Cyclone Classification over the South Atlantic Ocean in Centenary Reanalysis

Abstract

Since the beginning of the satellite era, only three tropical cyclones have been recorded over the South Atlantic Ocean. To investigate the potential occurrence of such systems since the 1900s, ERA20C, a centennial reanalysis, was utilised. This study first evaluates the performance of ERA20C in reproducing the climatology of all cyclone types over the southwestern South Atlantic Ocean by comparing it with a modern reanalysis (ERA5) for the period 1979–2010. Despite its simpler construction, ERA20C is able to reproduce key climatological features, such as frequency, location, seasonality, intensity, and thermal structure of cyclones similar to ERA5. Then, the Cyclone Phase Space (CPS) methodology was applied to determine the thermal structure at each time step for every cyclone between 1900 and 2010 in ERA20C. The cyclones were then categorised into different types (extratropical, subtropical, and tropical), and systems exhibiting a warm core at their initial time step were classified as tropical cyclogenesis. Between 1900 and 2010, 96 cases of tropical cyclogenesis were identified over the South Atlantic. Additionally, throughout the lifetime of all cyclones, a total of 1838 time steps exhibited a tropical structure, indicating that cyclones can acquire a warm core at different stages of their lifecycle. The coasts of southeastern and southern sectors of northeast Brazil emerged as the most favourable for cyclones with tropical structures during their lifecycle. The findings of this study highlight the occurrence of tropical cyclones in the South Atlantic prior to the satellite era, providing a foundation for future research into the physical mechanisms that enabled these events.

1. Introduction

Synoptic-scale cyclones are a common atmospheric system over the South Atlantic Ocean [1]. Extratropical cyclones are the most frequent type of these systems [1,2,3,4,5,6], often associated with the passage of cold fronts over the continent and mid-upper-levels waves travelling from the Pacific to the Atlantic Ocean [1]. This type of cyclone typically features a cold, asymmetric core, and stronger winds at higher altitudes. Subtropical cyclones, characterised by a hybrid core (warm at lower levels and cold at higher levels), also occur over the South Atlantic Ocean [7,8,9,10,11,12] and are responsible for strong winds and heavy rainfall along the southern and southeastern coasts of Brazil [8,10,13,14,15]. On the other hand, tropical cyclones are rare over the South Atlantic, with the register of only three systems since the beginning of the satelital era: Catarina (March 2004; [16]), Iba (March 2019; [17]), and Akará (February 2024; [18]). Hurricane Catarina originated from an extratropical cyclone that underwent tropical transition [19], Iba formed as a tropical cyclone and underwent subtropical and later extratropical phases during its lifecycle [17], and Akará was a result of a subtropical to tropical transition [18]. Note that synoptic-scale cyclones can have genesis in one category and keep this for the whole lifecycle or undergo another type, which provides the idea of a continuum [20,21,22,23,24,25,26].

While the genesis processes of extratropical cyclones over the South Atlantic Ocean are well understood (see [1] and references therein), the formation of subtropical and tropical cyclones remains a challenge. Some studies indicate that subtropical cyclones in this ocean basin can have genesis associated with a cold core cutoff low or a short-wave trough travelling at the mid-upper-troposphere, enhancing cyclonic circulation in regions with higher latent heat release at lower levels [7,8,9,10,11]. Low-level disturbance that initiates subtropical cyclones can be associated with horizontal shear induced by another system, typically the passage of a cold front [17,27], which, along with a mid-level low, maintains the cyclonic structure.

Subtropical cyclones can also originate from a warm seclusion of extratropical cyclones that follow the model of [28] extratropical cyclones. In this model, warm air is secluded at low levels in the cyclone center, and this condition helps to intensify the vertical gradient of turbulent heat fluxes and, consequently, diabatic processes. Raoni subtropical cyclone over the South Atlantic Ocean is an example of a subtropical cyclone development from a warm seclusion [29]. In terms of subtropical cyclone climatology in the South Atlantic Ocean [7,8,30], there is an annual average occurrence of ~7 events, most frequently found in southern and southeastern Brazil [8,11].

The occurrence of tropical cyclones in the South Atlantic Ocean is extremely rare due to its prevailing oceanic and atmospheric characteristics: cold sea surface temperatures and strong vertical shear of the horizontal wind at the 200–850 hPa layer [15,31,32]. Consequently, there is no established climatology for tropical cyclones in this region, with only three isolated events recorded. From these, two gained tropical features from transition (Catarina and Akará). For instance, in the Akará’s case, the primary driver of subtropical cyclogenesis was the passage of a cold front that created an environment with near-surface horizontal wind shear, contributing to most of the cyclonic relative vorticity in its genesis. The tropical transition occurred as the 200–850 hPa vertical wind shear weakened, and turbulent heat fluxes from the ocean to the atmosphere increased, enhancing diabatic processes that warmed the atmosphere [17].

Over the South Atlantic Ocean, little is known about the synoptic-scale cyclone transitions. Ref. [33] highlights the difficulty in predicting transitions, given that they are related to the nature of the storm in the initial phase of the cyclone and defining the type of cyclone based on the highest energy processes in it, such as the required amount of convection to differentiate subtropical and tropical cyclones. Ref. [33] also recommends the need for historical analysis of these transitions to obtain more universal criteria for classifying cyclones in different ocean basins. Cyclone classification is important because it can anticipate the weather phenomena associated with the different phases of cyclones (strong winds, heavy rain, sudden drops in temperature, etc.), allowing for the addition of important elements to weather forecasts and anticipating measures to prevent extreme weather events.

So far, tropical occurrences over the South Atlantic Ocean have been observed in the 21st century, which is characterised as the era of widespread use of satellite data for studying meteorological phenomena, particularly because they cover the oceans, where direct observations are less frequent. However, questions remain about the occurrence of tropical events in the past, when there were no adequate resources to observe such systems in more distant coastal areas.

Regarding past events, ref. [34] identified extratropical cyclones in two centenary reanalyses; NCEP-20C [35], and ERA20C [36]. They analysed the whole Southern Hemisphere and only the South Atlantic Ocean. For the whole hemisphere, the NCEP-20C shows a positive and statistically significant trend of the cyclone frequency, while the ERA20C indicates a negative trend of these systems. In ERA20C the frequency of cyclones did not show huge changes with the increase in the number of assimilated data, which can be a response of the near-surface winds assimilated in this reanalysis compared with NCEP-20C, which does not assimilate this variable. Ref. [5] compared an extratropical cyclone climatology in two reanalyses (ERA5 and CFSR/CFSv2) from 1979 to 2019. Over the South Atlantic Ocean, 93.1% of ERA5 cyclones have an identical correspondent storm in CFSR/CFSv2, but the latter presents stronger cyclones than ERA5.

A question that arises is how cyclones can be classified throughout their lifecycle. The most widely used theory for classifying cyclones is the Cyclone Phase Space [20,33]. CPS uses three primary parameters: (a) the thermal asymmetry (B; unit of metres) calculated based on the 900–600 hPa thickness to capture the strength of the cyclone’s frontal nature; (b) thermal wind at the lower troposphere (−|V_T^L|; 900 to 600 hPa); and (c) thermal wind at the upper troposphere (−|V_T^U|; 600 to 300 hPa). Before applying this, cyclones need to be identified and tracked. So, a file with the cyclone trajectory and another with geopotential height at 900, 600, and 300 hPa in grid points are used with the CPS algorithm, which defines the thermal structure of the cyclones in each timestep. One cyclone can present different phases across its lifecycle [20]. Hence, a challenge is to take the decision if a system is classified as a pure extratropical, subtropical, or tropical cyclone and when there are transitions. For this end, a third algorithm is necessary to analyse the CPS output during the lifecycle of each cyclone. Ref. [37] carried out this type of analysis for the cyclones over the South Atlantic Ocean in reanalysis and climate projections obtained from RegCM4 (Regional Climate Model version 4). Their main purpose was to identify tropical transition (TT), i.e., an extratropical or subtropical cyclone that undergoes a tropical phase. The authors evaluated specific criteria after 36 h from cyclogenesis. Systems that meet specific criteria were classified as cases of TT. According to the authors, the number of TT events and their spatial distribution vary significantly across each ensemble member, but on a decadal average, both in the present and future climate scenarios, the global climate models and RegCM4 show approximately 3 TT.

There are few climatological studies that classify cyclone types across different global oceanic basins, such as [38] for the entire globe, [39] for the eastern coast of Australia, [6,37] for the South Atlantic Ocean, and [40] for the Mediterranean Sea. Ref. [39], for instance, classified Australian east coast cyclones based on their thermal structure as warm core, hybrid, and cold core cyclones. They applied the CPS methodology to all time steps of all detected cyclones rather than to a selected subset of events. This approach, which will also be used in the present study, is useful for enhancing the understanding of the dominant thermal structure of cyclones in the various ocean basins of the globe.

Given the low number of tropical cyclones in the southwestern South Atlantic Ocean (hereafter South Atlantic Ocean), it is relevant to question whether events with such characteristics were recorded in periods prior to the satellite era. Advances in surface data assimilation and climate projection model experiments have enabled the development of historical datasets such as ERA20C, a century-long reanalysis, providing key atmospheric variables with relatively refined spatial resolution (~1.25°) over an extended period (1900–2010). In this context, this study has different aims: (a) validate the performance of ERA20C in reproducing the total cyclone climatology (independent of the cyclone type) compared to ERA5, a modern reanalysis; verify if the phases of cyclones are well captured by ERA20C; and identify if in the long term there are other occurrences of tropical cyclones over the South Atlantic Ocean. This study differs from previous ones because it is the first time that CPS methodology is applied to a centenary reanalysis for the South Atlantic Ocean.

2. Methodology

2.1. Study Area and Data

The study area encompasses South America and the South Atlantic Ocean. In Figure 1, the large area outlined by the black square, which also includes the South Pacific Ocean, represents the region used for tracking cyclones. Analyses will be conducted for specific subdomains—Southeast and South Brazil (SEB), Uruguay and extreme south of Brazil (URU), and Argentina (ARG)—as well as for the entire South Atlantic Ocean. Note that the latter (indicated by green lines) includes not only the ocean but also part of the eastern coast of South America (approximately 200 km inland), though it does not encompass the entire subdomains.

Synoptic-scale cyclones were identified and tracked in ERA20C [36] from the European Centre for Medium-Range Weather Forecasts (ECMWF) in the period 1900–2010. This centenar-reanalysis is based on the assimilation of surface pressure measurements from weather stations and wind data at 10 metres above the surface, providing the initial conditions for very short-term numerical forecasts (6 h) that can generate variables throughout the vertical structure of the atmosphere. The ERA20C was developed on a Gaussian grid with a spectral truncation of T159, with 91 vertical levels from the surface to the top of the atmosphere and 4 soil layers. For this study, horizontal wind components, geopotential height, and mean sea level pressure were measured every 6 h for the period analysis in 10 vertical pressure levels (1000, 925, 900, 850, 700, 600, 500, 400, 300, and 200 hPa). The obtained variables were interpolated using a bilinear method to a regular grid with a horizontal resolution of 1.125°.

Data from the ERA5 reanalysis [41] from ECMWF is also used. ERA5 is a more modern reanalysis that assimilates both conventional data from surface and altitude weather stations and indirect observations provided by meteorological satellites [41,42]. For the ERA5, the same 6-hourly ERA20C variables were obtained, but with a horizontal resolution of 0.25° and for the period 1979–2010. Both datasets were interpolated to the same regular latitude by longitude grid (1.5° × 1.5°) to perform the cyclone tracking.

The common period used for cyclone tracking was from 0000 UTC on 1 January 1979 to 1800 UTC on 31 December 2010. The analysis period begins in 1979 because the various reference studies were conducted from that year onwards, and it is limited to 2010, as that is the most recent data available from ERA20C.

2.2. Cyclone Tracking

The algorithm used to identify and track cyclones is based on the cyclonic relative vorticity [3]. Recently, this algorithm has been named S2R-VorTrack [43]. The first part of the algorithm consists of identifying candidate points for cyclonic vorticity minima using the nearest neighbour technique. Once identified, their positions are repeated in the next time step, and a new search is conducted around them to find the position at the time under analysis. Once the system’s displacement speed is known, it is used in the following time steps as a first guess for searching the relative vorticity minimum and giving the track. The algorithm’s output provides the date, latitude, longitude, and relative vorticity throughout the lifecycle of each cyclone.

S2R-VorTrack was applied to relative vorticity at 925 hPa, as in [6], in the area between 11° S–56° S and 97° W–4° E (Figure 1). Only grid points with cyclonic relative vorticity less than −1.0 × 10⁻⁵ s⁻¹ were considered by the algorithm to search the cyclones. In addition, our database includes only cyclones with a minimum lifespan of 24 h. In ERA20C the cyclones were identified from 1900 and in ERA5 from 1979. A common period of analysis is from 0000 UTC on 1 January 1979 to 1800 UTC on 31 December 2010. This is the same period used in previous studies, allowing us to perform comparisons.

With the tracking database, it is possible to compute several statistics such as the cyclogenesis density and trajectory density. Cyclogenesis density corresponds to the first time step of each cyclone that is located in a grid; this grid is divided into small boxes of 3° × 3°, and the number of cyclones in each small box is added, divided by the box area, and finally, divided by the number of years in the study. For the tracking density, the procedure is similar but considers all the time steps of the cyclones [44].

Figure 1 also presents the subdomains of Southeast-South Brazil (SEBrazil), South Brazil-Uruguay (Uruguay), and Argentina (Argentina) defined by [45] that will be used for the analysis in this study.

2.3. Cyclone Phase Space (CPS)

The Cyclone Phase Space (CPS; [20]) was used to identify the thermal structure of cyclones. Briefly, by knowing the position of each cyclone (latitude, longitude) and having the geopotential height at 900, 600, and 300 hPa, three parameters are calculated: thermal symmetry (B) and thermal wind at low ($-\left|V_T^L\right|$) and upper ($-\left|V_T^U\right|$) levels of the atmosphere, within a radius of 500 km from the centre of the cyclone. This radius was defined based on [46], who found that convergence in cyclones occurs between 4° and 6° from the centre of the system. Following [20], B is obtained as:

$$B=h{[{\left(\stackrel{=}{Z_{600\mathrm{hPa}}-Z_{900\mathrm{hPa}}}\right)}_R-{\left(\stackrel{=}{Z_{600\mathrm{hPa}}-Z_{900\mathrm{hPa}}}\right)}_L]}_{500 \mathrm{km}}$$

where Z is geopotential height, R and L indicate, respectively, the right and left sides of storm motion, and the overvar indicates the areal mean over a semicircle of radius 500 km. The integer h assumes the value +1 for the Northern Hemisphere and −1 for the Southern Hemisphere.

Knowing that the magnitude of the horizontal gradient of geopotential height increases (decreases) with height and, consequently, the intensity of the geostrophic wind, in a cold (warm) core cyclone [47], we can easily classify the thermal structure of the cyclones by using the thermal wind definition. Following [20], we can define the magnitude of the horizontal gradient of geopotential height as the difference between the maximum and minimum values of geopotential height on a given isobaric surface between the cyclone centre and the radius of 500 km:

$$\mathsf{\Delta }Z={(Z_{max}-Z_{min})}_p$$

The vertical derivative of ∆Z is equivalent to the thermal wind in a given layer. In the CPS formulation, the layers considered in the thermal wind calculus are 300–600 hPa and 600–900 hPa. Then,

$$-\left|V_T^L\right|={\displaystyle \frac{[\delta \left(Z_{máx}-Z_{min}{)}_{500 \mathrm{km}}\right]}{\delta ln p}}{|}_{900 \mathrm{hPa}}^{600 \mathrm{hPa}}$$

$$-\left|V_T^U\right|={\displaystyle \frac{[\delta \left(Z_{máx}-Z_{min}{)}_{500 \mathrm{km}}\right]}{\delta ln p}}{|}_{600 \mathrm{hPa}}^{300 \mathrm{hPa}}$$

Based on the thresholds of these parameters, one can understand the evolution of a cyclone over time and classify it as tropical, subtropical, or extratropical [38]. Extratropical cyclones, in a broad sense, have B >> 10, which indicates asymmetrical structure in lower levels, and $-\left|V_T^L\right|$ and $-\left|V_T^U\right|$ < 0, indicating that the wind intensity increases with the height increases. On the other hand, tropical cyclones have B < 10, $-\left|V_T^L\right|$ and $-\left|V_T^U\right|$ > 0.

2.4. Criteria Used to Classify Cyclones

The CPS methodology has been widely applied in various oceanic basins around the globe [8,9,38,39], but the thresholds representative of the systems in each place are still a challenge. As recommended by [33], historical analyses of cyclone phases are necessary to obtain more universal classification criteria across different oceanic basins. In this sense, after obtaining the CPS parameters, in the present study we classified the cyclones based on two sets of criteria following Figure 2. For subtropical phases, the thresholds of [39], applied in the South Pacific basin off the eastern coast of Australia, and those of [11], applied to the southwestern South Atlantic basin, were used (Figure 2a). For tropical phases, the thresholds from [37,39] were used, while for extratropical phases, the thresholds were the same as those from [39]. These thresholds (Figure 2) will be referred to as C01 [39], C02 [11], and C03 [34]. Our analysis includes only cyclones over South America and the South Atlantic Ocean (Figure 2).

In order to focus on the tropical systems, two types of analyses, using the criteria shown in Figure 2, were performed: (1) we searched for systems that had their cyclogenesis as tropical (first time step), and (2) for systems that in any time of their lifecycle obtained tropical features.

2.5. Analyses

We aim to know if the ERA20C climatology is able to reproduce the same patterns registered by ERA5. For this end, initially, we compared the climatologies of both datasets in a common period (1979–2010) and focused on: cyclogenesis and tracking density, annual cycle, intensity, and trends. For an analysis of the CPS parameters, we initially show dispersion diagrams of the three CPS parameters considering two periods: 1979–2010 for both ERA20C and ERA5, and 1900–2010 (long period) for ERA20C. Next, an analysis is performed to identify the number of time steps of all cyclones that showed characteristics in each cyclone type (extratropical, subtropical, and tropical); the same was performed for cyclogenesis. Finally, we show ERA20C performance in reproducing Hurricane Catarina (2004) compared to ERA5.

We highlight that the main sources of uncertainty in this study are associated with the tracking algorithm, which may fail to identify some cyclones or erroneously merge cyclone trajectories when they are in close proximity [3]. Additionally, the 500 km radius used to define the cyclone area in the CPS may not accurately represent meso-synoptic scale cyclones.

3. Results and Discussions

3.1. Validation Period: 1979–2010

3.1.1. Climatology, Including All Cyclone Types

Figure 3 shows the mean annual density of cyclogenesis, from 1979 to 2010, in both ERA20C and ERA5, with darker colours indicating regions of higher cyclone frequency. The spatial pattern presented in Figure 3 is quite similar between the ERA20C and ERA5. Both reanalyses highlight the three well-known cyclogenetic regions along the east coast of South America (SEB, URU, and ARG) documented in the literature ([1] and their references). However, ERA5 has more cyclogenesis occurrences over the continent around ~35° S–65° W and along the southern Brazilian coast (32–28° S), while ERA20C shows a more intense cyclogenetic core along the southern coast of ARG. The cyclogenetic core between 30° S and 25° S on the western side of the continent is related to coastal lows (another type of cyclogenesis; [45,48]), and the core between Bolivia and Paraguay represents the main region of thermal lows of South America [49].

The cyclogenesis processes in the different hotspots across the eastern coast of South America are presented in [1]. Here, we briefly summarise the main physical mechanisms. In the SEB region, cyclogenesis is dynamically triggered mainly by near-surface inverted troughs, shortwave troughs, and cutoff lows at mid-upper levels of the atmosphere; horizontal and/or vertical stratospheric intrusion; divergence associated with the upper-level subtropical jet; and the influence of the mid-upper-level semi-stationary trough due to the Andes Mountains. In the URU region, cyclogenesis is primarily driven by mid-upper-level troughs travelling from the South Pacific to the South Atlantic Ocean. In both regions, moisture flux convergence, supported by the continental low-level jet transporting moisture from tropical to subtropical latitudes and by northerly winds on the western flank of the South Atlantic Subtropical Anticyclone, provides a significant thermodynamic contribution to cyclogenesis. In these regions, diabatic processes are also important to enhance cyclone depth, as the heat fluxes from sea to atmosphere. In ARG, the main driver is the baroclinic instability and the regeneration process of some cyclones crossing the Andes from the South Pacific to the South Atlantic Ocean.

Figure 4 shows the annual mean track density for cyclones originating in each of the three cyclogenetic regions along the east coast of South America. The density is calculated by considering all time steps of the cyclones, allowing us to identify the preferential pathways of these systems. For each region, ERA20C reproduces the spatial pattern of track density registered in ERA5, having spatial correlation between 0.98 and 0.99. In SEB and URU, cyclones move to the southeast, while in ARG their trajectory is more zonal. Only in ARG (Figure 4c,f), ERA20C shows cyclones with slightly greater displacement than ERA5.

All the trajectory patterns are consistent with those obtained with ERA-Interim shown by [5,6,43]. It is important to note that ERA20C has a good performance even considering that it only assimilated surface observations (pressure, 10-m wind, and sea surface temperature) and in a lower number than ERA5.

Figure 5 shows the annual cycle of cyclogenesis calculated for each subdomain and the entire South Atlantic region (as indicated by the green box in Figure 1) for the common period 1979–2010. Overall, ERA20C represents both the phase and amplitude of cyclogenesis frequency throughout the year, similar to ERA5, albeit with minor differences. Specifically, ERA20C shows a smoother annual cycle in ARG (Figure 5c) and slightly higher interannual variability (greater standard deviation) in the frequency of cyclogenesis in SEB and URU (Figure 5a,b). For each subdomain, the ERA20C averages are very close to those of ERA5. For example, the annual mean in SEB is 27.7 and 27.8, respectively, in ERA20C and ERA5; ARG the differences increase a little, being 82.3 in ERA20C and 86.5 in ERA5. Therefore, in each subdomain and the South Atlantic, the annual mean of cyclones in ERA20C and ERA5 differs by less than 1%. Consistent with previous studies, both ERA20C and ERA5 identify austral summer as the most cyclogenetic season in SEB (~10 cyclones) and ARG (~21.9 and 23.4 cyclones) and austral winter in URU (15.7 and 16.1 cyclones) and the entire domain (108.1 and 109.6 cyclones). The drivers of these seasonal patterns are discussed in [1].

The frequency distribution of the cyclogenesis intensity (as measured by central relative vorticity) is shown in Figure 6. In each subdomain and the whole South Atlantic, the ERA20C presents a higher number of initially weaker cyclones (category between −1.0 and −2.0 × 10⁻⁵ s⁻¹) than ERA5, which is followed by an underestimation of stronger ones (category lower than −3.0 × 10⁻⁵ s⁻¹).

3.1.2. Trends and Interannual Variability

The time series of the annual number of cyclogenesis in each subdomain are shown in Figure 7, while the slopes and statistical significance of the trends are provided in

Table 1. Cyclogenesis annual trend (cyclones year⁻¹) for the three subdomains and the whole South Atlantic, considering the periods of 1900–2010 and 1979–2010. p-values at a 95% confidence level, using the Mann-Kendall test, are inside parentheses, with statistically significant trends highlighted in bold.

Periods	SEB	URU	ARG	South Atlantic
ERA20C 1900–2010	0.040 (0.006)	0.064 (0.012)	−0.177 (0.000)	−0.291 (0.000)
ERA20C 1979–2010	0.096 (0.414)	−0.089 (0.820)	−0.249 (0.162)	−0.470 (0.314)
ERA5 1979–2010	−0.015 (1.000)	0.103 (0.782)	0.005 (0.871)	−0.075 (0.314)

. The trends in all subdomains and the South Atlantic are statistically significant only in the ERA20C long period (Table 1).

The negative long-term (1900–2010) trend in ERA20C is also identified by [34], who used a different algorithm to track cyclones. In contrast, these authors demonstrated that the NCEP reanalysis family (NCEP20C for the long period, NCEP1, and NCEP2) shows positive and significant trends over the South Atlantic Ocean. According to [34], the positive trend in the NCEP reanalyses is directly related to the increase in near-surface data assimilated over time, allowing a better representation of the cyclone’s frequency. For ERA20C, the authors suggested that the continuous assimilation of near-surface wind over the oceans, which is not completed in NCEP20C, contributes to a better representation of cyclones and a more consistent pattern in cyclone frequency over time.

For short-term (1979–2010) trends, ERA20C and ERA5 present negative and insignificant trends, which aligns with the results of both ERA-Interim and ERA5 obtained by [34].

Considering the subdomains, ERA20C long period shows positive trends in SEB and URU and negative ones in ARG. For the short period, only SEB presented a positive trend, which is not statistically significant. It is interesting to notice that ERA5 has opposite trend signals to ERA20C in the short period: it shows a negative trend in SEB and positive ones in URU and ARG.

For the period 1979–2010, the interannual variability of cyclone frequency shows greater discrepancies between ERA20C and ERA5 in the SEB subdomain (Figure 7a), as indicated by the low correlation coefficient (0.28). In contrast, there is stronger agreement between the reanalyses for ARG and the South Atlantic, with correlation coefficients of 0.68 and 0.79, respectively (Figure 7c,d). The interannual variability of cyclones in URU presents a moderate correlation of 0.43 (Figure 7b).

Despite some differences, previous analyses have shown that ERA20C performs well in representing cyclone climatology (all types combined) in the South Atlantic Ocean, making it suitable for identifying cyclones’ phases since 1900.

3.2. Cyclone’s Classification

3.2.1. Dispersion Analysis of the CPS Parameters

Figure 8 presents the dispersion diagrams of the three CPS parameters (left side B vs. $-\left|V_T^L\right|$ and right side $-\left|V_T^U\right|$ vs. $-\left|V_T^L\right|$) computed for all time steps of the tracked cyclones across the South Atlantic, while Figure 9 is only for the time step of cyclogenesis. In both figures, the lines represent the reanalyses: ERA20C long and short periods and ERA5. The dispersion of the CPS parameters are similar in pattern in both reanalyses, even for the long period of ERA20C (Figure 8 and Figure 9). For B vs. $-\left|V_T^L\right|$, most parameters are concentrated in the first quadrant (Figure 8 and Figure 9a,c,d) and for $-\left|V_T^U\right|$ vs. $-\left|V_T^L\right|$, they concentrate in the fourth quadrant (Figure 8 and Figure 9b,d,f). It means that most of the time steps, the systems have asymmetric and deep cold core characteristics (B > 10, $-\left|V_T^L\right|$ < 0, and $-\left|V_T^U\right|$ < 0). Time steps across the cyclone’s lifecycle (Figure 8) and cyclogenesis (Figure 9) with a deep warm core ($-\left|V_T^L\right|$ > 0, and $-\left|V_T^U\right|$ > 0), but with symmetric or asymmetric features (B > 0), are found in the second quadrant of B vs. $-\left|V_T^L\right|$ and $-\left|V_T^U\right|$ vs. $-\left|V_T^L\right|$.

While in the diagram B vs. $-\left|V_T^L\right|$ the distribution of the parameters has a similar number in the second and third quadrants, in $-\left|V_T^U\right|$ vs. $-\left|V_T^L\right|$ the frequency is higher in the third quadrant. It means that the number of time steps showing symmetric or asymmetric features (B < 0) with a shallow warm core ($-\left|V_T^L\right|$ > 0, and $-\left|V_T^U\right|$ < 0) is higher than that with a deep warm core (Figure 8 and Figure 9). Notice that there are a huge number of time steps with B < 0 in the fourth quadrant of B vs. $-\left|V_T^L\right|$ (Figure 8a,c,e), which indicates the decayment of cold core cyclones.

The results from Figure 8 have a pattern similar to that of [39] for eastern Australia, except for the higher frequency of symmetric systems (third quadrant of Figure 8a,c,e).

3.2.2. Separating the Cyclone’s Types

The classification of cyclones’ phases (tropical, subtropical, and extratropical) using the criteria C01, C02, and C03 (Figure 2) is applied to each time step of their lifecycle, as well as only for cyclogenesis, across two different periods: (a) a long period from 1900 to 2010 in ERA20C, and (b) a common period from 1979 to 2010 for both ERA20C and ERA5 and is shown in

Table 2. Absolute and relative frequency (%, in parentheses) of the number of time steps in which cyclones exhibit extratropical, tropical, and subtropical phases identified by the criteria C01, C02, and C03 over the South Atlantic. In bold, the same information is presented but only for the first time step of the cyclones, i.e., cyclogenesis.

	ERA20C (1900–2010)	ERA20C (1979–2010)	ERA5 (1979–2010)
Extratropical C01, C02, C03	336,571 (71.4%) 32,990 (74.6%)	91,960 (69.8%) 9084 (72.5%)	95,042 (70.4%) 9423 (75.5%)
Subtropical C01	29,712 (6.3%) 1612 (3.6%)	9226 (7.0%) 484 (3.8%)	8225 (6.1%) 484 (3.9%)
Subtropical C02, C03	35,052 (7.4%) 1694 (3.8%)	11,069 (8.4%) 538 (4.3%)	10,608 (7.9%) 465 (3.8%)
Tropical C01	1838 (0.4%) 96 (0.2%)	511 (0.4%) 33 (0.3%)	733 (0.6%) 30 (0.2%)
Tropical C02, C03	6553 (1.4%) 393 (0.9%)	1789 (1.3%) 111 (0.9%)	2173 (1.6%) 128 (1.0%)
Others C01	103,018 (21.9%) 9516 (21.5%)	30,135 (22.8%) 2933 (23.4%)	31,035 (22.9%) 2538 (20.4%)
Others C02, C03	92,963 (19.8%) 9137 (20.7%)	27,014 (20.5%) 2801 (22.3%)	27,212 (20.1%) 2459 (19.7%)
Total Events	471,139 44,214	131,832 12,534	135,035 12,475
Ratio of cyclogenesis to total time steps (%)	9.4	9.5	9.2

.

The ratio of cyclogenesis to the total number of time steps throughout the cyclone lifecycle is similar in ERA20C and ERA5, approximately 9% (last line of Table 2), and ERA20C maintains this ratio also over longer periods. For the common period of the reanalysis, ~70% of the time steps exhibit extratropical characteristics in both ERA20C and ERA5. For cyclogenesis, there is a slight difference between them since 72.5% of the occurrences have extratropical characteristics in ERA20C and 75.5% in ERA5 (Table 2). These percentages are similar for the long period: 71.4% for all time steps and 74.6% for cyclogenesis. The predominance of the extratropical phase also aligns with previous studies, such as [37]. Considering the annual cycle of the cyclogenesis with extratropical characteristics, in all periods and reanalyses, there is a higher frequency during the austral winter (with a monthly mean of ~27) and a peak of frequency in January (Figure 10). It is also consistent with the literature [1,2,50].

The subtropical phase identified using the C01 criterion during the common period shows that 6.1% of the time steps exhibit this feature in ERA5 and 7% in ERA20C. However, both datasets report the same percentage (~4%) for subtropical cyclogenesis. The percentage of the subtropical phase across all time steps increases by approximately 1.5% in both reanalyses when the C02 and C03 criteria are applied. However, the percentage of cyclogenesis with a subtropical phase remains similar to that identified using C01. For the 1900–2010 period, ERA20C statistics are closer to ERA5 than those of ERA20C for the 1979–2010 period. Even though the C02 and C03 criteria identified more subtropical cyclogenesis (~+0.2%) than the C01 criterion [39], the annual cycle obtained for both is similar, with higher activity in summer and fewer events in winter (Figure 10). This same feature is obtained for ERA20C in the two periods and ERA5. The average annual number of subtropical cyclogenesis in the common period is ~15, which is higher than that obtained by [7,8], and it can be associated with the use of more flexible thresholds for the three CPS parameters in comparison with those authors.

As with the subtropical phase, the tropical phase during the common period (Table 2) shows that the C02 and C03 criteria identify a higher frequency of time steps during the cyclone’s lifecycle and cyclogenesis (~1% of all cases) with a warm core, compared to C01 (0.25%). ERA20C (1979–2010) generally shows a slight underestimation of ~0.1% compared to ERA5. For the long period, the percentages in ERA20C remain similar to those from 1979–2010. Tropical cyclogenesis, less frequent than the other two types, shows higher occurrence between November and February (Figure 10). Tropical cyclone climatology in the two reanalyses using the C01 criterion aligns with [37,51], who obtained ~1 tropical cyclone per year in simulations.

There is a great percentage of time steps that do not follow exactly the criteria C01 and C02-C03. These cases, referred to as “others”, have a ~1% higher frequency in C01 compared to C02-C03. During the common period, using all three criteria, ERA20C shows similar percentages to ERA5 for all time steps but overestimates cyclogenesis by 3% (Table 2). The annual cycle of “others” tends to follow the extratropical cyclones with lower variability (Figure 10).

The spatial distribution of all types of cyclogenesis, combined and separated for the cases of subtropical and tropical types, is shown in Figure 11. Extratropical cyclogenesis occurs throughout the South Atlantic Ocean in both ERA20C and ERA5 during the common period (Figure 11a). There is a high frequency of extratropical cyclogenesis along the eastern coast of South America, particularly off the southeastern coast of Argentina (~45° S; ARG) and Uruguay (URU). ARG stands out as the most cyclogenetic region due to the climatological meridional temperature gradient favouring near-surface baroclinic instability [1,2,52]. Additionally, the occurrence of some extratropical cyclogenesis north of 22° S is noteworthy, a phenomenon that is sparsely documented in the literature, as this region typically exhibits a weaker meridional temperature gradient.

The SEB region is the most favourable area for the occurrence of subtropical (Figure 11b) and tropical (Figure 11c) cyclogenesis. As shown by [53], this region has great genesis potential, as indicated by the GPI index of [54], which evaluates key atmospheric and oceanic variables for tropical systems. For both subtropical and tropical cyclogenesis, a considerable number of events are observed south of 40° S. Some of these cases might be mesoscale cyclones (polar lows), but further studies are required to confirm this. It is important to note that this is not an issue with the cyclone identification and classification method, as the results align with those reported in [38,39].

Finally, the spatial pattern described for the different types of cyclogenesis during the common period is also observed in the ERA20C long period (Figure 11 top).

For the South Atlantic domain, the frequency time series of the different cyclone types are presented in Figure 12. ERA20C shows negative trends for the time steps classified as extratropical in both the long and short periods, but only the long-period trend is statistically significant. On the other hand, the trend is positive in ERA5 (Figure 12a). For extratropical cyclogenesis, any dataset shows a statistically significant trend (Figure 12b). The subtropical phase is characterised by a positive trend in both reanalyses, with ERA20C being statistically significant (Figure 12c). Subtropical cyclogenesis, however, does not have a trend (Figure 12d). For tropical cases, no trend is observed, particularly for tropical cyclogenesis (Figure 12e,f).

The negative trend observed for the extratropical phase in ERA20C aligns with future projections under climate change scenarios [6,16]. According to [16], the decrease in the number of systems is accompanied by an increase in their intensities and associated precipitation, potentially causing issues in coastal regions.

3.3. Hurricane Catarina

According to the tracking of Hurricane Catarina presented by [18] using satellite estimates, Catarina’s genesis occurred at 1800 UTC on 19 March, and it dissipated at 1800 UTC on 28 March 2004. This system began as an extratropical cyclone and transitioned to a tropical cyclone at 0600 UTC on 25 March, maintaining its tropical characteristics until landfall between the coasts of Rio Grande do Sul (RS) and Santa Catarina (SC), two states of southern Brazil.

ERA5 records the cyclogenesis at 0600 UTC on 20 March, while in ERA20C, the genesis occurs at 1200 UTC on 21 March (Table S1 Supplementary Material). Thus, both reanalyses delay the cyclogenesis by 12 and 42 h, respectively, compared to the genesis timing indicated by [18]. On the other hand, cyclolysis occurs at 0000 UTC on 29 March in both reanalyses, aligning with the cited study. Although there is a difference in the initial date of the system in the two reanalyses, the common time steps in ERA5, ERA20C, and [18] show very close positions (geographic coordinates) of the low-pressure centre, with less than a 2° difference in latitude and longitude.

According to Figure 13c,f, the largest difference between ERA20C and ERA5 occurs during the reversal of Catarina’s trajectory on 22 March. In ERA20C, the system remains quasi-stationary over the ocean until it begins moving toward the continent (westward) on 25 March. In contrast, the trajectory in ERA5 shows a looping motion, similar to the tracks of [15,18]. Another difference is that in ERA5, the landfall occurs over both RS and SC, while in ERA20C, it is limited to the state of Santa Catarina.

Using the CPS, it is possible to classify the phases of the precursor cyclone to Catarina. In both reanalyses (Figure 13), the system starts with a cold core structure, being symmetric (B < 10) in ERA20C and asymmetric in ERA5. This structure persists until March 22. After this period, in ERA5, the system goes through a phase classified as “other”, acquires subtropical characteristics, and finally transitions to tropical. In ERA20C, however, the subtropical phase does not occur; the “other” phase predominates for a greater number of time steps compared to ERA5, after which the cyclone evolves into a tropical phase. Another difference between the reanalysis is that the $-\left|V_T^L\left|\times -\right|V_T^U\right|$ diagram shows the tropical phase with a deeper core in ERA5, as $-\left|V_T^U\right|$ reaches higher positive values compared to ERA20C. The decay (landfall) of Catarina in both reanalyses occurs during the tropical phase.

The synoptic characteristics of the environment in which Catarina develops are similar between ERA20C and ERA5. In both, the upper-level westerly flow (above the surface low-pressure system) is weakened or easterly over the surface low-pressure system (Figure 14a,c). To the north of the low, stronger westerly winds occupy a more zonal band in ERA20C (Figure 14b), while in ERA5, they exhibit a southwest-to-northeast orientation (Figure 14d). At 0000 UTC on 25 March (27 March), the central low pressure reaches 1012 hPa (1010 hPa) hPa in ERA20C and 1014 hPa (1012 hPa) hPa in ERA5. A vertical cross-section through the cyclone’s central latitude shows that cyclonic relative vorticity is more intense near the surface in ERA5 (Figure 14g,h) than in ERA20C (Figure 14e,f). Since relative vorticity depends on grid spacing, this greater intensity in ERA5 can be explained by this factor.

This case study highlights that ERA20C, despite being constructed with a smaller number of assimilated variables, adequately represents Hurricane Catarina in the CPS and its synoptic environment, providing confidence in the identification of tropical phases in climatological data since 1900.

4. Conclusions

This study tracked all types of synoptic-scale cyclones over the southwest South Atlantic Ocean to evaluate the performance of ERA20C in representing these systems compared to ERA5, a modern reanalysis, for the period 1979–2010. Subsequently, it aimed to classify the different phases of cyclones (extratropical, subtropical, and tropical) over a longer period (1900–2010) using ERA20C. The main results are summarised below.

• Validation Period: 1979–2010

Total cyclones: Initially, there were doubts about whether ERA20C could accurately represent the climatology of cyclones in the South Atlantic, given that its construction relies solely on surface pressure and near-surface wind observations [36], which are scarce data in this region. However, the results indicated that the cyclogenesis and trajectory densities of all cyclones were very similar between ERA20C and ERA5. Additionally, ERA20C and ERA5 presented very similar monthly and annual mean frequency of cyclones over the South Atlantic Ocean (391.7 in ERA20C and 389.9 in ERA5). The regions with the highest cyclogenesis density align with findings in the literature [3,45], showing higher density in ARG compared to the other subregions, SEB and URU.

Cyclone’s Classification: Both reanalyses showed that the predominant systems are cold core and asymmetric (~70%), with the distribution of CPS parameters at the time of genesis being similar to those observed throughout the entire lifecycle. Considering cyclogenesis, the annual average is 293.0 (303.9) cases for extratropical, 14.6 (15.5) for subtropical with C1 criteria and 17.3 (15.0) with C2–C3 criteria, and 1.0 (0.9) for tropical with C1 criteria and 3.6 (4.1) with C2–C3 criteria in ERA20C (ERA5). The coasts of southeast and part of south of northeast Brazil stand out as the most favourable for cyclones having subtropical and tropical structures along their lifecycle.

At this stage of the study, it was concluded that ERA20C is robust in representing cyclones in the South Atlantic Ocean, being useful to identify cyclone phases since 1900.

• Long Period Analysis: 1900–2010

A total of 96 cases of tropical cyclogenesis were identified over the southwest South Atlantic. Throughout the lifetime of all cyclones, a total of 1838 time steps exhibited a tropical structure, indicating that cyclones can acquire warm cores at different stages of their lifecycle. The time series of the cyclone’s phases during all time steps of these systems shows a negative and statistically significant trend to the extratropical type, a positive and statistically significant trend to the subtropical type, and no trends for the tropical type. Additionally, an analysis of Hurricane Catarina showed that ERA20C reproduces its transition from extratropical to tropical and the change in track from southeast to west at the time of transition, although ERA20C underestimates its lifetime and intensity. The synoptic environment shown by ERA5, where weak westerly winds at mid-levels predominated in the region, and the vorticity tube associated with Catarina are also found in ERA20C, with some small differences in intensity.

The analysis of Hurricane Catarina gives confidence to explore in future work the synoptic environments in which the 96 tropical cyclogenesis events identified in ERA20C developed. This analysis could be complemented by a dynamic study (analysing vorticity, heat and humidity budgets, and energy cycle) to define other important characteristics during the cyclone’s lifecycle. Additionally, the characterisation of the associated weather conditions (e.g., strong winds and precipitation) using historical data and media sources, as well as identifying the areas most affected near the Brazilian coast, can be performed. These additional studies will provide valuable knowledge that is currently lacking for the Atlantic Ocean basin.

Finally, it is important to note that while the CPS methodology is widely used for classifying cyclones, it has certain limitations. Representing a cyclone’s structure using only two atmospheric layers may not capture all the features necessary to identify the cyclone type accurately. This limitation can justify why some cases of tropical cyclones, where the warm core structure remains well-developed, do not show a positive value of $-\left|V_T^U\right|$, and as a result, such cyclones would not be classified as tropical by the CPS methodology. In our study, several cases of subtropical and tropical cyclones appear in midlatitudes, which also deserve more investigation. For mesoscale cyclones, such as those in the Mediterranean, the radius typically used in the analysis often needs to be adjusted to less than 500 km, which can complicate comparisons with larger-scale systems. These limitations underscore the need for collaborative efforts among researchers from different oceanic basins to refine the methodology and establish unified criteria that improve its applicability across various cyclone types and regions.