a. Passive microwave imagery

For labeling and documenting SE progressions, we select the 89–92-GHz microwave imagery as the primary source of information. Microwave imagery can observe deep convective features obscured by cirrus in geostationary infrared and visible imagery (Hawkins et al. 2006). Because of their long history of availability, microwave frequencies in the 89–92-GHz range have been typically used to monitor TCs. At these frequencies, ice particles within strong convection scatter terrestrial radiation, resulting in lower measured brightness temperatures (T_b; Aonashi and Ferraro 2020). Since SEs are often associated with deep convection, microwave imagery has often been used to identify SE formations and evolutions (Hawkins et al. 2006; Kuo et al. 2009; Kossin and Sitkowski 2009; Vaughan et al. 2020).

In using the 89–92-GHz range, we are able to capture the general evolution of convection in SEs, since much of this study involves high-intensity TCs associated with deep convective bands. However, there may be TCs that contain shallow convective features with minimal signatures in the 89–92-GHz imagery that would be more apparent in other observing frequencies, like the 37 GHz (e.g., Razin and Bell 2021). The lack of convective signatures in the 89–92-GHz imagery is still notable since they indicate convection that did not intensify and thus have a smaller impact on the TC (i.e., a weaker barrier effect). Since we believe the number of cases with these weaker convective signatures to be small, we reserve a study of these weaker convective signatures using the 37-GHz microwave imagery for the future.

For viewing the 89–92-GHz range, we utilize three conically scanning imagers (GMI, SSMIS, and AMSR2) from low-Earth-orbiting satellites. These imagers are chosen because of their constant and higher spatial resolutions, and dual polarizations. However, a disadvantage of low-Earth orbiting satellites involves inconsistent temporal and spatial coverage of storms. Table 1 summarizes these imagers, providing the associated polarization, and the frequency of each imager used in this study (Aonashi and Ferraro 2020). We discuss the fractional image coverage and temporal coverage in sections 4 and 5, respectively.

Table 1

Three conically scanning dual-polarization microwave imagers between 2016 and 2019 were used for the labeled dataset. All imager frequencies vary from 89.0 to 91.7 GHz with varying grid size resolutions and have horizontal and vertical polarization (Aonashi and Ferraro 2020).

View Table

b. Storm metrics and environmental data

In addition to microwave imagery, TC PRIMED provides storm metrics from the Automated Tropical Cyclone Forecast system (ATCF; Sampson and Schrader 2000) and environmental diagnostics derived from the fifth major global reanalysis produced by ECMWF (ERA5; Hersbach et al. 2020) at synoptic times (Slocum et al. 2022). The source of all data used in this study is TC PRIMED, which compiles the observations from their various original sources. Environmental and storm metrics are linearly interpolated to the time of each microwave overpass to provide analysis postlabeling. Table 2 details the original data sources for the calculated variables in TC PRIMED.

Table 2

Description of the calculated storm and environmental variables used in this study. The source of all data in this study is TC PRIMED, but the source listed here represents the original source used by TC PRIMED to perform these calculations.

View Table

The storm metrics in this study include maximum one-minute sustained winds at 10 m (i.e., intensity), the location of the TC center, storm motion, the radius of 5 kt (1 kt ≈ 0.51 m s⁻¹) winds at 850 hPa (R5)—which represents the general size of a TC and is based solely on infrared imagery and latitude (Knaff et al. 2014), and fractional R5—fraction of R5 over climatological R5 value as a function of intensity (fR5; Knaff et al. 2017). For reference, fR5 provides a size comparison of the TC with the global climatological size of a TC with an equivalent intensity. We evaluate R5 rather than the radius of maximum wind in this study because estimates of the radius of maximum winds from ATCF are not quality controlled, are not reanalyzed following each season (Knaff et al. 2021), and have large biases and uncertainties (Combot et al. 2020). The use of these homogeneous size estimates, e.g., R5 and fR5, is further justified because those are well related to the radius of maximum wind, 34-, 50-, and 64-kt wind radii when the current intensity is known (Knaff et al. 2016, 2017; Chavas and Knaff 2022). Future work could include the evaluation of other size retrieval methods such as the radius of maximum wind. Last, TC maximum wind speeds, position, and storm motion are based on final postseason best-track data (Razin et al. 2023b).

For environmental variables, we use the following azimuthally averaged large-scale diagnostics from TC PRIMED (Slocum et al. 2022): 200–850-hPa deep-layer vertical wind shear at 0–500-km radius, 200–850-hPa generalized shear at 0–500-km radius (Knaff et al. 2005), 200–850-hPa column average meridional wind υ, average 500–700-hPa middle-tropospheric RH and 300–500-hPa upper-tropospheric RH at 200–800-km radius, SST at 0–50-km radius, and thermal wind relation derived temperature advection (700–850 hPa) at 0–500-km radius.

The environmental variables here do not explicitly remove the TC vortex since an azimuthal average around the storm center is taken for the environmental variables. Slocum et al. (2022) demonstrates that calculated kinematic values with an azimuthal average and no vortex removal have low biases when evaluated to SHIPS large-scale diagnostics and dropwindsonde profiles. While some deviations may exist, the low-bias results indicate that the reanalysis used here can resolve nearby TC environments, especially for a global study of SEs and SE progressions.

Last, the storm metrics and environmental variables introduced above are reserved for the quantitative part of this study (sections 6b–7b) and are not used for the labeling (qualitative) discussed in sections 3 and 5.

a. Microwave image processing

We use conically scanning imagers with horizontal and vertical polarization (Table 1) from TC PRIMED since these imagers typically have the highest spatial resolution. Dual polarization allows for the creation of a single polarization-corrected temperature image (Cecil and Chronis 2018), which reduces the impact of low emissivity surface features, allowing TC internal features, such as SEs, to appear more distinct. We also consider how well the image is centered relative to the storm center and any resolution discrepancies between images. As an initial guess, TC PRIMED provides storm-centered imagery interpolated from the final best-track TC centers.

Next, we interpolate the microwave images to a polar grid with a grid spacing of 4 km in radius by 10° in azimuth. Since our dataset includes Northern and Southern Hemisphere storms, we orient the images relative to storm motion, which points to the top of the page, because convection and boundary layer convergence were shown in Shapiro (1983) to be concentrated ahead of storm motion. Shear orientation was not chosen here because most cases examined in this study are storms in low-shear environments (<5 m s⁻¹). Outer rainbands have a down-shear preference, but this trend is not as pronounced for low-shear environments (Corbosiero and Molinari 2002). Regardless, SE formations and ERCs are symmetric phenomena, and the orientation of the image does not impact the concentric determination of an SE, the coverage of the outer eyewall around the inner eyewall, or the results of the environmental quantities. Hence, the results of this study should be minimally impacted by the image orientation choice.

Last, we filter using a variational analysis technique with half-power filter wavelengths of 32 km in radius and 90° in azimuth (Mueller et al. 2006). The variational analysis assists in creating uniformity and minimizing discrepancies between imager resolution and scanning strategies, which ensures consistency in analyst labeling and homogeneity in features engineered for future applications, such as an SE detection machine learning algorithm.

b. Storm selection

Here, we discuss the random selection of seventy-two TCs with a lifetime maximum intensity of hurricane intensity or greater occurring in the North Atlantic, eastern and western North Pacific, and Southern Hemisphere between 2016 and 2019 to form the basis of the SE labels dataset. SEs between 2016 and 2019 are not limited to the storms in this dataset, but the random selection allows for a diverse set of SE formations that spans climatological and annual variabilities without having to document all 204 hurricane-strength TCs globally over that period. Table 3 details the storm distributions of our labeled dataset by season and basin. Note that the SE cases are not equally distributed by year or basin. This is due to the interannual variability of the occurrence of TCs with intensities that are typical for SE formation events, noting that Sitkowski (2012) found that 78% of major hurricanes with intensities above 49 m s⁻¹ formed at least one SE.

Table 3

Distribution of storms between 2016 and 2019 labeled for SEs and eyes sorted by season and basin. Note that this is only a subset of all storms evaluated between 2016 and 2019 and that SE formations are not limited to the storms evaluated in this dataset.

View Table

c. Eye labeling

Using microwave imagery, we define an eye as a region marked by a circular ring of lower T_b (the eyewall) near the storm center. Using this definition, we label each microwave image with a binary “yes” or “no” for both the presence of eyes and SEs. Along with each binary label, we include a subjective confidence level ranging from 1 (least confident yes) to 5 (most confident yes) and −1 (least confident no) to −5 (most confident no). Figure 1 features the subjective labeling process using four examples of the least to most confident cases for the presence of an eye.

View Full Size

Four examples of the eye “present” and eye “not present” labels using microwave imagery. Confidence levels for “no” labels range from −5 (most confident no) to −1 (least confident no) and confidence levels for “yes” labels range from 1 (least confident yes) to 5 (most confident yes). The upward-pointing black arrow indicates the direction of storm motion. Dark-blue and red colors represent lower T_b (ice scattering from convective bands), and the colors from light blue to yellow represent minimal ice scattering and areas outside the most intense convective bands. Shown are (a) a low-confidence eye present label, 1, of Humberto at 2301:14 UTC 14 Sep 2019 (intensity: 25 m s⁻¹); (b) a high-confidence eye present label, 5, of Lorenzo at 0905:26 UTC 26 Sep 2019 (intensity: 51 m s⁻¹); (c) a low-confidence eye not present label, −1, of Mangkhut at 0438:08 UTC 8 Sep 2018 (intensity: 20 m s⁻¹); and (d) a high-confidence eye not present label, −5, of Lorenzo at 0833:02 UTC 2 Oct 2019 (intensity: 39 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

In Fig. 1a, the ice-scattering signals (blue and red colors) feature a partial ring-like feature near the center of the storm. In this case, only moderate scattering (T_b > 200 K) exists near the TC core. Hence, we assign this yes eye with low confidence. In Fig. 1b, a pronounced symmetric ring of ice scattering (T_b < 200 K) surrounds the center of the storm. Therefore, we are highly confident an eye is present in Fig. 1b. In Fig. 1c, a small circular area of ice scattering is present near the center of the storm. However, the area of ice scattering does not extend much beyond 50 km with no other scattering features beyond the storm center. As a result, we believe that there is no eye present, but our confidence is low. In Fig. 1d, we are confident that there is no eye because of the minimal ice scattering present near the center and the lack of banding features within 100 km of the center.

d. Secondary eyewall labeling

Now that we have detailed the eye labeling method, we perform a similar confidence-based labeling method for SEs. First, to define an SE, we evaluate definitions from past works and note their limitations. For example, in Yang et al. (2013), the distance between the inner and outer eyewall sectors must be at least 50 km apart to distinguish between a spiral rainband or concentric eyewall, and the rainband must have a T_b ≤ 230 K, among other requirements. However, labeling SEs is an inherently subjective process, and not all cases may fit perfectly into a single set of requirements. Rigid rules may limit SE cases to certain characteristics that may not yet be known. For example, we find in this study that even clear-cut, high-confidence SEs can vary greatly in T_b since SEs can occur in a wide range of convective vertical depths. Hence, we do not define a T_b threshold.

We define an SE as an outer minimum in T_b that surrounds a defined eye with four or more octants (i.e., coverage ≥ 50%). Coverage ≥ 50% is our only requirement and does not include any other rigid requirements that would be difficult to adhere to. The coverage definition used in this study is consistent with past studies involving microwave imagery. Our definition is more relaxed than that of Kuo et al. (2009) and Kossin and Sitkowski (2009), which require a two-thirds and 75% coverage of the SE around the inner eyewall, respectively. However, we choose a more relaxed definition to allow both incomplete and complete SE formations to be included in our study, with the goal of capturing various SE evolutionary pathways not previously documented. We detail the methods of confidence-based labeling of SEs below.

Given that we do not include as many rigid SE definitions as in past studies, we mitigate subjectivity by using a confidence-based approach involving four experts to label the microwave imagery. The chosen dataset of storms is initially labeled by a single independent expert (“Expert 1”). Next, a second expert (“Expert 2”) independently labels the SE for the same dataset of microwave imagery. The two SE-labeled datasets are then placed into three groups: no SE, low confidence SEs (i.e., 1, 2, and 3), and high confidence SEs (i.e., 4 and 5). Any SE label discrepancies between Experts 1 and 2 that result in different groups are relabeled by two other independent experts (“Expert 3” and “Expert 4”). The resulting confidence level is then the average between Experts 2, 3, and 4. If Experts 1 and 2 agree that a label should result in a low-confidence yes, but disagree on the confidence level, the average value between Experts 1 and 2 is taken. However, if the average confidence level is 1.5, then the confidence from Expert 1 is taken. For discrepancies between Experts 1 and 2 within the high-confidence yes group, the confidence of Expert 1 is taken. For the no group, the confidence level is taken from Expert 1. The purpose of this multi-expert labeling system is to address the subjective nature of labeling SEs and the temporal shortcomings of microwave imagery. The most uncertain cases are denoted using low-confidence values in the SE labels dataset.

Figure 2 provides examples of various SE labels along with their respective confidences. In Fig. 2a, a robust circular inner eyewall (T_b < 200 K) is present around the center of the storm, and an outer band encloses four octants of the inner eyewall. In addition, a moat region, defined as an area with minimal ice scattering, is present between the inner and outer eyewall. However, near the bottom-left quadrant at around 100-km radius, the outer band spirals inwards and connects with the inner eyewall. Unlike spiral rainbands, SEs should be relatively concentric to the inner eyewall (Willoughby et al. 1982). Hence, we can only state with low confidence that there is an SE in this image, using the same confidence-based labeling method as in section 3c. In Fig. 2b, a pronounced circular inner eyewall (T_b < 200 K) with a similarly pronounced concentric SE is present. Since both eyewalls are pronounced and concentric around the entire storm, we are highly confident that an SE is present. Figure 2c shows a spiral band that encloses four octants of the inner rainband. This image is similar to the image in Fig. 2a; however, the moat region between the outer band and the inner eyewall is less distinct than shown in Fig. 2a, and the scattering in the outer band is not as pronounced (T_b > 200 K). As a result, we determine that no SE exists in this image, but with low confidence. Last, Fig. 2d features a circular inner eyewall with no moat region separating the inner eyewall from any outer banded features. Hence, we are highly confident that no SE exists in this image.

View Full Size

Same labeling format as in Fig. 1, but for SEs. Examples are shown of (a) a low-confidence SE “yes” label, 1, of Fengshen at 1638:15 UTC 15 Nov 2019 (intensity: 59 m s⁻¹); (b) a high-confidence SE yes label, 5, of Mangkhut at 0445:14 UTC 14 Sep 2018 (intensity: 75 m s⁻¹); (c) a low-confidence SE “no” label, −1, of Gita at 0506:16 UTC 16 Feb 2018 (intensity: 47 m s⁻¹); and (d) a high-confidence SE no label, −5, of Kammuri at 1724:02 UTC 2 Dec 2019 (intensity: 59 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

a. Progression and stage definitions

We define a progression as the evolution of a rainband from one stage to the next stage. Three stages exist (discussed in the next paragraph) and are centered around an SE formation. A pathway, which is discussed more in-depth in section 6a, is the full evolution of all three stages.

We define the first stage as the entrance stage, which is the first microwave image appearance of an outer convective band that leads to an SE formation. A convective band is defined as a curved spiraling band of lower T_b outside of an eye (as defined in section 3c). The second stage is when an SE is present, which is defined as the time between the first and last appearance of an SE on microwave imagery. The final stage is the exit stage, which is defined as the time of the first microwave image when only one eyewall is clearly visible after the appearance of an SE.

We remind readers that the three stages defined here share similarities and differences from the stages discussed in past works (Sitkowski et al. 2011; Houze et al. 2007; Vaughan et al. 2020). In Sitkowski et al. (2011), the stages are intensity based and the time of SE formation is not explicitly bound to any particular intensity stage. More specifically, the first stage in Sitkowski et al. (2011) is when an outer wind maximum can be detected in aircraft data, and this appears as a spiral rainband in the microwave imagery, which is similar to our defined Entrance. However, we now know that the intensity evolution may be more complex than originally thought (Kossin et al. 2023) and likely less accurate without aircraft reconnaissance. Other works have slightly different variants on the Entrance, as Houze et al. (2007) refers to the Entrance as an outer rainband, and Vaughan et al. (2020) refers to the Entrance as a stationary banding complex.

For this study, we aim to include all types of SE evolutions, regardless of whether the intensity evolution follows the Sitkowski et al. (2011) paradigm. Hence, the defined stages in this study are purely based on the appearance of convective bands and SE formation in the microwave imagery, not intensity or other environmental or storm variables.

b. Imagery considerations

Now that we have discussed the stage-based framework in section 5a, we consider various factors in the data that may impact our analysis. These impacts include the varying temporal resolution between imagers, the weakening of the tropical cyclone due to environmental conditions, and land interactions. In only using GMI, AMSR2, and SSMIS, some large temporal gaps exist in the imagery, with the longest time gap being roughly 34 h. However, the median temporal gap between images is 6 h and is sufficient for evaluating most SE evolutions. For cases with temporal gaps > 10 h, we mark the progression as “unable to be determined” and do not count these cases toward the yes or no replacement count. Since there is uncertainty in regard to the exact time of the stages due to the temporal gaps, we round the time lengths to 6-h intervals.

Also, when a TC is significantly weakening due to environmental conditions or land interactions, those SE cases still count toward the number of SE cases with no replacement but do not count toward any of the other SE exits and are recorded separately as a TC weakening case. TC weakening cases are recorded separately because rapid weakening can sometimes be associated with the disintegration of the TC structure, along with the SE, and are not reflective of typical exit-stage patterns.

Last, the convective band patterns identified and conclusions in the following subsections are purely based on the 89–92-GHz imagery. As discussed in section 2a, imagery of different channels represents different interactions between Earth’s radiation and the atmosphere and may result in different observed convective band patterns. These results may also vary from radar imagery, which has much greater temporal and spatial resolutions. However, the global coverage of microwave imagery motivates the choice of data. Future work to compare results between different channels is needed to fully understand SE progression across different vertical levels.

c. Entrance stage

In this section, we detail the appearance of the first stage (entrance) in microwave imagery. Similar to past studies (Houze et al. 2007; Vaughan et al. 2020), our entrances rely on the axisymmetrization of an outer rainband or stationary banding complex prior to SE formation. The first entrance type we have identified is an “Inner-to-Outer” rainband progression. The top row shows an example of this process in which a TC begins with a spiral band that extends inward toward the inner eyewall (Fig. 4a), and then the outer band detaches from the inner eyewall and becomes partially concentric with the inner eyewall (Fig. 4b). Similarly, in the microwave image example in the bottom row, we observe a spiral outer rainband (blue curve) that extends from the inner eyewall outward to a radius of ∼150 km (Fig. 4a), and then the spiral rainband detaches from the inner eyewall and forms an outer concentric band that partially encircles the inner eyewall 6 h later (Fig. 4b).

View Full Size

An Inner-to-Outer progression of an outer rainband leading to an SE. Shown are (top) a schematic of an ideal progression and (bottom) an example using an original microwave image of Typhoon Maria at (a) 1025:07 UTC 7 Jul 2018 (intensity: 64 m s⁻¹) and (b) 1638:07 UTC 7 Jul 2018 (intensity: 64 m s⁻¹). The black up arrow is the storm motion vector. Narrow black arrows point to locations of inner eyewall, SE, or spiral rainband.

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

Next, the second type of entrance we identified is the “Outer-to-Inner” progression shown in Fig. 5. In the top row, an outer rainband begins as a spiral rainband (Fig. 5a), transitions into a partial but more concentric band around the inner eyewall (Fig. 5b), and becomes concentric and partially wraps around the inner eyewall (Fig. 5c). The band that becomes the outer eyewall can either fully or partially enclose the inner eyewall. In the microwave imagery, Fig. 5 (bottom row), we also observe a spiral rainband that forms outside the TC core between 100 and 250 km to the bottom left of the inner eyewall (Fig. 5a), begins to contract and wrap around the inner core region between 100 and 200 km (Fig. 5b), and axisymmetrizes and partially encloses the inner core (Fig. 5c), becoming an SE between 100 and 150 km.

View Full Size

As in Fig. 4, but for an Outer-to-Inner progression SE entrance: (bottom) microwave image of Typhoon Lekima at (a) 0453:07 UTC (intensity: 45 m s⁻¹), (b) 1502:07 UTC (intensity: 61 m s⁻¹), and (c) 1703:07 UTC 7 Aug 2019 (intensity: 61 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

d. Secondary eyewall appearances

After the entrance stage, the next stage is the SE stage. Two types of SEs exist, “Full Concentric” (FC), which has ∼1.0 SE coverage around the inner eyewall, or “Half Concentric” (HC), which has greater than 0.5 SE coverage around the inner eyewall, but less than that of a FC-SE. The tracking of HC cases here allows for a more in-depth examination of SE characteristics, environmental conditions, and storm metrics, which are discussed in the following sections.

The top row of Fig. 6a is a schematic of an FC-SE, which is a circular SE that fully encloses the inner eyewall. The microwave image example featuring Typhoon Meranti in the bottom row of Fig. 6 also shows a clear circular SE that fully encloses the inner eyewall around 100 km. The top row of Fig. 6b shows a schematic of an HC-SE that partially encloses the inner eyewall. The microwave image example of Typhoon Hagibis in the bottom row shows a narrow partial band (blue color) that encloses the inner eyewall between 150 and 200 km but does not completely enclose the inner eyewall.

View Full Size

Same image format as in Fig. 4, but for the two types of SE appearances: (top) ideal schematic of a Full-Concentric SE (FC-SE) [in (a)] and a Half-Concentric SE (HC-SE) [in (b)] and (bottom) (a) An FC-SE for Typhoon Meranti at 1714:13 UTC 13 Sep 2016 (intensity: 85 m s⁻¹), and (b) an HC-SE for Typhoon Hagibis at 0938:11 UTC 11 Oct 2019 (intensity: 55 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

e. Exit stage

The last stage is the exit stage, which consists of two distinct categories: “No Replacement” or “Replacement.” In the No Replacement category, the inner eyewall does not dissipate during the entire SE formation process and remains the primary eyewall. However, several types of No Replacement exits can occur. The No Replacement exit types are “SE Fade,” which is the fading appearance of the SE on microwave imagery, “SE Merge,” which is the merging of the SE with the inner eyewall while the inner eyewall remains apparent, or “SE Spiral,” which is the loss of concentricity of the outer eyewall, becoming a spiral rainband.

The top row of Fig. 7 is a schematic of an SE Fade. The HC-SE partially encloses the inner eyewall (Fig. 7a), begins to fade (Fig. 7b), and then completely disappears (Fig. 7c). In the bottom row, the microwave image shows the HC-SE of Hurricane Florence between 100 and 200 km (Fig. 7a), and then it begins to fade (Fig. 7b), until the SE disappears about 11 h later (Fig. 7c). A slight remnant rainband remains between 200 and 250 km but is not an SE due to lack of coverage around the inner eyewall. Most importantly, the inner eyewall appearance does not change significantly and remains within 50 km during the SE Fade.

View Full Size

Same image type as in Fig. 4, but for an SE Fade, an SE exit pattern when no eyewall replacement occurs: (bottom) Hurricane Florence at (a) 0938:06 UTC 6 Sep 2018 (intensity: 49 m s⁻¹), (b) 1843:06 UTC 6 Sep 2018 (intensity: 38 m s⁻¹), and (c) 0516:07 UTC 7 Sep 2018 (intensity: 29 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

Figure 8 features an SE Merge, which is another SE exit that does not undergo replacement. In the top row, an SE encloses the inner eyewall (Fig. 8a), contracts toward the inner eyewall (Fig. 8b), and then merges with the primary eyewall (Fig. 8c). Similarly, the microwave image example in the bottom row shows the presence of a concentric eyewall between 50 and 100 km (Fig. 8a; note that this image is slightly off-centered to the bottom left). Next, the concentric eyewall contracts toward the inner eyewall 2 h later (Fig. 8b), and then the concentric eyewall merges with the inner eyewall (Fig. 8c). Throughout the process, the size of the moat decreases steadily until its disappearance in the last image. Note that the initial moat in Fig. 8a is small, but we do not define any minimum moat sizes for SEs and only follow the SE definition provided in section 3d. In this case, an Inner-to-Outer entrance was recorded at 0610:09 UTC 9 October 2019, which is a 10-h time gap before the first appearance of the SE given in Fig. 8a, hence it is possible the merge had begun earlier and a larger moat did exist. In addition, the appearance of the inner eyewall does not change significantly, meaning that the SE merged with the inner eyewall without replacing the inner eyewall.

View Full Size

Same image type as in Fig. 4, but for SE Merge, an SE exit pattern when no eyewall replacement occurs: (bottom) Typhoon Hagibis at (a) 1619:09 UTC 9 Oct 2019 (intensity: 70 m s⁻¹), (b) 1842:09 UTC 9 Oct 2019 (intensity: 69 m s⁻¹), and (c) 0559:10 UTC 10 Oct 2019 (intensity: 67 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

The last no-replacement subcategory is the SE Spiral. In this exit type, the SE reverts to a spiral rainband structure. In the top row, the schematic begins with an HC-SE (Fig. 9a), then the SE becomes noncircular (Fig. 9b), and last the SE becomes a spiral rainband (Fig. 9c). The microwave image example in the bottom row shows an HC-SE between 50 and 150 km (Fig. 9a) that evolves into a spiral rainband structure between 100 and 150 km (Fig. 9b) and then into a comma-shaped pattern connected with the inner core (Fig. 9c). An interesting finding is that after an SE Spiral exit occurs, the spiral rainband can become concentric again, forming another SE (usually HC-SE). These transitions between spiral rainbands and SEs can occur quickly (<12 h) and may not be captured in microwave imagery because of large temporal gaps. In addition to the temporal gap, the determination between spiral rainbands and SEs can be very subjective. Those cases are marked as low-confidence cases, and we refer readers to our confidence-based labeling methods above in section 3d for handling of such cases.

View Full Size

Same image type as in Fig. 4, but for SE Spiral, an SE exit pattern when no eyewall replacement occurs: (bottom) Typhoon Kong-Rey at (a) 0636:30 UTC 30 Sep 2018 (intensity: 34 m s⁻¹), (b) 1002:30 UTC 30 Sep 2018 (intensity: 35 m s⁻¹), and (c) 1658:30 UTC 30 Sep 2018 (intensity: 45 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

The final category in the SE exit stage is when the replacement of the inner eyewall by the SE occurs. In this case, the primary eyewall decays and becomes a relict, similar to ERCs documented in Sitkowski et al. (2011), completing our Classic Path evolution. In the top row, the schematic shows an FC-SE that encloses the inner eyewall (Fig. 10a), then the primary eyewall weakens into a relict eyewall (Fig. 10b), and finally the primary eyewall disappears from the imagery and the SE becomes the new inner eyewall (Fig. 10c). The microwave image example in the bottom row shows that an FC-SE surrounds a primary eyewall (Fig. 10a), then, 19 h later, the inner eyewall decays into a relict eyewall (Fig. 10b), and finally the inner eyewall disappears another 19 h later (Fig. 10c). If the final eyewall after the replacement is larger than the initial eyewall before SE formation, and the patterns shown in Fig. 10 are found, then we document an ERC occurrence with greater certainty. With the SE appearances of each stage discussed, we now can identify relevant SE characteristics and pathways.

View Full Size

Same image type as in Fig. 4, but for eyewall replacement: (bottom) Typhoon Maria at (a) 1013:09 UTC 9 Jul 2018 (intensity: 65 m s⁻¹), (b) 0459:10 UTC 10 Jul 2018 (intensity: 55 m s⁻¹), and (c) 0015:11 UTC 11 Jul 2018 (intensity: 48 m s⁻¹).

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

a. Most common secondary eyewall progressions and pathways

First, we begin by discussing the different SE progressions, which represent the change between the two stages. Since many pathways do not contain full temporal coverage and many different pathway combinations can exist, we identify the most common progressions to find the most common pathways. The pathways dataset consists of 64 and 87 SE pathways for the 4- and 3-threshold datasets, respectively. In total, 13 of 16 (81%; 4 threshold) and 19 of 22 (86%; 3 threshold) Inner-to-Outer cases progressed into an HC-SE, while the remaining cases progressed into an FC-SE. In contrast, 15 of 23 (66%; 4 threshold) and 17 of 31 (55%; 3 threshold) Outer-to-Inner cases progressed into an FC-SE. The increase in threshold favors a higher percentage of Outer-to-Inner cases to progress to FC-SE cases (55% → 66%), while the lower threshold favors the Inner-to-Outer to HC-SE cases progression (81% → 86%). Since the lower threshold subset includes more marginal SE cases, the trend here suggests that marginal SEs are more likely to be associated with an Inner-to-Outer to HC-SE progression.

For exit progressions, 15 of 22 (68%; 4 threshold) and 26 of 34 (76%; 3 threshold) HC cases resulted in no eyewall replacement. However, 16 of 21 (76%; 4 threshold) and 18 of 24 (75%; 3 threshold) FC cases resulted in an eyewall replacement. Here, raising the threshold to 4 reduces the HC dataset from 34 to 22 cases and decreases the percentage of HC to no replacement progressions from 76% to 68%, indicating that HC cases may be associated with more marginal SEs. However, raising the threshold to 4 for the FC cases only reduces the sample from 24 to 21 and only decreases the percentage of FC to replacement progressions from 76% to 75%, which is nearly no change. The small change shows a higher degree of confidence in identifying FC cases for both thresholds. In summary, the HC and FC trends discussed suggest that HC-SE → no replacement is likely associated with more marginal SEs, while FC-SEs to replacement are likely associated with higher confidence SEs.

Now that SE progressions are discussed, we examine common pathways, which feature the entire evolution of an SE from entrance to exit. Based on the SE progression trends, two common SE pathways are formulated, one associated with HC-SEs and one with FC-SEs. The first is an Inner-to-Outer entrance → HC-SE → No Replacement (“Path 1”), and the second is an Outer-to-Inner entrance → FC-SE → Replacement (Classic Path). We refer to the second path as the Classic Path because it is similar to the SE progression proposed in Sitkowski et al. (2011) and leads to an ERC.

We find that Path 1 and the classic ERC pathways occur with similar frequency. For a 4 threshold, there are 5 and 7 instances for Path 1 and Classic Path, respectively. For a 3 threshold, there are 8 and 7 instances for Path 1 and Classic Path, respectively. The counted instances for both paths require sufficient temporal coverage for the entire duration of the pathway, which reduces the number of cases when compared with the two-step evaluation above. Since the number of Classic Path cases remains at 7 cases for both thresholds, but the number of Path 1 cases decreases with a higher threshold, this suggests that marginal SEs are more likely associated with Path 1, while higher confidence SEs are more likely to be associated with the Classic Path. However, the small sample and absolute change are inconclusive and may indicate that these two pathways occur with relatively similar frequency. In addition, this analysis has not been replicated on other random samples to check for robustness, hence future studies to confirm and/or expand the identified SE progressions dataset are needed. Figure 11 summarizes the results of the progressions of the two most common pathways.

View Full Size

Summary of the two most common pathways of SE progressions that highlights the change between two stages. Percentages are based on the percentage of cases that progressed from the prior stage type to the next stage type; hence, the percentages in each column do not add up to 100%. The percentage above and below each arrow represents a 4 and 3 threshold, respectively.

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

Beyond the pathways detailed above, a third pathway exists (Outer-to-Inner → HC-SE → No Replacement). There are 4 and 8 instances (12% and 16% of HC cases) for the 4 and 3 thresholds of the third pathway, respectively. Using a 3 threshold, the third pathway has the same number of occurrences as Path 1, but with a 4 threshold, the third pathway has fewer occurrences. This suggests that the third pathway, similar to Path 1, is also likely associated with marginal SEs. When an Inner-to-Outer rainband develops, an HC-SE is likely to form, but an Outer-to-Inner rainband has only a slightly greater chance of forming an FC-SE than an HC-SE, 66%, and 55%, for the 4 and 3 thresholds, respectively (Fig. 11). However, given that the majority of Outer-to-Inner cases form FC-SEs, we highlight the Classic Path and Path 1 in Fig. 11 and the quantitative comparisons in the following sections.

b. Secondary eyewall characteristics

In addition to the common pathways, certain SE characteristics [e.g., the time of SE occurrence, SE duration as discussed in Trabing and Bell (2021), and R5] distinguish the two common SE pathways. We divide the SE cases by the replacement number, which is the number of replacements in a single TC, and the SE types, such as HC or FC. Next, we test for statistical significance between the groups using the time of SE duration, SE occurrence, and R5.

The SE duration is calculated by the difference in time between the first and last appearance of an SE on microwave imagery. Figures 12a and 12b are histograms of HC- and FC- SE durations for the 4 and 3 thresholds, respectively. The values above each bar represent the median number of microwave images available during each SE duration. The median values in Fig. 12 indicate that long-lasting SEs do contain a larger number of images to support their evaluations and a microwave image is available about every 6–12 h.

View Full Size

Histograms of SE durations for HC (orange) and FC (blue) cases using 12-h bin sizes for (a) 4 threshold and (b) 3 threshold. The SE time lengths are rounded and labeled to the nearest 6 h. For example, 6–12 h would indicate that the SE duration is labeled as either 6 or 12 h, and 18–24 h would indicate an SE duration of either 18 or 24 h. The values above the bars indicate the median number of microwave images within each SE time interval.

Citation: Weather and Forecasting 39, 1; 10.1175/WAF-D-23-0047.1

• Download Figure
• Download figure as PowerPoint slide

In terms of SE durations, we find that SE durations of 6–12 h are the most common, with the majority of HC-SEs lasting between 6 and 36 h. Interestingly, FC-SEs have a larger range of durations than HC-SEs, with an FC-SE from Typhoon Maria (2018) persisting up to 66 h (1752:05 UTC 7 July 2018 to 1152:05 UTC 10 July 2018). The 66-h FC-SE was evaluated using 8 microwave images, with an average temporal gap of 8.25 h, hence it is unlikely additional SE formations occurred within this time period. Interestingly, the change from the 3 to 4 threshold only affects the 12- and 24-h SE durations, especially the 12-h HC-SEs, while long-duration SEs (>24 h) are unaffected. This suggests that marginal SEs generally have shorter SE durations and may exist in environmental conditions that are less conducive to sustaining multiple eyewalls. While not examined in this study, future work to examine whether variations in shortwave heating due to diurnal variations played a role in these SE durations, as detailed in Trabing and Bell (2021), is needed.

Next, we examine the time of occurrence of the SE relative to the TC maximum lifetime intensity. Do SEs occur before or after peak intensity? The time of SE appearance is the number of hours before (negative) or after (positive) the peak TC lifetime intensity. Using time relative to intensity allows for a comparison of when an SE occurs between different storms. The mean and median values at the top section of Table 4 indicate that both HC and FC cases occur near the time of peak intensity, and this holds true for both the 3 and 4 thresholds. However, HC-SEs are more likely to form at the earlier and later stages of a TC life cycle than FC-SEs, since HC-SE cases have a larger standard deviation than FC cases and 25th and 75th percentile values that deviate farther from the time of peak intensity than the FC quartiles. While results in Table 4 represent an initial study on SE time of occurrence, we remind readers to proceed with caution when interpreting the results because of the aforementioned uncertainty resulting from large temporal gaps in the microwave imagery (discussed in section 5). Generally, varying the 3 and 4 thresholds do not significantly change the time of SE occurrences relative to peak intensity.

Table 4

The SE time of occurrence relative to the time of peak intensity for HC- and FC-SEs for the 4 threshold and 3 threshold (top half of the table). Negative and positive times respectively represent SE occurrence before and after the time of peak intensity. Also shown is the duration of SEs sorted by the number of SE replacements per TC for the 4 threshold and 3 threshold (bottom half of the table).

View Table

To investigate SE characteristics, we examine the SE cases that lead to eyewall replacement. In our study, 23 of 43 (53%; 4 threshold) and 26 of 58 (45%; 3 threshold) of known SEs underwent an eyewall replacement, while 5 of 23 (22%; 4 threshold) and 5 of 26 (19%; 3 threshold) of those replacement cases were a second (repeat) replacement in the same TC. The values in the bottom section of Table 4 indicate that second replacements occur longer than first replacements (28 vs 16 h, on average), which is consistent with Sitkowski et al. (2011). In addition, the 75th percentile of the second replacement is much longer than the first replacement (48 vs 24 h). A potential explanation for the longer second replacement duration may be attributed to storm size. R5 is generally larger for the second replacement (Table 5), hence, the radius of the inner eyewall is likely larger (Knaff et al. 2016), and has greater angular momentum, which results in a longer spin-down time. However, the median SE durations for the first or second replacement are both 12 h, but the standard deviation of the second replacement is much longer than the first replacement (25 vs 11 h). Varying the confidence threshold from 3 to 4 does not change the SE durations of first or second eyewall replacement either, but does reduce the sample size of first eyewall replacements from 21 to 18. This suggests that storms that produce a second eyewall replacement are more likely to exist in environmental conditions more conducive to high-confidence SEs. With the small sample and large standard deviation, these relationships are not statistically significant (two-sided Student’s t-test p value = 0.36 and 0.37 for 3 and 4 thresholds, respectively), but indicate a possible trend. We caution the reader that the small sample here does not preclude a confident conclusion, but rather serves as an initial finding. A future expanded SE progressions dataset is needed to conclusively determine the above results.

Table 5

The R5 of SEs sorted by the number of SE replacements per TC for the 4 threshold and 3 threshold.

View Table

Another aspect of SEs is storm size with respect to the replacement number. We reserve discussion of storm size with respect to the different SE types (HC and FC) to section 7a. Table 5 indicates that TCs are generally smaller during their first replacement when compared with the second replacement if evaluated using mean R5 (1562 vs 1591 km and 1524 vs 1591 km for the 3 and 4 thresholds, respectively). In addition, the 25th percentile and median indicate a larger R5 for the second replacement cases. However, the 75th-percentile value indicates that the first replacement is slightly larger than the second replacement (1648 vs 1661 km and 1648 vs 1660 km for the 3 and 4 thresholds, respectively) and the standard deviation for the first replacement is larger than the second replacement (152 vs 187 km and 152 vs 198 km for the 3 and 4 thresholds, respectively). Once again, the small sample leads us to a statistically insignificant result (two-sided Student’s t-test p value = 0.49 and 0.75 for the 3 and 4 thresholds, respectively), but with some indication that repeated replacements are associated with generally larger storms. R5 does not consider the climatological size with respect to intensity, hence to consider this, we also evaluate fR5. We find that the mean fR5 during second replacements is 1.06 for both the 3 and 4 thresholds and 1.00 (3 threshold) and 1.03 (4 threshold) during first replacements. However, given the small sample, fR5 was also statistically insignificant (two-sided Student’s t-test p value = 0.42 and 0.67 for 3 and 4 thresholds, respectively). In summary, FC-SEs generally occur closer to the time of peak intensity and repeat replacements generally have a longer SE duration and are larger in size.

a. Comparing environmental conditions between the two paths

We begin by testing variables for statistical significance between the two groups of entrance categories (Inner-to-Outer vs Outer-to-Inner) for both the 3 and 4 label thresholds. Table 6 details the mean values of each path and the p values at each stage. We find that the Outer-to-Inner entrances (Classic Path) are associated with stronger deep-flow υ (p value = 0.0157 and p value = 0.0403 for the 3 and 4 thresholds, respectively), and greater poleward component of storm motion (hereinafter referred to as poleward motion; p value = 0.0193 and p value = 0.0422 for the 3 and 4 thresholds, respectively) when compared with Inner-to-Outer entrance (Path 1). While all four p values satisfy α = 0.05, the p values for both variables using a 3 threshold are smaller than a 4 threshold. This trend is expected, since the lower 3 threshold includes more marginal SEs and the 4 threshold removes some of the less confident SE results, which suggests an association between weaker deep-flow υ and poleward motion for the marginal cases.

For the SE stage, Table 6 shows variables with statistical significance between HC and FC cases. Midlevel RH, upper-level RH, R5, fR5, and generalized shear are found to be statistically significant (p value < 0.05). Intensity is found to be marginally significant, with a p value < 0.1000 using a 4 threshold. Greater mid and upper-level RH, larger R5, larger fR5, and weaker generalized shear favor FC cases over HC cases. Interestingly, mid and upper-level RH are only significant when using a 3 threshold. Despite the removal of the marginal SEs in the 4 threshold subset, the mean midlevel RH for the Classic Path decreased slightly from 63.9% to 63.8%, while the mean midlevel RH for Path 1 increased from 59.3% to 59.7%. However, the small sample for the 4 threshold, despite the greater absolute difference between the two groups, explains the less significant result, and this also applies to upper-level RH. While not conclusive because of the sensitivity, it appears as though marginal SEs exist in lower RH environments in comparison with high-confidence SEs.

For R5 and fR5, Table 6 shows that these variables are significant for both thresholds, except for the fR5 with a 4 threshold. The R5 values indicate that Classic Path storms, those that have FC-SEs, are on average much larger than storms with a HC-SE. The higher p value for the 4 threshold is due to the removal of the least confident Path 1 SE cases, which increases the Path 1 mean R5 and decreases the sample size. However, one may argue that the larger R5 here is a reflection of the greater intensity within Classic Path storms. Hence, we evaluate fR5, which is the fractional R5 over climatological R5 value as a function of intensity (Table 2). The greater fR5 p values relative to the R5, despite significance using the 3 threshold and marginal significance using the 4 threshold, are unsurprising. This suggests that, while the Classic Path is likely associated with larger R5 and fR5, there is some intensity covariance in the larger storms for the Classic Path. This is supported in the greater intensity values for the Classic Path in comparison with Path 1 in Table 6, but only with marginal significance.

For the last SE-stage variable, stronger generalized-shear is associated with HC cases and is statistically significant for both thresholds. Table 6 shows that the mean Path-1 generalized shear for the 4 threshold is larger than the 3 threshold, which is surprising since we expect the marginal SE cases to have stronger shear. Regardless, both values are significant, and the stronger shear, in combination with lower mid and upper-level RH for HC cases, indicates that potential dry-air entrainment could be partially responsible for the SE asymmetry in HC-SEs.

Last, for the exit stage, Table 6 shows variables tested for statistical significance between No Replacement and Replacement cases. The null hypothesis is only rejected for intensity and temperature advection for the 3 threshold and deep-layer shear for both the 3 and 4 thresholds. On average, greater intensity and weaker deep-layer shear are associated with Replacement cases, and this holds true for both the 3 and 4 thresholds. As previously mentioned, the marginal SE cases removed with the higher 4 threshold result in a higher intensity for the No Replacement (Path 1) cases. However, this trend does not hold true for the mean intensity of the Replacement cases (Classic Path), which decreased from 59.8 to 59.0 m s⁻¹. When the label threshold is switched from 3 to 4, more No Replacement cases are removed than Replacement cases (32 → 20 vs 26 → 23, respectively). The decrease in mean intensity for the Replacement group is due to the removal of 3 cases, so the change is not necessarily notable. On the contrary, the increase in mean intensity for the Path 1 group from a 3 to a 4 threshold is more notable since 12 cases are removed. Hence, we attribute the greater p value for the 4+ threshold intensity to an increase in the Path 1 mean when marginal SE cases with weaker intensities are removed. We believe that the inner eyewall of a stronger TC may require a greater inflow to sustain its intensity, and the barrier effect, summarized in section 1 and discussed in Didlake et al. (2017), may also be stronger in FC-SE cases. An FC-SE has greater SE coverage, and this may more effectively suffocate the inflow toward the inner eyewall, promoting an eyewall replacement.

For deep-layer shear, both the 3 and 4 label thresholds meet the α = 0.05 criteria by a large margin. Unsurprisingly, Classic Path cases are associated with much lower mean shear values than Path 1. One would expect that greater shear would promote dry-air entrainment into the inner core of the TC, weakening the inner eyewall. However, it is also possible that the greater shear in No Replacement cases might cause the SE to be weaker, reducing the barrier effect that weakens the inner eyewall. It is also possible that the shear trend could covary with intensity since the Replacement Cases occur at higher intensity. However, we find this to be unlikely in section 7b, because shear values for Replacement Cases are extraordinarily low, even when compared with storms with stronger intensity.

Last, for temperature advection, we find that the mean Classic Path has a stronger 0- to 500-km radius cold-air advection (850–700 hPa) than Path 1, which is an interesting result. Cold advection may contribute to downward motion in the inner core region near the inner eyewall, weakening the inner eyewall and/or outer eyewall at different magnitudes. The connection between downward motion and inner eyewall weakening is discussed in Houze et al. (2007). In addition, cold-air advection at the 850–700-hPa layer could reduce atmospheric instability, suppressing convection of the inner eyewall. However, this trend is only statistically significant using the 3 threshold, which includes the most marginal SE cases, but the mean cold-air advection is weaker with the marginal SE cases. For a real-world example of the pathways discussed here, refer to the online supplemental material.

In summary, failed ERC attempts occur more commonly in marginal environments (e.g., stronger deep-layer shear) and motivate past SE identification algorithms such as Kossin and Sitkowski (2009) to use environmental variables as predictors.

b. Comparing the two paths with global environmental means

In this section, we compare the two pathways discussed above with the global TC PRIMED mean (hereinafter referred to as global mean). The global mean is evaluated using TCs equatorward of 30° with an intensity greater than 17.5 m s⁻¹. Similar to the analysis in section 7a, we compare the environmental variables with the global mean at the time of the three stages (entrance, SE, and exit) in Table 7.

Table 7

Comparisons of TC PRIMED global mean values for TCs equatorward of 30° with an intensity greater than 17.5 m s⁻¹ with Path 1 and Classic Path for both the 3 and 4 label thresholds; (3+) represents the 3 threshold and (4+) represents the 4 threshold. Variables with statistical significance (p value < 0.05) for either pathway are shown. Boldface values meet the α = 0.05 criterion. The “>” or “<” inside the parentheses indicates whether the mean of the variable in question is greater than or less than the TC PRIMED global mean. Refer to Table 2 for details on each variable.

View Table

We begin by discussing the most common significant variables that span across different times and both label thresholds. The resulting p value for intensity is 0.0000 for all stages at both label thresholds. This indicates that SE formations, regardless of the pathway, are more likely to occur in stronger TCs, which is consistent with past work (Hawkins et al. 2006; Kossin and Sitkowski 2009; Kuo et al. 2009). Next, R5 is significant across all stages and both thresholds relative to the global mean, indicating that larger storms are more likely to form SEs. However, larger storms may be related to SEs occurring in more intense TCs, which is reflected in fR5 when compared with the global mean. fR5 is only significant versus the global mean of 1.00 (by definition) for the Classic Path during the SE stage, indicating some covariance between R5 and intensity.

Next, generalized-shear is significant against the global mean for all stages and thresholds except for Path 1 during the SE stage using the 4 threshold. Similarly, deep-layer shear is significant for all stages and thresholds except Path 1 for the entrance stage using a 4 threshold. While we expected the exclusion of marginal SEs with the higher 4 threshold to reduce the mean shear values, this was not the case for the Entrance or SE stages. This suggests that the more confident set of HC-SEs, which proceeds after an Inner-to-Outer entrance, is in part driven by slightly stronger shear magnitudes that are closer to the global mean. To verify whether the shear values covary with intensity and TC life cycle, we compare the two SE categories using the 4 threshold with the global mean of storms with an intensity between 58 and 70 m s⁻¹. We find that even though the mean intensities of HC-SE and FC-SEs are lower than the global mean of storms between 58 and 70 m s⁻¹, the shear was still weaker during the two SE stages. Hence, it is unlikely the lower shear trend covaries with intensity.

Zonal storm motion is also significant for the entrance and SE stages for both paths and thresholds. The storms in the entrance and SE stages feature more westward storm motion than the global mean. Latitude is likely not the driving factor in the more westward storm motion since the mean latitude for these stages (18°–20°) is near the global mean latitude (18°). Similar to shear, we also verify whether the greater zonal storm motion trend exists when compared with the global mean of storms between 58–70 m s⁻¹ and also 49–58 m s⁻¹ and find that the quicker westward motion trend holds for both comparisons. Hence, we believe that the quicker zonal storm motion (westward) could aid in generating an entrance rainband that leads to an SE, but poleward motion distinguishes between whether an Outer-to-Inner entrance or Inner-to-Outer entrance occurs (section 7a). Since Inner-to-Outer entrances mostly lead to HC-SEs (Fig. 11), the poleward motion might be a critical first step in determining whether a progression will follow Path 1 or the Classic Path. With this in mind, greater westward and poleward motions favor an SE formation and Classic Path entrance, respectively.