Great strides have been made in our understanding of aerosol–cloud–radiation interactions; however, substantial uncertainty remains (Mhyre et al., 2013; Li et al., 2019), particularly in our understanding of the role of mineral dust during such interactions. Dust has a direct impact on climate since it is considered one of the most important light-absorbing particles in the atmosphere at shorter solar wavelengths (Sokolik and Toon, 1999). Recent studies show that by its radiation absorbing potential, dust can increase air temperature and reduce sea surface temperature at low altitudes (Sun and Zhao, 2020). Dust is a good ice-nucleating particle (DeMott et al., 2003), but its activity as cloud condensation nuclei (CCN) is less clear. Studies have shown that fresh dust is mainly hydrophobic and a poor cloud condensation nucleus (Rosenfeld et al., 2001), while others have shown that dust can act as a cloud condensation nucleus if it has aged and mixed with anthropogenic or organic aerosols (Fitzgerald et al., 2015). The latter observation is supported by laboratory studies that have found that the CCN activity depends on the mineralogical composition and mixing state of the dust particles (Sullivan et al., 2009; Gaston, 2020). Other studies have found that dust can commonly act as a cloud condensation nucleus even from the source, as it often has hygroscopic material, and thus can influence cloud formation (Twohy et al., 2009). In contrast to these findings, Denjean et al. (2015) saw that African dust transported from Africa to the Caribbean Basin remained mainly externally mixed and that dust particles did not take up significant amounts of water when exposed to up to 94 % relative humidity, a similar finding to that of Edwards et al. (2021) from African dust transported to Florida. The larger, super-micron dust particles act as giant CCN that can form large cloud droplets that lead to the earlier formation of raindrops (Rosenfeld et al., 2008).

In the Caribbean Basin, which is frequently inundated by mineral dust picked up by surface winds over the northern half of the African continent and transported westward, studies of aerosol–cloud–radiation interactions are limited, with most having been done on the islands of Puerto Rico and Barbados. Within the northeastern part of Puerto Rico is a tropical montane cloud forest (TMCF) known as Pico del Este (PDE). PDE has been an ideal location for studying clouds, as they form frequently and last for many hours (Eugster et al., 2006; Allen et al., 2008; Spiegel et al., 2014; Raga et al., 2016). Eugster et al. (2006) studied daily cloud properties as a function of net radiation and saw that denser clouds correlated with less radiation at the surface. Allan et al. (2008) observed that clouds affected by anthropogenic pollution had a larger number of droplets with smaller diameters. The results in Spiegel et al. (2014) support the observations in Allan et al. (2008) but with clouds impacted by African dust. In contrast, Raga et al. (2016) observed that meteorology was more important in cloud formation than the aerosol source, particularly during dust events. Cloud chemical properties have also been studied at this location (Weathers et al., 1988; Asbury et al., 1994; Gioda et al., 2009a, b, 2011, 2013; Reyes-Rodríguez et al., 2009; Valle-Díaz et al., 2016). Under the influence of African dust, there was an increase in non-sea salt calcium and other crustal and organic components when clouds were formed during periods of African dust (Valle-Díaz et al., 2016). There was also an increase in non-sea salt sulfate when under the influence of anthropogenic pollution (Gioda et al., 2009; Valle-Díaz et al., 2016).

The main objective of the study being reported here is to use measurements of cloud microphysical properties in a TMCF, complemented with aerosol measurements and back-trajectory analysis, to evaluate the relationship between the source of aerosols and the formation and evolution of clouds. The working null hypothesis is that at this location, the source of the aerosols will have a statistically insignificant effect on the cloud properties.

The times series of the microphysical properties of clouds that evolved in air with different air mass histories show that the size distributions of number and mass concentrations, their associated average Nc and MVD, and lifetimes are quite different. Clouds are formed on CCN as air masses cool and reach supersaturated conditions; hence, their characteristics will depend on the thermodynamics of the air as well as the physical and chemical properties of the aerosols carried by the air parcels.

The trends in the bulk properties derived from the size distributions, i.e., the Nc, LWC, and MVD, are metrics that can clarify differences in the behavior of clouds under the influence of different air mass histories, as illustrated in Fig. 7. In this figure, the deviations (anomalies) from the average values are shown as a function of time, where the averages are calculated for the period in which the air mass is arriving at PDE. The plotted time series begins 4 h prior to the air mass arrival and extends an additional 20 h. Starting and ending values of -100 % demarcate the time periods of the air masses with individual durations.

Due to the arrival of the AO air at PDE at the same time as the cloud was forming, the concentration deviation (Fig. 7a), ΔNc, from the average is negative, increasing to an approximately +25 % deviation that lasts 8 h before once again decreasing below the average. The ΔLWC (Fig. 7b) follows a similar trend, correlating closely to the ΔNc; however, the maximum ΔLWC is +75 % compared to +30 % for the ΔNc. This difference in the maximum deviations is a result of changes in the ΔMVD (Fig. 7c), whose trend follows those of the ΔNc and ΔNc but reaches a maximum deviation of 20 %. Since LWC ∼ (Nc)(MVD)3, the ΔLWC will experience a much larger change with deviations in MVD than in Nc, and by using the same calculation used in the propagation of errors, i.e., the root summed square (RSS), ΔLWC ∼ ΔNc + 3ΔMVD = 80 %.

The AD air masses arrived at PDE and encountered forming cloud whose concentration is 50 % higher than the average during the 20 h residence of the new air, but the ΔNc gradually decreases for 8 h, at which time it becomes more negative, reaching a minimum of -90 %, 12 h after this air mass reached the mountaintop. During the next 6 h, the deviation trends back towards the average and is nearly zero at the end of this air masses residence at PDE. Unlike the trends in the AO-impacted clouds, not only do the ΔNc and ΔLWC trends track one another, they vary by approximately the same percentage during the first 5 h, while the ΔMVD remains constant at 25 %. This indicates that the changes in LWC are almost completely a result of variations in Nc for the first 5 h, after which the much sharper decrease in the ΔLWC (Fig. 7b) is due to the same rapid decrease in the ΔMVD (Fig. 7c).

When the air of AS history reaches the mountaintop that is already in cloud, the Nc, LWC, and MVDs are 75 %, 125 %, and 25 % higher, respectively, than their averaged values over the period and maintain these values for a couple of hours before decreasing. The ΔNc and ΔLWC decrease to a minimum of -35 % over 8 h, while the ΔMVD decreases rapidly to 0 % in less than an hour and remains constant for 8 h. During the remaining time that the AS air is reaching PDE the ΔNc increases from negative to positive, but even though the ΔLWC trend is positive, the ΔMVD is not, so that the LWC never again exceeds the average. This underscores the importance of the droplet size in controlling the LWC.

The trends in the ΔNc, ΔLWC, and the ΔMVD when the clouds are under the influence of the air masses with the SE history are the opposite of those experienced under AD and AS air masses. The initial deviations in all three of these parameters is -100 % from the average because the initial cloud droplet concentration that the air encountered was very low. The initial increase in the ΔMVD is not reflected in the ΔNc or ΔLWC because of the low concentrations. Once the cloud begins developing, however, the ΔNc and ΔLWC continue increasing, along with the ΔMVD until the end of the SE residency.

To encapsulate in a few sentences the physics behind the trends that we see in the deviations of the cloud microphysical parameters: (1) when ΔNc increases, new droplets are activated; (2) when ΔLWC is increasing, it can be due to increases in ΔNc and ΔMVD (droplets growing in size by condensation or coalescence); (3) when ΔNc decreases, droplets are being removed by evaporation or precipitation; and (4) when ΔLWC is decreasing, it may be a result of decreasing Nc and MVD.

In this context, the impact of the air masses with four different histories is as follows:

AO – activation and growth of new droplets followed by eventual evaporation due to entrainment and mixing of subsaturated air;

Following this summary, the trends in the microphysical properties suggest that only the AO and SE air masses introduce new CCN, whereas neither of the air masses of African history have CCN in sufficient concentration or difference in properties to generate new cloud droplets.

Previous studies at this location looking at cloud properties influenced by different air masses and aerosol types have come to mixed results. Eugster et al. (2006) and Spiegel et al. (2014) arrived at the conclusion that the presence of African dust particles could be acting as cloud condensation nuclei and thus must be the main aerosol responsible for different cloud microphysical properties. However, our results show otherwise and are in line with those seen in Raga et al. (2016), whose results show that air mass history, mainly precipitation and time below 500 m prior to reaching the sampling site, had a better relationship with the number of CCN.

As previously mentioned, not only are the properties of the aerosols important when assessing the effect of the air masses, but the dynamics and thermodynamics of these air masses are also potentially of relevance. The Puerto Rico mountaintop clouds are formed from rising and cooling air brought about by surface heating that leads to convection, orographic lifting or by cold frontal passage. Only the first two mechanisms occurred during the summers of 2013–2015. Given that surface heating is a daytime phenomenon, the clouds that formed during the incursions of air masses with AO and SE histories, around 16:00 LST on both occasions, are likely the result of convective activity. In addition, in both these cases, the clouds persisted well into the evening, even after these particular air masses were no longer in residence. Hence, the clouds that were already on the mountain during the SD and AS incursions were also most probably those that had formed earlier in the day by convective activities. So before finalizing our conclusions about how the clouds in the four cases were altered by the changing air mass histories, it is important to evaluate if there were obvious differences in the winds or thermodynamic variables.

Figure 8 summarizes the wind speeds and directions prior to and during the incursions of air masses, where the vertical dashed lines separate the time periods before and after the new air arrived. Other than the average wind speed being somewhat lower when the air is from the African Sahel, there are no obvious shifts in the velocity that can be associated with differences in the cloud properties. A significant increase in wind speed might lead to more turbulence and entrainment, or a large change in direction could be associated with a different magnitude of forcing; however, since neither of these shifts occurred, it suggests that atmospheric dynamics do not play a significant role in the cloud evolution in these cases.

The thermodynamic history of the air masses, i.e., the temperature and relative humidity (RH), tells a different story as is illustrated in Fig. 9, where the back-trajectory analysis summarizes how these two parameters evolved in time prior to the arrival of air at PDE.

As expected, the temperature histories follow those of altitude that were drawn in Fig. 4b, i.e., the African air masses, SD and AS, had been at lower altitudes and warmer temperatures 7 d prior to reaching PDE, while the AO and SE over the Atlantic were much higher and colder. All the trajectories converge on an average temperature of 18∘ ± 2∘ 24 h prior to arrival at PDE so that there are no differences that might impact the subsequent temperature of the cloud when mixed with this air.

The RH histories of the air masses are distinctly different from one another (Fig. 9b) and reflect to some degree both their temperature and altitude trends, in particular the air mass with the SE history that was within 1000 m of the ocean surface throughout most of its 7 d travel. The proximity to this water vapor source kept its RH above 80 % from 3 d back until reaching PDE. The decrease in RH at 20 h is a result of its ascent as it approached the island.

The RH trends of the AO, SD, and AS air masses are more puzzling and would require an analysis beyond the scope of the current study to explain the rapid increase in the RH of these air mass beginning at 30 h that leads to RH values above 90 % when these parcels reach the measurement site. If the RH had been much lower than 90 % when arriving at PDE, mixing it into the existing cloud would have led to more rapid evaporation of the droplets and dissipation of the cloud. However, as the time series of the size distributions shows, the dissipation happened hours after the arrival of the air masses, so mixing will decrease the concentration, not due to the removal of droplets by evaporation but just by mixing in droplet-free air.

It is important to acknowledge that we have imposed conditions on the analysis that might have biased the results towards the higher RH air at the time of arrival or 1 h before; however, as the time traces show in Fig. 9b, there was already a strong, positive trend of humidification many hours prior to the air mass arrivals.

To conclude the discussion, we return to the analysis of the accumulation of precipitation along the air mass trajectories that was first introduced in Fig. 4c. The rainfall that is reported from the HYSPLIT model is an estimate based on the humidity fields and atmospheric cooling derived from the meteorological fields that are used when the model is run. Although these are not observed values, previously published comparisons between in situ and satellite measurements have shown that it is a reasonable approximation that can be used to evaluate precipitation along the trajectories. In the present study, we are interested in how precipitation affects the population of CCN in the air mass that will eventually reach Puerto Rico. The model predicts that the majority of rainfall is happening in the SD and AS air masses more than 5 d prior to their arrival at PDE. This suggests that some fraction of the dust and biomass burning aerosols that might have been picked up over Africa will be removed by precipitation, well before they will have a chance to influence the mountaintop clouds in Puerto Rico. On the other hand, if this precipitation is a result of deep convection, while rainfall might remove some CCN, the vertical air motions that lead to the formation of clouds and precipitations will also transport CCN that are not washed out. This latter mechanism of aerosol transport is particularly important for those air masses with AO and AS histories whose altitudes from 7 d forward never brought them near the surface as was seen with the SD and SE air masses.

The evidence that argues for the insertion of new aerosols into the air masses by deep convection lies in the RH histories because the convection that would bring particles from the boundary layer into the free troposphere would also inject additional water vapor. Indeed, we see that 6 d back in the AS history, at the same time that the accumulated rain increases, the RH also increases, so that both water vapor and particles are being lifted vertically at the same time some rainfall is removing particles. On the other hand, the increase in rainfall accumulation experience by the AO and SD air masses is not accompanied by substantial RH increases.

Thus, to revise our former observations about the impact of these four types of air masses with different histories, the clouds that are impacted by AO air masses experienced an increase in concentration that can be associated with new CCN activation; however, given the evidence that these air masses had not been near the ocean surface, or received additional particles over the previous 7 d, it is likely the increase in RH of this air mass as it approached Puerto Rico that provided additional moisture to the existing cloud. This fresh influx of water vapor conceivably could reactivate CCN residuals leftover from evaporating droplets. The evaluation of winds, temperature, and humidity do not change our original conclusion about the SE air masses that are bringing anthropogenic CCN from the boundary layer that subsequently led to higher concentrations of droplets and that the primary influence of the SD and AS air is to dilute and eventually dissipate the clouds with no new droplets activated.

In an effort to expand the current database of measurements related to the interaction of aerosol particles transported from the African continent to the Caribbean and their subsequent impact on cloud formation and evolution, we conducted measurement campaigns on the Caribbean island of Puerto Rico during the summers of 2013, 2014, and 2015. On frequent occasions, the island was clearly inundated with air masses from the Saharan desert and African Sahel. Measurements were made at a coastal location of the aerosol optical properties and during these same periods, cloud microphysical properties were measured at a mountain site that was downwind from the coastal site. The wind velocity, temperature, RH, and rain rate were measured at both sites. Based on the averaged wind directions, neither of these sites were influenced by anthropogenic emissions from local sources.

The aerosol optical properties and particle chemical composition measured at the coastal site were used to identify 16 high-dust and nine low-dust events of African dust. During high-dust periods, the scattering and absorption coefficients, AAE, and nss-Ca2+ ion concentrations were higher and the SAE were lower than for low-dust periods.

The HYSPLIT back-trajectory model was used to identify air masses with four distinctly different characteristics based on their geographic locations prior to arriving at the Pico de Este (PDE) measurement site: midlatitude Atlantic Ocean (AO), Saharan desert (SD), African Sahel (AS), and arrival from southeast of Puerto Rico (SE). Time periods were selected for analysis during which these air masses were arriving continuously over durations ranging from 12–20 h when there was a cloud present on the mountain peak. The cloud droplet size distributions were measured with a droplet spectrometer over the size range from 5–90 µm, from which the microphysical properties, total number concentration (Nc), liquid water content (LWC), and median volume diameters (MVDs) were derived as a function of time.

Based on the atmospheric dynamics and thermodynamics and the trends in the microphysical properties, we arrive at the following conclusions regarding the impact of these air masses with four different histories on cloud evolution:

The AO air masses influence the reactivation of existing CCN as a result of the mixing of high-humidity air with the resident cloud and not due to the introduction of new CCN.

These conclusions do not support those by Spiegel et al. (2014), who concluded that microphysical properties of clouds at PDE were significantly altered by African dust. They do support the arguments put forward by Raga et al. (2016), who used similar measurements at PDE from the summer of 2011 to conclude that the meteorology has a larger impact on the cloud microphysics than the properties of the CCN.

Clearly, the conclusions posted here are somewhat speculative in nature and require a much more detailed and long-term measurement program, coupled with cloud and chemistry models to validate these speculations. Nevertheless, the results are consistent with previous studies and, most importantly, provide a well-documented set of measurements to enhance the current data set of similar observations in the Caribbean.

Currently in progress is the next step to extend this study with a more comprehensive suite of sensors over a period that covers several years in order to take into account the year-to-year variability in synoptic and mesoscale weather patterns that modify the trajectories of dust and pollution transported to the Caribbean. Following the passage of Hurricane María, in September 2017, 2 years after the current study ended, both the mountain and coastal research sites were destroyed with all the equipment. As a result, through funding from the National Science Foundation, these sites have been rebuilt and new cloud and aerosol instrumentation purchased and installed. These new measurements, along with simulations with the Weather Research Forecast (WRF) will directly address the questions raised in the current study and begin moving the conclusions from the realm of speculation to statistically supported facts.