The ERA5 variables are tri-linearly interpolated to a vertical profile (1000–300 hPa, every 7 hPa; Fig. a) at the geographical location of the BCO every hour in the EUREC4A period (16 January to 20 February 2020) and every third hour in the quasi-climatological period (January and February 2010 to 2020). For the case study analyses (Sects.  and ), we average or sum the variables over each day and two vertical layers, namely a lower-tropospheric (1000–650 hPa) and a mid-tropospheric (650–300 hPa) layer. The 24 h time window is chosen because the associated large-scale conditions typically remain similar within 1 d (Fig. ). Moreover, the cloud radiative effect should be assessed combining daytime and nighttime conditions. The two vertical layers are defined such that both contain 50 vertical data points and are motivated by the fact that the maximal vertical extent of the low-level cloud layer during the campaign is at about 650 hPa (Fig. ). This simple definition of temporal and vertical boundaries to obtain a summarizing statistics is also applied at the interannual timescale to investigate the robustness of the identified links between the large-scale circulation and the local conditions. We repeated the above-described steps for four additional points shifted zonally and meridionally from the BCO by 0.5∘ and found that the results for the five locations differ only marginally (not shown). We therefore decide to focus in this study on the profiles directly above the BCO only. The following variables are computed with ERA5 data and partly also with measurements.

Integrated water vapor (IWV). We calculate the IWV for the two layers (IWV1000–650 hPa, IWV650–300 hPa) using the equation IWVpbottom-ptop[m]=1gρl∫pbottomptopqdp, with specific humidity (q), density of liquid water (ρl), pressure (p), and the gravitational acceleration (g). We derive the IWV from ERA5 and the radiosonde measurements . Daily mean IWV values are used in both case studies (Sects.  and ).

Total ice (liquid) water. For ERA5, total ice (liquid) water is calculated by adding total column cloud ice water and the vertically integrated specific snow (rain) water content, which is obtained following the procedure for IWV. ERA5's total ice (liquid) water is compared to CERES' ice (liquid) water path, in the following also referred to as total ice (liquid) water. Total ice water is used in the TMD case study (Sect.  and in Supplement 1 together with total liquid water).

Precipitation totals. ERA5 daily precipitation totals are obtained from adding up convective and large-scale precipitation. Daily precipitation totals are also retrieved from Vaisala WXT-520 measurements and are used in the overview of the EUREC4A period (Sect. ) and the EDI case study (Sect. ).

Net cloud radiative effect (CRE). Following and , we calculate the net cloud radiative effect (CRE) as the sum of the longwave (LWCRE) and shortwave (SWCRE) cloud radiative effect. The LWCRE results from the difference of the top-of-atmosphere thermal radiation under clear-sky conditions and the one under cloudy conditions (clear-sky minus cloudy) and analogously for the SWCRE and solar radiation. Using this convention, negative values represent a radiative cooling due to the presence of clouds, and positive values represent a warming. As the cloud radiative forcing is very inhomogeneous in space (see Figs. S1.7–S1.8), an area-weighted mean over the domain 10–20∘ N, 50–60∘ W is calculated. The domain's size is chosen similar to , who used 10–20∘ N, 48–58∘ W, but here the borders are shifted to the west to include Barbados. The domain-mean CRE allows a comparison with the results from and is used in both case studies (Sects.  and ). The CRE was also derived from the satellite product CERES .

Three-dimensional backward trajectories are computed with the Lagrangian analysis tool LAGRANTO , starting from the interpolation points defined above (Fig. b). The calculations are based on the ERA5 wind fields and determine the air parcels' position (longitude, latitude, pressure) during the 10 d before arrival. A set of variables is interpolated along the trajectories, namely, specific humidity (absolute values and anomalies relative to the three-dimensional campaign mean field), relative humidity, liquid/rain/snow/ice water content, and surface evaporation. We extract each air parcel's position 4 d before arrival (Fig. ; see Fig. S1.1 for other time steps) to get an impression of the pathway variability across the vertical profile and during the EUREC4A time period. The choice of 4 d is inspired by the study of , who showed that the transport during this pre-arrival time window is essential for understanding moisture anomalies induced by the large-scale circulation in the trades. Typically, multiple airstreams (coherent bundles of trajectories with similar characteristics) arrive in each of the two layers during 24 h. For each airstream, defined in Sects.  and , we compute the mean specific humidity (per air parcel) and the contribution to the IWV of the arrival layer (Tables  and ).

The findings of this study are mainly based on ERA5 reanalyses; therefore, some words about their validity are needed. Several comparisons between ERA5 and measurements of the local conditions in Barbados, made in the context of the two case studies (e.g., Figs. , , , and S1.3–S1.8), show that ERA5 reproduces the variability fairly well (including the timing of local extremes) but differs in terms of absolute values. ERA5 underestimates IWV, especially in the lower troposphere, and has difficulties in capturing the extremely high values of specific humidity observed by the BCO radiosondes , e.g., from 22 to 24 January 2020, in both layers (Fig. a, b). This lower-tropospheric dry bias of ERA5 has been noticed previously . Besides this systematic bias, the otherwise good agreement of IWV variability does not come as a surprise, as the soundings from the nearby Grantley Adams International Airport of Barbados (but not from the BCO) are assimilated in ERA5 .

Precipitation totals (Figs. c, b) are underestimated by ERA5 compared to the in situ measurements of the meteorological station WXT-520, which is most probably an effect of ERA5’s spatial and temporal resolution. While the reanalysis yields an average over a model grid box at every hour, the WXT-520 provides a point measurement every 10 s. A direct comparison is therefore difficult. Similarly, ERA5 produces total ice water values that are too low (Figs. d, b) with CERES as a reference. Note, however, that most of the time CERES total ice water information in the vicinity of Barbados is missing and only spatially interpolated data are available (red dots in Fig. d). Thus, the explanatory power of CERES total ice water shown in Fig.  is very limited, and it is hard to judge which of the two data sets is closer to real conditions.

On the hourly timescale, ERA5's underestimation of atmospheric ice in mid- to high-level clouds which potentially reduce the cooling effect of low-level clouds; see results in an underestimation of the cloud radiative warming during local nighttime (see, e.g., 14 February in Fig. d, Supplement 1) compared to CERES. During local daytime, ERA5 often yields a stronger cloud radiative cooling, which might be the result of low-level cloud fractions that are too high (as observed, at least, in one case; see Fig. S1.7). Eventually, these differences lead to deviating daily mean CRE for ERA5 and CERES (Figs. , , Supplement 1). Important to note, however, is that time steps with strong anomalies in ERA5 are also identified as anomalies in CERES, e.g., on 22 January and 14 February 2020. Moreover, based on the observed differences in absolute values, we assume that the anomalies of IWV, precipitation, total ice water, and (on 22 January) the cloud radiative cooling, discussed in the two case studies, were stronger in reality than indicated by ERA5.

Even harder to evaluate is the accuracy of the backward trajectories calculated with ERA5 wind fields. As a first step, we compared the horizontal ERA5 wind components to the ones measured by the BCO radiosondes (Fig. S1.4), which showed that the two data sets agree in terms of variability and absolute values. To assess the vertical velocities, we compared the mean pressure velocities of the trajectories during the last 24 h before arrival (i.e., the period when the trajectories are already in the vicinity of Barbados) to measurements valid over a circular area with a diameter of ∼175 km; from previous field campaigns (Fig. S1.5). The pressure velocities derived from the trajectories are mostly within the range of roughly -5 to 5 hPah-1 determined by their Fig. 3 using dropsonde measurements. Moreover, we know from previous studies that ERA5 agrees reasonably well with dropsonde-derived divergence measurements, which directly link to vertical motion . Thus, we are confident that the ERA5 wind fields near Barbados are close to the real conditions and therefore lead to representative trajectories. The informative value of ERA5 trajectories was recently also demonstrated by , who showed that ERA5 trajectories are physically more consistent (e.g., improved potential temperature conservation along trajectories in the stratosphere) than ERA-Interim trajectories, most probably due to a higher temporal resolution of the wind fields. Finally, could explain measured variations of stable water isotope signals in Barbados with ERA5 backward trajectories, which supports the usefulness of the trajectories.

On 22 January 2020, a trailing cold front, visible as a cloud band loosely connected to a cyclone over the subtropical North Atlantic (Fig. a), reached Barbados. The resulting mesoscale cloud organization in the vicinity of the island can be described as cloud bands typical for the fish cloud pattern following the classification of . The fish cloud broke up into smaller fragments the day after its passage over Barbados, while traveling further into the tropics (not shown). On 22 January 2020, a saturated layer developed in the early morning near 800 hPa, which grew vertically up to 600 hPa at the expense of the dry air aloft, as can be seen in the profiles from the balloon soundings (Fig. b) and the cloud radar's measurements (Fig. c). The precipitating humid layer persisted until shortly before local midday (confirmed by personal observation on-site).

The lower-tropospheric moist layer, formed by the passage of the cold front on 22 January 2020, can be clearly identified as strongly anomalous compared to the rest of the campaign in the daily mean IWV1000–650 hPa (Fig. a). The cold front led to prolonged precipitation periods in Barbados such that the precipitation total on 22 January 2020 is also strongly anomalous compared to the rest of the campaign (Fig. b). This is consistent with the study by , who showed that fish clouds are associated with high rain amounts but not necessarily high rain rates. Both anomalies, IWV1000–650 hPa and precipitation totals, are more pronounced in the measurements than in ERA5 (Fig. a, b).

The fish cloud had a net cooling effect over Barbados in the daily mean (Fig. c), leading to strongly negative CRE values of about -38 Wm-2 in ERA5 and -47 Wm-2 in CERES on 22 January 2020. This reflects that the cooling shortwave effect during daytime dominated the continuous but comparably weak warming longwave effect (Fig. ). Note that this effect is spatially confined to the cloud band (Fig. S1.7). If averaged over a 10∘ × 10∘ domain (10–20∘ N, 60–50∘ W; red box in Fig. a), the CRE on 22 January 2020 reduces to values of about -27 Wm-2 in ERA5 and -8 Wm-2 in CERES, which is still anomalously negative compared to the rest of the campaign according to ERA5 but not according to CERES (Fig. c). Looking at the spatial distribution of the CRE in the considered domain (see exemplary time step in Fig. S1.7), we note that ERA5 overestimates the presence of liquid/low-level clouds compared to CERES, leading to a stronger cloud radiative cooling in ERA5. Generally, ERA5 shows 10 to 20 Wm-2 more negative net cooling over the 10∘ × 10∘ domain compared to CERES or the satellite-based study of their Fig. 5. Deviations of a similar magnitude have been found in other studies comparing the CRE derived from reanalysis data to satellite-based products their Figs. 3 and 4. In the next subsection, we discuss to what extent the cold front's movement into the tropics and the associated EDI contribute to the above-described anomalous local conditions.

The trajectories leading into the anomalously moist layer (1000–650 hPa) over the BCO on 22 January 2020 (Fig. ) can be bundled into five coherent airstreams. To assess these airstreams, we first identify the trajectories with a strong descent and assign them to three EDI airstreams (the leading L in the airstreams' names stands for the arrival in the lower troposphere):

L-EDIwcb – trajectories with a strong EDI descent (≥400hPa(48h)-1), which also experienced a warm conveyor belt (WCB)-like ascent ≥600hPa(48h)-1; following from the lowermost to the upper troposphere prior to the EDI descent, identified during the 10 d before arrival;

Overall, the transport into Barbados' lower troposphere on 22 January 2020 was influenced by an extratropical upper-level Rossby wave breaking over the central North Atlantic roughly 4 d earlier (Figs.  and S2.4 and animation in Supplement 3), associated with an upper-level PV streamer (red contour in Fig. ) and an intense surface cyclone underneath black contour in Fig. , using the detection method of. A region of dynamically driven large-scale subsidence is found upstream of the PV streamer (Fig. a–c), as expected from quasi-geostrophic theory . From 17–19 January 2020, the three L-EDI airstreams (L-EDIwcb, L-EDIs, and L-EDIw; brown upward-/downward-facing triangles and rhombi in Fig. ; brown lines in Fig. ) were steered into this region at upper levels (p∼ 500 hPa) and rapidly descended into the cold sector of the extratropical cyclone (Fig. a), where they met the L-LT airstream (green left-facing triangles in Fig. a).

The descent caused a massive decrease in relative humidity along the trajectories (Fig. c), as the L-EDI air parcels were adiabatically warmed while staying dry (Fig. f), reflecting limited mixing with the environment. Simultaneously, ocean evaporation beneath the L-EDI air parcels was reinforced (blue contours in Figs. b; d). On 20 January 2020 the EDI-like descent was completed and the L-EDI penetrated the cloud layer, which is marked by a sudden increase in relative humidity (Fig. c) and liquid water content (Fig. e). Once the airstream arrived in the sub-cloud layer, the specific humidity continued to increase (Fig. f), while the liquid water content decreased (Fig. e), which likely contributed to the L-EDI air parcels' moistening through hydrometeor evaporation. The enhanced surface evaporation starting with the EDI's descent resulting from the advection of dry unsaturated air over the ocean surface, e.g., and the subsequent increase in specific humidity approximately 24 h later is typical for EDIs over the North Atlantic their Figs. 8 and 11.

The L-LT air parcels (remaining in the lower-tropospheric layer during the 5 d before arrival) moved towards lower latitudes along the western edge of the surface cyclone (Fig. a) while slightly descending (green line in Fig. a). They experienced a gradual increase in relative (green line in Fig. c) and specific humidity (Fig. f), which set in earlier than for the L-EDI airstreams. On 19–20 January 2020, the L-LT's moistening (Fig. f) temporarily flattened due to mixing with the L-EDI airstreams. At the same time, the PV streamer broke up and a PV cutoff formed northeast of Barbados (Fig. c–d). The low-level cyclonic circulation induced by the PV cutoff pushed the L-LT and the L-EDI air parcels (located west of the cutoff) southwards (Fig. d). While further descending, the L-LT and L-EDI air parcels caught up with the cold front. The cold front is identifiable in a region with a strong horizontal gradient in equivalent potential temperature (Fig. S2.4) behind a line of high total column water (Fig. b–d), which is associated with low-level convergence and precipitation (Fig. S2.5). At roughly 20∘ N, the L-LT and L-EDI air parcels entered the trade-wind region and eventually reached Barbados at low levels (1000–650 hPa).

The remaining L-MT air parcels (slowly descending from above 650 hPa) moved predominantly within tropical latitudes (purple line in Fig. b) and descended moderately (Fig. a), following a typical Hadley-type transport pathway. This airstream arrived at the top of the lower troposphere in the region of the inversion and most likely mixed only marginally with the cloud layer air, as shown by the low values in liquid water content (Fig. e) and specific humidity (Fig. f).

In summary, 45  % of the 1200 air parcels arriving between 1000 and 650 hPa on 22 January 2020 belong to the L-EDI airstreams (Table ), contributing to 58 % of the IWV1000–650 hPa. The L-LT contains 20 % of the trajectories and makes up 26 % of the IWV1000–650 hPa (Table ). Lastly, L-MT holds 35 % of the trajectories, which contributes 16 % of the IWV1000–650 hPa (Table ). Thus, we conclude that the EDI air parcels were the key players in the formation of the low-tropospheric moist anomaly on 22 January 2020.

In the following, we assess how frequently EDIs occur in general and how often they are related to local conditions similar to the ones on 22 January 2020. For this, we extract all EDI days from January and February 2010–2020 (quasi-climatological period). The remaining days are referred to as non-EDI days. An EDI day is identified if at least 5% of the trajectories arriving from 00:00 to 23:00 UTC between 1000 and 650 hPa over the BCO descended 400 hPa (48 h)-1 within the 4 d prior to arrival. The considered time span of 4 d ensures that only EDIs subsiding over the North Atlantic (corresponding to the extratropical dry intrusion regime in , their Fig. 3a) and not over Europe/Africa (corresponding to the extratropical trade-wind regime in , their Fig. 3b) are selected. Typically, air parcels from the extratropical trade-wind regime require about 6 d to traverse the North Atlantic. The selection criterion is subjectively chosen but ensures that the well-documented EDI cases on 22 January 2020 and on 31 January 2018

Here we only identify the onset of this event due to using a slightly different EDI selection criterion and a smaller set of trajectories. Note that the results' sensitivity to the EDI selection criteria are addressed in Fig. S1.2.

Applying this criterion, we find that EDIs arriving at Barbados are comparatively rare. Only 44 (∼ 7 %) of the total 652 d in the quasi-climatological period are identified as EDI days (Table ). The local conditions at the BCO on EDI days range from extremely dry with no precipitation to extremely moist with high precipitation totals and varying CRE (Fig. ). In the considered period, EDIs occur on 6 (18 %) of the 33 driest days according to IWV1000–650 hPa, on 8 (24 %) of the 33 wettest days according to total precipitation, and on 7 (21 %) of the 33 days with the lowest CRE. Other (presumably local) processes are responsible for the rest of the three variables' extremes. Thus, even though EDIs arriving in Barbados are relatively rare, they can trigger extremes of either sign in IWV1000–650 hPa, and they occur in connection with heavy precipitation and anomalously low CRE more frequently than their climatological occurrence frequency indicates.

On EDI days, the two variables, IWV1000–650 hPa and precipitation totals, are more strongly positively correlated than on non-EDI days (Table ). The relation between CRE and precipitation or IWV1000–650 hPa is only marginally stronger on EDI days than on non-EDI days. Nevertheless, we observe a shift towards lower daily mean CRE values on EDI days (with a mean of -28 Wm-2) compared to non-EDI days (with a mean of -18 Wm-2). In particular, the EDIs with extremely high IWV1000–650 hPa and daily precipitation totals (in the upper right corner of Fig. a, b) consistently have an enhanced cloud radiative cooling. Two processes in connection with EDIs contribute to a strengthening of the cloud radiative cooling. Due to the EDI's subsidence, we expect an anomalously dry free troposphere (no mid-/upper-level clouds), which strengthens the longwave radiative cooling and explains low CREs on days with a dry lower troposphere. Additionally, the EDI can be associated with a fish cloud (as on 22 January 2020), which strengthens the shortwave radiative cooling and leads to low CREs on days with a moist lower troposphere. Anomalously high CREs on EDI days are most likely caused by other processes higher up in the atmosphere such as cirrus clouds, which affect the CRE but have no impact on the low-level variables. These overall stronger links between precipitation, IWV1000–650 hPa, and CRE emerge due to the large-scale nature of the perturbations induced by EDIs.

We assume that the difference between the extremely moist and wet EDIs (in the upper right corner of Fig. a, b) and the remaining EDIs arises from mesoscale details of how the EDIs modulate the dynamical properties of the associated trailing cold fronts, which can be best described by the low-level divergence in the region of Barbados (Fig. b). As we know from theory, a frontogenetic front is characterized by a strong dipole in low-level divergence associated with the frontal ageostrophic circulation. Behind the cold front, namely in the cold sector in which the EDI spreads out, low-level divergence dominates (see also Supplement 2 for the horizontal and vertical sections across the horizontal divergence field associated with a cold front). Thus, a strongly negative daily mean divergence (i.e., convergence) on an EDI day possibly indicates that Barbados was heavily influenced by the front's warm side, while a strongly positive value hints towards a prolonged influence of the front's cold side and the following cold sector. To test this hypothesis, we study the transport history of the five EDIs with the most negative and the most positive low-level divergence, referred to as EDIcon and EDIdiv, respectively (Fig. , listed in Table ) and the synoptic situation of the two most extreme cases (Fig. ).

EDIcon trajectories descend strongly (Fig. a), reach cloud tops 2–3 d before arrival (peak in liquid water content, Fig. c), dive into the boundary layer, and experience an important increase in liquid water content when passing through the front and ascending again on the warm side 1 d before arrival (see exemplary case in Figs. S2.4–S2.5). Due to their closeness to the surface, EDIcon trajectories efficiently trigger ocean evaporation (Fig. b), which contributes to a rapid increase in specific humidity (Fig. d). Unlike EDIcon, EDIdiv trajectories show a slowdown of the descent when reaching ∼ 800 hPa (Fig. a; see exemplary case in Figs. S2.6–S2.7). From then on liquid water content remains high (Fig. c), suggesting that the trajectories stay in clouds and do not reach the sub-cloud layer. Consequently, EDIdiv trajectories are associated with lower values of surface evaporation (Fig. b) and specific humidity (Fig. d) than EDIcon. The synoptic situations on the EDI day with the lowest and highest low-level divergence (and two additional cases discussed) confirm the above-described mechanisms.

Strongest EDIcon event. An event in the EDIcon category with strong low-level convergence ahead of a frontogenetic cold front occurred on 27 January 2010 with the following characteristics: divergence = -15.8×10-6 s-1, IWV1000–650 hPa = 33.8 mm, precipitation = 3.2 mm, and CRE = -29.8 Wm-2. At 12:00 UTC on 27 January 2010, Barbados was located on the warm side of the cold front in a region of strong low-level convergence (Fig. a). The thick red contours in Fig. a show ERA-Interim fronts identified with a horizontal gradient in equivalent potential temperature at 850 hPa larger than 3.5 K(100km)-1 . The EDI air parcels in this case moved from the cold sector through the front, most likely due to the ageostrophic circulation associated with the front. This idea is supported by the Q vectors (arrows in Fig. a) which point from low to high equivalent potential temperature near the front, indicating frontogenesis, as expected (see Sect. 1) when a deformation flow (the EDI) acts on a horizontal temperature gradient (the front). Consequently, the ageostrophic circulation at the front is upheld shown by the neighboring bands of low-level divergence and convergence in Fig. b; note that a different pressure level is shown than in Fig. a to enable a direct comparison to while it passes Barbados. The dynamics of the front directly impacts the observed local conditions. The low-level convergence leads to an accumulation of moisture, increasing the IWV1000–650 hPa value. The ascending motion on the front's warm side promotes cloud formation, reducing the shortwave radiative input and potentially initiating precipitation. Strong convergence at 950 hPa was previously linked to the fish cloud pattern and thus to strong cloud radiative cooling . Indeed, we find a distinct cloud band directly over/close to Barbados on the 5 considered EDIcon days (exemplary cases in Fig. S2.1).

Strongest EDIdiv event. An event in the EDIdiv category with low-level divergence behind the cold front occurred on 3 February 2013 with the following characteristics: divergence = 22.4×10-6 s-1, IWV1000–650 hPa = 19.2 mm, precipitation = 0 mm, and CRE = -0.8 Wm-2. At 06:00 UTC on 3 February 2013 (Fig. c), the island was located in the cold sector behind a cold front (Fig. c) in an environment of strong low-level divergence (Fig. d), leading to the exact opposite characteristics than on 27 January 2010, i.e., low IWV1000–650 hPa, cloud-free conditions, and no precipitation. However, it would be misleading to relate EDIs arriving together with the cold sector exclusively to cloud-free conditions. Often the cold sector features typical open or closed convection cells e.g.,exemplary cases in Fig. S2.2, which might further explain the large variability of CRE on EDI days.

The main difference between events in the EDIcon and EDIdiv categories is the relationship between the cold front and the EDI. In the first case, the EDI overtakes the cold front by entering the boundary layer on the cold front's cold side and subsequently ascending on the front's warm side. In the second case, the front and the EDI are spatially and temporally separated, with the front propagating faster than the EDI and the EDI entering the boundary layer only when arriving in Barbados. However, the fact that the EDI catches up with the cold front is not sufficient for wet conditions to arise. We found at least one case on 31 January 2018 discussed in, when the EDI air parcels arrive in Barbados together with the warm air ahead of the cold front, but the distinct dipole structure of low-level divergence at the front had already dissolved (front in a frontolytic state). A plausible cause for these contrasting situations is the upper-level forcing and the corresponding strength of the surface cyclone which is notably stronger for EDIcon days than for EDIdiv (Figs. , S2.3) and therefore supports the stronger deformation and frontogenesis.

In at least two of the analyzed cases (Figs. S2.3–S2.5), the trailing cold front in the subtropics (independent on whether they fall into the EDIcon or EDIdiv category) adopts characteristics similar to katafronts (see introduction). The warm and moist subtropical boundary layer air does not ascend along the backward tilted frontal surface (as typical for anafronts) but remains ahead of the surface cold front. As a result, precipitation (if present) falls in a narrow band (the so-called fish cloud) along the surface cold front.

Overall, we demonstrated that EDIs descending from the midlatitude jet stream region into low latitudes disturb the trade-wind region's environmental conditions such as CRE, precipitation, and IWV1000–650 hPa. If the EDI catches up with the cold front, this can lead to a low-tropospheric moist anomaly, the occurrence of the fish cloud pattern, and enhanced radiative cooling. Due to their large-scale flow forcing, EDIs are good candidates for studying the cloud–circulation coupling in detail also in coarse-resolution (0.5–1∘ horizontal grid spacing) climate simulations.

On 14 February 2020, Barbados was covered by a mid-tropospheric shelf cloud detrained from cumulonimbi over South America (shown in Sect. ) forming a large-scale nearly closed altocumulus layer extending from 5∘ N over South America northwards to about 18∘ N (Fig. a). The large-scale dimension of the mid-level cloud layer is emphasized in the Poldirad cross section (Fig. d), where it spans the full width of 120 km. The shelf cloud persisted throughout 14 and 15 February and dissipated on 16 February 2020 (not shown). The saturated layer associated with these mixed-phase clouds (Fig. b) was relatively deep and extended between the 0 ∘C isotherm near 600 and 300 hPa at -30 ∘C (Fig. b). A distinct dry layer separated the shelf cloud from the shallow trade-wind cumuli beneath (Fig. b–d). This dry layer was moistened through the evaporation of hydrometeors falling from the shelf cloud. In some cases the precipitation reached the surface, e.g., 20 km away from the Poldirad (Fig. d). In other cases it completely evaporated, as shown in the cloud radar image (Fig. c) in the form of downward-propagating reflectivity signals that, however, do not reach the shallow trade-wind cumuli. The evaporative cooling in the dry layer most probably helped to maintain a stable layer near the 0 ∘C isotherm as discussed in, e.g.,.

We assume that this moistening process created favorable conditions for the vertical growth of shallow convection beneath. Indeed, individual convective cells penetrating the dry layer from below are visible from the large radar reflectivity values in the two radar images, e.g., after 20:00 UTC above Barbados (Fig. c) and about an hour earlier between 80 and 85 km away from the Poldirad (Fig. d). In these situations enhanced shallow convection injected moisture upward into the shelf cloud, potentially contributing to the persistence and considerable northward extension of this layer. At least in two cases, at 08:00 and 14:00 UTC, such a convective cell pushing through the trade inversion was observed shortly after a moistening event of the dry layer by evaporation of hydrometeors falling out of the altocumulus layer (Fig. c). This temporal succession suggests that evaporatively driven downdrafts might have acted as an initiating process for more vigorous shallow convection at low levels e.g.,. However, we cannot distinguish between the temporal evolution and the advection of clouds (low-level easterlies and mid-level southerlies, Fig. b) in the images from the vertically pointing cloud radar (Fig. c). Thus, mid- and low-level features that appear in close temporal succession are not necessarily causally linked.

Above Barbados, the shelf cloud caused a positive anomaly in the daily mean of the IWV650–300 hPa (Fig. a) and the total ice water (Fig. b) compared to the remaining campaign days. Similar to the first case study, the anomalies are stronger in the measurements than in the reanalysis. Both data sets indicate an enhanced cloud radiative cooling above Barbados and over the 10∘ × 10∘ domain (Fig. c). The absolute CRE values (at the BCO -79 Wm-2 in ERA5 and -61 Wm-2 in CERES; over the domain -34 Wm-2 in ERA5 and -19 Wm-2 in CERES) are lower in ERA5 than in CERES. This observation fits well with the higher total ice water recorded by CERES and the general overestimation of the cloud radiative cooling by ERA5, already observed in the first case study. In contrast to the first case study, however, the CRE anomaly is also negative over the domain (lowest two boxplots in Fig. c). In the following subsection, we examine the transport history of the air parcels that arrive at mid-tropospheric levels and outline the role of tropical dynamics.

The air parcels that arrived in Barbados in the mid-tropospheric layer (650–300 hPa) on 14 February 2020 originate from different, mainly tropical and subtropical locations (Fig. a). To identify the pathways that eventually led to the moist anomaly, we define four airstreams based on the trajectories’ pressure evolution (each trajectory is assigned to one airstream only and the leading M in the airstreams' names stands for the arrival in the middle troposphere):

M-trades – trajectories that enter the mid-tropospheric layer from below;

The M-trades air parcels started their journey near the African west coast (green left-facing triangles in Fig. a) and crossed the North Atlantic along the southern edge of a persistent anticyclone, centered at roughly 30∘ N. They traveled at low altitudes until reaching South America (Fig. b), where they rapidly ascended (green line in Fig. a) in deep convective systems associated with the ITCZ (shown by the low cloud temperatures in Fig. d). The ascent is associated with a drop in specific humidity (Fig. e), resulting from cloud formation. Near the melting layer (dashed red line in Fig. ), the air parcels were detrained and spread northwards embedded in a developing shelf cloud (Fig. a). At the cloud's northern tip, some air parcels left the cloud and penetrated the dry layer beneath (Fig. b), presumably due to the subsidence induced by radiative cooling their Fig. 16. Consequently, the airstream's liquid water content temporarily decreased (Fig. c) until clouds formed again due to the outflow from cumuli congesti over South America on 12 February 2020 (Figs. c, c). The specific humidity remained largely unaffected by this process (Fig. e). However, these nearly constant values of specific humidity became increasingly anomalous as the M-trades airstream moved northwards into climatologically drier regions (Fig. f).

Shortly before arrival in Barbados, ice (Fig. d) and snow formed likely due to ascent along the slightly tilted isentropes. The air parcels ascended well above the 0 ∘C isotherm, which might have initiated heterogeneous freezing (Fig. d; by meridionally shifting the vertical cross section (not shown), we checked that the cloud is a large-scale shelf cloud and not just the upper part of a tilted deep convective tower which intersects the cross section). Simultaneously, the cloud was precipitating into the dry layer beneath (dashed blue contours in Fig. d). The distinct pattern of strong downward winds (orange contours in Fig. d, already noted in Fig. c) below a precipitating cloud points towards evaporatively driven downdrafts, supporting the hypothesis formulated in Sect. .

The M-local airstream was already located over South America in the mid-tropospheric layer 10 d prior to its arrival (purple right-facing triangles in Fig. a). Around 10 February 2020, M-local merged with M-trades (green and purple lines in Fig. a, b). After that, the two airstreams show very similar transport histories, and therefore we refer back to the detailed description in the preceding paragraphs. Together, M-local and M-trades contain 76 % of the air parcels that arrive at mid-tropospheric levels on 14 February 2020 and brought 93 % of the IWV650–300 hPa (Table ).

Therefore, the remaining airstreams (M-UT1, M-UT2) only contributed little in terms of air parcels and moisture content (Table ). The few M-UT1 air parcels rose over the eastern Pacific (light brown upward-facing triangles in Fig. ) from 8–10 February 2020 (light brown line in Fig. a). The rapid ascent brought them to subtropical latitudes (Fig. b) and resulted in a substantial decrease in specific humidity (Fig. e). When the M-UL1 airstream finally arrived in Barbados, it had dried out almost completely (Table , Fig. e).

The M-UL2 airstream exited the subtropical jet and approached from upper levels slightly north of Barbados or featured a Hadley-cell-like descent at tropical latitudes (dark brown downward-facing triangles in Fig. ). The trajectories in the subtropical jet completed almost a full circle around the globe within 10 d, starting their journey over continental North Africa (Fig. a). As M-UL2 stayed at upper levels (dark brown line in Fig. a), its water content (Fig. c–e) remained close to zero throughout the 10 d. M-UL1 and M-UL2 make up 24 % of the air parcels and hold 7 % of the IWV650–300 hPa (Table ). Thus, even though the two airstreams happened to arrive in the mid-tropospheric layer on 14 February 2020, they were not relevant for the observed moist anomaly. In other words, the mid-level moist anomaly above Barbados was essentially produced by outflow from the tropical shelf cloud, which was fed by airstreams from the North Atlantic trades and tropical South America.

The characteristic features of the most relevant airstreams (M-trades, M-local) involved in the formation of the mid-tropospheric moist anomaly above Barbados on 14 February 2020 are the advection from low latitudes and the detrainment at mid-tropospheric altitudes. To assess the importance of this transport pathway, we select all days from the quasi-climatological period on which at least 25 % of the trajectories arriving at 00:00–23:00 UTC between 650 and 300 hPa over the BCO traveled at tropical latitudes (minimal latitude <10∘ N) and never overshot the arrival layer (minimal pressure >300 hPa) within the 10 d before arrival. The latitudinal criteria are chosen based on the climatological position of the ITCZ (in the vicinity of which mid-level detrainment might take place), which is south of 10∘ N in January and February according to their Fig. 4g. We consider the full 10 d time span of the trajectories, because we assume that horizontal transport at mid-tropospheric levels is slow compared to other levels (due to the change from low-level easterlies to upper-level westerlies). The TMD selection criteria are subjective but serve the purpose to identify days with a transport history similar to 14 February 2020. Note that the demanded percentage of trajectories meeting the criteria for a tropical detrainment day is higher than for an EDI day (Sect. ), because the M-trades and M-local airstreams in Sect.  contained more trajectories than the L-EDI airstreams in Sect.  . The TMD (and EDI) frequency during the quasi-climatological period for other percentage thresholds can be studied in Fig. S1.2.

Applying this definition, we identify 65 TMD days (∼ 10 % of the quasi-climatological period), which however belong to fewer events, which typically last for several consecutive days (Table ). Most (53) TMDs lead to an anomalously moist middle troposphere above Barbados, occasionally associated with the formation of ice crystals (Fig. b; the 2 d with the highest total ice water are 14 and 15 February 2020). Twenty (61 %) of the TMDs belong to the 33 d with the highest values of IWV650–300 hPa (Fig. a) and 17 (52 %) to the 33 d with the highest total ice water (Fig. b). It appears that the mid-tropospheric humidity conditions do not strongly influence the CRE over Barbados (Fig. a, Table ). Thus, the case observed on 14 February 2020 during EUREC4A represents a very rare outlier with a CRE of -79 Wm-2.

The TMD trajectories leading to the few (12) dry anomalies (TMDdry) in the middle troposphere start at and ascend to lower pressure levels/higher altitudes (Fig. a) than the ones leading to moist anomalies (TMDmoist). Another characteristic of TMDdry trajectories is their excursion to comparably high latitudes (Fig. b), which matches one possible formation pathway of mid-tropospheric dry layers over the North Atlantic, described earlier by . One might argue that these cases do not qualify as TMDs, because their ascent and detrainment (if at all) happened earlier than the 10 d before arrival. Nevertheless, we can learn from them that for the formation of a mid-tropospheric moist anomaly (through a TMD), it is crucial that the air parcels start at low altitudes (Fig. a), where specific humidity is higher (Fig. c), and are lifted only as far as the cooling and subsequent condensation does not dry them out completely. The lower the altitude at which the air parcels are detrained, the moister do they arrive in Barbados (Fig. ; note the high correlation between IWV650–300 hPa and minimal pressure).

The quasi-climatological analysis linked tropical mid-level detrainment to mid-tropospheric moist anomalies above Barbados. We established a strong relation between the detrainment height and the magnitude of the moist anomaly. However, the amount of water vapor between 650 and 300 hPa appeared to be largely unrelated to CRE. The lack of a clear link between mid-tropospheric moisture/clouds fits well with existing literature, showing that the CRE of mid-level clouds depends on the cloud phase and thickness of the cloud, as well as the underlying surface and additional cloud layers at different altitudes .