds. The phase of the annual cycle [e.g.,Camargo et al., 2007] is realistically simulated; however, the number of TCs produced by SP-CCSM is significantly larger, especially during the observed intense activity season.

Figure 2.

(a) Histogram of TC intensity measured by the lifetime maximum wind speed and (b) the annual cycle of TC activity. Black bars correspond to CCSM and gray bars to SP-CCSM.

[11] Since the only difference between the two models is the way in which clouds are represented, the remainder of the paper is focused on understanding the role of clouds in the tropical cyclogenesis. TC activity strongly depends on the scale interactions in the climate system [Ooyama, 1982]. As mentioned in the introduction, the atmospheric large-scale circulation has to favor the westward propagating African waves blue or MJO-related disturbances in the form of disorganized thunderstorms to organize into a self-sustaining, synoptic-scale vortex and cloud-scale processes to transfer the ocean heat flux into the developing vortex. We will investigate the impact of representing cumulus convection on the elements mentioned above as the main players in the formation and maintenance of the tropical storm systems.

[12] In the North Atlantic basin, TCs with intense and sustained surface winds, refereed to as hurricanes when the wind speed is 64 kt or more, form when the SST is in the 26.5°C–30°C range [Lajoie and Walsh, 2010] and this warm water extends away from the equator where the Coriolis effect is of sufficient strength. While the surface temperature plays a significant role in the cyclogenesis, the intensity of TCs is partially related to the subsurface ocean structure. Figure 3compares the SST and the upper ocean heat content (OHC) climatologies for August-September-October (ASO) simulated by the two models. Overall, the SSTs in the CCSM (Figure 3b) are colder than in SP-CCSM (Figure 3a) and only over a small region do they exceed the value of 26.5°C. In the SP-CCSM (Figure 3d) the vertically integrated OHC in the upper 300 m is larger than in CCSM (Figure 3e). Similarly, the thermocline in the SP-CCSM is deeper than in CCSM (not shown). Case studies [e.g.,Shay et al., 2000; Shay, 2009] suggest that a warm, deep ocean regime favors a rapid intensification of storms. The distribution of genesis points in the vicinity of 10°N in CCSM and weak intensity is consistent with the shoaling of the thermocline seen in Figure 3f. The weak TC activity in the Caribbean simulated by CCSM can be due to colder SSTs in this region.

Figure 3.

(a–c) Sea surface temperature (°C) and (d–f) upper ocean heat content (°C) ASO climatology simulated by SP-CCSM and CCSM, and the difference between the two models (SPCCSM-CCSM). The 26.5°C isotherm in the SST plots is highlighted. The shading in the difference plots denotes the 90% confidence level.

[13] If the CCSM tropical circulation has systems of disturbed weather that have the potential of evolving into hurricanes, as they move over cold waters the disturbances will weaken. It is also possible that once formed the disturbances cannot move poleward because of their location close to the equator where the Coriolis acceleration is very small [Chan, 2005]. The temperature of the ocean's surface is partially influenced by turbulence in the ocean mixing layer and solar radiation reaching the surface. The representation of ocean mixing processes is the same in both models. Representation of clouds residing over the area is different. Stratocumulus clouds, which tend to form over the relatively cold open ocean, reduce the downward radiation and further cool the SST by maintaining a positive feedback. The amount of low-level cloud over the North Atlantic basin during the ASO is larger in the CCSM simulation than in the SP-CCSM (not shown).

[14] Other factors contributing to the intensification of tropical disturbances are the vertical wind shear and the available moisture in the region of converging winds. Large values of the wind shear allow the angular momentum and condensation heat to escape from the disturbed region and the developing cyclone weakens by spreading over a large area [Knaff et al., 2004]. Figure 4shows that CCSM simulates vertical wind shear with amplitudes larger than in the SP-CCSM.

Figure 4.

(a–c) ASO climatology of the magnitude of the vertical wind shear (850 hPa–250 hPa, and (d–f) the relative humidity (vertical integral of the 850–700 hPa layer, simulated by SP-CCSM and CCSM, and the difference between the two models (SPCCSM-CCSM). The shading in the difference plots denotes the 90% confidence level.

[15] In the SP-CCSM the CRMs allow a dynamical representation of updrafts and downdrafts inside the cloud ensemble; thus the multi-scale modeling approach prevents the horizontal winds from the surrounding cumulus free environment to become very different from the zonal wind inside the cumulus clouds.

[16] The distribution of relative humidity in the 850–700 hPa layer indicates that CRMs are more efficient in moistening the lower troposphere than the conventional parameterization since at the surface the difference between the relative humidity in the two models does not exceed 5% (not shown). The low relative humidity at the surface in the CCSM yields to a dry environment conductive to evaporation and more water vapor can evaporate from the surface. However, relationship between the model temperatures in the 850–700 hPa layer is reversed from that at the surface shown in Figure 4, and the SP-CCSM is colder. This result suggests that cumulus convection is the main factor through which the atmosphere becomes humidified and thus more favorable to hurricane triggering.

4. Conclusions

[17] The individual influences of the mechanisms discussed here were proved in highly idealized studies. This experiment is unique because it combines all of them in a realistic framework and it shows that the representation of cumulus convection in numerical models is the key element for the successful simulation of the dynamical and thermodynamical mechanisms involved in the TC activity. The explicit representation of cloud processes results in increased moistening of the lower troposphere, which is more favorable to cyclogenesis. This conclusion is consistent with the increased TC frequency in the cumulus parameterization with enhanced cumulus mixing [Zhao et al., 2012]. Cumulus convection plays the concertmaster role by tuningthe effects of the ocean-atmosphere interaction into a rapid intensification of the resulting disturbances. Representation of cumulus convection by conventional parameterization divides the cloud-scale processes into shallow and deep systems and requires closure assumptions for each scheme. The transition from shallow to deep convection depends on the relationship between the moisture and temperature of the cloud and the environment [Wu et al., 2009]. Cumulus parameterization in CCSM does not include a representation of the transition mechanisms. These results have implications for a broad range of climate applications. The ability of SP-CCSM to capture some salient features of the TC activity opens a new chapter for the necessity of global climate models in studying the conditions that favor hurricanes or comparable tropical storms. Unlike regional climate models [Wang and Sobel, 2011], in global models the atmospheric teleconnections from the equatorial Pacific and Sahel [Pielke and Landsea, 1999; Kim et al., 2009; Goldenberg and Shapiro, 1996] are directly included in the large-scale circulation, which interacts with the tropical deep convection. Concurrently, climate simulations with accurate representation of global tropical cyclones statistics will have a smaller number of uncertainties when used for climate projection scenarios. Importantly, the results we report here are extendable to the other components of the climate system affected by the TC activity. The upper thermocline simulated in SP-CCSM is deeper than in CCSM simulation. Due to the coarse resolution of the ocean model one can only speculate that there is a relationship between the two phenomena but further sensitivity studies are required to shed light on this hypothesis.

Acknowledgments

[18] This research was supported by the NSF under grants 0830068, ATM-0425247, and AGS-1119269. Computing resources were provided by the Computational and Information System Laboratory at NCAR. The author is grateful to Kerry Emanuel for applying the downscaling analysis and insightful discussion during the preparation of the paper. I would like to thank Wayne Schubert for his suggested improvements of the paper. Two anonymous reviewers provided useful comments on an early draft.

[19] The Editor thanks the two anonymous reviewers for assisting in the evaluation of this paper.