The inner life of the Atlantic Intertropical Convergence Zone

The Intertropical Convergence Zone (ITCZ) is a central component of the atmospheric general circulation, but remarkably little is known about the dynamical and thermodynamical structure of the convergence zone itself. This is true even for the structure of the low‐level convergence that gives the ITCZ its name. Following on from the major international field campaigns in the 1960s and 1970s, we performed extensive atmospheric profiling of the Atlantic ITCZ during a ship‐based measurement campaign aboard the research vessel SONNE in summer 2021. Combining data collected during our north–south crossing of the ITCZ with reanalysis data shows the ITCZ to be a meridionally extended region of intense precipitation, with enhanced surface convergence at its edges rather than in the center. Based on the location of these edges, we construct a composite view of the structure of the Atlantic ITCZ. The ITCZ, far from being simply a region of enhanced deep convection, has a rich inner life, that is, a rich dynamical and thermodynamic structure that changes throughout the course of the year, and has a northern edge that differs systematically from the southern edge.


INTRODUCTION
It is hard to form a mental picture of Earth's climate without imagining the Intertropical Convergence Zone (ITCZ).Therein air rises, defining the upward branch of the Hadley cell, and precipitation forms.The energy released through the formation of precipitation balances the atmosphere's radiant heat loss to space, setting the atmosphere in motion.While, as this description shows, it is easy to picture the ITCZ as an integral part of the general circulation, it is harder to form a mental picture of the ITCZ itself and its own integral parts.
Recent research has disproportionately focused on the ITCZ's outer life.By this we mean it posed questions that implied a distant, or external, point of view.Examples include the following: where is it?(Waliser & Gautier, 1993); why is it there?(Bischoff & Schneider, 2016;Kang et al., 2008;Mikolajewicz et al., 2018;Philander et al., 1996;Riehl, 1954); how and why does this position vary?(Hohenegger & Stevens, 2022;Voigt et al., 2016); and why do climate models find it so hard to simulate these features?(Bellucci et al., 2010;Lin, 2007;Tian & Dong, 2020).A much older literature adopting a more inner point of view exists, as reviewed by Klocke et al. (2017), which attempts to understand the relationship between the ITCZ regions of low winds, or the doldrums, and the corresponding atmosphere and ocean characteristics in general.But as sails succumbed to steam a hundred years ago, interest in the ITCZ's inner life waned, to the point where it seems as though there is nothing left to describe, let alone explain.This is a surprising state of affairs, all the more so when it is contrasted with what we know about other circulation systems, and how their inner life plays such a vital role in their outward expression.
We are not the first to comment on this state of affairs.Wodzicki and Rapp (2016) raised similar issues, as did Klocke et al. (2017), who went so far as to speculate that the inner life of the ITCZ might control its outer life.This argument is in line with the findings of Ferreira and Schubert (1997) and Wang and Magnusdottir (2006), who show that the convection within the ITCZ can also cause its breakdown.Following up on these ideas, we initiated a program of ship and aircraft-borne observations to study the inner life of the ITCZ.This program began with extensive atmospheric profiling of the Atlantic ITCZ on board the research vessel Sonne in the summer of 2021.During this first campaign we were struck by the rich dynamical and thermodynamical structure of the inner ITCZ itself, reminiscent of the observation that "the inner life has its soft and gentle beauty" (V.Woolfe).This, together with features reminiscent of those observed by Klocke et al. (2017), motivated a more systematic study, as presented in this article.
Starting with an analysis of low-level convergence, which after all gives its name to the ITCZ, we realized that there is considerable ambiguity as to the latitudinal structure of the convergence zone and how it varies with longitude.Even within a single rain belt (or band of convergence), it remains unclear whether the convergence zone is defined better by a single line of convergence in the center of the rain belt or by multiple lines of convergence with pronounced convergence marking the edges of the rain belt.
Most studies adopt the former, central convergence line, definition.See, for example, the historically based diagram of the ITCZ in fig. 1 of Nicholson (2018), or Wodzicki and Rapp (2016), who used monthly averaged velocity fields to identify the ITCZ as a single line of low-level convergence and determine the ITCZ extent via the northern and southern limits of the precipitation surrounding it.This prevailing view, however, contrasts with the earlier literature: for instance, Fletcher (1945) noted that the ITCZ "will frequently break down into two or more lines of convergence".Also, Flohn (1951) schematized a single ITCZ over the Atlantic as being delimited by two edge intensified bands of precipitation, each presumably supported by its own line of enhanced convergence.Subsequent studies, based largely on data from satellites and large field programs, confirm the more complex nature of low-level convergence, which frequently appears to be composed of two lines of surface convergence: see, for example, Chen and Ogura (1982), Frank (1983), and Huaman et al. (2022).Chen and Ogura (1982), for example, at times found two lines of convergence over the East Atlantic ITCZ using data collected during the Global Atmospheric Research Program (GARP) Atlantic Tropical Experiment (GATE).In addition to the convergence line caused by the surface confluence of the trade winds, they documented a second convergence line around 3.5 • south of this confluence line associated with speed convergence.More recently, as simulations on grids fine enough and over domains large enough to represent both the inner and outer life of the ITCZ have become possible, they too hint at similar features.For example, in fig. 1 of Klocke et al. (2017), the Atlantic ITCZ is visible as a region of low surface wind speeds, with the northern and southern edges characterized by an abrupt decrease in wind speed and thus presumably convergence.The existence of sharp edges in column water vapor and correspondingly in precipitation, although not directly in surface convergence, was shown by Mapes et al. (2018), Masunaga and Mapes (2020), and Masunaga (2023).
To summarize, while the structure of low-level convergence has been discussed in a few studies, there is still no clear consensus on how best to describe it.Theoretical and modeling studies usually emphasize a single convergence line (see e.g., Hohenegger & Jakob, 2020, for a recent example), whereby observations tend to document a more complex structure, often better idealized as a double convergence line, as shown schematically in Figure 1a and b respectively.To the extent that convergence within the ITCZ is associated with several lines of convergence that extend only a few degrees of longitude (Weller et al., 2017), our limiting cases in Figure 1 encapsulate the idea that these lines tend to form and dissipate in the center, or toward the edges, of the rain belt.This then leads us to ask: is the ITCZ better idealised as a rain belt centered on a single line of convergence, or rather a rain belt straddling a pair of edge-intensified bands of convergence?
To answer this question we first analyze observations collected during our transect of the East Atlantic ITCZ on board the research vessel (RV) Sonne in summer 2021.These observations are then used to verify the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis of meteorological data for the period of the research cruise, based on which the reanalysis is used to explore how well our findings from a single transect generalize to other times, seasons, and locations using reanalysis data.Data sets are introduced in Section 2. In Section 3 we introduce a method to distinguish between central convergence and edge convergence, and to composite meteorological fields about convergence features that may shift in space and time.The resulting composite structure of low-level convergence as well as the composite structures of a number of other dynamic and thermodynamic variables important for deep convection are presented in Section 4, including their sensitivity to longitude and season.We summarise our results and draw a tentative sketch of the ITCZ in Section 5.

DATA
Our analysis is based on observations collected during a single north-south crossing of the Atlantic ITCZ in boreal summer on board the German research vessel Sonne, as well as 20 years of reanalysis data and estimates of surface precipitation based on satellite measurements.These data sets are described in more detail below.

Observational data collected during ship campaign
In summer 2021 we took advantage of the research cruise Mooring Rescue 2 (SO284), using the German research vessel Sonne, to make intensive measurements within and about the Atlantic ITCZ.The cruise was led by Geomar Helmholtz Centre for Ocean Research Kiel-and targeted the tropical Atlantic region; for more details see Brandt et al. (2021).Our focus is on data we collected during a period of intensive observations during the north-south crossing of the Atlantic ITCZ at about 20 • -23 • W during the time period from July 9-13, 2021.During the crossing, we launched a total of 32 radiosondes from the working deck (Schulz et al., 2022).The launch frequency ranged from two to three hours, and the latitude of the research vessel at the radiosonde launch times is shown in Figure 2a.The radio sounding system we used consisted of Vaisala radiosondes of type RS41-SGP, helium-filled balloons with parachutes (Totex TA200-No.088),a launching station, and a data-acquisition system (Vaisala Sounding Processing Subsystem SPS311G).The Vaisala RS41-SGP is a standard meteorological radiosonde and measures temperature, humidity and pressure at a temporal resolution of 1 s during its ascent and descent.These measurements were complemented by the set of continuously running on-board measurements.In particular, we use disdrometer measurements of precipitation rate (Parsivel 2 -Ott) as well as retrievals of column water vapor derived from the on-board Global Navigation Satellite System data as described in Bosser et al. (2021).In addition, we use near-surface measurements of wind-speed and direction (Ventus-Lufft), measured approximately 21.5 m above mean sea level, air temperature and humidity (EE33-E+E), measured approximately 24.5 m above mean sea level, pressure (PTB330-Vaisala), measured approximately 14 m above mean sea level, and sea temperature (Temperaturfühler 2020-METEK, formerly TH-Friedrichs), measured approximately 2 m below mean sea level, provided by the on-board weather station of the German National Meteorological Service.The on-board measurements of wind-speed and direction were used to calculate the meridional wind speed shown in Figure 2, for example.

Multiyear gridded data
In addition to the observational data, we analyzed the 5th Generation of the ECMWF Reanalysis of meteorological data (ERA5, Hersbach et al., 2018) and precipitation estimates as provided by the Integrated Multi-satellite Retrievals for GPM (IMERG) data set.In both cases we analyzed data from August 18, 2001-August 17, 2021.
Although we focus primarily on the boreal summer ITCZ, we use data from all seasons for both ERA5 and IMERG to also examine the seasonal dependence of our results.
From ERA5 we examined 10-m zonal and meridional wind speed, from which we additionally calculate divergence and vorticity.We also analyze total precipitation, convective available potential energy (CAPE), convective inhibition (CIN), total column water vapor (CWV), and sea-surface temperature (SST), as well as surface pressure and 2-m temperature.For all these fields we use the hourly product regridded to a regular latitude-longitude grid of 0.25 • .To assess the suitability of the ERA5 for our purposes we first compare it with the direct, ship-based observations.Figure 2 shows the meridional wind speed v for the period of the ITCZ crossing according to both ERA5 and on-board measurements.As the ship campaign data are available at a much higher temporal and spatial resolution than the reanalysis data, we first calculate 75-minute averages.For the typical speed of the ship, 12 knots, this corresponds approximately to a 0.25 • spatial resolution.In Figure 2a, the meridional wind speeds in ERA5, taken along the ship track, show overall good agreement with the on-board measurements.Disagreement between the ship-based measurements and the reanalysis data is apparent around 9 • N, during the period when the ship was approximately stationary.At that time the wind direction changed to a southerly direction for about two to three hours before the arrival of a cold-pool signal.The surface air temperature as well as the wind-speed measurements on board then showed signals of a number of cold pools which formed just to the north of us.Taking the meridional gradient of v as a proxy for convergence, convergence peaked at the time when the first cold-pool signal arrived (approximately at 9.5 • N) and then a second time, at the end of the last cold-pool signal (8.5 • N).Comparison with the along-track measurements of air temperature and meridional wind speed from ERA5 shows that, while a significant decrease in air temperature is also present in the reanalysis data, the variability in the meridional wind speed is significantly lower, with weak but mostly southerly winds (not shown).As, in the following, we use zonally averaged reanalysis data (averaged over 3 • in longitude) to study latitudinal cross-sections of the ITCZ, we show the time evolution of the thus averaged meridional wind speed in Figure 2b. Figure 2b also shows the ship-based measurements as a function of latitude and time, which illustrates that, although (for simplicity) we tend to treat the ship-based measurements as if they were a latitudinal cross-section at a particular time, the meridional field changed during the time of the crossing, as, for example, is already evident from the above discussion regarding the cold-pool signals.
We first test how well our observations generalize by analyzing the July and August climatology of ERA5 over a region extending from 20 • -23 • W, as this brings perspective to the ship-based measurements.
The choice of longitudinal range for the purpose of averaging is primarily a trade-off between oversampling fragmented convergence lines that might occur if the region is chosen too narrowly and smoothing the gradients that might result from choosing a region that is too broad.The latter consideration arises from the fact that individual convergence lines rarely extend across the entire Atlantic ocean, especially not along a single degree of latitude.We choose a longitudinal extent of 3 • , corresponding to about 330 km, based on the results of Weller et al. (2017).Their research has shown that a significant fraction of precipitation over tropical oceans is associated with coherent convergence lines that are at least 300 km long.To assess the impact of this choice on our results, we repeated the analysis below with regions with widths of 1.5 • and 6 • , but found no qualitative changes (not shown).
As shown by Liu and Xie (2002), a double ITCZ structure is visible in the surface wind field almost all the year round in the Atlantic.To exclude the southern ITCZ (see their fig.2b), from our analysis we choose a dynamic range of latitude values, extending 10 • towards the north and 10 • towards the south of the latitude value at which precipitation peaks as determined from IMERG; see Table 1.
For July, the analysis domain, therefore, extends from 2.5 • S-17.5 • N and for August from 1.75 • S-18.25 • N.
Later, to see if our findings generalize across the basin and to other seasons, we extend our analysis to the three additional regions from 23 • -26 • W, 26 • -29 • W, and 29 • -32 • W, as well as to the months of January and February.

METHOD
Our primary goal is to answer the question of whether or not the convergence in the ITCZ is better described by a convergence line marking the center of the rain belt or by convergence lines marking the northern and southern edges of the rain belt.Toward this end, in the following, we propose a method to distinguish between these two cases based on the near-surface meridional wind field.We use the meridional wind field, rather than the convergence field, as the latter is much more noisy.Use of the meridional wind was also attractive by virtue of its simplicity and it's direct link to the meridional overturning circulation of the Hadley cell.
Figure 3a shows that the meridional wind speed of the southerly trade winds increases up to about 5 • N in July and August, which is similar to what Liu and Xie (2002) find for the East Atlantic based on scatterometer data.Wallace et al. (1989) also find a distinct peak in meridional wind speed in the east Pacific, which they relate to the underlying meridional SST gradient.The peak of the East Atlantic meridional wind speed persists throughout the year, even though the amplitude changes and the location of the peak shifts between about 5 • N in boreal summer and 5 • S in boreal winter.The region of mean convergence in Figure 3a is bounded in the south by the local maximum in meridional wind speed and in the north approximately by the location where the meridional wind speed changes sign.As expected, the region of mean convergence also coincides with the region of significant precipitation; see Figure 3b.We expect the latitudinal structures shown in Figure 3 to be smeared out as the synoptic variability in the position of the ITCZ is as large, or larger, than the width of the rain belt, for instance due to the passage of an African easterly wave (Chen & Ogura, 1982;Frank, 1983).Thus, while Figure 3 suggests that the meridional wind field is suitable for studying the convergence field, we cannot answer the question about whether convergence is intensified at the edges of the rain belt or in the center of the rain belt, based on this.
If the convergence in the ITCZ is marked by a central convergence line, we expect one latitude with exceptionally strong convergence in the center of the rain belt; likewise, if the convergence is marked by convergence lines at the edges of the rain belt, we expect exceptionally strong convergence at at least two, well-separated, latitudes.Assuming that low-level convergence and divergence in the ITCZ are mainly due to changes in the meridional wind speed, v, with latitude, y, we have convergence where  y v(y) < 0 and divergence where  y v(y) > 0 and the two cases can be idealized as shown in the bottom row of Figure 1.In Figure 1, the location of a step in the meridional wind speed marks the location of (infinitely) strong convergence,  y v = −∞, while a constant wind speed marks a region with neither convergence nor divergence, that is,  y v = 0.The two cases can be understood as follows.
• Central convergence: a single, central convergence line results when trade winds from the northeast to north meet trade winds from the southeast to south in the center of the rain belt, that is, convergence due to the confluence of the trade winds; see Figure 1c.
• Edge convergence: two convergence lines result when there is an abrupt slackening of the southerly trade winds, that is speed convergence at the southern edge of the rain belt, prior to meeting the northerly trade winds marking the northern edge of the rain belt; see Figure 1d.
Regarding the edge convergence case, we also tested for the possibility that the southern edge of the rain belt is associated with confluence and the northern edge associated with speed convergence in the northerly trade winds, but this case proved to be very rare, and is not shown.Hence, a second line, if it exists, must be associated with speed convergence, south of the confluence line.This is also the situation documented by Chen and Ogura (1982), based on the GATE data, and as described in the Introduction.Note that, while the edge convergence case requires the presence of at least two convergence lines, it does not exclude the presence of additional convergence lines inside the rain belt caused, for example, by cold pools.
Once we have identified the latitude of the convergence line, or lines, we composite about this latitude to explore the composite structure of the ITCZ presumed to be associated with these features.The extent to which (i) a simple, wind-based, definition of the convergence lines is associated with marked convergence, and (ii) the familiar structure of the rain bands emerges from compositing about these zones of convergence, tests our method for identifying the convergence zone and the relevance of these convergence features for rain belts.
Compositing is based on the locations of the points denoted by y south and y north in Figure 1.Point y north marks the latitude of confluence of the trade winds of the two hemispheres and thus the latitude where the meridional component of the wind velocity v changes sign.In particular, we define y north as the northernmost latitude where v is zero, that is, v(y = y north ) = 0 with v < 0 for all y > y north .Point y south aims to detect the latitude of speed convergence.It is based on the idea that the southerly trade wind speed increases towards the north but then abruptly reduces at y south .To this end, we define point y south as the northernmost point at which v > k ⋅ v max .For k = 1.0, the condition for y south would be the same as identifying y south with the location of the maximum of the signed meridional wind speed.As we expect the location of strongest convergence to occur north of v max , in the following we use k = 0.8.We repeated our analysis with values of k in the range 0.7-0.95 and found our main findings to be robust to this choice.If the velocity field is approximately described by either of the two depictions shown in Figure 1, the case central convergence corresponds to y south ≈ y north while the case edge convergence corresponds to the case y south ≪ y north with the width of the ITCZ simply given by w ITCZ = y north − y south .Given the way our method is set up and that the meridional wind speed has to be continuous, unlike that sketched in Figure 1, y north has to be larger than y south .We therefore need to define a critical minimum width w crit to distinguish between the two cases.
Here we set w crit = 200 km ≈ 1.8 • , that is, to the spatial scale separating the meso-beta from the meso-alpha scale.The rationale is that cases where the two convergence lines are separated by a distance smaller than or comparable to the scale of a single cloud cluster or squall line, typical atmospheric features of the meso-beta scale, should still be considered as corresponding to the central convergence case.Using w crit as an approximate boundary between the two cases, we can examine the frequency distribution of w ITCZ to answer our original question of whether the ITCZ is better described by a single or a double convergence line.
As this method makes a number of assumptions, we need to verify these assumptions by analyzing the meridional wind speed.With the above definitions, we define a coordinate shift such that where y is simply the standard latitudinal coordinate.This shift ensures that y south is located at y shift = 0 and v(y To account for synoptic variability in the ITCZ width, that is, w ITCZ = y north − y south , we define y scale at each time step such that In the y scale based coordinate system, v(y scale = 0) = v(y = y south ) = k ⋅ v max , as before, but also v(y scale = 1) = v(y = y north ) = 0.In the following we first explore the structure of the meridional velocity for the shifted variable, that is, y shift , for which we additionally condition w ITCZ to be within an interval [w, w + Δw].We show that the structure of the ITCZ is similar, in a scaled sense, irrespective of its width.This then motivates a subsequent analysis, where we explore this structure based on the scaled variable y scale .Based on y shift and y scale , we evaluate our method by assessing whether y south and y north are regions of strong convergence.
We note that the method described above is biased, as v can only decrease below a certain threshold, that is, below k ⋅ v max or zero, if  y v < 0, which increases the likelihood for ∇ ⋅ v(x, y) < 0. In fact, the probability of v sinking below any given threshold increases for steeper negative gradients, that is, more negative values of  y v.More formally, even if the value of  y v at any location were entirely random,  y v at the specific location where it crosses a certain threshold would be drawn from a conditional probability distribution, with the condition being that  y v is sufficiently small for v to decrease below the given threshold.This will bias the assessment of the relative strength of convergence at y south and y north compared with other locations if one simply compares the mean divergence at y south and y north with the mean divergence at other locations.
This bias can be partly removed by comparing the convergence strength at y south and y north with the convergence strength at a latitude y considering only times when ∇ ⋅ v(x, y) < 0, that is, when ignoring all instances where the surface velocity field at latitude y is divergent.In the following, therefore, we will consider not only the composite of the divergence field to assess whether there is increased convergence at y south and y north but also the composite of the divergence field conditioned on ∇ ⋅ v(x, y) < 0. We refer to this field as divergence (< 0 only).Conditioning on instances where the surface velocity field is convergent compensates for the fact that  y v has to be negative at y south and y north but does not compensate the bias towards steeper gradients.Therefore, we test our method in two additional ways.
First, instead of using the meridional wind field to test whether y south and y north are marked by particularly strong convergence, we use the convergence field directly to identify locations of peak convergence.We employ the SciPy Virtanen et al. (2020) "find_peaks" package (version 1.8.1) and the convergence field calculated based on ERA5, −∇ ⋅ v(x, y), which is then zonally averaged as before.For each instance in time, we detect the number of prominent peaks along with their respective latitude, prominence, and height.To identify a local maximum in convergence as a peak, we set the required prominence and the height of the peaks to be equal to the 90th percentile of the convergence field.For instances where we find at least two peaks in convergence, we composite the corresponding meridional wind fields based on the latitudes of the two most prominent peaks.This allows us to test whether the wind-speed profile assumed in Figure 1b can be recovered.Additionally, we compare the detected latitudes of the locations of peak convergence with the latitudes we find for y south and y north .An apparent disadvantage of this method is that the results, especially concerning the number of detected peaks, will depend on the criteria we set for the height and prominence of the peaks.
Second, we use the column water-vapor field to identify the margins of the moist Tropics (Mapes et al., 2018;Masunaga, 2023;Masunaga & Mapes, 2020) to test whether surface convergence peaks at the margins or in the center.To this end, we simply redefine the locations y north and y south based on the northernmost and southernmost latitudes where CWV crosses the 50-mm contour, respectively, before calculating the mean composite of the divergence field, analogous to what has been described above.The advantage of this method is that it makes the margin detection independent of the surface wind or the convergence field.The disadvantage is that the region of the moist Tropics, as identified by the moist margins, is not identical to the ITCZ but rather is the moist envelope expected to contain the deep convection associated with the ITCZ.
Finally, we note that, independent of what method we use to identify y north and y south , our rescaling is not limited to the velocity or convergence field but can also be applied to any other field.We will use this below to test how the location of y south and y north relates to the tropical rain belt and, moreover, to construct a schematic representation of the ITCZ.

Detected latitudes of the southern and northern convergence line
Applying the meridional wind speed based method introduced above to the ship campaign and the reanalysis data, we first determine the latitudes of y south and y north .Applying our method to meridional wind speeds measured during the north-south crossing, we find y south at 6.1 • N and y north at 9.2 • N, resulting in a width of 3.1 • .The latitudes of y south and y north are indicated in Figure 2b.The detected latitudes of y south and y north for the months of July and August for the 20 years of reanalysis data are shown in Figure 4, together with the resulting widths (y north − y south ).The median value for y south is 6.5 • N, that for y north 10.8 • N, and that for the width 4.1 • .Based on the value of w crit = 1.8 • introduced above, we find that only in about 10% of the cases is the ITCZ better described in terms of the central convergence case.Comparison of the ERA5-based distribution of y north and y south with the corresponding values for the ship campaign suggests that we captured a typical realization during our north-south crossing; see also Figure 4. We can also compare these distributions broadly with the day-to-day time evolution of the latitude of the East Atlantic ITCZ during GATE as analyzed by Frank (1983).Based on the location of maximum convection, they find that the ITCZ latitude propagated between 2.5 • N and 13 • N, with a peak at 6.5 • N.These latitudes also contain the majority of the detected values of y south and y north , and compare especially well with the distribution of y south .
The latitudes of y south and y north as detected from the on-board measurements as well as almost all of the detected latitudes in the reanalysis data are located north of the Equator.This agrees with our expectations, as the Atlantic ITCZ is located north of the Equator in boreal summer.If we extend our analysis domain to 10 • S, we find a weak, secondary peak in the occurrence of y south between 5 • S and 9 • S (not shown).Comparison with Liu and Xie (2002) suggests that we can associate these instances with the southern branch of the Atlantic double ITCZ.

4.1.1
Convergence strength at y south and y north Before discussing the implications of Figure 4 further, we test the assumption that y south and y north mark locations of strong convergence and capture the location of the tropical rain belt in the boreal summer ITCZ.This assumption is verified by Figure 5, which shows the zonal  mean latitudinal dependence of the 10-m meridional wind speed, divergence, divergence (<0 only), and precipitation field.The top row of Figure 5 shows the zonal mean as a function of y shift (see Equation 1), after conditioning on different widths as indicated by the different colours.
In the bottom row, the zonal mean is shown as a function of y scale (see Equation 2).We find that, depending on the width, the meridional wind speed (roughly) resembles one or the other of the schematic wind speeds shown in the bottom row of Figure 1.In particular, Figure 5a shows that, for instances when the ITCZ is narrow, that is, the value of w ITCZ is small, the velocity is marked by wind speeds that quickly drop below zero (as in Figure 1c), while for large values of w ITCZ an intermediate plateau of low wind speeds becomes more and more apparent (as in Figure 1d).Independent of the width, the meridional southerly trade winds increase on approaching the ITCZ while the northerly trade winds decrease on approaching the ITCZ.
The corresponding plot of the divergence field shows that the strongest convergence occurs approximately at y = y south and y = y north in a two-peaked structure.As noted above, our method is biased, as y south and y north are generally expected to be locations of convergence rather than divergence.As shown in Figure 5c, we find preferential convergence at these locations, even when ignoring all instances of divergence (i.e., all instances where the divergence field is positive) before calculating the zonal mean.Finally, when considering the zonally averaged distribution of precipitation, it is apparent that the locations of y south and y north correspond roughly to the southern and northern edge of the region of intense precipitation, respectively, with the majority of precipitation occurring between y south and y north and a maximum located at y south , as was also documented for the GATE data (Chen & Ogura, 1982).As we did not use the precipitation field in our determination of y south and y north , this result, together with the two-peak structure of the convergence field, suggests that this method correctly identifies a southern edge of the ITCZ (y south ) as well as a northern edge of the ITCZ (y north ).Another, in our opinion, quite beautiful indication for the validity of our method can be seen in Figure 6, where we show satellite images for the times when the research vessel was near y north (Figure 6a) and y south (Figure 6b).In both images, the ITCZ is clearly visible as a region of enhanced deep convection, with a clear northern edge on July 10 at the time when the ship was at y ship = y north , and a southern edge on July 11 when y ship = y south .While the absence of any deep convection inside the ITCZ in Figure 6 makes the presence of two convergence lines particularly apparent during the time of our crossing, we do not usually expect this to be the case, as precipitation is not solely confined to the edges (Figure 5h).More generally, therefore, we expect the confluence and speed convergence line to be, if at all, visible in satellite images as two lines of clouds marking the southern and northern edges of the region of deep convective clouds.

Peak convergence strength
How many convergence lines do we usually find in the reanalysis data and-if we find more than one-do we recover the assumed meridional wind profile when conditioning on the locations of the most prominent convergence lines?Examining, as before, the boreal summer ITCZ in the eastern Atlantic and identifying convergence peaks according to the criteria defined in Section 3, we find no convergence line in about 5% of the times, one line in 29%, and more than one line in 66% of the times.Again, this result suggests that the ITCZ largely corresponds to the case of edge convergence, although the frequency of the central convergence case estimated here is higher compared with the estimate based on our main method.
If we restrict our analysis to the times when at least two peaks in convergence are detected, we can calculate a composite of the meridional velocity based on the latitudes of the two strongest convergence lines.As shown in Figure 7a, the resulting composite closely resembles the one sketched in Figure 1, with the southern convergence line close to the maximum velocity and the northern convergence line close to zero.In contrast to the previous method, this is not set a priori.Calculating the mean prominence of the detected southern (−210 × 10 −6 m⋅s −2 ) and northern convergence lines (−204 × 10 −6 m⋅s −2 ), respectively, we find that the values are very comparable, suggesting that there is no single "dominant" location of convergence.In fact, the slightly stronger prominence of the southern convergence line is in agreement with the results shown in Figure 5. Figure 7b shows that the distributions of the latitudes of the southern and northern convergence peaks closely resemble the ones found above, though the total number is lower, as we exclude cases with fewer than two lines (roughly one third of the instances).Also in close agreement with the previous results is the median width, found to be 3.8 • rather than 4.1 • .

Convergence strength at the moist margins
Finally, the result of our moist margin-based compositing method is shown in Figure 8.In Figure 8a, we show the conditional composites of CWV itself.As anticipated from the location of the moist peak in the bimodal distribution of Mapes et al. (2018) (see their fig.1), the moist plateaus in CWV measure around 55 mm.Intriguingly, as the width of the moist region increases, the distribution displays a growing asymmetry, with peak CWV values close to the southern moist margin.In Figure 8b, we illustrate the corresponding composites for the 10-m divergence strength.We observe that the southern and northern moist margins effectively delineate the region of mean convergence (interior) from that of mean divergence (exterior).Consistent with the hypothesis of edge convergence, the convergence does not peak in the center of the moist plateau, but at its edges.Interestingly, we find an asymmetry in the convergence strength, with convergence at the southern edge being stronger than at the northern edge, increasingly so as the distance between the edges increases.While the peaks of divergence are much less pronounced compared with the previous two methods, as expected due to the fact that CWV and the surface wind field change on different time scales, the existence of a double-peaked structure in convergence provides an important, and independent, indication for edge convergence.Taken together, these results allow us to answer our original question of whether the low-level convergence in the ITCZ is best described by central convergence or edge convergence for the East Atlantic ITCZ in boreal summer.According to the method based on the meridional wind speed, the width of the ITCZ is larger than w crit in more than 90% of the cases.Considering the distribution of ITCZ widths for widths greater than w crit in more detail, we find that w ITCZ varies between 2.00 • (10th percentile) and 6.25 • (80th percentile) and that the convergence has a double-peaked structure even for the smallest widths shown in Figure 5, with the range of [2.00 • , 2.25 • ] corresponding to the range of the 10th-15th percentiles.The method based on convergence peaks detects at least two convergence lines in about 66% of the cases.This still indicates that the ITCZ is better described by edge convergence in the majority of cases, although the sensitivity to the set criteria for peak detection somewhat reduces the significance of this result.To summarize, convergence in the East Atlantic ITCZ in boreal summer is, in the majority of cases, best described by the edge-convergence-like scenario, where convergence peaks at the edge rather than the center of the tropical rain belt.

Latitudinal profiles of the Atlantic ITCZ
In the following, we use the method based on the meridional wind speed to compute latitudinal profiles of a number of dynamic and thermodynamic quantities relevant for, or associated with, convection using IMERG, ERA5 as well as our campaign data to derive a mean structure of the East Atlantic ITCZ in boreal summer.The mean structure of these fields with respect to the locations of peak convergence is shown in Figure 9.
Figure 9a shows the latitudinal profile of precipitation as measured by IMERG.These latitudinal profiles can be directly compared with the corresponding profiles for ERA5 precipitation rates in Figure 5d.As also found for the ERA5 precipitation data, the majority of precipitation occurs for y in [y south , y north ] and, again, the distribution is asymmetric, with the strongest precipitation rates close to the southern edge, that is, y south .
The latitudinal distribution of CWV, Figure 9b, as expected, demonstrates that the region of elevated CWV is concentrated within the inner ITCZ, with CWV decreasing markedly south of y south and north of y north .The CWV is, similarly to the precipitation, asymmetrically distributed with a maximum near y south .This is also true for the single cross-section as measured on board the research vessel.Overall, these results are in excellent agreement with our expectation of the ITCZ, commonly known to be the region of high precipitation rates and also of high column water vapor, but at the same time they highlight an asymmetry between its northern and southern edges.
An asymmetry is also observed in other variables, e.g.CAPE (Figure 9c) and CIN (Figure 9d) have different characteristics north and south of the ITCZ.In particular, the region north of the ITCZ is characterized by high values of convective inhibition and simultaneously low values of CAPE.However, as we approach the position of y north from the north, CAPE increases sharply while CIN decreases rapidly.In fact, CAPE maximizes at y north and then decreases slowly towards y south and more rapidly south of y south .While the region within the ITCZ is marked by the lowest values of CIN, CIN south of the ITCZ is significantly smaller than north of the ITCZ.Together with the results for precipitation and column water vapor, this suggests that the atmospheric conditions north of y north are comparatively stable (high CIN, low CAPE), the conditions between y south and y north are the most unstable and thus favorable for convection (low CIN, high CAPE), and the conditions south of y south are more neutral (low CIN, low CAPE).
Whereas the single crossing measured during the ship campaign appears to capture a characteristic state of the ITCZ for most variables, the CAPE values we calculated from the radiosondes are about a factor of five higher than the mean values shown in the middle column (note the changed y scale in the right column in Figure 9c).However, these values are at least approximately consistent with the ERA5 values measured along the ship track, as shown by the gray line in the right column of Figure 9c.Further investigation shows that CAPE mostly peaks at the northern edge, that is, near y north , with maximum values that often do not exceed 300 J⋅kg −1 , but that a rapid increase to CAPE values exceeding 2000 J⋅kg −1 occurs at irregular intervals (not shown).Similarly, though not quite as pronounced as for CAPE, we observe much higher values of CIN during the crossing.CIN (north of the Equator) is the only variable that shows a significant difference between ERA5 values measured along the ship track and our soundings.
The compositing of the zonal wind speed (Figure 9e) also identifies surface westerlies, which are largely confined to the latitudes between y south and y north .This corresponds well with the conceptual picture developed long ago, by Flohn (1951).
Compositing the surface air temperature (Figure 9f) and the sea-surface temperature (Figure 9g) demonstrates that the ITCZ is located in the region with the highest sea-surface temperature.This result corresponds well to the general idea that the location of the ITCZ, the formation of convection, and so on, are roughly determined by the local SST maximum.While the surface air temperature outside the ITCZ closely follows the sea-surface temperature, within the ITCZ the surface temperature shows a weak but clearly visible depression.Considering additional data collected during the ship campaign suggests that this depression might be related to the combined effect of cold pools, with single cold-pool events visible in the right column of Figure 9f.
Considering the surface pressure, Figure 9h, we find, as expected, that the ITCZ is marked by a weak depression in pressure with the absolute minimum at the northern edge of the ITCZ, that is, at y north .While the surface pressure decreases from south to north, we note that the gradient in pressure almost vanishes close to y south .While, on average, we also see a weak depression in the pressure measured F I G U R E 9 Left column: zonal mean fields of ERA5 or IMERG after shifting the latitudes (as for top row in Figure 5).Middle column: zonal mean fields of ERA5 after rescaling the latitudes (as for bottom row in Figure 5).Right column: corresponding rescaled values for the ship campaign, (black) on-board measurements and (gray) reanalysis data.The different fields are (a) surface precipitation rate, (b) total column water vapor, (c) CAPE, (d), CIN, (e) zonal wind speed, (f) surface air temperature, (g) sea-surface temperature, and (h) surface pressure. (a) during the ship campaign, these measurements are difficult to interpret through the noise of the semidiurnal tide.
Finally, we can compare these findings with the East Pacific ITCZ, as recently studied during the Organization of Tropical East Pacific Convection (OTREC) campaign from August-October 2019 (e.g., Fuchs-Stone et al., 2020).Huaman et al. (2022), analyzing the East Pacific ITCZ's circulation system using nine research flights, find two locations of surface meridional wind convergence, one at around 5 • N and one at around 8 • N.This description is reminiscent of the edge convergence case in this study.While their southern convergence peak aligns with our description, attributed to meridional wind-induced speed convergence, the northern convergence peak is, in their case, also due to speed convergence, contrasting with our analysis.Both studies show that deep convection and precipitation peak at the edge with the strongest convergence, which is the northern edge in the eastern Pacific and the southern edge in the eastern Atlantic.In addition, there is consistency in the relationship between precipitation and other meteorological variables.Our findings of precipitation and column water vapor peaking opposite to CAPE align with the anticorrelation observed during OTREC (Fuchs-Stone et al., 2020).This comparison, therefore, highlights both similarities and differences between the East Atlantic and East Pacific ITCZs.
To summarize, the locations of y south and y north , calculated solely based on the meridional velocity, are also important reference points for the atmospheric variables considered here and the following schematic description of the East Atlantic ITCZ in July and August arises.The ITCZ, not surprisingly, coincides with a region of high near-surface air temperatures, substantial column water vapor, and low surface pressure.Considering the structure more closely, however, we find interesting asymmetries.In particular, the atmosphere within the ITCZ is unstable, but stable north and neutral south of the ITCZ.In addition, the precipitation rate as well as column water vapor, both high within the ITCZ, peak at the southern edge of the ITCZ.As these results are limited to the months of July and August as well as the East Atlantic, we will now study how this picture of the ITCZ compares with that for other locations and seasons.

4.3
Seasonal and longitudinal dependence

4.3.1
Seasonal cycle of y south and y north Since the ITCZ moves north and south during the year, we expect the positions of y south and y north also to exhibit a seasonal cycle.This is indeed confirmed by the data, as demonstrated by Figure 10, which shows the mean positions of y south and y north over the course of the year for different longitudinal bands (so far, we have only considered the longitudinal band from 20 • -23 • W).We find that the latitudes of y south and y north are primarily a function of time of year, rather than longitude.While y south and y north shift by about 10 • during the course of the year, the width of the ITCZ (y north − y south ) remains remarkably  (a) (b) constant and only varies between about 3.5 • and 5.5 • .It is also apparent that the northward propagation of y south and y north is different from the corresponding southward propagation.The ITCZ jumps across the Equator, but more so in boreal spring than in austral spring.Also, the differences between the different longitudinal bands is most pronounced during the northward propagation and, while y north (almost) always remains north of the Equator, y south is located south of the Equator from December-April.
To indicate the location of the tropical rain belt in Figure 10, we also include the median latitudes of maximum precipitation.Comparison of the seasonal cycle of maximum precipitation rate and the location of y south and y north shows that, on the one hand, the location of convergence and the location of the tropical rain belt shift in phase but, on the other hand, the amplitude of the latitudinal propagation of the rain belt is considerably smaller.Throughout most of the year, the precipitation maximum tends to be located closer to the location of y north rather than y south .Only from about July-October do we find the precipitation maximum to occur between y south and y north or even closer to y south , as shown above in Figure 9.The seasonal dependence of the latitudinal profiles will be discussed further below.

Longitudinal dependence of latitudinal profiles
As already apparent from the seasonal cycle of maximum precipitation in Figure 10, the structure of the ITCZ changes during the course of the year.Comparing the structure of the ITCZ during July and August with the structure for different months throughout the year, we find that the structure of the ITCZ is most different from Figure 9 during January and February.In the following, we will therefore contrast the mean structure during these two periods (i.e., July/August versus January/February).To this end, we show the scaled versions (y scale ) of the dynamic and thermodynamic fields as a function of the different longitudinal bands for July/August in Figure 11 and January/February in Figure 12.Figures 11 and 12 show all dynamic and thermodynamic variables shown in Figures 5 and 9 and, in addition, near-surface Starting our discussion with the latitudinal dependence in Figure 11, we find that in summer our main conclusions for the East Atlantic ITCZ hold for the entire tropical Atlantic ITCZ.In fact, we find very little dependence on the longitudinal band considered for most variables, with hardly any difference in the meridional wind speed, divergence, convergence, precipitation, CWV, and OLR TOA .Thus, in July and August, the method introduced above detects an ITCZ with increased surface convergence at the edges of the tropical rain belt, throughout the tropical Atlantic.Moreover, considering the composite structure of OLR TOA shows that the deepest convection inside the tropical rain belt occurs close to y south .While we also find little systematic dependence on longitude for CIN, we note that the peak in CAPE close to the location of y north discussed above is much more pronounced in the latitudinal band of 20 • -23 • W than further west.For the latitudinal profile of near-surface vorticity, we note a very pronounced peak at the location of y north with a slight increase in maximum vorticity as we move towards the west.Pronounced systematic differences are found, however, for the zonal wind speed, surface pressure, and the two temperature variables.
Surface air temperature as well as sea-surface temperature increase towards the west, more so outside the ITCZ.This is not surprising, as it is a well known fact that the East Atlantic is marked by strong sea-surface temperature gradients, while the West Atlantic is marked by weak sea-surface temperature gradients.It is, however, surprising that, while the sea-surface temperature inside the ITCZ increases slightly towards the west, the air temperature inside the ITCZ is independent of longitude.Compared with the mainly symmetric longitudinal dependence of surface temperature, the longitudinal dependence of surface pressure differs north and south of the ITCZ.While the surface pressure south of y south decreases slightly toward the west, surface pressure also increases towards the west north of y north , which is not what one would expect if sea-surface temperatures were imprinting themselves on the surface pressure field.In addition, to the north of y north , easterlies strengthen toward the west, while within the ITCZ the westerlies vary little with longitude.
It is interesting to compare these results with the simple dynamical model of Lindzen and Nigam (1987), at least qualitatively.The key idea of Lindzen and Nigam (1987) is that a considerable fraction of the near-surface winds in the Tropics can be explained solely by SST-induced pressure gradients.In their model, SST temperature differences lead to differences in the boundary-layer temperature which, in turn, impacts the surface pressure.Assuming the meridional flow to be a balance between the pressure gradient force, the Coriolis force, and friction, they derive a set of equations for the horizontal wind speeds.Comparing the longitudinal dependence of surface pressure and surface temperature, we find that only the region south of the ITCZ changes as expected from Lindzen and Nigam (1987).North of the ITCZ, we see a simultaneous increase in surface pressure and surface temperature from east to west, which is clearly in contradiction to Lindzen and Nigam (1987).The winds, however, follow the pressure gradients as would be expected from geostrophy; this is most apparent north of y north , where longitudinal variations in zonal winds and pressure gradients are most pronounced.

Seasonal dependence of latitudinal profiles
In January and February, the structure of the ITCZ is most different from the boreal summer ITCZ just described.Some aspects of the structure of the boreal winter ITCZ shown in Figure 12 are the latitudinal mirror images of Figure 11.For example, precipitation rate, column water vapor, and OLR TOA now peak at the northern, rather than southern, edge of the ITCZ, close to y north , while CAPE now peaks close to y south for all longitudinal bands apart from the one extending from 20 • -23 • W. CIN values in boreal winter are also, on average, somewhat higher south of the ITCZ, though this difference is far less pronounced compared with the north-south CIN differences in boreal summer.There are also other aspects that are not simply mirrored in the latitudinal direction.Importantly, while the location of the Equator in summer is south of y south , the location of the Equator is now not north of y north but, on average, in between y south and y north .This is consistent with the fact that we no longer see westerly winds between y south and y north and there is a significantly reduced vorticity close to y north .As expected for the monsoon, analyzing the zonal wind field further, we find a pronounced signal for westerly winds only in July, August, and September (not shown).Also, while convergence strength at y south and y north is about equal in boreal summer, convergence is now much more pronounced at y north than at y south .The convergence peak at y south is, moreover, less sharply defined, with high convergence values towards the north.This more blurred-out convergence structure, with higher convergence north of y north in the west than the east, might explain why regions of high precipitation, column water vapor, and OLR TOA extend beyond the location of peak convergence and also their systematic increase from east to west, mainly north of y north .Even the northerly local maximum in surface air temperature, which closely coincides with the location of y north in Figure 11j, is somewhat displaced towards the north in Figure 12j, even though sea-surface temperatures appear to peak at y north .Finally, it is interesting to consider the spatial distribution of surface pressure.Figure 12i shows a low-pressure region which extends from y south to y north .Interestingly, we see basically no dependence on longitude, which, given the longitudinal dependence of sea-surface temperature, one would expect from Lindzen and Nigam (1987).Differences in SST, imprinted on surface pressure, therefore cannot explain the longitudinal dependence seen in the zonal wind speed.To summarize, the location of y south plays a role in the thermodynamic and dynamical fields we consider, although less so in boreal winter than in boreal summer.

DISCUSSION AND CONCLUSION
In this study, we peer into the inner life of the Atlantic ITCZ by combining observational data collected during a north-south crossing of the Atlantic ITCZ in summer with multiyear reanalysis data.We set out to answer the question of whether convergence in the Atlantic ITCZ is generally characterized by one line of convergence about which a broader rain band is centered, or whether the rain band instead straddles two edge-intensified bands of convergence.We detect both cases but find that, most of the time, the ITCZ has two lines of convergence: a speed convergence line in the south (y south ) caused by deceleration of the southerly trade winds and a confluence line in the north (y north ), where southerly and northerly trade winds meet.
This study focuses primarily on the ITCZ in the eastern Atlantic in July and August.During this time of year and in this region, the mean locations of the speed convergence line and the confluence line are at about 7 • N and 11 • N in latitude, respectively, marking the southern and northern edges of the region of intense precipitation.We, therefore, refer to the two convergence lines as the southern and northern edges of the ITCZ.The edges of the ITCZ as defined also describe locations where significant changes in other variables related to convection occur.Not surprisingly, the ITCZ coincides with regions of low surface pressure and high sea-surface temperature, with an atmosphere that is generally more unstable to convection inside the ITCZ than outside.As has been noted before, the inside of the boreal summer ITCZ is marked by westerly winds.What is striking, however, is that the structure of the ITCZ is not symmetric.This is true within the ITCZ as well as when contrasting the regions north and south of the ITCZ.Away from the ITCZ, the atmosphere is stable to convection north of the ITCZ and more neutrally stratified to the south.Inside the ITCZ, we find, for example, that precipitation rate and column water vapor, both generally high within the ITCZ, actually peak at the southern edge of the ITCZ during July and August.Comparison with the top of the atmosphere outgoing longwave radiation suggests that this is also the location of the deepest convection.These results show little longitudinal dependence and are summarized in a tentative sketch of the Atlantic ITCZ in boreal summer shown in Figure 13a.Comparison with a sketch of the East Pacific ITCZ based on the OTREC campaign (August-October 2019) shows striking similarities, even though precipitation and column water vapor peak at the northern, rather than southern, edge of the ITCZ (see fig. 4d in Huaman et al., 2022).
While the properties of the Atlantic ITCZ considered here change only slightly with longitude, we find significant changes with season.In particular, the relative importance of the two convergence lines changes throughout the year.While we find that the positions of the convergence lines shift in phase to the position of the tropical rain belt, we find that the precipitation maximum is closer to the location of the confluence, that is, the northern edge of the ITCZ, rather than the location of speed convergence, that is, the southern edge of the ITCZ, for most of the year.Our results for the Atlantic winter ITCZ are summarized in Figure 13b.This figure makes it apparent that the structure of the ITCZ in January-February (boreal winter) is not merely a latitudinal reflection of the ITCZ in summer.First of all, the Equator is now mostly located between the speed convergence and the confluence line, in line with the absence of westerly surface winds.Also, and importantly, the low-level convergence is much more pronounced at the location of confluence compared with the location of speed convergence.This is consistent with the finding that the location of speed convergence plays a less important role for the other dynamic and thermodynamic variables considered.Although the entire ITCZ still marks a region characterized by significant atmospheric instability, sea-surface temperature and precipitation rates now peak around the confluence line.These results might indicate the existence of additional convergence lines north of the confluence line in boreal winter when conditions are more symmetric with respect to the Equator.The properties of the boreal winter ITCZ will, therefore, be investigated in more detail in a future study using observational data we collected during the ARC research campaign in January and February 2023 aboard the German RV Maria S. Merian.
Comparison with top of the atmosphere outgoing longwave radiation indicates that the peak in precipitation marks the location of deepest convection, suggesting that the vertical structure of convection differs between the two edges.As the vertical structure of convection or vertical motions may be highly biased in reanalyses (e.g., Huaman et al., 2022), a detailed investigation of the vertical structure of the atmosphere at the edges, potentially relevant for understanding why precipitation appears to favor the winter side of the ITCZ, is beyond the scope of the current study.To investigate the vertical profiles of deep convection and, more generally, the atmospheric state in the region of the Atlantic ITCZ is one of the key goals of the campaign Organized Convection Experiments in the tropical Atlantic (ORCESTRA), which will take place in the tropical Atlantic in August and September 2024.
Considering the importance of the Atlantic ITCZ for the Atlantic climate, and perhaps other regions as well, surprisingly little is known about its structure or inner life.However, this study shows that the Atlantic ITCZ in fact has a rich, and surprisingly asymmetric, dynamic and thermodynamic structure.This work raises more questions than it answers; we hope that it will open a door for further research that will lead us to understand better whether processes regulating the inner life of the ITCZ might hold the key to a better understanding of its outer life.

ACKNOWLEDGEMENTS
This work was supported by the Deutsche Bundesministerium für Bildung und Forschung (BMBF) through the support of the RV Sonne cruise no.SO284.For support during the cruise, the authors are grateful to Peter Brandt and the GEOMAR Helmholtz Centre for Ocean Research Kiel as well as Tilo Birnbaum and his crew.The authors also acknowledge the OCEANET-Atmosphere team from the Leibniz Institute for Tropospheric Research (TROPOS), in particular Ronny Engelmann and Annett Skupin, for the Parsivel 2 disdrometer measurements.
The entire radiosonde data set collected during the RV Sonne cruise no.SO284, including the data used in this study, has been published by Schulz et al. (2022) and is publicly available at https://zenodo.org/record/7051674#.Y6wxduzMI-Q.Hersbach et al. (2018) was downloaded from the Copernicus Climate Change Service (C3S) Climate Data Store.The results contain modified Copernicus Climate Change Service information from 2021.Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.
We also acknowledge the use of imagery from the NASA Worldview application (https://worldview .earthdata.nasa.gov),part of the NASA Earth Observing System Data and Information System (EOSDIS).
Finally, the authors thank the anonymous reviewers and the associate editor, Dominique Bouniol, for their helpful comments, which significantly improved the quality of the article.Open Access funding enabled and organized by Projekt DEAL.

1
Illustration of (a) central convergence and (b) edge convergence with respect to the rain belt in the Atlantic summer ITCZ with the location of the Equator indicated by the dashed gray lines.Convergence at y north results from the confluence of the trade winds and that at y south from speed convergence due to an abrupt slackening of the southerly trade winds.In the central convergence case, the southerly trade winds increase in wind speed up to the latitude of confluence (y south = y north ).In the edge convergence case, y south and y north are well separated by a region of decreased wind speed.Top row: schematic top-down perspective of surface winds, convergence, and precipitation.The gray arrows represent surface winds, the blue lines indicate the location of strongest convergence, and the shaded gray areas indicate the region of high precipitation rates.Bottom row: schematic of meridional wind speeds.The regions around the locations of (infinitely) strong convergence,  y v = −∞, at y south and y north are highlighted in blue.
Meridional near-surface wind speed during the time of the ITCZ crossing (2021-07-09T00:00:00Z until 2021-07-13T00:00:00Z).(a) meridional velocities along the ship track shown as a function of latitude for the on-board measurements (75-min averages; black) and the reanalysis data (0.25 • resolution; gray).The two horizontal lines denote the locations of y south at 6.1 • N and y north at 9.2 • N based on the on-board measurements.(b) The time evolution of the zonally averaged (20 • -23 • W) meridional velocities from reanalysis data (background colors) as well as the location of the RV Sonne during the launching times of the radiosondes (scatter points) and the on-board wind measurements (colors of scatter points).
The zonal mean (20 • -23 • W) 10-m divergence and (a) meridional wind field and (b) hourly averaged precipitation rate for July and August, calculated from 20 years of ERA5.
Histograms of (a) the latitudes of y south and y north and (b) the widths (calculated as y north − y south ) based on the reanalysis data for the months of July and August for the years 2001-2021 (20 • -23 • W).The vertical lines indicate the corresponding latitudes of the on-board measurements.
Zonal mean, 20 • -23 • W, of the ERA5 (a) meridional wind speed at 10 m, (b) divergence at 10 m, (c) divergence (<0 only) at 10 m, and (d) precipitation rate.For the top row, the latitudes at each instance in time are shifted before calculating the zonal mean such that the location of y south corresponds to y shift = 0.The different colors indicate the range of widths considered: [w, w + Δw], corresponding to the (10th-15th), (35th-40th), (60th-65th), (85th-90th) percentiles, each with a sample size of roughly 1000 instances.For the bottom row, latitudes are rescaled such that the locations of y south and y north correspond to y scale = 0 and y scale = 1.The range of widths considered is [2.0 • , 6.5 • ], corresponding to the 10th and 80th percentiles, and the sample size is approximately 20.000.The vertical gray dashed line indicates the location of the Equator, calculated as the median value of (top row) −y south and (bottom row) −y south ∕w ITCZ .
Satellite images of the tropical Atlantic from GOES-16.Both images were taken during the time period of the ship campaign, with the location of the ship indicated in yellow (date and time are shown in the title).Image (a) was taken at the time of entering the ITCZ and (b) was taken when leaving the ITCZ.
a) Zonal mean (20 • W to 23 • W) of the July and August ERA5 meridional velocity conditioned on y south and y north (solid line) and on the locations of peak convergence (dashed line) for w in the range of the 60th to 65th percentile.(b) Northern and southern latitudes of the two most prominent convergence features for all instances when at least two peaks are detected (solid lines).For comparison, the locations of y south and y north are shown in shading.Zonal mean (20 • -23 • W) of the July and August ERA5 (a) column water vapor and (b) divergence at 10 m.As in the top row of Figure 5, the latitudes at each instance in time are shifted before calculating the zonal mean, but here with respect to the latitude of the southern moist margin (column water vapor equal to 50 mm).The different colors indicate the range of widths considered: [w, w + Δw].The crosses indicate the latitudes of peak convergence.
10 Seasonal cycle of the ITCZ.Median latitudes of y south (solid) and y north (dotted) are shown for different longitudinal bands (different colors).The gray crosses show the median location of the precipitation maximum (IMERG) across all longitudinal bands (gray horizontal markers) and the maximum and minimum locations among the bands considered (gray vertical markers).

F
August after rescaling the latitudes as in the bottom row of Figure 5.The different colors indicate the different longitudinal bands over which the zonal mean is calculated.The fields are (a) 10-m meridional wind speed, (b) 10-m divergence, (c) 10-m divergence (<0 only), (d) surface precipitation rate, (e) total column water vapor, (f) CAPE, (g) CIN, (h) 10-m zonal wind speed, (i) 10-m vorticity, (j) temperature of the (solid) air at 2 m and (dashed) the sea surface, (k) top of the atmosphere outgoing longwave radiation, (l) surface pressure.

Figure
Figure11but for January and February.

F
I G U R E 13 Latitudinal sketch of the Atlantic ITCZ for (a) boreal summer and (b) boreal winter.The arrows indicate meridional wind speed and direction, crosses indicate easterly wind speed, and dots indicate westerly wind speed.Colours indicate sea-surface temperatures from cold (dark blue) to hot (dark red).Atmospheric stability based on CAPE and CIN values is indicated as either stable, neutral, or unstable.Clouds indicate the location of the strongest surface precipitation rate and cloud height is estimated based on outgoing longwave radiation.