The Dynamical Tropopause Location as a Potential Predictor for North Atlantic Tropical Cyclone Activity

1. Introduction

Tropical cyclones (TCs) have the potential to lead to a loss of life and cause massive damage when making landfall. The damage is likely to increase with a larger and wealthier population concentrated in coastal areas (Pielke et al. 2008). TCs are projected to become more destructive (Grinsted et al. 2019) and intense TCs may become more frequent [e.g., Bender et al. (2010)] in a warmer climate.

Vertical wind shear, i.e., the vertical shear of horizontal wind, has long been linked to TC intensity. Increased vertical wind shear has been found to impede intensification by tilting the TC vortex and causing an outward flux of potential vorticity (PV) and equivalent potential temperature, thereby weakening the cyclone (Frank and Ritchie 2001). Further, vertical wind shear aids in ventilating the TC core by entraining relatively dry environmental air into convective updrafts (Emanuel et al. 2004; Tang and Emanuel 2010, 2012) and into the boundary layer (Riemer et al. 2010; Riemer and Laliberté 2015). Entrainment thereby reduces latent heating and buoyancy, and thus hinders convection. Furthermore, entrainment reduces the equivalent potential temperature in the convective updrafts, which leads to outflow at lower altitudes and thereby warmer temperatures. This reduces the intensity of the TC following the heat engine model of Emanuel (1986). Relative humidity at various pressure levels throughout the lower and midtroposphere has been linked to TC intensity and intensification rates, and reduced relative humidity has been found to reduce TC activity (e.g., Kaplan and DeMaria 2003; Emanuel et al. 2004; Hendricks et al. 2010; Wu et al. 2012). The ventilation index of Tang and Emanuel (2012) includes the combined effect of humidity and vertical wind shear. It has been found useful in the reconstruction of the annual cycle of North Atlantic TC activity, in particular the typically sharp increase in activity from July to August (Yang et al. 2021). Further, the ventilation index shows significant correlation with TC frequency and rapid intensification [i.e., an increase in maximum wind speed of at least 15.4 m s⁻¹ in 24 h, see Kaplan and DeMaria (2003)] frequency within numerical modeling frameworks (Tang and Camargo 2014).

The large potential for damage that TCs can cause gives rise to the need for seasonal forecasting of TCs, which is an endeavor that has been undertaken for several decades. In the North Atlantic, the earliest attempts by Gray (1984a,b) used El Niño events, quasi-biennial oscillation phases and sea level pressure anomalies in the Caribbean during spring and early summer to forecast hurricane activity with useful skill. Since then, this approach of statistical forecasting has been continuously developed (e.g., Klotzbach and Gray 2004; Saunders and Lea 2005; Klotzbach 2007). Statistical forecasting generally makes use of the relationship between observable phenomena and environmental factors relevant to TC activity. For example, positive El Niño phases correlate with an increase in vertical wind shear and a decrease in TC activity in the tropical North Atlantic (Aiyyer and Thorncroft 2006).

Dynamical seasonal forecasts, where a numerical model is used to predict the future state of the atmosphere, can also be used to predict TC activity. In a purely dynamical approach, TCs within the forecast are tracked and evaluated directly. Even at low resolutions, climate models can produce TC-like structures, though the simulated TCs are larger and weaker than observed TCs (Manabe et al. 1970; Bengtsson et al. 1982). Still, while individual TCs are unrealistic, they provide useful information for forecasting the geographical and seasonal distribution of TCs (Bengtsson et al. 1995). Dynamical forecasting has since improved greatly, and provides skillful results (e.g., Thorncroft and Pytharoulis 2001; Vitart et al. 2007; Vecchi et al. 2014; Zhang et al. 2019). Dynamical forecasts can also be used in combination with statistical modeling in a hybrid approach, where the large-scale environment produced by the dynamical forecast is used to make statistical forecasts. Operational TC forecasts have generally been found to have good skill in predicting the number of TCs, and some provide useful information on landfall locations and regional activity (Klotzbach et al. 2019).

Forecasts for the 2013 North Atlantic hurricane season were consistently indicating an above average activity in terms of number of named storms, hurricanes and major hurricanes. In reality, there were 14 named storms, which is slightly above average, but only two category-1 hurricanes and no major hurricanes, which is far below average. Zhang et al. (2016) proposed that an increased number of anticyclonic Rossby wave breaking events reduced the humidity in the tropics through mixing with extratropical air and increased vertical wind shear. This prevented cyclones from intensifying despite other environmental factors being conducive to intense cyclones. Zhang et al. (2016) investigated the development of Rossby waves using PV, and drew special attention to the peculiar extreme equatorward position of the 2 potential vorticity unit (PVU; 1 PVU = 10⁻⁶ K kg⁻¹ m² s⁻¹) contour on the 350-K isentropic surface during August. The 2-PVU contour can also be used to identify stratospheric PV streamers, which are elongated filaments that protrude from the high-PV stratosphere into the low-PV troposphere. These, in turn, can be used as a proxy for anticyclonic Rossby wave breaking (Wernli and Sprenger 2007; Béguin et al. 2013; Sprenger et al. 2017).

PV streamers and associated Rossby wave breaking events can be reliably identified (e.g., Bowley et al. 2019; Papin et al. 2020), and Rossby wave breaking has been studied in reanalysis data (e.g., Postel and Hitchman 1999; Scott and Cammas 2002; Abatzoglou and Magnusdottir 2006; Wernli and Sprenger 2007). However, predicting Rossby wave breaking based on numerical models may show biases in frequency and location (Barnes and Hartmann 2012). The location biases are not simply a direct result of biases in the general tropopause location, but of where along the tropopause models produce these events. The bias in Rossby wave breaking frequency is largest in the North Atlantic region, and the bias of the 2-PVU contour (i.e., the tropopause) latitude has been found to be lower than that of the Rossby wave breaking latitude during summer in climate models (Béguin et al. 2013).

The link between TC activity, which is often described in terms of accumulated cyclone energy (ACE) (Bell et al. 2000), and PV streamers as a proxy for Rossby wave breaking has been investigated in several publications. Li et al. (2018) found that Rossby wave breaking events are generally uncorrelated with ACE on a short time scale of 8 days, using their entire data dataset from 1985 to 2013, but also found that they correlate during a number of individual years, which may point toward interannual variability of correlation. On a seasonal time scale, ACE correlates substantially and negatively with Rossby wave breaking frequency and the area of the involved PV streamers (Zhang et al. 2017), and a combined measure which takes into account frequency, size and the magnitude of the anomaly of PV streamers (Papin et al. 2020) on the 350-K isentropic surface. Rossby wave breaking can also enable tropical cyclogenesis (Takemura and Mukougawa 2021) via tropical transition, where a precursor extratropical cyclone develops tropical characteristics, such as a warm core and axisymmetry (Davis and Bosart 2004). McTaggart-Cowan et al. (2013) found that tropical cyclogenesis via tropical transition accounts for over a third of tropical cyclogenesis events in the North Atlantic basin. Despite the potential for cyclogenesis, it appears that the overall effect of an increased Rossby wave breaking frequency is detrimental to TC activity in terms of ACE (Zhang et al. 2017). TCs formed via tropical transition have been found to be less predictable than TCs formed via other genesis pathways (Wang et al. 2018), which suggests that changes in Rossby wave breaking frequency may affect predictability throughout a TC season.

The aim of this study is to assess if the latitude of the 2-PVU contour on multiple isentropic surfaces can potentially act as a predictor for seasonal ACE. The effects on environmental vertical wind shear and midtropospheric relative humidity are assessed and compared to the effects caused by Rossby wave breaking. Furthermore, it is investigated if changes in the position of the 2-PVU contour have an impact on other TC-related metrics concerning number and lifetime.

The following sections are organized as follows. Section 2 describes the methods and data used in this study. Section 3 describes the statistical link between the 2-PVU contour latitude and ACE. Section 4 describes the impact that the 2-PVU contour latitude exerts on environmental variables, and section 5 describes the exerted impact on TC count and lifetime metrics. Section 6 summarizes the results and states the key conclusions.

2. Data and methods

For the purpose of this study, the North Atlantic hurricane season is defined as beginning at 0000 UTC 1 June and ending at 0000 UTC 1 December. Only the North Atlantic basin is considered.

HURDAT2 data are further used to derive other metrics for seasonal activity. The first is the number of individual days per season where a cyclone of hurricane strength (categorized in HURDAT2 as HU) exists within the North Atlantic basin (hurricane days). The second is the number of individual days per season where a cyclone of at least tropical storm strength (categorized in HURDAT2 as TS or HU) exists within the North Atlantic basin (storm days), which thus includes hurricane days. The third is the number of named storms. The fourth is the number of named storms that make landfall, where named storms that make landfall multiple times are counted only once. The fifth is the lifetime of TCs, where the number of entries at synoptic hours and at which the TC is categorized as TD (tropical depression), TS or HU is used. All named terms that contain the word “storm” refer exclusively to TC systems, and not any other storm system, within this study.

Genesis location density is defined by first dividing the North Atlantic basin into squares of 5° side length. HURDAT2 data are then used to obtain the first entry for each cyclone in a set of years, as motivated in the corresponding section. The resulting number of cyclones per square is divided by the number of included cyclones, as this allows for a more meaningful comparison of two distinct sets of years with a different number of total cyclones.

ERA5 reanalysis data (Hersbach et al. 2020) are used to obtain PV, wind and relative humidity data. PV is interpolated to a 0.5° × 0.5° grid and to isentropic surfaces from 345 to 365 K in 5-K intervals. The algorithm used in Sprenger et al. (2017) is used to identify the 2-PVU contour on isentropic surfaces in terms of equidistant points along the contour. The latitude of the 2-PVU contour, here dubbed ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, is then calculated as the mean latitude of these points within a zonal window of 5° half-width at individual longitudes. ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ thus describes the smoothed 2-PVU contour latitude at every longitude on the considered isentropic surfaces.

The PV streamer detection algorithm of Sprenger et al. (2017), which is a refined version of the Wernli and Sprenger (2007) algorithm, is used to obtain PV streamer frequencies with a resolution of 0.5° × 0.5°. Other methods of obtaining PV streamer frequency, in particular the one used in Papin et al. (2020), identify the centroid of the streamer, and obtain from it the geographical distribution of the frequency. In comparison, the method of Sprenger et al. (2017) identifies all points on the specified grid where a PV streamer is present, and not only the centroid position. The frequency is defined as the ratio of time steps at which a PV streamer is present at a given location and the total number of time steps in the considered time interval. PV streamers are used as a proxy for anticyclonic Rossby wave breaking.

Various pressure levels have been used to describe midtropospheric humidity. Within this study, it is quantified by total precipitable water between 850 and 200 hPa in order to not be limited to a single or only few pressure levels. Here 850 and 200 hPa are chosen be consistent with the levels used in Zhang et al. (2016).

The main development region (MDR) is defined as the region from 10° to 20°N and from 20° to 80°W (Goldenberg and Shapiro 1996). The MDR is split into an eastern MDR (EMDR) and a western MDR (WMDR), with the border being at 50°W. Further, a high intensity region (HIR) is defined to represent the region where major hurricanes occur, based on Knapp et al. (2010). The corners of this region are at 30°N, 50°W; 30°N, 100°W; 21°N, 100°W; 10°N, 83°W; and 10°N,50°W, such that it encompasses the Gulf of Mexico and the western tropical North Atlantic basin. Both regions are shown in Fig. 1.

Mean 2-PVU contour in September on the 345–365-K isentropic surfaces using 1980–2017 climatology (solid) and 2013 (dashed) ERA5 data between 100° and 20°W. Gray regions denote the HIR and MDR, with the darker shading indicating the overlap of the two regions.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Mean 2-PVU contour in September on the 345–365-K isentropic surfaces using 1980–2017 climatology (solid) and 2013 (dashed) ERA5 data between 100° and 20°W. Gray regions denote the HIR and MDR, with the darker shading indicating the overlap of the two regions.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

Mean 2-PVU contour in September on the 345–365-K isentropic surfaces using 1980–2017 climatology (solid) and 2013 (dashed) ERA5 data between 100° and 20°W. Gray regions denote the HIR and MDR, with the darker shading indicating the overlap of the two regions.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

3. ${\overline{\mathbf{\Phi }}}_{\mathbf{TP}}$ as a predictor for ACE

This section explores ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ as a simplified predictor for seasonal ACE in place of other PV-related metrics, such as area and frequency of Rossby wave breaking events [e.g., Zhang et al. (2016)], or PV streamer index (Papin et al. 2020). Figure 1 shows the September mean 2-PVU contours for the ERA5 (1981–2017) period and for the year 2013. Previous studies focused heavily on the 350-K isentropic surface, and found that Rossby wave breaking has an effect on ACE on a subseasonal and seasonal time scale (Zhang et al. 2016, 2017; Papin et al. 2020), but not on a time scale of 8 days (Li et al. 2018). Here, it is explored whether ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ calculated for monthly mean PV fields on the 350-K isentropic surface also correlates well with ACE, and whether this correlation is stronger on other isentropic surfaces.

Figure 2 shows the correlation coefficient of monthly mean ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ at longitudes from 90° to 30°W and from June to September with seasonal basin-wide ACE. Only the 350- and 360-K isentropic surfaces are shown, as the correlations below 350 K and above 360 K are weaker.

Linear correlation coefficients of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces with seasonal ACE using ERA5 and HURDAT2 data from 1980 to 2017 from June to September using a 5° half-width.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Linear correlation coefficients of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces with seasonal ACE using ERA5 and HURDAT2 data from 1980 to 2017 from June to September using a 5° half-width.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

Linear correlation coefficients of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces with seasonal ACE using ERA5 and HURDAT2 data from 1980 to 2017 from June to September using a 5° half-width.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

On the 350-K isentropic surface during September there is substantial correlation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with ACE (r = 0.67) with a maximum at 71°W. This is roughly where the climatological TC track density (1979–2007) is at a maximum, and where the most intense cyclones are located (Knapp et al. 2010). Therefore, environmental changes in this region affect many cyclones, and changes in TC intensity can have a higher impact on ACE due to the squared dependence of ACE on wind speed. Further, 71°W is about 5° to the west of where Zhang et al. (2016) identified the highest Rossby wave breaking frequency in August, and substantially farther to the west of where Papin et al. (2020) identified the climatological maximum of PV streamer frequency during September. However, both use the PV streamer centroid to designate the position of the PV streamer and Rossby wave breaking. The centroid position is expected to be farther east than the minimum latitude associated with a PV streamer. As shown in Fig. 3, which is discussed in section 4c, the maximum correlation is also farther to the west of the maximum PV streamer frequency when using the method of Sprenger et al. (2017), which is not as far east as that found in Papin et al. (2020). In the low-activity 2013 season, the mean 2-PVU contour shows an equatorward deflection at around 75°W in September. At this location, the PV streamer frequency has been found to be increased in 2013 (Papin et al. 2020). It thus appears that ACE is influenced by ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, and by the frequency of PV streamers, specifically in a region where PV streamers are not maximally frequent in a climatological sense, and where many potentially intense cyclones occur.

ERA5 1980–2017 climatological PV streamer frequency on the (top) 350- and (bottom) 360-K isentropic surface during (left) August and (right) September.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

ERA5 1980–2017 climatological PV streamer frequency on the (top) 350- and (bottom) 360-K isentropic surface during (left) August and (right) September.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

ERA5 1980–2017 climatological PV streamer frequency on the (top) 350- and (bottom) 360-K isentropic surface during (left) August and (right) September.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

On the 360-K isentropic surface, the correlation coefficient of September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with ACE is largest (r = 0.78) at 74°W. While the correlation coefficient is very close to that on the 350-K isentropic surface, it is stronger than on the 350-K isentropic surface and remains stronger over a broader region. The reason for this difference is explored in the next section.

On the 350-K isentropic surface, June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is substantially correlated with ACE, which is stronger than the correlations of July and August ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with ACE. This is likely not due to a direct effect of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on ACE during June, as only about 1% of ACE is produced in June in a 1980–2017 mean using HURDAT2 data, while about 5% and 24% are produced in July and August, respectively. Further, June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ correlation with June monthly ACE is weaker than with seasonal ACE (not shown). A more plausible explanation is thus that ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is autocorrelated. PV streamers typically do not persist for multiple months, and individual Rossby wave breaking events typically do not correlate with ACE on short time scales (Li et al. 2018). It is therefore assumed that autocorrelation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is due to other underlying features, such as the correlation of Rossby wave breaking events with teleconnection indices described by Bowley et al. (2019).

The ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ autocorrelations are summarized in Table 1. As the question at hand is whether strong correlation of June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with seasonal ACE is due in part to autocorrelation and strong correlation of September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with seasonal ACE, the longitude where the correlation coefficient is maximal is chosen for every month and isentropic surface, as denoted in Table 1.

Table 1.

Autocorrelation coefficients of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces between months at their respective longitude where ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ correlates maximally with seasonal ACE, and said longitude. The listed longitude is the central longitude of the 5° half-width zonal window within which ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is calculated.

View Table

On the 350-K isentropic surface, June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ correlates with September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with a correlation coefficient of r = 0.43. This is similar to the autocorrelation of the PV streamer index in June–July with that in August–November found by Papin et al. (2020). July and August ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ correlate similarly with September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$.

On the 360-K isentropic surface, the autocorrelation coefficient of June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is 0.4, but it should be noted that June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ at 74°W is essentially uncorrelated with September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ at 74°W (r < 0.2). Autocorrelation coefficients of July and August with September are 0.35 and 0.67, respectively. The increased autocorrelation of August with September is reflected in the increased correlation of August ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with seasonal ACE at the western edge of the investigated longitudinal range, while the opposite is observed for July, where a reduction in autocorrelation accompanies a reduction in correlation with seasonal ACE. The substantial correlation of June and July ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with seasonal ACE are thus concluded to be due to autocorrelation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. The correlation of August ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with seasonal ACE may be due to two factors. First, August contributes substantially more to seasonal ACE than June and July, which allows for a direct influence on ACE that is on a relevant scale compared to the full season. Second, the autocorrelation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ from August to September implies that effects of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on ACE during September are inherently linked to August ${\overline{\mathrm{\Phi }}}_{\text{TP}}$.

In summary, September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 360-K isentropic surface could be a potent predictor for seasonal TC activity due to the broad region of strong correlation. Notably, the correlation is slightly stronger than those found by Zhang et al. (2017) and Papin et al. (2020) between several Rossby wave breaking metrics on the 350-K isentropic surface, such as frequency and area, and ACE in the North Atlantic basin. Both Zhang et al. (2017) and Papin et al. (2020) also relied on reanalysis data to come to these findings, but based on 1979–2013 and 1979–2015 data, respectively. Since the correlation is particularly strong in September, ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ as a predictor would be mostly relevant for midseason forecasting, where the lead time is the shortest. June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350-K isentropic surface may be of some use for preseason forecasting of ACE, due to autocorrelation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. However, the correlation with ACE is substantially weaker than that found on the 360-K isentropic surface for September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$.

4. The link between ${\overline{\mathbf{\Phi }}}_{\mathbf{TP}}$ and environmental variables

The mechanisms by which ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ influences ACE are not necessarily the same across isentropic surfaces, and the location where the influence is exerted may also differ. This section shows that vertical wind shear and relative humidity are affected by ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, that ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350-K isentropic surface influences ACE via similar pathways as ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 360-K isentropic surface does, and that the affected regions differ, but partially overlap.

Zhang et al. (2016) argue that Rossby wave breaking leads to a reduction in humidity and an increase in vertical wind shear in the tropical North Atlantic. Multiple linear regression analysis based on 1980–2017 ERA5 data is thus used to assess whether there is a link between September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, 850–200-hPa precipitable water, and 200–850-hPa vertical wind shear. The goal is not to use this analysis to make any direct predictions, but only to argue that a dependence between these quantities exists. September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ at 71°W is used on the 350-K isentropic surface, and September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ at 74°W is used on the 360-K isentropic surface, as this is where the correlations with seasonal ACE are largest (see Table 1). Adjusted R² values, which quantify the variance in the predicted variable that is explained by the variance in the predictors, are referred to as R² values for simplicity. The p value is listed for the R² value and the regression coefficients, which represents the probability that the null hypothesis, i.e., there being no link between the predictors and the explained variable, is true based on the statistical test. For example, a p value of 0.05 would thus signify a 95% confidence that a detected link truly exists, as the probability that the null hypothesis is true is only 5%. HIR, MDR, EMDR, and WMDR mean values of vertical wind shear and precipitable water are considered separately for the linear regression analysis.

a. The HIR and the MDR

Table 2 summarizes the regression results for the HIR and MDR. Using only vertical wind shear as a predictor and using the 350-K isentropic surface, both regions show substantial correlation (R² > 0.4, rows 1 and 2) of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with the mean vertical wind shear in the respective region. Using precipitable water as an additional predictor slightly increases the R² value in the HIR, but has only a very weak effect in the MDR. The p value associated with precipitable water is somewhat low for the HIR (p < 0.1, row 3) and very large for the MDR (p > 0.3, row 4). Changes in ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ thus seem to be linked to changes in precipitable water in the HIR, though not as strongly as to vertical wind shear. In contrast, it cannot be said with confidence that ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is linked to changes in precipitable water in the MDR.

Table 2.

Multiple linear regression results for September mean ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ predicted by September mean vertical wind shear (VWS) and precipitable water (PW), listing the model, regression coefficients (β_n), the p value of the coefficients (p_βn), the adjusted R² value, and the p value of the corresponding F test. Indices for ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ denote the isentropic surface in K, and alphabetical indices for precipitable water and vertical wind shear denote the region within which they are averaged. ERA5 data for 1980–2017 are used.

View Table

The lack of correlation with precipitable water in the MDR differs strongly from the findings of Zhang et al. (2016), who state that humidity changes are linked to Rossby wave breaking events on the 350-K isentropic surface. However, this apparent contradiction can be reconciled. Zhang et al. (2016) use the month of August, and not September, in their analysis. Repeating the multiple linear regression analysis for August reveals that using precipitable water as a sole predictor yields a similar R² value as when using vertical wind shear as a sole predictor (R² = 0.23 and 0.27, respectively). Further, using precipitable water as the sole predictor for September results in a lower R² value (R² = 0.15), indicating that the link between ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ and precipitable water is substantially stronger in August than in September. Precipitable water as the sole predictor is significantly linked to ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ at the 95% level in all of these cases, but still very weak compared to vertical wind shear as the sole predictor in the MDR, as is evident from the R² values.

A plausible reason for the decline of precipitable water as a predictor for ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is that Rossby wave breaking is most active during summer. It peaks in July in the North Atlantic basin (Postel and Hitchman 1999; Abatzoglou and Magnusdottir 2006; Papin et al. 2020), and is responsible for mixing dry air into the tropics (e.g., Zhang et al. 2016). A reduction in these events could lead to less mixing, and thus a reduced impact on precipitable water in the tropical North Atlantic in September as compared to August. Further, there is also a reduction in vertical extent of Rossby wave breaking events from August to September (Abatzoglou and Magnusdottir 2006). As the presence of a PV streamer is not required for a horizontal PV gradient to exert an influence on the wind field, and horizontal PV gradients are weaker in summer than in autumn (Postel and Hitchman 1999), vertical wind shear can thus become the dominant predictor instead of precipitable water. Notably, even though the North Atlantic extratropics are generally drier in boreal autumn than in boreal summer (Gettelman et al. 2006), the humidity change from August to September does not appear large enough to compensate for the reduction in mixing due to reduced PV streamer activity.

Using the 360-K isentropic surface (rows 5–8), vertical wind shear remains the dominant predictor. However, using precipitable water as an additional predictor in the MDR increases the R² value from 0.335 to 0.389, and the p value associated with precipitable as a predictor is quite low with p = 0.014. Precipitable water thus appears to be responsive to changes in ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 360-K isentropic surface in the MDR. Section 4b argues that this response is not present in the entire MDR, and section 4c elucidates the cause for this response.

b. The WMDR and EMDR

Table 3 shows the multiple linear regression results of the WMDR and EMDR subregions. Using the 350-K isentropic surface and vertical wind shear as the sole predictor (rows 1 and 2), the correlation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with vertical wind shear is substantially stronger in the WMDR than in the EMDR. As argued in the previous section, the WMDR is where the correlation coefficient of September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with ACE is maximal, and to the west of where the maximum in PV streamer frequency is located (see Fig. 3, which is discussed in more detail further below). The vertical wind shear response in the WMDR thus seems to be sensitive to westward displacements of PV streamer frequency as was the case in 2013 (Papin et al. 2020), while vertical wind shear in the EMDR appears to be affected less. The reduced effect on the EMDR compared to the WMDR is a result of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350-K isentropic surface being used at 71°W, i.e., where the correlation with ACE is strongest, which is within the WMDR.

Table 3.

As in Table 2, but using the EMDR and WMDR.

View Table

Using precipitable water as an additional predictor in the WMDR reduces the R² value very slightly, and has an associated p value in excess of 0.4. Precipitable water changes due to changes in ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350-K isentropic surface are thus concluded to be negligible.

Using the 360-K isentropic surface and vertical wind shear as the sole predictor, the R² values decrease substantially in both subregions. However, using precipitable water as an additional predictor in the WMDR, the resulting R² value of 0.48 is rather close to those predicting the 350-K isentropic surface in the WMDR (R² = 0.46) and in the entire MDR (R² = 0.49). Therefore, when using the 360-K isentropic surface, both vertical wind shear and precipitable water in the WMDR are important predictors for ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. The reason for this is explored in section 4c.

d. ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles on the 360-K isentropic surface

As 38 years are used within this study (1980–2017), the 9 years with the northernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ and the 9 years with the southernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ positions on the 360-K isentropic surface roughly correspond to, and are from here on referred to, as ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. They are contrasted to highlight the difference between low (southern) and high (northern) values of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. Only the 360-K isentropic surface is used to derive these quartiles, as this is where correlation with seasonal ACE is strongest in September (see Fig. 2), and the specification of the isentropic surface is omitted for brevity.

Figure 4 shows the PV streamer frequency for the southernmost and northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. The southernmost quartile shows a substantially higher PV streamer frequency than the northernmost quartile, though the maximum is located close to the Pacific region. The northernmost quartile shows two comparatively weak maxima within the WMDR. More southern positions of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ thus appear to be associated with enhanced mixing of extratropical and tropical air, though at some distance from the MDR. Figure 5 shows 1980–2017 climatological September relative humidity at 600 hPa and the difference between the northernmost and southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile. The 600-hPa level is chosen as this is the level used for the ventilation index, which depends on the difference between saturated entropy and environmental entropy. For the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile, where PV streamer frequency is reduced but has maxima within the MDR, relative humidity is still increased near the boundary between the WMDR and the EMDR. This indicates that the reduction in PV streamer frequency in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile favors higher relative humidity, and thus TC intensification, in the WMDR more than the westward displacement of a comparatively strong maximum in the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile. On a different note, 600-hPa relative humidity changes substantially and significantly between the equator and 10°N between the quartiles. This implies that ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ could be linked to changes in the intertropical convergence zone (ITCZ), both in position and strength. However, this is not further explored here.

Mean September PV streamer frequency of ERA5 1980–2017 (top) southernmost and (bottom) northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. The MDR is outlined in black.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Mean September PV streamer frequency of ERA5 1980–2017 (top) southernmost and (bottom) northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. The MDR is outlined in black.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

Mean September PV streamer frequency of ERA5 1980–2017 (top) southernmost and (bottom) northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. The MDR is outlined in black.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

(top) Mean September 500-hPa relative humidity of the ERA5 1980–2017 period and (bottom) the difference of the northernmost and southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. Yellow contours denote the 90% (dotted), 95% (dashed), and 99% (solid) confidence levels. The MDR is outlined in black.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

(top) Mean September 500-hPa relative humidity of the ERA5 1980–2017 period and (bottom) the difference of the northernmost and southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. Yellow contours denote the 90% (dotted), 95% (dashed), and 99% (solid) confidence levels. The MDR is outlined in black.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

(top) Mean September 500-hPa relative humidity of the ERA5 1980–2017 period and (bottom) the difference of the northernmost and southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. Yellow contours denote the 90% (dotted), 95% (dashed), and 99% (solid) confidence levels. The MDR is outlined in black.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

Figure 6 shows a September mean 80°–50°W (i.e., the WMDR) zonal mean of zonal wind speed, the 2-PVU contour and the 350- and 360-K isentropic surfaces for the 1980–2017 period and the southernmost and northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. Both in climatology and in the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile, westerly winds are dominant at upper levels in the WMDR. This places the WMDR downstream of the PV streamer frequency maximum at upper levels in the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ case. Even though it is displaced to the west of the WMDR, the region of high PV streamer frequency thus still impacts the WMDR. Further, there is large-scale subsidence at upper and midlevels in the WMDR zonal mean (not shown), which allows mixing by Rossby wave breaking to the west of the WMDR to ultimately affect 600-hPa relative humidity within the WMDR. In contrast, upper-level winds in the WMDR in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile are predominantly easterly. The region where the quartile difference in 600-hPa relative humidity is largest within the MDR is therefore upstream of the maximum in PV streamer frequency for the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ (see Figs. 4 and 5). The northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile thus appears to be more humid because the effect of breaking waves does not easily propagate further into the MDR, but instead propagates toward its western boundary. As relative humidity is increased throughout most of the WMDR in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile, the reduction in PV streamer frequency is concluded to be of higher importance to 600-hPa relative humidity within the WMDR than the relative displacement of high PV streamer frequency regions between quartiles.

(left) ERA5 1980–2017 climatological September mean 80°–50°W zonal-mean zonal wind speed (black contours, from −8 to 8 m s⁻¹, negative contours are dashed), 350- and 360-K potential temperature contours (red), and 2-PVU contour (blue line). (center) As in in the left panel, but for the nine southernmost September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$. (right) As in the left panel, but for the nine northernmost September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$. Blue crosses show 2-PVU contour intersections with potential temperature contours, and the gray area denotes the meridional extent of the MDR.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

(left) ERA5 1980–2017 climatological September mean 80°–50°W zonal-mean zonal wind speed (black contours, from −8 to 8 m s⁻¹, negative contours are dashed), 350- and 360-K potential temperature contours (red), and 2-PVU contour (blue line). (center) As in in the left panel, but for the nine southernmost September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$. (right) As in the left panel, but for the nine northernmost September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$. Blue crosses show 2-PVU contour intersections with potential temperature contours, and the gray area denotes the meridional extent of the MDR.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

(left) ERA5 1980–2017 climatological September mean 80°–50°W zonal-mean zonal wind speed (black contours, from −8 to 8 m s⁻¹, negative contours are dashed), 350- and 360-K potential temperature contours (red), and 2-PVU contour (blue line). (center) As in in the left panel, but for the nine southernmost September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$. (right) As in the left panel, but for the nine northernmost September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$. Blue crosses show 2-PVU contour intersections with potential temperature contours, and the gray area denotes the meridional extent of the MDR.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

The reversal of upper-level wind direction between quartiles also has implications for vertical wind shear. The 850-hPa winds are easterly in the climatological mean as well as in the discussed quartiles. As the 200-hPa winds in the WMDR are mostly easterly in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile, there is no reversal of wind direction with height, and vertical wind shear is thus greatly reduced. Near-surface winds are reduced in magnitude compared to the climatological mean, which also aids in reducing vertical wind shear. The southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile shows an increased magnitude in zonal wind at 200 hPa in the MDR compared to the climatological mean, as well as an increase in the magnitude of near-surface winds. Vertical wind shear is thus increased due to an increase in magnitude at both levels used in the vertical wind shear definition, and the reversal of wind direction with height. The resulting difference in vertical wind shear between the two quartiles is shown in Fig. 7, along with the 1980–2017 climatology. There is indeed a clear reduction in vertical wind shear in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile as compared to the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile. This is in a location where there is a climatological maximum in vertical wind shear in excess of 12 m s⁻¹. A northward displacement of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ thus reduces vertical wind shear throughout the WMDR, and specifically also in a region with climatologically rather pronounced vertical wind shear, which is beneficial to TC intensification. This is consistent with Aiyyer and Thorncroft (2006), who found an above average number of TCs during years where the 200-hPa flow in the MDR, which they defined as being slightly larger than used here, is easterly.

As in Fig. 5, but showing 200–850-hPa vertical wind shear.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

As in Fig. 5, but showing 200–850-hPa vertical wind shear.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

As in Fig. 5, but showing 200–850-hPa vertical wind shear.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

The combined effect of vertical wind shear and relative humidity, as quantified by the ventilation index, is shown in Fig. 8. Climatologically, the ventilation index has a local maximum off the northern coastline of South America, which is where there is also a climatological maximum in vertical wind shear. The climatological ventilation index is largest in the MDR at the northern boundary of the EMDR, where vertical wind shear is high and relative humidity is low. The quartile difference shows a sizeable decrease in ventilation index for the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ compared to the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ off the coastline of South America and throughout a large portion of the WMDR. In this region, vertical wind shear is reduced for the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, as described above, and SST is increased for the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ (not shown), which could cause a reduction in the ventilation index by increasing the potential intensity. The local maximum in the quartile difference in the EMDR, while intriguing, is only significant at the 80% level, which is not deemed sufficient to warrant further investigation here.

As in Fig. 5, but showing ventilation index in the MDR. The 99% confidence level is omitted to make the figure easier to read.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

As in Fig. 5, but showing ventilation index in the MDR. The 99% confidence level is omitted to make the figure easier to read.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

As in Fig. 5, but showing ventilation index in the MDR. The 99% confidence level is omitted to make the figure easier to read.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

The connection between September mean ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, vertical wind shear and midtropospheric humidity is summarized in Fig. 9 in the form of a pseudoschematic. A poleward displacement of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is associated with reduced vertical wind shear and slightly increased relative humidity in favor of increased TC activity, while the opposite is the case for an equatorward displacement. The effect is mainly present in the WMDR.

September mean 2-PVU contour position for 1980–2017 (solid black), the years with the nine most equatorward September mean ${\overline{\mathrm{\Phi }}}_{\text{TP,}360}$ (dashed red) and the nine most poleward September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$ (dashed blue) using ERA5 data and the WMDR (gray region). Arrows indicate poleward (blue) and equatorward (red) displacements of the 2-PVU contours relative to the 1980–2017 climatological 2-PVU contour at 75°W (the vertical dashed black line denotes the 75°W longitude line). WMDR mean 200–850-hPa vertical wind shear and 500-hPa relative humidity are given as text in the figure.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

September mean 2-PVU contour position for 1980–2017 (solid black), the years with the nine most equatorward September mean ${\overline{\mathrm{\Phi }}}_{\text{TP,}360}$ (dashed red) and the nine most poleward September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$ (dashed blue) using ERA5 data and the WMDR (gray region). Arrows indicate poleward (blue) and equatorward (red) displacements of the 2-PVU contours relative to the 1980–2017 climatological 2-PVU contour at 75°W (the vertical dashed black line denotes the 75°W longitude line). WMDR mean 200–850-hPa vertical wind shear and 500-hPa relative humidity are given as text in the figure.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

September mean 2-PVU contour position for 1980–2017 (solid black), the years with the nine most equatorward September mean ${\overline{\mathrm{\Phi }}}_{\text{TP,}360}$ (dashed red) and the nine most poleward September mean ${\overline{\mathrm{\Phi }}}_{\text{TP},360}$ (dashed blue) using ERA5 data and the WMDR (gray region). Arrows indicate poleward (blue) and equatorward (red) displacements of the 2-PVU contours relative to the 1980–2017 climatological 2-PVU contour at 75°W (the vertical dashed black line denotes the 75°W longitude line). WMDR mean 200–850-hPa vertical wind shear and 500-hPa relative humidity are given as text in the figure.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

e. Redundancy of isentropic levels as predictors

The ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is linked to vertical wind shear in the WMDR on both the 350- and 360-K isentropic surfaces, and thus there may be autocorrelation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ between isentropic surfaces. This is substantiated by Abatzoglou and Magnusdottir (2006) and Bowley et al. (2019), which both show that PV streamers can have a substantial vertical extent, and that the horizontal extent on the 360-K isentropic surface can greatly exceed the horizontal extent on the 350-K isentropic surface. However, Abatzoglou and Magnusdottir (2006) also show that during September in the North Atlantic basin, the vertical extent of PV streamers is decreased relative to the vertical extent in summer. Due to their counting method, it becomes clear that there are PV streamers on the 360-K isentropic surface that are not present on the 350-K isentropic surface. As ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is not only determined by PV streamers, it should also be noted that ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces can be controlled by the same underlying feature, such as a tropical upper-tropospheric trough [TUTT, Fitzpatrick et al. (1995)]. This, however, is not necessarily always the case. A quantification of how often an underlying feature controls ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on both isentropic surfaces is not provided here, and speculation concerning such a quantification is intentionally avoided, as it is not deemed necessary for the following analysis. In light of these circumstances, ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces could be at least partially redundant when used to predict the square root of ACE.

Multiple linear regression analysis is again used to assess whether using both isentropic surfaces is superior to the use of only one. The square root of ACE is predicted instead of ACE to de-trend the residuals of the regression. This is necessary to ensure that the regression properly captures the relation between the dependent variable and the predictors. The latitudes where the correlation coefficient of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with seasonal ACE is maximal during September, as denoted in Table 1, are used for this analysis. The results are summarized in Table 4.

Table 4.

As in Table 2, but predicting the square root of ACE with September mean ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic levels, and using HURDAT2 to calculate ACE.

View Table

Using the 360-K isentropic surfaces yields a far larger R² value than using the 350-K isentropic surface (R² = 0.50 versus R² = 0.32). When both isentropic surfaces are used, the R² value only increases very slightly. Further, the p value associated with the 350-K isentropic surface as a predictor is rather large (p > 0.2). There is thus no tangible benefit to using both isentropic surfaces, and the 360-K isentropic surface appears to be the superior predictor.

5. Impact on storm number and lifetime

From the definition of ACE, it follows that not only a reduced intensification via environmental variables can impact ACE, but also a reduction in named storm number or hurricane lifetime. This section assesses whether changes in September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces are linked to changes in named storm number or hurricane lifetime. Further, the number and ratio of TCs that make landfall is assessed, as this is when TCs pose the greatest and most immediate threat.

Table 5 summarizes the correlation coefficients of September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with the number of storm days, hurricane days, and named storms, the average hurricane lifetime, the number of TCs that make landfall, and the ratio of TCs that make landfall to the number of named storms throughout the entire North Atlantic hurricane season. The entire season is considered for these variables instead of only September, as this avoids the need for an arbitrary definition of how to treat systems that cross from one month into another, and because most of the climatological seasonal activity (>85% of ACE) occurs from August to October.

Table 5.

Correlation coefficients of September mean ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces (as per subscript) with storm days (SD), hurricane days (HD), number of named storms (NS), the lifetime of hurricanes (LT_HU), the number of TCs that make landfall (LF), and the ratio of TCs that make landfall to the number of named storms (q_LF). A single asterisk denotes significance at the 95% level, and two asterisks denote significance at the 99% level.

View Table

For all metrics, the difference from one isentropic surface to the other is generally rather small, with the ratio of TCs that make landfall showing the largest increase in correlation coefficient from 0.26 to 0.36. As all coefficients are positive, a more poleward ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is favorable for TC formation and intensification for all considered metrics. Zhang et al. (2016) found that Rossby wave breaking frequency correlates with the North Atlantic hurricane frequency (r = −0.47) and the tropical storm frequency (r = −0.39). The correlation coefficients of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on both the 350- and 360-K isentropic surfaces with the number of named storms are stronger in magnitude, as expected from the stronger correlation of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with ACE. Zhang et al. (2017) found that Rossby wave breaking frequency is correlated with hurricane count (r = −0.67) and the number of named storms (r = −0.48). They further found that Rossby wave breaking frequency correlates strongly with ACE (r = −0.73), which is close to the value found using September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. The correlation with named storms is therefore stronger when using ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ instead of Rossby wave breaking frequency, but this does not appear to affect the correlation with ACE.

The change in correlation coefficient with the ratio of TCs that make landfall is of particular note, as the associated p value decreases from 0.11 on the 350-K isentropic surface to 0.02 on the 360-K isentropic surface. The change in the number of TCs that make landfall is therefore not only a result of there being more named storms when ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is more poleward, but TCs are also more likely to make landfall.

The number of storm days and hurricane days can also be considered for September only, as these metrics do not cross from one month into the next. On both considered isentropic surfaces, the correlation coefficients are reduced slightly. On the 350-K isentropic surface, they are reduced to 0.58 and 0.52 for storm days and hurricane days, respectively, down from 0.67 to 0.60 for the entire North Atlantic hurricane season. On the 360-K isentropic surface, they are similarly reduced to 0.55 and 0.66 for storm days and hurricane days, respectively, from 0.66 to 0.69 for the entire North Atlantic hurricane season. The number of storm days and hurricane days are linked to the intensity of cyclones, which in turn is linked to the response of vertical wind shear and relative humidity patterns to ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. As this response also exists in months other than September, as already mentioned for August, and because ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ is autocorrelated, it is concluded that autocorrelation aids in producing the high correlation between September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ and seasonal mean number of storm days and hurricane days.

The lifetime of hurricanes can be affected by the location of their genesis. The northernmost and southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles on the 360-K isentropic surface years are compared in their respective genesis location distributions. The difference in seasonal genesis location density is shown in Fig. 10 in the top panel. Differences in genesis location density are rather small throughout the EMDR (32% of TCs formed in the EMDR for both quartiles), while the WMDR is more active for more poleward ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ (29% of TCs in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile and 18% of TCs in the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile formed in the WMDR). A more equatorward ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ favors the region north of the WMDR, i.e., cyclogenesis at a more poleward location.

Difference in genesis location density of the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile and the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile for (top) all TCs within the North Atlantic hurricane season and (bottom) TCs formed in September. The region outlined in black is the MDR.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Difference in genesis location density of the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile and the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile for (top) all TCs within the North Atlantic hurricane season and (bottom) TCs formed in September. The region outlined in black is the MDR.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

Difference in genesis location density of the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile and the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile for (top) all TCs within the North Atlantic hurricane season and (bottom) TCs formed in September. The region outlined in black is the MDR.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

The bottom panel of Fig. 10 shows the difference in the equivalent genesis location density using only TCs that were formed in September. The favored cyclogenesis location to the north of the MDR for more equatorward ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years is present in September genesis locations as well. The EMDR is a slightly favored genesis location for more poleward ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ (42% of TCs in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile and 38% of TCs in the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile formed in the EMDR), with the central EMDR in particular being favored in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile. TCs therefore show a tendency to be generated not only closer to the extratropics, but on average also closer to the North American coastline in years with a more equatorward ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. The differences in genesis location density are consistent with a reduced lifetime for a more equatorward ${\overline{\mathrm{\Phi }}}_{\text{TP}}$. While a pattern emerges, it is not exceptionally strong, which is consistent with the comparatively low correlation coefficient of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with hurricane lifetime (see Table 5).

Table 6 shows the metrics used for correlations in Table 5, with the absolute numbers listed for the northernmost and southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartiles. The corresponding tracks are shown in Fig. 11. In total, the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years produced 41 TCs starting in September, and the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years produced 42 TCs starting in September. While the difference seems small, the number of produced hurricanes is 22 and 32, respectively. This is consistent with the forecast failure of 2013 and the analysis of Zhang et al. (2016), which indicated that while TCs are generated in the presence of high Rossby wave breaking activity, their intensification is impeded. Note that Table 6 provides numbers for the entire season, so while the number of TCs in September is similar in both quartiles, the number of TCs throughout the season still differs substantially. For a more northern ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, there is an increase in landfall events particularly in the Gulf of Mexico, but there appears to be a reduction of events along the eastern coast of North America. This is consistent with genesis being favored in the region north of the MDR for the southernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ positions, as TCs generated in that region do not typically enter the Gulf of Mexico, but preferentially make landfall along the eastern coast of North America. The reduction in landfall opportunities is consistent with the lower number of landfall events and the reduced landfall ratio for the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile in Table 6. The northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile shows an increase in cyclogenesis in the EMDR, as also seen in Fig. 10. Although the ratio of TCs that make landfall is increased in the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile, many of the TCs originating in the EMDR recurve without making landfall, which increases their lifetime compared to TCs forming farther west that make landfall, as also reflected in the increased hurricane lifetime in Table 6. It is thus visible that changes in genesis location density affect landfall location and frequency, as well as hurricane lifetime.

Tracks of TCs generated in September for (top) the nine southernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years and (bottom) the nine northernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Tracks of TCs generated in September for (top) the nine southernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years and (bottom) the nine northernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

Tracks of TCs generated in September for (top) the nine southernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years and (bottom) the nine northernmost September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ years.

Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0479.1

• Download Figure
• Download figure as PowerPoint slide

Table 6.

Number of storm days (SD), hurricane days (HD), and named storms (NS), the lifetime of hurricanes (LT_HU, in number of synoptic time steps), the number of TCs that make landfall (LF), and the ratio of TCs that make landfall to the number of named storms (q_LF) for the northernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile (${\overline{\mathrm{\Phi }}}_{\text{TP}}^N$) and the southernmost ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ quartile (${\overline{\mathrm{\Phi }}}_{\text{TP}}^S$).

View Table

6. Summary and conclusions

The metric ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ was introduced to describe the influence of the latitude of the 2-PVU contour on an isentropic surface on TC activity. Using ERA5 and HURDAT2 data, the correlation of monthly mean ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with TC activity as quantified by ACE was found to vary greatly from month to month, and strongly depends on the longitude of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, as well as the isentropic surface. Particularly strong correlations of ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350- and 360-K isentropic surfaces with ACE were found during the month of September in the western North Atlantic basin. Weaker but still substantial correlation was found during the month of June throughout the central North Atlantic basin for ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350-K isentropic surface. It is thus concluded that September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 360-K isentropic surface shows potential as a useful predictor for seasonal ACE in midseason hybrid-approach forecasting (i.e., a combination of statistical and dynamical forecasting). This, however, depends on the capability to predict ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ with a lead time of about 2 months, which is not addressed here. June ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350-K isentropic surface has some potential to be a useful predictor in preseason forecasting, though not to the same extent as September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ for midseason forecasting.

Multiple linear regression analysis was used to assess the link between September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$, 200–850-hPa vertical wind shear, and 850–200-hPa precipitable water in several subregions of the North Atlantic basin. It was found that ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 350-K isentropic surface affects vertical wind shear in the MDR and in the HIR, but that there is no tangible effect on precipitable water in the MDR. ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ on the 360-K isentropic surface affects mainly vertical wind shear in the western MDR, but also has a tangible effect on vertical wind shear in the eastern MDR and the HIR. In the western MDR, there is a substantial effect on precipitable water. It was argued that this change is due to a region of high PV streamer frequency on the 360-K isentropic surface intruding into the western MDR from August to September. It is concluded that these impacts allow September ${\overline{\mathrm{\Phi }}}_{\text{TP}}$ to influence ACE by changing environmental factors relevant to TC intensification.