A variety of measurements are used in this study, mainly performed at the Amazon Tall Tower Observatory (ATTO). The site is located in central Amazonia (2°08′48′′ S, 59°00′19′′ W, 150 m a.s.l.), about 150 km northeast of Manaus, the largest nearby city, with over 2 million inhabitants. Surrounded by a nearly pristine forest with an average canopy height of 35 m , the ATTO site is ideal for studying biosphere-atmosphere interactions in a tropical ecosystem. Atmospheric measurements are performed at two 80 m high towers, one equipped with an aerosol inlet positioned at a height of 60 m, and at the ATTO tower, which is 325 m tall, allowing an assessment of the atmospheric composition at multiple height levels above the canopy. All instruments are located in containers at the base of the towers, where temperature and humidity are controlled, and operate with a 2.5 µm cutoff size, encompassing the fine mode and part of the coarse mode particles. Additional information on the ATTO site can be found elsewhere .

In this study, the measurements at the ATTO site were performed by numerous instruments, continuously sampling air from heights of 60 and 325 m between May 2018 and December 2023, in 30 min averages. The period from March to November 2020 was excluded from the dataset due to a lack of maintenance on-site and calibrations of the instruments caused by the COVID-19 pandemic.

In this study, the definitions of the wet and dry seasons are a vital part of the methodology. In the literature, the usual definitions for the seasons at ATTO were defined by , with the wet season comprising the months of February to May, while the dry season spans from August to November. However, as the changes in the ENSO phase and other large-scale systems affect the precipitation regimes in the Amazon, the best description of the atmosphere allows for variable start and end dates for the seasons. The present study solely divides the year into two sections based on the aerosol concentrations, without transitional periods. Thus, the transitional periods can also be characterized as showing wet and dry season conditions based on the aerosol concentrations found during December, January, June, and July. For this purpose, the number concentration of accumulation mode particles at 325 m was chosen as the primary indicator of a change in the atmospheric patterns related to wet and dry seasons. This choice of indicator variable is based on the scavenging and deposition of accumulation mode particles during rainfall. This is much more common during the wet season, which means that the average number concentration of accumulation mode particles should steadily increase at the beginning of the dry season.

The choice of the 325 m height data was based on the better representation of the larger scale of the atmosphere at this height, whilst the 60 m measurements are more indicative of local processes within the canopy. Here, the transition dates were defined by the 1-week moving mean of the accumulation mode number concentration and a threshold determined by the 90th percentile of this variable during the fixed wet season (February to May). Once the moving mean becomes larger than the threshold and does not return to lower values, the dry season starts, and when it falls below the threshold without surpassing it again, the wet season starts. This method proved quite effective, as the 90th percentile threshold avoids ambiguity since the moving mean does not cross this value many times during the transition period, and the obtained transition dates roughly divide the year into two seasons of similar duration, as shown in Fig. S3. There was slight variation among the years, with the dry season beginning usually in early June and the wet season starting in early December. The large-scale ENSO events had some impact, as the dry season of 2023 (a strong El Niño year) was the longest, starting in late May and ending only in late December.

A cluster analysis procedure was performed to distinguish periods with similar aerosol optical properties associated with different particle sources and atmospheric processes. This analysis consists of grouping data points that show similar values of independent variables. For this study, the chosen clustering variables were the scattering and absorption Ångström exponents, and the eBC concentration, all measured at the 325 m height. These variables are not directly correlated, as both SAE and AAE are intensive properties and come from independent measurements of separate aspects of particle-radiation interactions, while the eBC concentration evaluates the amount of pollution in the atmosphere. This variable choice was motivated by the Ångström matrix, which is a well-known method of separating possible particle populations based on intensive properties, as well as the availability of measurements of the aerosol particles' optical properties. The choice of the 325 m measurements was based on a desire to represent a larger footprint area and larger-scale atmospheric conditions, whereas the ones from 60 m are more affected by local canopy processes and the nocturnal boundary layer. To minimize the influence of outliers, the data were organized into 3 h averages, similarly to , before applying the cluster algorithms and manually verified for occasional instrument failures. After these procedures, the final dataset consisted of over 10 000 samples, equivalent to 3.5 years of measurements, between May 2018 and December 2023.

There are many different clustering methods, each more suited to different applications and objectives within the algorithm, the data, or the clusters themselves . This study employed the k-means algorithm, which performed the best on commonly used tools to evaluate cluster assignments , and more details about the algorithm itself can be found elsewhere . Here, the sample set consists of arrays of the three predictor variables (SAE, AAE and eBC), which have been normalized through a Z-score calculation. This normalization is necessary for the k-means algorithm to make all variables equivalent, as they have different distributions and units.

Although the k-means clustering method can be applied simultaneously to the entire dataset, the main processes and drivers of aerosol population changes differ between the wet and dry seasons. Therefore, the dataset was first divided between wet and dry seasons, and the clustering process was performed separately for the two periods before naming the clusters. Furthermore, despite the need for variable equivalency within the algorithm, the best description of an aerosol population might not consider all variables equally. Although the eBC concentration directly measures the amount of pollution in the atmosphere, both the AAE and SAE are indirectly related to aerosol composition and size distribution, respectively. Weights were then attributed to each variable in each season before applying the k-means algorithm. During the wet season, low particle counts in the optically active range cause very low absorption and scattering coefficients, which greatly increase the uncertainties and fluctuations of the AAE and SAE values. Due to these higher uncertainties on the intensive properties, it was chosen to increase the importance of the eBC concentration for the algorithm by a factor of 1.5 during the wet season, applied directly to its normalized values. Similarly, for the dry season measurements, the size distributions for different conditions are more similar due to the substantial concentrations in the accumulation mode contributing to the scattering of radiation , which motivated the choice of decreasing the importance of the SAE values by a factor of 0.5 during this period before clustering.

Albeit this is an unsupervised method, meaning that the data points are not labeled before the execution of the algorithm, the number of clusters has to be chosen first. This complex task was performed carefully because the clusters need to have some physical meaning. The k-means algorithm was applied multiple times to each sample set, varying the number of clusters from 2 to 8, as shown in the example run of Fig. S4. The evaluation of the optimal number of clusters was made using silhouette scores (s), and the possibility of a physical interpretation of the clusters. For the wet season analysis, the silhouette score was highest in the 2-clusters case, with a clear physical meaning for each one, and the consideration of one more cluster did not add much information, which led to the final choice of 2 clusters for the wet season. Similarly, in the dry season, the s values of the two and three cluster cases were the highest. Including the third cluster improved the interpretation of the results, resulting in choosing 3 clusters for the dry season.

The sensitivity study shown in Fig. S5 corroborates the choices mentioned above for the attributed weights. For the wet season classification into two clusters, the eBC weight practically had no effect for values above 1.0. The separation of only the relatively rare, more polluted periods from the overall cleaner background typical of the wet season, is one of the desired results of this clustering effort. This is already achieved for an eBC weight marginally above 1.0, which indicates that the chosen value of 1.5 is within the desired range. For the dry season, the occurrence of African plume events is relatively scarce as well, with less than 15 % of the days being flagged as such by . In this scenario, these events are identified only for lower SAE weights, with the chosen value of 0.5 representing the most accurate separation from the regional biomass-burning episodes.

Throughout the year, the eBC concentrations at ATTO vary significantly, with the cleanest months being April and May (with median values of 0.025 µgm-3 at 325 m), while the highest concentrations were observed between August and October (median of 0.36 µgm-3). Furthermore, the eBC concentration was slightly higher at 325 m at times, with a statistically significant difference being observed during pristine-like conditions, in line with the findings of . The aerosol intensive physical properties shown in Fig.  also hint at a more profound change in the aerosol population throughout the year. The SAE is lower at 60 m during the whole year due to the consistent presence of coarse mode aerosols closer to the canopy , mainly from primary biogenic sources. The higher SAE values observed during the dry season result from a larger fraction of fine mode aerosols at both heights, mostly of anthropogenic origin, while the minimum observed between January and March at both heights indicates the higher abundance of coarse mode particles, marking the peak of Saharan dust influx .

During the wet season, there is significant variability in the SAE at 325 m, represented by the large variations in the interquartile ranges, which could result from low scattering coefficients, increasing the uncertainty of the derived exponent. In contrast, the annual cycle of the AAE is less pronounced, with lower values generally occurring at the dry-to-wet season transition and the beginning of the wet season (December to February). This is due to a larger BC fraction in the absorbing component of the aerosols, as during this time of the year, the intrusion of BC-rich plumes from Africa is frequent. The similar AAE values between the two heights indicate that no local sources of absorbing particles are relevant. The similarity of eBC at both heights is really impressive, showing that eBC is controlled by sources and processes that act similarly at 60 and 325 m, such as long-range transport of absorbing particles, unlike the findings of , obtained for a large urban area, where the observed height gradient of BC concentration is indicative of nearby ground sources of pollution. Furthermore, the diurnal cycles of these variables are shown in Fig. S6, where it was noted that both the SAE and eBC exhibit minimal variation in comparison to their seasonal cycle. Lastly, the diurnal trend in the AAE is more notable, which could result in a similar diurnal cycle for the cluster fractions, but this was not observed in the results, as the difference between clusters is larger than the magnitude of the diurnal cycles of the considered variables.

The separation of the data into 5 clusters (2 in the wet season and 3 in the dry season, as detailed in the methodology) is shown in the Ångström matrix in Fig. a, with the dot colors indicating each cluster, obtained from the 325 m measurements. The purple and blue clusters were assigned to the wet season data, while the green, yellow, and red clusters were within the dry season. Furthermore, Fig. S7 shows the Ångström matrix for the 60 m measurements highlighted with the colors that represent each cluster.

During the wet season, the association of samples into the blue cluster occurred solely on relatively high eBC concentrations (> 0.15 µgm-3), with low mean values of SAE and AAE, while the purple cluster indicates a cleaner atmosphere. Furthermore, the annual cycle shown in Fig. b indicates that blue cluster occurrence is quite common from December to February and almost nonexistent during April and May, consistent with LRT events carrying Saharan dust and biomass-burning particles from North Africa at the beginning of the wet season. These conditions are also characterized by low SAE and AAE values (medians of 1.0 and 1.13, respectively), indicating a large contribution of coarse mode particles and low BrC fractions, consistent with aged particles from faraway sources. Hence, from hereon, the blue cluster is referred to as “wet season LRT conditions”, and the purple cluster as “near-pristine conditions”. Figure S7 also shows that during near-pristine periods the SAE rarely surpasses 2.0 at 60 m, highlighting the role of larger particles found at this height. During the analyzed wet seasons, 81 % of the data was considered near-pristine conditions, while 19 % of the available measurements were classified as LRT events. This indicates that while the purple cluster defines the typical background conditions for the wet season, there might be some diluted LRT signal within these samples, as the rigorous pristine conditions defined in the literature are much rarer. Pristine conditions are defined in the literature by periods with absorption by aerosols below the instrument's detection limit, which is an eBC concentration below 0.011 µgm-3 for 30 min averages. These conditions are a subset only of the near-pristine cluster, with the lowest eBC and particle concentrations, comprising around 11 % of the conditions outlined by the purple cluster, or on average 35 d yr⁻¹.

In the dry season, red cluster samples show the highest AAE values and eBC concentrations, followed by yellow and green. Green cluster measurements are usually associated with relatively low eBC concentrations within the dry season, meaning there is little impact from biomass burning or other anthropogenic sources during these periods, which are more common at the beginning of the dry season (mainly June and July). This represents another period of “background conditions”, although for the dry season in this case, which could be characterized by different aerosol sources such as distinct biogenic activity. During the core of the dry season, from August to October, the more polluted clusters are more frequent, with occasional green cluster points intercalated, indicating lower pollution concentrations usually occurring after rainfall. The difference in the seasonality of these clusters, along with back trajectories calculated for continuous periods within each one (Fig. S8) and their mean values of AAE and SAE, indicate that the yellow cluster might be related to LRT conditions, with African biomass-burning particles.

Similarly, the red cluster is more associated with regional BB pollution episodes, as these periods show higher AAE values, indicative of a freshly emitted aerosol plume with a higher BrC fraction. The red cluster is more concentrated in October and November, suggesting a higher importance of South American BB in the latter part of the dry season, in agreement with . Thus, the green cluster is named “dry season low-impact conditions”, the yellow represents “dry season LRT events”, and the red cluster indicates periods with the influence of “regional pollution episodes”, as summarized in Table . The centroids of each cluster are also shown, which allows for the inclusion of more measurements into the same clusters in the future, by calculating the closest centroid to each sample, after the season separation. During the analyzed dry seasons, 66 % of the available measurements were classified as low-impact periods, 19 % as LRT events, and 15 % as regional pollution events. The abnormally large fraction of yellow-classified points in September might be caused by a lack of data during Sep 2022 and 2023 due to failures in the 325 m aethalometer (Fig. S2). During the dry season of the referred years, transport of regional or even local pollution from biomass burning was more common than in previous years, with the 2023 El Niño influencing the precipitation regime in Amazonia, increasing drought severity and the number of fires in the region . If the data had been available for these months, the contribution of red cluster points occurring in September would possibly be larger.

The aerosol populations represented by each cluster exhibit different physical and chemical properties. Figure a,d,g compares the distribution of the predictor variables among the clusters with measurements from the 325 m height, where the lowest eBC concentrations were associated with near-pristine conditions with a median value of 0.05 µgm-3. This cluster represents the background conditions of the wet seasons, with little or no influence of pollution. Meanwhile, the highest eBC concentrations were classified into the more heavily polluted clusters of the dry season (yellow and red), both averaging almost 0.50 µgm-3, which is consistent with the influence of biomass-burning aerosols. A large SAE dispersion also characterizes near-pristine conditions due to the higher uncertainty caused by very low scattering coefficients and a prominent Aitken mode (Fig. S9a, f). At the same time, the periods influenced by regional emissions of biomass-burning particles show the highest AAE values, with a median of 1.59. The number of accumulation mode particles, fine mode volume, and PM_2.5 mass increased from clusters 1 to 5, and the scattering coefficient followed the same trend, as expected.

During the dry season, LRT conditions showed lower SSA values than the regional pollution episodes, with median values and interquartile ranges of 0.82(0.79–0.85) and 0.88(0.85–0.90), respectively (Fig. j). The statistical significance of this difference was evaluated using both the Wilcoxon rank-sum test, which yielded a virtually null p-value, and a Z-test of the means, which resulted in a Z-score above the 99 % significance level, indicating that the SSA distributions for dry season LRT and regional BB influences are indeed different. This hints at a higher contribution of BC absorption during events of LRT and agrees well with the values obtained by . As shown in Fig. l, the fraction of the absorption at 440 nm attributed to BrC is on average 11 % during dry season LRT and 13 % during episodes of regional BB aerosols. This is a statistically significant difference pointing to the same conclusion, although such a small difference might not fully explain the radically contrasting AAE values between the two clusters. Amazonian BB aerosols are also known to have smaller BC cores , which is a factor in the AAE value of pure black carbon particles, being larger than 1.0 for diameters between 100 and 200 nm .

The average chemical composition of the non-refractive PM₁ aerosols during these events shows a higher eBC and inorganic fraction for the LRT conditions. Only a 3 % contribution of eBC and 12 % of inorganic compounds were found during the influence of fresh biomass-burning plumes. In contrast, dry season LRT events show 6 % of eBC and 14 % of inorganics (Fig. i). Furthermore, despite having similar eBC mass concentrations, the regional pollution episodes show a much higher aerosol load, with a 120 % increase in the accumulation mode number concentration, a 103 % increase in the fine mode volume, and between 50 % and 60 % in the PM_2.5 mass and scattering coefficient, in comparison with the LRT events, as shown in Fig. b, e, h, k.

LRT events in both seasons also showed the highest values of both real and imaginary parts of the refractive index. The real part, related to radiation scattering, reached values of 1.40(1.34–1.49) and 1.35(1.30–1.53) for LRT events in the wet and dry seasons, respectively, in contrast to mean/median values below 1.3 for the remaining periods (Fig. c). The imaginary part, indicative of light absorption, registered median values and interquartile ranges of 0.047(0.039–0.058) during wet season LRT events and 0.030(0.022–0.046) during dry season LRT conditions, much higher than the average values between 0.010 and 0.020 for the other clusters (Fig. f). These differences are statistically significant, and indicate that the aerosols arriving in central Amazonia during LRT events from Africa are highly efficient in absorbing radiation and much more refringent than those of local/regional biogenic origin.

Aerosols in central Amazonia are known for having a very stable chemical composition, in terms of the overall fraction of each species, despite the substantial differences in concentration between the wet and dry seasons . Pristine periods, a subset of the near-pristine cluster, consist of particles with high organic content (82 %) and low eBC fraction (1 %), while the more polluted periods of the dry season also show an organic fraction above 80 %. On the other hand, LRT events in the wet season are unique because their average aerosol chemical composition is significantly different, with much higher eBC (17 %) and inorganic (26 %) fractions. Comparing the periods classified as LRT in both seasons, the LRT conditions in the wet season also show much lower SAE and SSA values, consistent with highly absorbing particles and a significant presence of dust and coarse mode aerosols. The differences between these clusters can be attributed mainly to two factors. Firstly, the varying position of the Intertropical Convergence Zone (ITCZ) throughout the year, primarily bringing dust from the Sahara Desert during the Amazonian wet season and biomass-burning particles from Southern-Hemisphere Africa during the dry season. Secondly, the frequency of precipitation leading to wet deposition during the wet season and the different amounts of regionally influenced aerosols in the atmosphere upon the arrival of the African plume. During the dry season, the low precipitation leads to an increased lifetime of aerosols in the atmosphere, and the higher aerosol load of regional origin still dominates the physical and chemical properties observed at the site, whereas the lower concentrations found in the wet season lead to an increased importance of the African plume to the overall aerosol population.

Dry season low-impact periods, represented by the green cluster, are frequent during the transition months, mainly June and July, as well as November and December. In the wet-to-dry transition, precipitation decreases gradually, keeping the aerosols in the atmosphere for extended periods, leading to increased particle concentrations. However, biomass-burning aerosols have virtually no impact during these months, keeping the eBC concentrations relatively low. In contrast, precipitation rapidly increases during the dry-to-wet transition, leading to frequent wet deposition of aerosols and lowering the eBC concentration. Transitioning to wet season conditions with frequent rainfall and low particle concentration is quicker. This difference in the transition mechanisms favors the classification of points in the dry-to-wet transition period toward wet season clusters.

Furthermore, low-impact dry season and near-pristine wet season aerosols exhibit remarkably similar intensive properties, with relatively high SAE and SSA values, and the single scattering albedo reaching 0.86(0.84–0.89) and 0.86(0.80–0.91), respectively. The characteristic refractive indices of these two clusters are also similar, with relatively low values in both the real and imaginary parts, averaging around 1.28–0.014i. This indicates that the aerosol populations are quite similar, with the main differences being the average lifetime of the particles in the atmosphere and the total aerosol load, which is 3 to 4 times higher during low-impact dry season periods. Despite their similarities, the BrC fraction of the absorption is significantly higher during near-pristine and pristine wet season conditions, reaching 15 % and 21 %, respectively, against a median of 11 % for low-impact dry season periods. The importance of newly formed absorbing aerosols may be larger during the cleanest periods , as anthropogenic pollution in the atmosphere has little to no impact. The presence of absorbing fungal spores, more abundant in the wet season, might also contribute to this difference, as many spores are pigmented with melanin .

This section explores the similarities and differences observed for the analyzed variables at the two height levels among the identified clusters. The median 60 and 325 m PNSDs for the conditions outlined by each cluster are shown in Fig. S9, with the fitted parameters of the aerosol size distributions given in Table . The accumulation mode mean geometric diameter is larger at 325 m for all clusters, highlighting the occurrence of upward transport and aging throughout the year. The difference between the 60 and 325 m Dacc is larger during dry season LRT events, reaching 13 nm in these periods. This finding yields similar conclusions reached by , but now extended for the entire year, and the obtained values mostly agree with , showing consistency on these studies. Overall, the accumulation mode contributes to a larger volume fraction of the fine mode aerosols at 325 m, especially during near-pristine conditions, when the Sub-50 nm and Aitken modes are more prevalent in both heights. A larger accumulation mode also characterizes Saharan dust events, shifted towards the largest diameters with the peak located at 180.6 nm for the measurements at 325 m and at 176.2 nm for the particles measured at the 60 m height.

As previously discussed in Sect. , the coarse mode particle concentration is always higher at 60 m, which is the cause for the lower overall SAE. Additionally, the SAE gradient is more significant during near-pristine conditions, as shown in Fig. a, due to the lower aerosol load in the fine mode at both heights leading to the exacerbation of the coarse mode relative contribution difference between 60 and 325 m. The AAE is similar between the two heights throughout the year, despite minor differences found during near-pristine conditions, when it is higher at 60 m, and in regional pollution episodes, when it is higher at 325 m. The significance of these differences was evaluated using the Wilcoxon rank-sum test, which yielded virtually null p-values. Further verification was achieved using a Z-test of the means, confirming Z-scores above the significance level of 95 % for both cases, indicating that these differences are indeed significant. These differences could indicate a larger contribution of pigmented biogenic particles such as fungal spores closer to the canopy, which becomes statistically significant during cleaner periods when the influence of other particle sources is diminished.

The SSA is lower at 325 m at all times, as the aerosol population closer to the canopy is less absorbing. This difference is more prominent during the wet season, with greater importance of the aging processes of primary particles to the overall aerosol population . The more advanced stages of aging and thicker particle coating at 325 m are consistent with a higher aerosol volume at tower top height, found on most periods. The larger differences were found again in the wet season, with a fine mode aerosol volume 38 % higher in near-pristine conditions and 55 % higher during Saharan dust events. For the dry season clusters, low-impact conditions and regional BB episodes showed the highest differences in aerosol volume towards higher values at 325 m. At the same time, the pattern was reversed during LRT events, as higher volumes were observed at 60 m. The higher aerosol load during the dry season and the higher average lifetime of particles in the atmosphere lead to smaller SSA and fine mode volume gradients.

The chemical composition of the aerosols at different heights shows minor differences that might be related to canopy sources, vertical transport processes, and advection of air parcels with different origins. The organic fraction is identical at both heights during the dry season, while a difference of, on average, 4 p.p. (percentage points) was found during the wet season, with higher values at 60 m. This shows that biogenic emissions from below the canopy are increasingly important during the wet season. Moreover, sulfate concentrations and fractions were significantly higher at 325 m during wet season LRT events, while smaller differences occurred for the remaining conditions, highlighting the occurrence of Saharan dust intrusion. Finally, statistical significance was achieved for larger ammonium and nitrate fractions observed at 325 m, which could be attributed to biomass-burning aerosols transported together with the dust in LRT events or the formation of NO_x by lightning at higher levels , as the differences appear to be more prominent during the rainiest periods, or in-canopy reactions with emitted VOCs .

Figure a and b compares the 60 and 325 m refractive indices. The real part has a more pronounced vertical gradient during the wet season, with higher median values closer to the canopy of 1.47 and 1.58 during near-pristine and LRT events, respectively. In contrast, medians of 1.29 and 1.41 were obtained at 325 m, highlighting the effectiveness of coating in reducing the refringence of aerosols. Once again, the statistical significance of these claims was verified with the Wilcoxon rank-sum test, which resulted in virtually null p-values. Dry season measurements yielded a less intense vertical gradient and more stable values throughout the season, especially at 60 m height. Furthermore, the imaginary part is almost always higher at 325 m, strengthening the current understanding of more absorbing aerosols at the top of the tower. Despite this, some variations occur, with low-impact conditions showing no statistically significant difference between the two heights (at a median of 0.014). At the same time, LRT events in both seasons have the largest gradient (0.051 at 325 m and 0.038 at 60 m during Saharan dust events, and 0.035 against 0.025, respectively, during dry season African BB influx).

The brown carbon fraction of the total absorption at 440 nm also shows large differences among the clusters regarding its vertical profile (Fig. c). The largest gradients were found during near-pristine conditions when the median BrC contribution is more than 50 % higher at 60 m, agreeing with previous results . The processes that lead to this difference can be elucidated by evaluating the diurnal cycles (Fig. d, e), as the BrC fraction increases during the night at 60 m, following the emission of absorbing organic compounds by the forest, from primary biogenic aerosol particles , which are more prevalent close to the canopy . Tropical fungi are known to emit their spores at night when humidity is high , to protect them from the intense UV radiation. As an additional protection measure, most spores use melanin as a pigment, which makes them highly absorbing in the 440 nm range , and might be a factor in the BrC gradient and diurnal profile.

In addition to the emission cycle, the nocturnal stable boundary layer development traps these particles at lower levels. The stable layer also favors the dry deposition of accumulation mode particles, as the total eBC concentration at 60 m drops during the night. The BrC fraction then decreases after sunrise due to photochemical degradation and the start of turbulent processes and vertical transport. In contrast, at 325 m the BrC contribution increases during the early morning and afternoon, as BrC-rich aerosols are transported upwards by turbulent vertical mixing. However, more polluted conditions of the dry season showed a reversed gradient, with higher BrC fractions at 325 m, although still at very low values around 10 %. These periods also show mainly unclear and noisy diurnal cycles, as there are few simultaneous measurements available at both heights.

In the previous sections, we showed how the clustering separation was able to distinguish different atmospheric conditions by grouping periods with similar aerosol populations. For these different particle groups, the definition of mass scattering efficiency (MSE) is the amount of scattering that a unit of aerosol particle mass can produce. The MSE can also be used to characterize the particle populations, as it indicates the scattering power of an aerosol type and relates mass concentration to visibility and aerosol optical thickness. Eq. () describes the linear relation between the mass concentration and scattering coefficient, 3σscat,λ=αscat,λ×(MNR-PM1+M0), where σscat is the scattering coefficient measured by the nephelometers, αscat is the mass scattering efficiency, MNR-PM1 is the non-refractive fine mode mass, estimated by the ACSM, and M0 is the aerosol mass that scatters radiation but cannot be measured by the ACSM, mainly part of the coarse mode between 1 and 2.5 µm, as well as dust particles and BC. This is a widely used method, applied by previous studies , but the assumption that M0 is approximately the same for all considered points is flawed, especially during periods with large coarse mode contributions, which most-likely leads to overestimations of the MSE values. Note that Eq. () is wavelength-dependent, so the following analyses were performed considering only the scattering coefficients at 525 nm. For the linear regressions applied to the datasets corresponding to each cluster, the scattering coefficient and aerosol mass uncertainties are important steps to achieve the correct values of αscat, and they were determined following previous works. and determined that Aurora and TSI nephelometer measurements at 1 min time resolution show a 3.0 % uncertainty, which converts to 0.22 % in 3 h averages, as utilized in this study. This contribution is combined with a fixed value of 0.13 Mm-1, related to the instruments' lower detection limit. Similarly, the mass uncertainties are comprised of a relative term of 1.5 % at the utilized time resolution , and an absolute value term of 0.08 µgm-3 related to the detection limit . These errors were combined with a statistical uncertainty, represented by the standard deviation of the means for the 3 h averages for both scattering and aerosol mass. In addition, to avoid biases that can occur with linear fitting methods that consider uncertainties on both variables, the x-axis uncertainty was converted to the y-axis iteratively, and a standard least squares fitting was applied.

Figure shows the correlation plots between the aerosol mass and scattering coefficient measured at 60 and 325 m during the conditions flagged by each cluster, while Table  summarizes the values found by the linear regressions for both heights, which were statistically significant for all cases with the fitted coefficients showing p<0.05. The correlation is higher for the dry season clusters due to the higher aerosol mass and reduced relative uncertainty, as well as an increased fine mode fraction, which leads to a better estimation of the total aerosol mass by the ACSM. The average MSE at 325 m is always lower than at 60 m, as the presence of freshly emitted primary aerosols is more dominant closer to the canopy. These particles are very efficient in scattering radiation, while more advanced aging stages and larger particle coating are responsible for a decrease of MSE at higher levels.

On top of this general trend, distinct features can be observed for each cluster. The MSE at 325 m was nearly identical during the wet season clusters, reaching (3.96 ± 0.07) m2g-1 at near-pristine conditions and (4.09 ± 0.17) m2g-1 for LRT events. In fact, when performing a Z-test of these values, statistical difference is not achieved within a 95 % confidence interval (Z=0.7). These results show remarkably stable values despite the very different aerosol compositions of the particle populations that comprise these clusters. This stability, however, is not observed at 60 m, where the mean MSE values increase from (5.99 ± 0.14) m2g-1 during near-pristine conditions to (7.34 ± 0.54) m2g-1 during Saharan dust events, representing a 23 % average rise from one cluster to the other. Note that these values are probably overestimated due to the large coarse mode contributions during wet season conditions, especially at 60 m. Similarly, low-impact and dry season LRT conditions also showed nearly the same mass scattering efficiencies at 325 m, at (3.60 ± 0.05) and (3.70 ± 0.20) m2g-1, respectively, compatible within the 95 % confidence interval, and around 9 % lower than the wet season average. Regional pollution episodes showed slightly higher scattering efficiency, as the aerosol population is comprised of more freshly emitted particles, resulting in an average MSE of (4.43 ± 0.10) m2g-1.

Furthermore, the linear coefficients of the regressions yielded different amounts of dust/coarse mode mass, which the ACSM did not measure. Clusters 2 and 4, associated with LRT events, showed higher M0 values at 325 m, of (1.38 ± 0.09) and (3.19 ± 0.45) µgm-3 respectively, in comparison with the background conditions, indicating a high contribution of coarse mode and dust particles during these events. Wet season near-pristine, dry season low-impact conditions, and regional pollution episodes registered lower values of M0, indicating a relatively lower impact of coarse mode and dust particles at the 325 m height during these conditions.

Low-impact conditions showed the lowest MSE at both heights, with an average value of (5.11 ± 0.05) m2g-1 at 60 m, agreeing with the values found by previous studies for the cleanest periods of the year. A direct comparison with near-pristine conditions (mean MSE of (5.99 ± 0.14) m2g-1) is possible, as the aerosols that compose these periods have similar intensive properties. As previously stated, the main difference between these two clusters is the average lifetime of the particles in the atmosphere due to the less frequent wet scavenging by rainfall. This increased lifetime during low-impact dry season periods allows for aging processes to occur, lowering their MSE compared to near-pristine periods. Aerosol coating is usually comprised of organic compounds and water vapor, both of which act as an absorbent layer, decreasing the scattering efficiency of the particles , as well as increasing the aerosol volume and mass faster than the scattering cross-section. Furthermore, LRT events significantly affect the scattering efficiency at 60 m, especially in the wet season, with both wet and dry season LRT events showing higher MSE values (reaching (7.34 ± 0.54) and (5.54 ± 0.18) m2g-1, respectively) than during low-impact conditions, in contrast to what was found for the 325 m height. Based on these differences, one can postulate that the more aged aerosols are almost permanently found at 325 m. In contrast, at 60 m, the particle population is less aged and thus more responsive to sudden changes in chemical composition and lifetime of the arriving particles from LRT, resulting in a more variable MSE at different events. Furthermore, the dry season scattering coefficients and NR-PM₁ masses are more correlated than the wet season measurements at 60 m, as coarse mode particle bursts, mainly comprised of bioaerosols , are frequent at this height during the wet season, when fungi are more active. The ACSM does not measure these particles, but they do scatter radiation, leading to the events shown as apparent vertical lines in Fig. f, g. These measurements also show the lowest values of SAE within each cluster, reinforcing the higher importance of coarse mode particles during these periods (Fig. S10).