Abstract

Changes in tropical cyclone (TC) frequency under anthropogenic climate change are examined for 13 global models from phase 5 of the Coupled Model Intercomparison Project (CMIP5), using the Okubo–Weiss–Zeta parameter (OWZP) TC-detection method developed by the authors in earlier papers. The method detects large-scale conditions within which TCs form. It was developed and tuned in atmospheric reanalysis data and then applied without change to the climate models to ensure model and detector independence. Changes in TC frequency are determined by comparing TC detections in the CMIP5 historical runs (1970–2000) with high emission scenario (representative concentration pathway 8.5) future runs (2070–2100). A number of the models project increases in frequency of higher-latitude tropical cyclones in the late twenty-first century. Inspection reveals that these high-latitude systems were subtropical in origin and are thus eliminated from the analysis using an objective classification technique. TC detections in 8 of the 13 models reproduce observed TC formation numbers and geographic distributions reasonably well, with annual numbers within ±50% of observations. TC detections in the remaining five models are particularly low in number (10%–28% of observed). The eight models with a reasonable TC climatology all project decreases in global TC frequency varying between 7% and 28%. Large intermodel and interbasin variations in magnitude and sign are present, with the greatest variations in the Northern Hemisphere basins. These results are consistent with results from earlier-generation climate models and thus confirm the robustness of coupled model projections of globally reduced TC frequency.

1. Introduction

The state of the science summary paper by a World Meteorological Organization (WMO) Expert Team (Knutson et al. 2010) reported that late-twenty-first-century model projections indicate decreases in tropical cyclone (TC) frequency consistently across models from different global centers, with values ranging from −6% to −34% globally. They reported a lack of consistency for individual cyclone basins, with changes of ±50% or more projected by various models. With data from the generation of coupled atmosphere–ocean climate models that feature in phase 5 of the Coupled Model Intercomparison Project (CMIP5) now available, we examine the output of 13 CMIP5 models to determine whether the Knutson et al. findings on TC frequency are substantiated by this later generation of models, by investigating the projected changes in the frequency of tropical cyclones at the end of the twenty-first century under a high emission scenario [representative concentration pathway 8.5 (RCP8.5); van Vuuren et al. 2011].

For details of the CMIP5 experimental design, the reader is referred to Taylor et al. (2012). TCs are detected directly in the climate model data using the Okubo–Weiss–Zeta parameter (OWZP) methodology of Tory et al. (2013a,b,c). This method determines TC occurrence through detection of the large-scale atmospheric state under which tropical cyclogenesis occurs. This means that it is not dependent on the climate model's capability of resolving the TC. At the core of the method is the requirement for enhanced levels of low-deformation vorticity to be present in the formation environment. The low-deformation vorticity as determined from the modified Okubo–Weiss parameter termed OWZ is

 
formula

Here, is a normalized Okubo–Weiss parameter, is the vertical component of relative vorticity, is the Coriolis parameter, and is the absolute vorticity. The normalized Okubo–Weiss parameter is defined as

 
formula

where the numerator is the Okubo–Weiss parameter and E and F represent the stretching and shearing deformation, respectively,

 
formula
 
formula
 
formula

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

A detailed assessment of the detection method performance [applied to 20 yr of European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Re-Analysis (ERA-Interim) data and verified against observed TCs] is given in Tory et al. (2013a). This performance is summarized in the first row of Table 1. The authors note that no detection method is perfect, and as a consequence any discussion of model performance regarding TC frequency should acknowledge that the performance relates to the combination of the model's ability to generate TC-like circulations and the detector's ability to identify them. It follows that there is value in using different detection techniques in assessing TC frequency in climate models and that reported detections reflect both the model and detection scheme.

Table 1.

Thresholds (boldface) constituting the OWZP and performance measures (italics) for the OWZP combinations discussed in this paper. The OWZP developed and used in Tory et al. (2013a,b,c) appears in the first row, and the three modifications discussed in section 2b and the appendixes appear in the subsequent rows. These include limiting the Coriolis contribution to OWZ to a latitude of 20° (OWZ f ≤ 20°) and excluding storms from the analysis that form poleward of a 200-hPa jet (no subtropical TCs). The performance relates to application of the OWZPs to ERA-Interim data and verification against 20 yr of International Best Track Archive for Climate Stewardship (IBTrACS) data. “Hit (%)” represents observed TC tracks with a matching detected track. “False alarm (%)” represents a ratio of false detections to observed TCs. The critical success index (CSI) is the ratio of hits to the sum of observed storms plus false alarms (a perfect score = 1.0). Also shown are relative humidity (RH), wind shear (Wsh), and specific humidity (SH). Subscripts in the threshold labels refer to pressure levels (hPa).

Thresholds (boldface) constituting the OWZP and performance measures (italics) for the OWZP combinations discussed in this paper. The OWZP developed and used in Tory et al. (2013a,b,c) appears in the first row, and the three modifications discussed in section 2b and the appendixes appear in the subsequent rows. These include limiting the Coriolis contribution to OWZ to a latitude of 20° (OWZ f ≤ 20°) and excluding storms from the analysis that form poleward of a 200-hPa jet (no subtropical TCs). The performance relates to application of the OWZPs to ERA-Interim data and verification against 20 yr of International Best Track Archive for Climate Stewardship (IBTrACS) data. “Hit (%)” represents observed TC tracks with a matching detected track. “False alarm (%)” represents a ratio of false detections to observed TCs. The critical success index (CSI) is the ratio of hits to the sum of observed storms plus false alarms (a perfect score = 1.0). Also shown are relative humidity (RH), wind shear (Wsh), and specific humidity (SH). Subscripts in the threshold labels refer to pressure levels (hPa).
Thresholds (boldface) constituting the OWZP and performance measures (italics) for the OWZP combinations discussed in this paper. The OWZP developed and used in Tory et al. (2013a,b,c) appears in the first row, and the three modifications discussed in section 2b and the appendixes appear in the subsequent rows. These include limiting the Coriolis contribution to OWZ to a latitude of 20° (OWZ f ≤ 20°) and excluding storms from the analysis that form poleward of a 200-hPa jet (no subtropical TCs). The performance relates to application of the OWZPs to ERA-Interim data and verification against 20 yr of International Best Track Archive for Climate Stewardship (IBTrACS) data. “Hit (%)” represents observed TC tracks with a matching detected track. “False alarm (%)” represents a ratio of false detections to observed TCs. The critical success index (CSI) is the ratio of hits to the sum of observed storms plus false alarms (a perfect score = 1.0). Also shown are relative humidity (RH), wind shear (Wsh), and specific humidity (SH). Subscripts in the threshold labels refer to pressure levels (hPa).

Of the 13 CMIP5 models examined, the TC detections in 5 models do not reproduce reasonable current climate TC climatology. Of the remaining eight models, the results of the earlier generation of models are reaffirmed with a global reduction of TC frequency, but with high uncertainty in the response in three Northern Hemisphere basins.

As discussed, the purpose of the study is to test the robustness of the results from earlier-generation models of a projected decrease in tropical cyclone frequency. This is confirmed. However, an unexpected result is a large number of higher-latitude tropical-type vortices in the late-twenty-first-century projections of several global models. Inspection revealed that these cyclones have baroclinic processes operating at their development stage. Various methods were examined to remove these cyclones from consideration. The most effective is to classify as “subtropical” any detection that occurs poleward of the subtropical jet stream.

The paper is structured as follows: Section 2 describes the methodology, including three modifications introduced since the OWZP was first published, and the models used. Detailed descriptions of the modifications and the implications for the results are provided in appendixes. Section 3 presents and discusses the results of the modified OWZP applied to the selected CMIP5 models. This includes the ability of the models to reproduce and the OWZP detector to detect the current climatology of TCs, as well as the changes in TC frequency in the future (warmer) climate. Conclusions are given in section 4.

2. Methodology

a. OWZP TC detection

The methodology determines the existence of TCs directly from the climate model output based on the OWZP detector and tracker introduced by the current authors in Tory et al. (2013a,b,c). The OWZP method has three features that distinguish it from other methods:

  • It is based on the observations that the parameter space defining the conditions under which TCs form is large scale in nature, on the order of 1000 km (McBride and Zehr 1981, Davidson et al. 1990). Thus, rather than detect the mesoscale structure of TCs directly, it detects the environment that favors their formation.

  • Circulations with the dynamic potential to support TC formation are identified and tracked in the climate model output. The identification is based on (model-resolution independent) threshold values of OWZ [Eq. (1)] on the 850- and 500-hPa pressure levels. These threshold values are used to identify low-deformation vorticity (near-solid-body rotation) on the scale of a few hundred kilometers throughout the low–middle troposphere, which is deemed to be necessary for TC formation (Tory et al. 2013c). A TC is declared when these OWZ thresholds plus thermodynamic and vertical wind shear thresholds are satisfied for at least 48 h. This combination of thresholds and conditions constitutes the OWZP.

  • The development of the method and threshold tuning was performed in current atmosphere reanalysis (ERA-Interim) data (Tory et al. 2013a). The method is applied to the climate models without any further adjustment or tuning.

The detection procedure is described in detail in Tory et al. (2013c) and is summarized here in the following steps:

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

  2. Each grid point is assessed to see if it is likely to be part of a circulation that would support TC formation.

  3. The “likely” grid points from step 2 are grouped into individual storms at an instant in time.

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

  5. Each storm track is assessed to see whether it satisfies a set of conditions for a minimum consecutive time period of 48 h. These conditions include a set of thresholds listed in Table 1.

b. Modifications to the OWZP

Three modifications have been introduced to the OWZP since Tory et al. (2013a) was published. A central argument of that paper was that the OWZP was developed independent of the climate models to which it would be applied, and the main goal was to develop a detection system that is globally applicable and independent of model grid resolution. To maintain independence from climate models, any changes to the original OWZP method must be subjected to a rigorous performance assessment. In this paper, the modified OWZPs were applied to the same 20 yr of reanalysis data and each individual detected TC was verified against the same observational dataset used in the original development of the method (Tory et al. 2013a,c). Only changes that provided improved performance were considered for implementation.

The first modification dealt with a dynamical concern after it was discovered that an anticyclonic circulation poleward of 20° could potentially satisfy the OWZ threshold of 50 × 10−6 s−1 (the threshold value used on the 500-hPa level). While it is extremely unlikely that an anticyclonic circulation would ever be detected as a TC using the OWZP methodology, a decision was made to cap the Coriolis contribution to the OWZ [Eq. (1)] to 20° latitude. The details are provided in  appendix A.

The second modification was inspired by suspected unrealistic results when the OWZP-detection method was applied to CMIP5 models, in which a high number of high-latitude detections appeared in some models. Investigations revealed that nearly all of these high-latitude storms could be classified as subtropical systems. As subtropical systems are also of great interest, the investigation is reported in detail in  appendix B, where an objective method for separating tropical from subtropical systems is introduced and the performance of the OWZP in identifying subtropical storms is discussed.

The third modification emerged from a retuning exercise after the subtropical systems were excluded from the analysis in order to focus on tropical circulations alone. This resulted in an increase in the 950-hPa specific humidity threshold (Table 1), with a corresponding small improvement in performance. Implications of this change are discussed in  appendix C.

c. CMIP5 model data

The CMIP5 models used are from several institutions around the globe and are listed in Table 2. The model resolution is listed in the last column and is defined as the number of grid points in the longitudinal and latitudinal direction available in the CMIP5 model output files. Models were chosen based on data availability and ease of data processing. Analysis of additional models is ongoing. The CMIP5 experiments include a set of experiments referred to by Taylor et al. (2012) as long-term (century time scale) integrations with a starting condition based on multicentury preindustrial conditions. The current paper uses the output of two sets of experiments for each model. The first is the set defined by Taylor et al. (2012) as the historical experiment, used here to evaluate the ability of the models to simulate the large-scale structure of TCs in the current climate. The second set is that defined by Taylor et al. as the RCP8.5 experiment. This corresponds to a high atmospheric emissions scenario. It was chosen because the aim here is to determine the robustness of the pre-CMIP5 projections of decreased TC frequency. For a first evaluation, it is sensible to examine the highest emissions scenario to best address the low signal to noise ratio in TC projections, by maximizing the expected signal.

Table 2.

List of all CMIP5 models investigated in this study, their expanded names, their institutions, and their given resolutions. The resolution is defined here as the number of horizontal grid points in the model output files at the CMIP5 distribution centers.

List of all CMIP5 models investigated in this study, their expanded names, their institutions, and their given resolutions. The resolution is defined here as the number of horizontal grid points in the model output files at the CMIP5 distribution centers.
List of all CMIP5 models investigated in this study, their expanded names, their institutions, and their given resolutions. The resolution is defined here as the number of horizontal grid points in the model output files at the CMIP5 distribution centers.

d. Observed TC climatology

The current TC formation climatology for the period 1970–2000 is given in Fig. 1. The TC formation location is specified as the first position of any circulation in the observed TC database (IBTrACS; Knapp et al. 2010) with a 10-min maximum sustained wind speed of at least 17 m s−1 at any 0000 UTC time during its lifetime.1 This definition does exclude some weak TCs in the North Atlantic basin, where the tropical storm wind speed threshold is a 1-min maximum sustained wind speed of 17 m s−1. The IBTrACS dataset incorporates cyclone best-track data from all global forecast offices and warning centers. For those cyclones tracked by more than one warning center, there are alternative versions of the track and TC intensity. In such cases, we use the track and intensity measure from the forecast center with WMO responsibility for that TC basin.

Fig. 1.

Genesis positions of TCs observed in IBTrACS for the period 1970–2000.

Fig. 1.

Genesis positions of TCs observed in IBTrACS for the period 1970–2000.

3. Results: Tropical cyclone projections

The TC formation locations, as determined by the OWZP algorithm for each of the models examined, are shown in Figs. 2 and 3. In each figure, the left panel is for the period 1970–2000 from the “historic experiment” and the right panel is for the period 2070–2100 from the CMIP5 RCP8.5 experiment. Examining current climate in the left panel, the OWZP TC detections in eight models (shown in Fig. 2) reproduce the observed TC climatology reasonably well (within ±50%; Tables 3 and 4). In comparison, the OWZP TC detections in the five models shown in Fig. 3 are significantly less than the observed TC climatology, with the historical climate TC detections ranging from about 10% to 28% of observed TC numbers. Using the OWZP algorithm on the output of CMIP3 models, Tory et al. (2013b) examined the seasonal distribution and longitudinal and latitudinal distributions of cyclone occurrence and found them to be accurate representations of the distributions in nature. The exception was in the North Atlantic basin where TC detections were very low. The North Atlantic low-detection rates have been attributed to the inability of climate models to adequately represent African easterly waves, which provide the majority of TC precursors in the that basin (e.g., Thorncroft and Hodges 2001). The same is generally true for the CMIP5 models in Figs. 2 and 3 where North Atlantic TC detections are low, with the exception of one model, MIROC5. (The mean annual frequency of OWZP-detected TCs in current climate for the 13 models is shown in Table 3.) While the geographic spread of the TC detections bears a close resemblance to climatology for both sets of models (Figs. 2 and 3, left panels), reliability concerns over the very low TC-detection rate led to a subjective decision to eliminate the models featured in Fig. 3 (marked by asterisks in Table 2) from further analysis.

Fig. 2.

OWZP-detected TC formation positions for (left) the late-twentieth-century simulations (1970–2000) and (right) the late-twenty-first-century simulations (2070–2100) using the modified OWZP described in section 2b. The 1970–2000 TC detections in these eight CMIP5 models were within ±50% of the equivalent observed global TC numbers.

Fig. 2.

OWZP-detected TC formation positions for (left) the late-twentieth-century simulations (1970–2000) and (right) the late-twenty-first-century simulations (2070–2100) using the modified OWZP described in section 2b. The 1970–2000 TC detections in these eight CMIP5 models were within ±50% of the equivalent observed global TC numbers.

Fig. 3.

As in Fig. 2, but for the five CMIP5 models with very low numbers of TC detections (approximately 10%–30% of the observed global TC frequency).

Fig. 3.

As in Fig. 2, but for the five CMIP5 models with very low numbers of TC detections (approximately 10%–30% of the observed global TC frequency).

Table 3.

Mean frequency (1970–2000) of TC formation observed (IBTrACS) and for each model (historical) for: globe, hemispheres [Northern Hemisphere (NH) and Southern Hemisphere (SH)], and for basins [south Indian Ocean (SI), west South Pacific (SP), east South Pacific plus South Atlantic (ESPSA), north Indian Ocean (NI), west North Pacific (WNP), east North Pacific (ENP), and North Atlantic (NA)]. Asterisks denote models excluded from the projection analysis because of their very low–detection rates (section 4). The IBTrACS annual TC numbers listed here reflect the exclusion of some weak North Atlantic TCs (section 2d) and the exclusion of storms that formed poleward of a 200-hPa jet (section 2b and  appendix B).

Mean frequency (1970–2000) of TC formation observed (IBTrACS) and for each model (historical) for: globe, hemispheres [Northern Hemisphere (NH) and Southern Hemisphere (SH)], and for basins [south Indian Ocean (SI), west South Pacific (SP), east South Pacific plus South Atlantic (ESPSA), north Indian Ocean (NI), west North Pacific (WNP), east North Pacific (ENP), and North Atlantic (NA)]. Asterisks denote models excluded from the projection analysis because of their very low–detection rates (section 4). The IBTrACS annual TC numbers listed here reflect the exclusion of some weak North Atlantic TCs (section 2d) and the exclusion of storms that formed poleward of a 200-hPa jet (section 2b and appendix B).
Mean frequency (1970–2000) of TC formation observed (IBTrACS) and for each model (historical) for: globe, hemispheres [Northern Hemisphere (NH) and Southern Hemisphere (SH)], and for basins [south Indian Ocean (SI), west South Pacific (SP), east South Pacific plus South Atlantic (ESPSA), north Indian Ocean (NI), west North Pacific (WNP), east North Pacific (ENP), and North Atlantic (NA)]. Asterisks denote models excluded from the projection analysis because of their very low–detection rates (section 4). The IBTrACS annual TC numbers listed here reflect the exclusion of some weak North Atlantic TCs (section 2d) and the exclusion of storms that formed poleward of a 200-hPa jet (section 2b and appendix B).
Table 4.

As in Table 3, but for 2070–2100 with the RCP8.5 scenarios. Statistically significant changes in TC frequency (95% confidence) are indicated for the eight models with reasonable annual TC frequencies by boldface and italic fonts, representing decreasing and increasing TC frequency, respectively.

As in Table 3, but for 2070–2100 with the RCP8.5 scenarios. Statistically significant changes in TC frequency (95% confidence) are indicated for the eight models with reasonable annual TC frequencies by boldface and italic fonts, representing decreasing and increasing TC frequency, respectively.
As in Table 3, but for 2070–2100 with the RCP8.5 scenarios. Statistically significant changes in TC frequency (95% confidence) are indicated for the eight models with reasonable annual TC frequencies by boldface and italic fonts, representing decreasing and increasing TC frequency, respectively.

The first finding of this study then is that, of the 13 models considered, 8 reproduce the OWZP-determined large-scale TC formation conditions at a frequency similar to observed TC counts.

a. Models with detection rates within 50% of the observed TC climatology

The right panels of Fig. 2 show the TC genesis locations for each model over the last 31 yr of the twenty-first century according to the CMIP5 high emission scenario RCP8.5. Visually, it can be seen that the geographic patterns are similar in the future climate and that there is no fundamental change in TC behavior. The actual counts for each cyclone basin and for the two hemispheres for each model are shown in Table 4. The numbers of detections that differ significantly from the historical period are in boldface (significantly less) and italics (significantly more). Here, significance refers to 95% confidence and was calculated using the bootstrap resampling method of (Efron and Tibshirani 1991).

While there is a small increase in higher-latitude TCs in some models, there is no evidence of a systematic poleward shift in TC detections. A global reduction in TC-detection frequency is evident in the results shown in the right panels of Figs. 2 and 3 and is consistent with the data in Table 4 (i.e., the higher frequency of boldface than italic text). This result supports the findings of the previous generations of climate models as summarized by Knutson et al. (2010). Figure 4 quantifies the changes and the distribution by cyclone basin. It shows the percentage change for the globe, for the Northern and Southern Hemispheres, and for each TC basin. Results are shown only for the eight models whose TC-detection climatology is within ±50% of the observed current climate annual TC frequency. All eight models show a global reduction ranging from 28% (GFDL CM3) to 7% (GFDL-ESM2M). In addition, all the models show a reduction for the Southern Hemisphere, and each Southern Hemisphere TC basin. For the Northern Hemisphere, the responses range from a 4% increase in cyclone frequency (GFDL-ESM2M) to a 25% decrease (GFDL CM3). For the four Northern Hemisphere cyclone basins (north Indian Ocean, west North Pacific, east North Pacific, and North Atlantic), there are very large differences in both the sign and the magnitude of the projected changes from model to model. These results are consistent with Villarini and Vecchi (2012), who investigated changing tropical storm (TS) frequency in the North Atlantic basin, using a statistical downscaling technique applied to 17 CMIP5 models. They found that, while there was an increase in TS frequency in the first half of the twenty-first century, the results were of ambiguous sign over the entire century.

Fig. 4.

Percentage change in mean TC frequency between the late-twentieth and late-twenty-first centuries for the CMIP5 models deemed to have reasonable global TC climatology (i.e., within 50% of that observed). Changes that are significant at 95% and 90% confidence levels are indicated by asterisk and plus symbols, respectively.

Fig. 4.

Percentage change in mean TC frequency between the late-twentieth and late-twenty-first centuries for the CMIP5 models deemed to have reasonable global TC climatology (i.e., within 50% of that observed). Changes that are significant at 95% and 90% confidence levels are indicated by asterisk and plus symbols, respectively.

When the statistical significance of the changes is taken into account, the general conclusion of a decreasing TC frequency in a warmer climate is further strengthened. Globally, hemispherically, and in all Southern Hemisphere basins, all statistically significant changes (95% confidence) show decreases in TC frequency. In the Northern Hemisphere basins, the result is not as consistent, with one-third of the statistically significant changes (95% confidence) showing an increase in frequency.

b. Models with very low detection rates

As noted in the introduction, TC detections are a product of the model's ability to generate TC-like circulations and the detectors ability to identify them. It follows that the low numbers of detections in the models featured in Fig. 3 could indicate a problem with the detector. One potential issue is grid-resolution dependence. While the detection system was designed to minimize grid-resolution dependence, it would be difficult to construct a system that was truly independent. Indeed, Table 2 shows that, of the five suspect models (marked with asterisks), none are of high resolution; instead, there is one low-resolution model and four medium-resolution models. However, there are very large differences in TC-detection numbers between some models of very similar resolution. For example, of the medium-resolution models (with 128 or 144 longitudinal grid points), four produce less than 50% of the observed climatology and four produce very realistic annual TC numbers (BCC-CSM1.1 and the three GFDL models). Furthermore, of the three highest-resolution models (with 256 or 288 longitudinal grid points) TC detections in one (MIROC5) are approximately double that of the other two (CNRM-CM5 and CCSM4). These results suggest that, if there is a grid-dependent bias in the OWZP-detection scheme or grid dependency in model circulation development, it is likely to be small compared to other factors. Furthermore, a manual investigation of instantaneous wind plots, including an entire year of FGOALS-g2 data, showed no suggestion of a systematic misidentification of TC-like circulations. Instead, there were very few tropical cyclonic circulations. Investigation into the highly variable numbers of TC detections between models is ongoing.

4. Conclusions

Using the OWZP TC-detection methodology of Tory et al. (2013a,b,c), it has been shown that 8 of the 13 CMIP5 models examined demonstrate the current TC climatology well (to within 50% globally). The OWZP method makes no attempt to identify TC-like circulations. Rather, it detects the large-scale dynamical flow conditions (mainly a deep layer of near-solid-body rotation) within which TCs occur in nature. Additional thermodynamic conditions and a vertical wind shear threshold must also be met for a minimum period of 48 h before a TC is deemed to have formed. Applied to climate models, the OWZP method examines whether these large-scale dynamical and thermodynamic conditions are present in the climate model fields.

Three modifications were made to the OWZP since it was first published in Tory et al. (2013a,b,c). The first put a cap on the Coriolis contribution to the OWZ [Eq. (1)], to eliminate the remote possibility of an anticyclonic circulation satisfying the OWZ thresholds. The second eliminated storms that form poleward of a 200-hPa jet, and the third involved the retuning of the 950-hPa specific humidity threshold after the previous two modifications had been implemented.

TCs that form poleward of a 200-hPa jet are considered to be subtropical storms in this paper. They account for about 2% of TCs we analyzed in the IBTrACS database between 1989 and 2008. Verification of the OWZP-detection system showed that the system does not perform well detecting storms identified as TCs in the IBTrACS database that were found to be subtropical. Because the purpose of the OWZP-detection system is to identify storms that are truly tropical in origin, this result is not problematic. The result does demonstrate that the OWZP-detection system cannot be used to identify subtropical storms, which must therefore be excluded from the analysis.

We conclude that the very low numbers of detections in five of the CMIP5 models is due to the fact that these models rarely simulate the OWZP large-scale flow conditions that Tory et al. (2013a,c) associated with TC formation in ERA-Interim. For the remaining eight models, a reduced TC-detection frequency globally and for the Southern Hemisphere is evident during the late twenty-first century compared with the twentieth-century simulations. This result is synthesized in the summary diagram of Fig. 4. The reduction is fairly consistent between models for the separate Southern Hemisphere TC basins. In contrast, for the Northern Hemisphere cyclone basins there is disagreement between models as to both the magnitude and the sign of changes in TC frequency under anthropogenic climate change. These results are consistent with those from the previous generation of climate models, as summarized by Knutson et al. (2010). In particular, we find the TC frequency projection conclusion from the earlier generation of models, of a decrease in global TC numbers, carries across to the new generation of models.

Acknowledgments

We thank Andrew Dowdy, Charlie Lok, Chis Landsea, and two anonymous reviewers for their valuable insight. We acknowledge the Pacific Australia Climate Change Science and Adaptation Planning program (PACCSAP) project for supporting this work. PACCSAP is funded by AusAID, in collaboration with the Australian Government Department of Climate Change and Energy Efficiency, and delivered by the Australian Bureau of Meteorology and the Commonwealth Scientific and Industrial Research Organisation (CSIRO).

APPENDIX A

OWZP Modification 1: Capping of Coriolis Contribution to OWZ

Since publication of Tory et al. (2013a), we became aware that an anticyclonic circulation poleward of 20° could potentially satisfy the OWZ threshold of 50 × 10−6 s−1 (the threshold value used on the 500-hPa level; Table 1) because the Coriolis contribution to the cyclonic absolute vorticity exceeds this amount at latitudes poleward of about 20°. Furthermore, the Coriolis contribution to OWZ at latitudes poleward of 25° exceeds the threshold value used on the 850-hPa level. Of course, only weak anticyclonic circulations with anticyclonic relative vorticity magnitude less than the difference between the OWZ threshold and the Coriolis value could potentially be identified. To illustrate, note that ranges between 0 and 1, and for simplicity consider a Northern Hemisphere example. From Eq. (1), , and it follows that if a particular threshold TH is satisfied, which can be rearranged to show that the magnitude of the anticyclonic relative vorticity must be greater than the difference between the Coriolis value and the threshold .

While the possibility of an anticyclonic circulation satisfying all thresholds is very unlikely, it was decided to ensure all formation environments have a minimal cyclonic (relative) circulation at 850 hPa, by capping the Coriolis contribution to OWZ at 20° latitude. This change effectively reduced the OWZ values of all circulations poleward of 20°. While the change contributed to a small reduction in detected TCs, it also resulted in a small improvement in the OWZP TC-detection performance due primarily to a reduction in false alarms at higher latitudes (cf. first and second rows in Table 1).

With this modification, the latitudinally varying Coriolis contribution to OWZ is limited to within 20° of the equator. One might ask if a latitudinally varying Coriolis contribution is necessary [i.e., if could be used in place of in Eq. (1) and the thresholds retuned]. We performed a very similar exercise in response to a reviewer's question while preparing Tory et al. (2013c) and concluded that the varying Coriolis contribution is very important within 20° of the equator. The question was whether the Okubo–Weiss parameter [the numerator in Eq. (2)] could be used instead of OWZ in the OWZP. We tested to simplify the retuning process, as it has the same units as OWZ. It turns out that is very similar to a zero-Coriolis OWZ (see Fig. 2 of Tory et al. 2013c), for low-deformation vorticity (i.e., when E and F are not dominant).

When the thresholds were retuned to maximize overall global TC-detection performance the false alarm rate was found to be 11 times that of the miss rate within 10° of the equator, compared with a near one-to-one false alarm/miss rate elsewhere. Because most detections lie within 10°–20° of the equator, the detection scheme was effectively tuned to these latitudes. The same tuning argument applies to the OWZ, which does not suffer from poor performance within 10° of the equator. This result led us to conclude that the latitudinally varying Coriolis contribution is very important at these latitudes. However, because both schemes ( and OWZ) demonstrated similar performance at higher latitudes (when tuned to maximize performance), we concluded that the variable contribution of Coriolis at higher latitudes is not especially important, a result consistent with our performance assessment of modification 1.

APPENDIX B

OWZP Modification 2: Subtropical Storms

The first application of the OWZP to 13 CMIP5 models with and without modification 1 ( appendix A) produced a high number of high-latitude detections (poleward of about 30°) in some models with the highest numbers in the late twenty-first century. This is evident in Figs. B1 and B2, which show the TC formation locations as determined by the OWZP described in Tory et al. (2013a,b,c) for each of the models examined. Most of the high-latitude detections are marked by red stars (described below). The high number of high-latitude detections in six of the models (CNRM-CM5, CCSM4, BCC-CSM1.1, MIROC5, CanESM2, and IPSL-CM5A-MR) was surprising as only small numbers of high-latitude detections were present in the CMIP3 results of Tory et al. (2013b) and the ERA-Interim results of Tory et al. (2013a).

Fig. B1.

As in Fig. 2, but for the OWZP settings described in Tory et al. (2013a). The red stars indicate those storms forming poleward of a 200-hPa jet.

Fig. B1.

As in Fig. 2, but for the OWZP settings described in Tory et al. (2013a). The red stars indicate those storms forming poleward of a 200-hPa jet.

Fig. B2.

As in Fig. B1, but for the five CMIP5 models with very low numbers of TC detections (approximately 10%–30% of the observed global TC frequency).

Fig. B2.

As in Fig. B1, but for the five CMIP5 models with very low numbers of TC detections (approximately 10%–30% of the observed global TC frequency).

A manual analysis of the synoptic environment for a selection of these storms suggested that they had baroclinic origins. There is a long history in the research literature concerning the separation of tropical cyclones from baroclinic systems, and a common term used is “hybrid systems,” which includes certain tropical cyclone characteristics and certain characteristics of cyclones developed through baroclinic instability. Recent major studies on the topic include those of McTaggart-Cowan et al. (2013) and Mauk and Hobgood (2012), which include thorough reviews of the topic. McTaggart-Cowan et al. introduced five categories of large-scale TC formation environment, ranging from nonbaroclinic or tropical through to the most baroclinic referred to as “strong tropical transition” based on varying degrees of baroclinicity. Our analysis of the OWZP performance for each category showed that the OWZP performed well for most TCs deemed to have developed in a baroclinic environment. The result also showed that baroclinic formation environments are not confined to higher latitudes.

Inspection of the synoptic environment of a selection of high-latitude vortices identified by the OWZP algorithm in the future climate runs showed that nearly all formed poleward of a 200-hPa jet. Here, a jet is defined where the 200-hPa wind speed exceeds 25 m s−1 and the zonal wind component exceeds +15 m s−1. More specifically, the tropics/subtropics interface is defined as the jet axis, which is defined as the latitude where, in the poleward direction, the wind speed first begins to decrease in magnitude. Thus in the current study, we have chosen to define tropical cyclones that develop on the poleward side of a 200-hPa subtropical jet stream as subtropical storms. Our reasoning for identifying these systems as baroclinic draws on the synoptic experience of one of the authors (John McBride) over several decades. It is supported by the widely used concept whereby the tropics is defined as the zone of a high tropopause separated from the extratropics by a discontinuity to a lower-altitude tropopause, with the separation being marked by a tropopause fold and the subtropical jet. This concept appears in the literature on tropopause folds (e.g., Keyser and Shapiro, 1986) and on stratosphere–troposphere exchange (e.g., Holton et al. 1995; Stohl et al. 2003) and was also proposed by McBride and Frank (1999; see their Fig. 9 and related discussion). Further support comes from the recent literature on the tropical tropopause layer [TTL; reviewed by Fueglistaler et al. (2009)] and from studies of the expanding tropics in warmer climate: for example, Lucas et al. (2012), who use latitudinal discontinuity in tropopause height to define the edge of the tropics.

As noted above the OWZP performance was assessed for the individual TC categories of McTaggart-Cowan et al. (2013), using the ERA-Interim data with which the OWZP was originally developed and tuned (Tory et al. 2013a). The OWZP performed well for all categories except the strong tropical transition category (category “S”). Not surprisingly, excluding these storms from the performance assessment, with an increase in the 950-hPa specific humidity SH950 threshold,2 led to increased OWZP performance. However, this was not a satisfactory outcome because many correctly detected TCs were also eliminated.

Exclusion of the objectively determined subtropical storms led to a greater performance improvement than excluding the McTaggart-Cowan et al. S storms. When applied to the CMIP5 models, the majority of high-latitude detections disappeared (red stars in Figs. B1 and B2). Interestingly, while the majority of the observed storms we identified as subtropical fitted into the McTaggart-Cowan et al. S category, they accounted for less than half the total number of S storms (i.e., more than half of the storms in the most highly baroclinic category) formed in the tropics.

The OWZP was not designed to identify subtropical storms, and indeed our analysis of these particular storms showed the OWZP performed poorly with a 77% miss rate and 35% false alarm rate. We speculate that the high miss rate is due to a combination of factors including the following: storms developing in sheared or dry environments excluded by the OWZP thresholds (Table 1) and storms developing without sufficient vertical alignment throughout the full 48-h minimum development period.

While Figs. B1 and B2 seem to show an increased frequency and poleward shift of OWZP-determined storms that developed poleward of a 200-hPa jet, it is important not to draw any conclusions from these results about potential changes to subtropical storm behavior in a warmer climate. Instead, it is very possible that the increased numbers of detected subtropical storms in the future climate runs represent a reduced subtropical storm miss rate, as more subtropical storms will satisfy the SH950 threshold because of lower-tropospheric moistening over the warmer oceans.

The preceding discussion argues that the OWZP was not designed to identify subtropical storms and does not perform well in identifying these storms. By defining the tropics to be a dynamic region that lies between two 200-hPa jets, one in each hemisphere, subtropical storms can easily be eliminated from the analysis.

APPENDIX C

OWZP Modification 3: Retuning the 950-hPa Specific Humidity Threshold

The original purpose of the SH950 threshold was to eliminate storms that, while satisfying the relative humidity thresholds, were unlikely to condense sufficient moisture to drive the TC genesis process. This threshold also served the purpose of eliminating the majority of mid- to high-latitude baroclinic circulations.

After removing subtropical storms (i.e., storms that formed poleward of an upper-tropospheric jet;  appendix B) the SH950 threshold became less important, because the majority of otherwise “cooler” detections were eliminated as they appeared on the poleward side of the subtropical jet. A small number of nontropical circulations were still detected at longitudes where the subtropical jet was not well defined. Most of these storms were eliminated by the SH950 thresholds, even when set to modest values (e.g., 10 g kg−1). The global OWZP performance is not particularly sensitive to the SH950 threshold, because only a small fraction of the storms occur at latitudes where the SH950 threshold is limiting (e.g., poleward of 30°). It follows that the major role of the SH950 threshold is to distinguish TCs from non-TCs near the tropics/subtropics interface. Table 1 shows very minimal difference in performance when a threshold of 12.3 g kg−1 was used (as in Tory et al. 2013a,b) compared with no threshold used at all (cf. rows 3 and 5 of Table 1). Optimal performance was found with a threshold of 14 g kg−1, which is the value used in this study.

Projected changes in global TC occurrence in the CMIP5 models are also insensitive to the choice of the SH950 threshold. For example, the projected changes in TC numbers differed by 0.1% when the SH950 threshold was removed in the CCSM4 model (not shown). Detections greater than 30° latitude (high-latitude detections) are more sensitive to SH950 thresholds. While it is beyond the scope of this paper to investigate higher-latitude detections in any detail, because of considerable interest in the question of whether TCs will form further poleward in a warming climate, we make the following points:

  1. Between 1989 and 2008, there were about 2.5 yr−1 observed high-latitude TCs (forming equatorward of the subtropical jet).

  2. Using an SH950 threshold of 14 g kg−1, the OWZP detected this observed number of TCs in both ERA-Interim and in the current climate CCSM4 model.

  3. Removing the SH950 threshold from the CCSM4 model led to about 5 yr−1 high-latitude detections in the current climate.

  4. Regardless of SH950 threshold, the number of high-latitude detections in the CCSM4 model increased by 2.0 yr−1 in the future climate.

The results from CCSM4 suggest there will be a small increase in high-latitude TCs in a future climate, which is independent of the SH950 threshold choice. However, the change in high-latitude TCs, expressed as a percentage of current climate detections, is sensitive to the SH950 threshold choice. While the 14 g kg−1 SH950 threshold gives the most realistic results in ERA-Interim and in the CCSM4 current climate, the threshold will perhaps only be valid in a future climate if the thermodynamic environment near the tropics/subtropics interface does not change significantly (e.g., the current thermodynamic environment expands poleward with any poleward movement of the tropics/subtropics interface).

REFERENCES

REFERENCES
Davidson
,
N. E.
,
G. J.
Holland
,
J. L.
McBride
, and
T. D.
Keenan
,
1990
:
On the formation of AMEX tropical cyclones Irma and Jason
.
Mon. Wea. Rev.
,
118
,
1981
2000
.
Efron
,
B.
, and
R.
Tibshirani
,
1991
:
Statistical data analysis in the computer age
.
Science
,
253
,
390
395
.
Fueglistaler
,
S.
,
A. E.
Dessler
,
T. J.
Dunkerton
,
I.
Folkins
,
Q.
Fu
, and
P. W.
Mote
,
2009
:
Tropical tropopause layer
.
Rev. Geophys.
,
47
,
RG1004
,
doi:10.1029/2008RG000267
.
Holton
,
J. R.
,
P. H.
Haynes
,
M. E.
McIntyre
,
A. R.
Douglass
,
R. B.
Rood
, and
L.
Pfister
,
1995
:
Stratosphere–troposphere exchange
.
Rev. Geophys.
,
33
,
403
440
.
Keyser
,
D.
, and
M. A.
Shapiro
,
1986
:
A review of the structure and dynamics of upper level frontal zones
.
Mon. Wea. Rev.
,
114
,
452
499
.
Knapp
,
K. R.
,
M. C.
Kruk
,
D. H.
Levinson
,
H. J.
Diamond
, and
C. J.
Neumann
,
2010
:
The International Best Track Archive for Climate Stewardship (IBTrACS) unifying tropical cyclone data
.
Bull. Amer. Meteor. Soc.
,
91
,
363
376
.
Knutson
,
T. R.
, and
Coauthors
,
2010
:
Tropical cyclones and climate change
.
Nat. Geosci.
,
3
,
157
163
.
Lucas
,
C.
,
H.
Nguyen
, and
B.
Timbal
,
2012
:
An observational analysis of Southern Hemisphere tropical expansion
.
J. Geophys. Res.
,
117
,
D17112
,
doi:10.1029/2011JD017033
.
Mauk
,
R. G.
, and
J. S.
Hobgood
,
2012
:
Tropical cyclone formation in environments with cool SST and high wind shear over the northeastern Atlantic Ocean
.
Wea. Forecasting
,
27
,
1433
1448
.
McBride
,
J. L.
, and
R.
Zehr
,
1981
:
Observational analysis of tropical cyclone formation. Part II: Comparison of non-developing versus developing systems
.
J. Atmos. Sci.
,
38
,
1132
1151
.
McBride
,
J. L.
, and
W. M.
Frank
,
1999
:
Relationships between stability and monsoon convection
.
J. Atmos. Sci.
,
56
,
24
36
.
McTaggart-Cowan
,
R.
,
T.
Galarneau
Jr.
,
L.
Bosart
,
R.
Moore
, and
O.
Martius
,
2013
:
A global climatology of baroclinically influenced tropical cyclogenesis
.
Mon. Wea. Rev.
,
141
,
1963
1989
.
Stohl
,
A.
, and
Coauthors
,
2003
:
Stratosphere–troposphere exchange: A review, and what we have learned from STACCATO
.
J. Geophys. Res.
,
108
,
8516
,
doi:10.1029/2002JD002490
.
Taylor
,
K. E.
,
R. J.
Stouffer
, and
G. A.
Meehl
,
2012
:
An overview of CMIP5 and the experiment design
.
Bull. Amer. Meteor. Soc.
,
93
,
485
498
.
Thorncroft
,
C.
, and
K.
Hodges
,
2001
:
African easterly wave variability and its relationship to Atlantic tropical cyclone activity
.
J. Climate
,
14
,
1166
1179
.
Tory
,
K. J.
,
S. S.
Chand
,
R. A.
Dare
, and
J. L.
McBride
,
2013a
:
The development and assessment of a model-, grid-, and basin-independent tropical cyclone detection scheme
.
J. Climate
,
26
,
5493
5507
.
Tory
,
K. J.
,
S. S.
Chand
,
R. A.
Dare
, and
J. L.
McBride
,
2013b
:
An assessment of a model-independent tropical cyclone detection procedure in selected CMIP3 global climate models
.
J. Climate
,
26
,
5508
5522
.
Tory
,
K. J.
,
R. A.
Dare
,
N. E.
Davidson
,
J. L.
McBride
, and
S. S.
Chand
,
2013c
:
The importance of low-deformation vorticity in tropical cyclone formation
.
Atmos. Chem. Phys.
,
13
,
2115
2132
.
van Vuuren
,
D. P.
, and
Coauthors
,
2011
:
The representative concentration pathways: An overview
.
Climatic Change
,
109
,
5
31
.
Villarini
,
G.
, and
G. A.
Vecchi
,
2012
:
Twenty-first-century projections of North Atlantic tropical storms from CMIP5 models
.
Nat. Climate Change
,
2
,
604
607
.

Footnotes

1

The OWZP method was developed using data spaced 24 h apart. For valid comparison, the observed data used were also 24 h apart. In this paper we continue to use model data at 24 h intervals, despite CMIP5 data being available at 6-hourly intervals. This decision was made for two reasons: (i) to reduce the very considerable processing time and (ii) to avoid revalidation of the OWZP technique at the higher time frequency. We note that the OWZP is being run in real time at 12-hourly intervals with very promising results, but the results are yet to be objectively verified.

2

The SH950 threshold of 12.3 g kg−1 [used in Tory et al. (2013a,b)] had been lowered from a preliminary value of about 15.0 g kg−1 in order to increase the number of detections of observed TCs in the North Atlantic that we now classify as subtropical. The 12.3 g kg−1 value was the best compromise between detecting as many observed North Atlantic TCs as possible while minimizing the number of false detections at higher latitudes in all basins.