Observation and modeling of a historic African dust intrusion into the Caribbean Basin and the southern U.S. in June 2020

This study characterizes a massive African dust intrusion into the Caribbean Basin and southern U.S. in June 2020, which is nicknamed the “Godzilla” dust plume, using a comprehensive set of satellite and ground-based observations (including MODIS, CALIOP, SEVIRI, AERONET, and EPA Air Quality network) and the NASA GEOS global aerosol 20 transport model. The MODIS data record registered this massive dust intrusion event as the most intense episode over the past two decades. During this event, the aerosol optical depth observed by AERONET and MODIS peaked at 3.5 off the coast of West Africa and 1.8 in the Caribbean Basin. CALIOP observations show that the top of dust plume reached altitudes of 6-8 km in West Africa and descended to about 4 km altitude over the Caribbean Basin and 2 km over the U.S. Gulf coast. The dust plume degraded the air quality in Puerto Rico to the hazardous level, with maximum daily PM10 concentration of 453 μg m-3 25 recorded on June 23. The dust intrusion into the U.S. raised the PM2.5 concentration on June 27 to a level exceeding the EPA air quality standard in about 40% of the stations in the southern U.S. Satellite observations reveal that dust emissions from convection-generated haboobs and other sources in West Africa were large albeit not extreme on a daily basis. However, the anomalous strength and northern shift of the North Atlantic Subtropical High (NASH) together with the Azores low formed a closed circulation pattern that allowed for accumulation of the dust near the African coast for about four days. When the NASH 30 was weakened and wandered back to south, the dust outflow region was dominated by a strong African Easterly Jet that rapidly transported the accumulated dust from the coastal region toward the Caribbean Basin, resulting in the record-breaking African dust intrusion. In comparison to satellite observations, the GEOS model well reproduced the MODIS observed tracks of the https://doi.org/10.5194/acp-2021-73 Preprint. Discussion started: 3 March 2021 c © Author(s) 2021. CC BY 4.0 License.

Earth and the sun (about one and a half million miles above the Earth's surface) (Marshak et al., 2018). Featured in the image is a dense dust plume over the Caribbean Basin followed by another just off the African coast in the eastern North Atlantic Ocean. These two dust plumes are about 5000 km apart but appears to be comparable in the intensity. The dust over the Caribbean Basin during this period has attracted considerable interests from scientific community and media because of its 70 huge extent and massive amount, so-called the "Godzilla" dust plume (https://phys.org/news/2020-06-sahara-blanketscaribbean-air-quality.html), and "a dust plume to remember" (https://earthobservatory.nasa.gov/images/146913/a-dust-plumeto-remember) for its extraordinary characteristics. Francis et al. (2020) examined the atmospheric circulation characteristics that drove the formation and transport of this dust storm. In this study, we will use a variety of remote sensing and in situ observations and simulations with the NASA Goddard Earth Observing System (GEOS) model to characterize the gigantic 75 dust plume and assess its impact on the air quality in the southern U.S. Specifically, we will: (1) characterize the evolution of the three-dimensional structure of the dust plumes along their cross-ocean transit , (2) place the intensity of the "Godzilla dust plume" in a context of the last two decades, (3) understand major synoptic processes that resulted in the gigantic dust intrusion into the Caribbean Basin, (4) assess its impact on particulate matter (PM) air quality in the southern U.S., and (5) evaluate the Goddard Earth Observing System (GEOS) model simulation of the dust event with the observations. 80 The rest of the paper is organized as follows. Section 2 describes the data and model we use to characterize the dust event,

MODIS aerosol optical depth
The MODIS instruments onboard both the NASA Terra (morning) and Aqua (afternoon) satellites, acquire near global, daily observations of aerosols with a wide swath of ~2330 km. Because of its wide spectral range and the simplicity of the dark ocean surface, MODIS dark-target (DT) algorithm (Remer et al., , 2020Levy et al., 2013) has the capability of retrieving 95 AOD with a relatively high accuracy, as well as information on particle size (in the form of Angstrom exponent, effective radius or fine-mode fraction -FMF). The FMF measures the contribution of fine-mode particles to total AOD at 0.55 µm . In the operational DT aerosol retrieval dust is assumed to be spherical, which introduces errors in the https://doi.org/10.5194/acp-2021-73 Preprint. Discussion started: 3 March 2021 c Author(s) 2021. CC BY 4.0 License. aerosol retrievals downwind of the dust source regions. Most recently, an enhanced DT retrieval algorithm has been developed to improve dust retrievals by accounting for non-sphericity of dust particles (Zhou et al., 2020a). It has been shown that this 100 enhanced dust retrieval algorithm significantly improves the retrievals of AOD and FMF over ocean (Zhou et al., 2020b). For this study exclusively, the enhanced DT algorithm has been applied to the identified dust scenes over ocean from June 10-30, 2020. Although the DT algorithm is also applied to retrieval AOD over vegetated lands, it does not retrieve aerosol over deserts because of interference of strong surface signal. The Deep Blue (DB) algorithm was initially developed to retrieve AOD and other aerosol properties over bright surfaces and then extended to vegetated lands and oceans , which 105 complements the DT retrievals. The DT and DB products have been combined, on the basis of their performance in reproducing the Aerosol Robotic Network (AERONET) observations, to characterize the global aerosol system (Levy et al., 2013). For this study, we aggregate the enhanced DT over-ocean retrievals into 1°x1° grids. Over land, we use MODIS Collection 6.1 daily data. We also combine MODIS AOD at 550 nm from Terra and Aqua to acquire a better spatial coverage of daily aerosol distribution than each satellite alone. When both Terra and Aqua have AOD retrievals, they are averaged. In this study, we 110 will use the AERONET data to validate MODIS AOD retrieval for this intense dust event. The AERONET is a ground-based network with equipped well-calibrated Sun photometers that have been measuring AOD (with an accuracy of 0.01) and retrieving a set of particle properties around the globe .

CALIOP aerosol extinction profiles
CALIOP is a two-wavelength, polarization lidar onboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite 115 Observation (CALIPSO) satellite with an equator-crossing time of about 1:30 PM and 1:30 AM, and a 16-day repeat cycle.
Since June 2006, CALIOP has been almost continuously collecting high vertical resolution (e.g., 30 m) profiles of the attenuated backscatter by aerosols and clouds at 532 nm and 1064 nm wavelengths along with polarized backscatter at 532 nm between 82ºN and 82ºS (Winker et al., 2009). Currently, CALIOP is the only spaceborne lidar on orbit that provides this key information about the vertical distribution of aerosol. The unprecedented long data record of CALIOP aerosol profiles 120 accumulated over more than a decade has contributed to a revolutionary understanding of aerosols in the Earth system. It is worth noting that CALIOP can detect aerosol layers in clear sky, below thin cirrus clouds, and above opaque low-level clouds during both day and night, although the nighttime data have better accuracy than the daytime data (Winker et al., 2010). In this study, we will use the CALIOP version 4.20 aerosol extinction profile data at a nominal horizontal resolution of 5 km supplemented by the vertical feature masks in both daytime and nighttime, which represents significant improvements over 125 the previous data versions . We only use high quality aerosol data with the Cloud Aerosol Discrimination (CAD) score between -100 and -90 following Yu et al. (2019). SEVIRI onboard the Meteosat Second Generation (MSG) satellite series in geostationary orbit (36,000 km) and centered at (0ºN,0ºE) provides images of Europe and Africa at a frequency of every 15 min, day and night (Schmetz et al., 2002). This 130 allows for monitoring the genesis and movement of dust clouds at high temporal resolution (Schepanski et al., 2007;Ashpole and Washington, 2012). The brightness temperature (BT) at 10.8 µm and two BT differences (between 8.7 µm and 10.8 µm, and between 12.0 µm and 10.8 µm) are rendered to red-green-blue (RGB) beams to highlight the presence of dust and different cloud phases (deep clouds, middle clouds, and low clouds) (Lensky and Rosenfeld, 2008;Brindley et al., 2012). In this study, we use SEVIRI RGB imagery to illustrate the genesis and movement of mesoscale convective systems, haboobs, and dust 135 plumes from other sources.

PM concentrations from EPA air quality network
EPA of the United States has established a comprehensive network across the nation (including Puerto Rico, and the U. S. Virgin Islands) to monitor the outdoor air quality of ozone, PM, and other chemical species. In this study, we will use the measured daily PM2.5 data in June 2020 over nine southern states of the U.S., including Florida, South Carolina, Georgia, 140 Alabama, Mississippi, Louisiana, Arkansas, Oklahoma, and Texas. This wide swath of states captured the major influence of the massive dust intrusion on air quality. Unfortunately, most of EPA measuring sites in Puerto Rico were not active during the period of this study, except Canato where PM10 (PM with aerodynamic diameter of smaller than 10 µm) concentration was measured during the June 22-30 period. Given that in the southern U.S. the EPA network currently only collects PM10 concentration at very limited number of sites, our analysis will focus on PM2.5. 145

GEOS simulations of aerosol
The NASA GEOS is a global Earth system model that includes components for atmospheric circulation and composition, ocean circulation and biogeochemistry, land surface processes, and data assimilation (Rienecker et al., 2011). The coupled atmospheric constituent module within the GEOS architecture most relevant to this study is an aerosol module based on the Goddard Chemistry Aerosol Radiation Transport (GOCART) model . GOCART simulates major 150 components of aerosols (with diameter between 0.02 and 20 µm) and some gaseous precursors, including dust, sea-salt, sulfate, nitrate, ammonium, organic carbon, black carbon, SO2, dimethyl sulfide, and NH3 (Chin et al., 2002Ginoux et al., 2001;Bian et al., 2017). The model is run in a replay mode, with meteorological fields being taken from the Modern-Era Retrospective analysis for Research and Applications -version 2 (MERRA-2) reanalysis (Gelaro et al., 2017) every six hours.
The model has a horizontal resolution of 1ºx1º and 72 layers in the vertical. The GEOS hourly outputs of aerosol are used in 155 this study. Note that the model run does not assimilate satellite data of aerosol.
In the GOCART dust modeling, bulk dust emissions are calculated online based on 10-m wind speed and a predetermined dust source function. The dust source function is a dynamic one that uses the topographic depression and the dynamic surface bareness derived from the satellite observations (Ginoux et al., 2001;Kim et al., 2013). This dynamic dust source function accounts for the seasonal and interannual variations of the surface bareness and soil moisture, which improves 160 simulated temporal variation of dust aerosols over some semi-arid areas (Kim et al., 2013). Currently, dust particle size distribution (PSD) in GEOS model is described with five size bins (i.e., 0.2-2 µm, 2-3.6 µm, 3.6-6 µm, 6-12 µm, and 12-20 µm in diameter) (Ginoux et al., 2001;. The size distribution of emitted dust is empirically prescribed following Tegen and Fung (1994). Emitted dust is transported by winds and removed from the atmosphere via gravitational settling, dry deposition by turbulence, and scavenging by large-scale and convective rain. The gravitational settling is 165 calculated with an assumption of spherical particle following a method as described in Ginoux et al. (2001). The model parameterizes large-scale in-cloud and below-cloud scavenging as a function of rainfall production rate and precipitation fluxes, respectively, and the scavenging in convective updrafts as a function of the updraft mass flux. Dust optical properties in the model are based on the Meng et al. (2010) database that incorporates Mie, T-Matrix, DDA, and geometric optics (depending on size parameter), as described in Colarco et al. (2014). The shape distribution presently used is the 170 spheroidal distribution proposed by Dubovik et al. (2006).

Observational characterizations of the dust event
In this section we use satellite and ground-based observations to characterize the dust event, including the evolution of trans-Atlantic dust plumes, strength of the dust intrusion event in the context of last two decades, impacts of the dust intrusion event 175 on air quality in Puerto Rico and the southern U.S., and synoptic meteorological conditions controlling the dust event.

Evolution of the trans-Atlantic dust plumes
Horizontal variations of trans-Atlantic dust plumes are characterized by MODIS aerosol retrievals. Figure 2 shows the MODIS daily AOD maps from June 13 to 27 at a frequency of every other day (a full day-to-day variation of AOD can be seen in an animation in Supplementary Online Material). Here MODIS observations from both Terra and Aqua are combined to represent 180 daily AOD with a better spatial coverage than either alone. Overlaid on the AOD map is horizontal wind vectors at about 4km altitude from the MERRA2 reanalysis. Clearly seen in these maps are the dust plumes of as wide as 2500 km (confined within 5ºN-30ºN latitude belt) being transported across the tropical Atlantic Ocean in a meandering path and ultimately reaching the Gulf of Mexico and the southern U.S. A discontinuity along the West African coastline reflects difference between the MODIS DT and DB algorithms. In the early days (June 13-15), the dust plume was largely confined to the African coastal region (east 185 of 35ºW), which is consistent with the presence of a strong meridional wind component in the region. This coastal accumulation of dust led to a peak AOD of about 3.5 on June 17. Although the plume had already started moving westward on June 17 as a result of a much weakened meridional wind, the rapid ventilation of dust away of the African coast took place on June 18. On June 19, the plume extended from the African coast to 50ºW with more dust coming out of West African deserts. The dust https://doi.org/10.5194/acp-2021-73 Preprint. Discussion started: 3 March 2021 c Author(s) 2021. CC BY 4.0 License. plume front was swirling around a weak anticyclone with its front moving northward to nearly 30ºN. In the following days, 190 the dust plume drifted south and reached the northern coast of South America on June 21. The plume with its front at 70ºW was followed by another narrow dust plume located near the coast of West Africa with AOD generally smaller than 1. It appears that significant dust in the plume had been deposited into the ocean during the period of June 19-21. Some new dust sources on were also evident over West Africa (e.g., southern Algeria, Mali, and Mauritania). On June 23 dual dust plumes appeared on the map, the primary "Godzilla" dust plume over the Caribbean Basin (centered around 15ºN and 68ºW) and the  In West Africa and along the coast, the top of the dust plume is at 6-8 km, which is higher in the north than in the south. This dust plume top altitude is higher than the climatology of summertime extreme dust events (~5 km) (Huang et al., 2010). The 225 intense dust layers stay above the low-level clouds (light grey shading) (Figures 5a and 5b). Also, the heavy dust layer attenuates the CALIOP beam entirely so that no signal (black shading) is apparent below 2 km in some locations ( Figure 5b).
After being transported to the Caribbean Basin, the top of the dust plume is at about 4 km and the dust layer appears to mix with marine aerosol in the boundary layer. The mixing leads to the maximum extinction near the surface. Because the aerosol loading was significantly reduced through deposition processes along the transport, totally attenuated features do not exist over 230 the Caribbean Basin. The CALIOP high-resolution measurements also show fine structures in the dust plume, including several sandwiched layers of high aerosol extinction of greater than 0.5 km -1 between 1.5 km and 4 km near the African coast and about 0.3 km -1 between 1 and 3 km in the Caribbean Basin.

3.1.2
Impacts on air quality in Puerto Rico and the southern U.S. Figure

A historic event in the past two decades and its synoptic control 265
The June 2020 event of African dust intrusion into the Caribbean Basin and the Americas is a historic one projecting above the climatology from the past two decades, as registered in the MODIS/Terra data record since 2000 (Figure 9). We carried out regional analysis of MODIS Terra daily AOD since 2000 in seven regions as defined in Figure Figure 9b-9h. In each region, daily AOD for January -June 2020 is marked as red dots and lines, with the evolution of daily AOD from June 10 to 30, 2020 being elaborated in the inset. For visual clarity, we present the 2000-2019 daily AOD climatology in the form of the 20-year average (black line) plus its range (grey vertical bar). Clearly, the dust event in June 2020 has the highest AOD over the past two decades over the North African coast (b), the southern Caribbean Basin (c), and the northern Caribbean Basin (d). In the northeast coast of South America (e), the dust transport to 275 this region peaks in March-June with a minimum in August-November, which is determined by the seasonal migration of ITCZ (Yu et al., 2015a(Yu et al., , 2015bProspero et al., 2014). Despite this, the 2020 June event had the second highest AOD over the past two decades and was the highest in June. The Gulf of Mexico (f) and the tropical eastern Pacific Ocean (g) are highly impacted by biomass burning smoke from the central America in spring. Although the June 2020 dust event had lower AOD than for some extreme springtime biomass burning events, it was indeed the highest in June. Moreover, it is very rare for 280 African dust to make it into the tropical eastern Pacific because observations have suggested a Central American barrier to dust transport (Nowottnick et al., 2011). Therefore, for all these six regions affected by trans-Atlantic dust transport, the June 2020 dust is an historic event over the past two decades when seasonal variations of dust and smoke transport are factored in.
On the contrary, the MODIS AOD over the Saharan deserts (h) does not indicate that daily dust emissions from North Africa were particularly large in early and mid-June. In fact, it was smaller than AOD in late May and June 6-8, 2020. Although the 285 2020 June AOD was higher than the climatological average in June, it was not the highest. An analysis in West Africa (10ºN-30ºN, 17ºW-10ºE), which is a part of the SAHD and likely the major source region for this dust event, displays similar AOD variations (see Figure S2).
Given that the dust loading in source regions in June 2020 were large albeit not historic (Figure 9h and Figure S2), the observed historic intrusion of African dust into the Caribbean Basin and the southern U.S. should have been modulated by 290 meteorological conditions. The North Atlantic subtropical high (NASH), also known as the Bermuda-Azores high is a semipersistent synoptic system that affects the meteorology and atmospheric circulations in West Africa and tropical Atlantic Ocean. The variation in NASH location and intensity would affect how the dust is transported across the tropical Atlantic Ocean. Here we analyze the MERRA2 meteorology associated with the dust episode by focusing on geopotential height and wind. Figure 10 displays the evolving spatial patterns of the geopotential height and wind vectors at 600 hPa from June 14 to 295 June 19. On June 14, the subtropical high was centered at (45ºW, 43ºN) with a maximum height of about 4500 m. This ridge system was accompanied by a low-pressure system or trough to its southeast around the Azores and an extensive high-pressure system (~4550 m) over West Africa. This setting of synoptic systems created an unfavorable atmospheric circulation condition for trans-Atlantic transport of dust. At the lower latitudes (south to ~20ºN), West Africa was dominated by strong northeasterly winds, which rapidly exported dust from Sahara-Sahel transit to the eastern Atlantic Ocean. But the easterly veered to the north 300 in the coastal ocean (15ºW-35ºW), to the east at the northern fringe of the African continent (30ºN-35ºN), and eventually to the south in central Africa. This created a nearly closed atmospheric circulation system over West Africa and the eastern North Atlantic Ocean that could recirculate and trap the dust in the West African coast. The unfavorable synoptic systems persisted through June 15 and 16, although they were gradually weakened. By June 17-19, the subtropical high weakened further and drifted southward; meanwhile the trough over Azores was gradually filled up. The mid-latitude westerly pushed southward 305 along the African coastline and broke up the closed atmospheric circulation over West Africa and the coastal ocean. As a result, dust outflow region was dominated by a strong African Easterly Jet (AEJ), which would favor the rapid transport of the accumulated dust from the African coast toward the Caribbean Basin.
Satellite observations corroborate the above analysis of the potential control of the synoptic systems on distributing African dust. As shown earlier in Figures 2 and 3, MODIS AOD started to build up on June 13 but a majority of the dust did 310 not transport westward beyond the 35ºW until June 18. The highest AOD near the coast occurred on June 17. Moreover, the dust distribution modulated by the synoptic systems can be vividly displayed in an animation of SEVIRI full-disk RGB dust imageries once every 30 min over the June 12 -25 period (https://doi.org/10.5446/51548). Additionally, the animation in the SOM clearly shows the evolution of haboobs and their radial outflow behaviour, driven by outflows from convective downdrafts, which is not always evident in the still images. Here we show a sequence of SEVIRI still images (zoomed in North 315 Africa) at 12Z of June 14-19, 2020 to illustrate the day-to-day evolution of the dust plumes (Figure 11). In these images, magenta, dark red, orange, and dull pink denotes dust, deep clouds, middle clouds, and low clouds, respectively. On June 14, SEVIRI detected two dust plumes (Figure 11a). One plume originating from the southern Mauritania was dispersed over a small coastal area (22ºW-16ºW and 12ºN-20ºN). The other dust plume was originated from a haboob developed over Niger due to strong downdrafts associated with a mesoscale convection system (dark red). The dust plume was situated north of the 320 track of the convective system and was trailing the rapid moving deep clouds because of the much weaker wind speed than in the convective core (refer to Figure 10). The convective systems swept swiftly across West Africa and reached the coastal ocean by early hours of June 15. This formed an extensive dust belt between 15ºN-22ºN that extended from Niger to the coast of Mauritania, as shown in Figure 11b. The haboob-generated dust mixed with that produced from West African deserts and stayed over coastal water (east to 30ºW and 15ºN -30ºN). The extensive dust belt continued to proceed towards the ocean on 325 June 16 and more dust was accumulated into the coastal region (east to 40ºW, Figure 11c). These images clearly show that dust emerging from the continent accumulated over the coastal region for more than three days, yielding the heaviest dust plume on June 17. Then this amplified dust plume was ventilated out of the coastal region by the easterlies on June 18 and 19 ( Figures 11e and 11f), leading to the historic intrusion of African dust into the Caribbean Basin and southern U.S. Note also that additional dust plumes from haboobs (June 18) and other West African sources (June 19) were added to the trans-Atlantic 330 transport.
The above analysis suggests that the strength and location of NASH plays an important role in modulating the trans-Atlantic dust transport during this historic dust intrusion event. It is intriguing to compare the June 2020 NASH with other years. Figure 12 compares the June geopotential height at 600 hPa between 2020 (a) and 1980-2019 climatology (b). Clearly, the NASH in June 2020 was stronger and located further north in comparison to the 40-year climatology. As shown in (c), the 335 geopotential height in 2020 is more than 80 m higher than the climatology. South to this high anomaly is a low anomaly that extends from Bermuda to western Europe, with the lowest taking place off the coast of West Europe and the second lowest between Azores and Canary Islands. Over West Africa, the geopotential height in 2020 is higher than the climatology by up to 20 m over the northwestern Africa. Over the last four decades, the 2020 geopotential height over the high anomaly center (60ºW-30ºW, 35ºN-50ºN) is the second highest, slightly lower than 2006 (d). This analysis suggests that the subtropical high 340 in June 2020 was highly anomalous in both the intensity and position.

GEOS model simulations of the dust intrusion event
In section 3.1, we have characterized the evolution of the historic dust plume in three dimensions associated with synoptic systems and assessed its impact on air quality in the southern U.S. by using a set of satellite and ground-based observations.
Here we assess to what extent the GEOS model can reproduce the observed characteristics of this historic event. Similar to 345 Figure 9, we analyze GEOS AOD from January 1, 2000 to June 30, 2020 on the regional basis (see Figure S3 in SOM). It shows that although the model characterizes the June 2020 event as a historic one over the North African coast (NAFC) and the southern Caribbean Basin (SCRB), the magnitude is more than a factor of 2 smaller than the MODIS AOD. Similar to the MODIS observations, the GEOS AOD over the desert (SAHD) during the event is not historically high. Unlike the MODIS observations, GEOS simulations of AOD over the other four regions are not the highest even after accounting for seasonal 350 variations of dust and smoke transport. In the following, we further compare the GEOS simulations of aerosol threedimensional distributions with MODIS and CALIOP observations over dust source region and along the trans-Atlantic transport route.

Dust source region
As discussed earlier and displayed in the SEVIRI animation, the major source of the Godzilla dust plume is associated with 355 intense haboobs generated by a strong and fast-moving convective system over the southern Sahara from June 13 to June 15.
How does the GEOS model perform in simulating haboobs associated with mesoscale convective systems? Figure  Although CALIOP reveals the elevated dust plume (either above clouds or totally attenuated features) with the highest 370 extinction at the altitude of 4-6 km, the GEOS model displays a rapid decrease of aerosol extinction with increasing altitude.
Both comparisons confirm that the model with a horizontal resolution of one degree has a grand challenge to realistically simulate the mesoscale convection and haboobs. The model substantially underestimates dust loading over the desert, implying a very substantial underestimate of dust emissions. The model also drifts the dust plume northwards and fails to pump up dust from the surface to higher altitudes for ensuing long-range transport. These modeling deficiencies affect the 375 simulation of trans-Atlantic dust transport as discussed in next section.

3.2.2
Trans-Atlantic dust transport more pronounced with increasing transport distance. To further quantify the difference between GEOS and MODIS, we create the Hovmöller diagrams for GEOS AOD and AOD difference between MODIS and GEOS (MODIS -GEOS), as shown in Figure 16. The GEOS AOD Hovmöller diagram clearly shows that the model reproduces the distinct trans-Atlantic dust plume tracks as observed by MODIS (Figure 3). However, the GEOS substantially underestimated the MODIS observations. For the 385 primary or "Godzilla" dust plume, the MODIS AOD is higher by up to 1 (corresponding to a factor of 2) near the African coast and by up to 0.6 (corresponding to a factor of 5) in the Caribbean Basin than the model simulation. The increasing MODIS and GEOS discrepancy with increasing transport distance suggests that GEOS model removes the dust too efficiently from the atmosphere, consistent with previous finding (Yu et al., 2019;Kim et al., 2014). For the secondary dust plume with weaker intensity, the GEOS model performs better; generally, MODIS AOD is larger than GEOS AOD by a factor of no more than 2. 390 A more complete view of MODIS and GEOS AOD evolution during June 10-30 period is displayed in an animation (https://av.tib.eu/media/50830). Finally, the long-term GEOS model simulations do not show that the Godzilla dust plume is historic over the past two decades.
The vertical structure of the "Godzilla" dust plume exhibits striking differences between GEOS and the CALIOP observations, as shown in Figures 17 and 18 descends at a rate of about 500 m d -1 (~20 m hr -1 ), which agrees well with the climatology of the extreme dust events Huang et al., 2010). GEOS model misses or substantially underestimates the elevated dust plume, although it generally agrees better with CALIOP at lower altitudes. During June 22 -25 and in the west Atlantic Ocean and Caribbean 410 Basin, the dust plume continues descending with distance, mixing with background marine aerosol in the boundary layer, and touches the surface. Compared to the tropical eastern Atlantic Ocean, the CALIOP-GEOS discrepancy becomes much larger in the lower atmosphere ( Figure 18). When integrating aerosol extinction in the vertical column, the CALIOP to GEOS AOD ratio increases from 1.43 near the coast (June 16) to 1.84 in the middle ridge (June 20), and 3.46 in the Gulf of Mexico (June 25), suggesting that the CALIOP-GEOS discrepancy increases with distance. This feature is consistent with that between 415 MODIS and GEOS as revealed and discussed earlier (Figure 16). The missing of the elevated dust layer by GEOS over the upwind ocean and desert regions contributes to the large discrepancies observed in the downwind regions as the dust plume descends. It is also possible that CALIOP observed high values of aerosol extinction in the lowest ~500 m layer may be prone to interference by surface signal and/or cloud contamination. When the lowest 500 m layer is excluded in the calculation of AOD, the CALIOP to GEOS AOD ratio ranges from 1.54 to 3.84, slightly larger than that for the whole column. Excluding 420 the lowest 500 m layer does not reduce the discrepancy between CALIOP and GEOS.

Conclusions
We have used a set of remote sensing observations, including MODIS, CALIOP, SEVIRI, and AERONET, to characterize the three-dimensional evolution of the gigantic African dust intrusion into the Caribbean Basin and southern U.S. in late June 2020 (June 13-27, 2020). For this gigantic dust event the aerosol optical depth broke the MODIS record of the past two decades, 425 with AOD more than 3.5 at the African coast and 1.8 in the Caribbean Basin. The dust plume, originating from the convectively generated haboobs over sources in West Africa (mainly Niger, Mali, and Mauritania), was lifted from the desert surface to altitudes of up to 6-8 km, which is higher than the 5 km for the climatological summertime extreme dust events (Huang et al., 2010). Due to the persistence of a closed atmospheric circulation system over West Africa, the large but not extreme daily dust loading from Sahara accumulated in the African coastal region (east to 35ºW) for about four days. The average transport speed 430 of the dust plume is 1000 km d -1 , which agrees very well with the climatology of summertime extreme dust events Huang et al., 2010). During trans-Atlantic transport the top of the dust plume descended from 6-8 km over the West African coast to about 4 km altitude over the Caribbean Basin and 2 km over the U.S. Gulf coast. The descent of dust plume imposes important implications for air quality in the Caribbean Basin and the southern U.S. In Puerto Rico, the Godzilla dust plume caused a record-breaking PM10 concentration of 453 µg m -3 . The dust intrusion into the southern U.S. raised the 435 PM2.5 concentration to a level exceeding the EPA air quality standard in about 20% and 40% of the EPA stations in nine southern states on June 26 and 27, respectively. The poorest air quality with PM2.5 as high as 74 µg m -3 occurred in the Florida panhandle region and western Texas.
The analysis of MERRA2 meteorology suggests that the unfavorable ventilation condition and the resultant dust accumulation along the African coast in the early stage of the dust storm was associated with the anomalous strength and 440 northward shift of the North Atlantic subtropical high (NASH) that was accompanied by the low-pressure system over the Azores and the high-pressure system over West Africa. In fact, June 2020 had the second strongest NASH over the past four decades, only slightly weaker than the 2006 record. When the NASH became weaker and wandered back to south, the dust outflow region was dominated by the African Easterly Jet (AEJ), which carried the accumulated dust plume rapidly and maintaining its high concentrations from the coastal region toward the Caribbean Basin within four days, resulting in the 445 extraordinary dust loading observed. Our results do not fully agree with what Francis et al. (2020) found on the atmospheric drivers of the dust storm. For example, they argued that the development of a subtropical high off the coast of West Africa generated anomalously strong northeasterlies over Sahara (19º-30ºN, 20º-0ºW) that caused continuous dust emissions over four days and high dust loading in the eastern tropical Atlantic Ocean. Our analysis of the SEVIRI dust images showed that intense haboobs swept through Niger-Mali-Mauritania corridor (south of 20 º N generally) and contributed significantly to the 450 dust event. The dust emissions associated with these haboobs cannot be adequately explained by the large-scale meteorology used in Francis et al. (2020), because the reanalysis cannot capture such strong winds accurately (Cowie et al., 2015;Roberts et al., 2017) and their focused dust source region is largely outside the corridor of the intense haboobs identified in the SEVIRI images. We also found that unique synoptic setting associated with anomalous NASH strength and position created the closed atmospheric circulations over West Africa and its adjacent coastal ocean for several days, which trapped the continuously 455 emitted dust in the African coast. In addition, Francis et al. (2020) found that the AEJ was much strengthened by the anticyclonic circulation associated with the anomalous sub-tropical high, which favored a rapid westward transport of dust toward the Americas. However, our estimated trans-Atlantic transport speed of 1000 km d -1 is more or less the same as the speed for the summertime dust events during 2003(Huang et al., 2010, suggesting that the strong AEJ in June 2020 was unlikely to be a major factor for the highest-in-record dust detected in the Caribbean Basin. 460 In comparison to satellite observations, the GEOS model substantially underestimated dust loading over the desert, which were strongly related to emissions from haboobs. The model also did not lift up enough dust to the middle troposphere for ensuing long-range transport. These deficiencies are likely resulted from unrealistic representations of moist convection, haboobs, and the vertical transport of dust in the model, possibly related to the model's coarse horizontal and vertical resolutions. As a result, the model largely failed to capture the satellite-observed elevated dust plume along the cross-ocean 465 track and underestimated the dust intrusion into the Caribbean Basin and the Americas by a factor of 4 or more for AOD. Nevertheless, the model reproduced the plume track reasonably well on a daily basis, suggesting that large-scale meteorological fields that drove the aerosol transport modeling are accurate. Assimilating satellite observations of aerosol optical depth into the model can significantly improve the model's prediction of column aerosol loading Buchard et al., 2017). Given the substantial differences in the aerosol vertical distribution between GEOS and CALIOP, 470 however, if the assimilation only normalizes the modeled vertical distribution by the column AOD, the assimilation will continue to put too much of the dust in the lower layers. This may continue to artificially enhance the dust deposition along the transport path and introduce high bias in the surface dust concentration, which is of concern for air quality applications.
Modeling improvement needs to focus on developing more realistic representations of moist convections, haboobs, and the vertical transport of dust (e.g., Roberts et al., 2018). 475 This work demonstrates that haboobs and convective systems over Africa have the ability to impact conditions far downstream. It is vital that models possess a capability of simulating convective outflows driving dust uplift, followed by accurately redistributing this emitted dust vertically throughout the Saharan boundary layer up to ~6-8 km as the haboobs decay. This study shows that if models are not able to represent dust up to the high observed altitudes over source regions, the resulting long-range transport will be incorrect. O'Sullivan et al. (2020) recently found that modelled summertime dust in the 480 tropical Eastern Atlantic region was too low in the atmosphere compared to in-situ aircraft observations and that part of the problem was that the coarser dust particles were both not lifted to high enough altitudes and also settled out of the atmosphere too rapidly. It is clear that in order to improve dust models' ability to represent dust transport, efforts are needed to improve the representation of processes controlling dust uplift (such as haboobs), dust redistribution through the Saharan boundary layer and processes controlling their emission, transport and deposition, as a function of size. It is vital that future evaluations 485 incorporate observations of vertical distribution of dust in order to fully understand and evaluate dust models.

Data availability
All datasets of aerosol and meteorology were obtained from a variety of sources with public access: The MODIS aerosol data

Supplements
Supplementary figures referred in the paper are provided online.

Author Contributions
HY and MC conceived the study. HY, QT, LZ, QS, YS, and DK analyzed satellite and surface observations as well as model outputs. YZ and RCL produced the MODIS enhanced dust retrievals for the event. HB performed the GEOS simulations. YP and CLR provided SEVIRI dust imagery. All co-authors participated in discussion of the analysis. The paper was written by HY and commented and revised by all co-authors. 505

Competing interests
The authors declare that they have no conflict of interest.