The data used for this study and presented here originated from the measurement site at Mindelo, which is a city on the northwest coast of the island São Vicente belonging to the Cabo Verde Islands (Fig. ). They are located downwind the Saharan desert and in the trade wind zone with the predominant wind direction being northeast in the lower altitudes. The measurement site at the Ocean Science Center Mindelo (OSCM, 16.878° N, 24.995° W, 10 ma.s.l.) is a coastal site. The marine influence is high for the entire island. Due to the northeasterly trade winds, the measurement site is also affected by the anthropogenic activity on the island. In addition to the measurements at Mindelo, long-term observations of a sun photometer operated in the Aerosol Robotic Network AERONET; on the island Sal (similar latitude like São Vicente, but somewhat more east, see Fig. ) were used for a long-term climatology, setting the measurements from Mindelo into a larger context.

At the OSCM, the setup of an ACTRIS aerosol and cloud remote sensing facility has started in June 2021 in the framework of JATAC , which was initiated by the European Space Agency (ESA). The ground-based component of JATAC, called ASKOS , took place in three intense phases in September 2021, June 2022, and September 2022. Several international institutions were involved with the main goal of collecting synergistic measurements for a quality-assured reference data set for the calibration and validation activities of ESA's satellite Aeolus . Aeolus was equipped with a wind lidar operated at 355 nm, delivering mainly wind products . Additionally, aerosol retrievals were available . During this campaign, Aeolus was measuring directly over Mindelo each Friday evening at around 19:30 UTC see, e.g..

Amongst others, the station is equipped with the multiwavelength-Raman-polarization lidar PollyXT and a CIMEL Sun Sky Lunar photometer of type CE318-T, which is also part of AERONET . Both instruments were already used for a previous study of , where their capabilities are described in detail. For the study presented here, the lidar measurements of the particle backscatter coefficient and the particle linear depolarization ratio, each at 355, 532, and 1064 nm, and of the extinction coefficient and the lidar ratio at 355 and 532 nm were relevant as well as the backscatter-related and extinction-related Ångström exponents between the different wavelengths. The PollyXT system is equipped with two receiving telescopes, one for far-range (FR) and one for near-range (NR) measurements. While the FR measurements reach higher altitudes above the lidar full overlap at 800 m above the lidar;, the NR measurements are more accurate in lower altitudes up to 2 km above the lidar and reach closer to the ground than the FR measurements full overlap at 120 m above the lidar;. Whenever available, NR measurements were preferred to characterize the lower atmosphere. Uncertainties of the lidar-derived optical properties are the statistical error of the particle extinction coefficient, a minimized systematic relative error of 15 % for the particle backscatter coefficient and a constant absolute error of 0.02 at 355 nm and of 0.01 at 532 and 1064 nm for the particle linear depolarization ratio, as intensively described in . The errors of the lidar ratio and of the Ångström exponent were calculated via the analytical Gaussian error propagation. From the AERONET sun photometer, the level 2.0 aerosol optical depth (AOD) was used for comparison with the lidar measurements and with the long-term climatology of the AOD from Sal. Uncertainties for a newly calibrated sun photometer are <±0.01 for wavelengths larger than 440 nm and <±0.02 for shorter wavelengths

The above-described PollyXT lidar is, in addition to the ACTRIS network, operated in the international network PollyNET . Within PollyNET, vertical profiles of the aerosol optical properties are derived and calibrated automatically via the PollyNET processing chain . To retrieve accurate profiles, the raw signal needs to be averaged over a larger cloud-free period (usually around 1 h), which is searched for by the implemented cloud screening algorithm. Clouds are detected using the signal gradient method, which identifies large slopes in the vertical profile of the photon count rate, occurring at the cloud base. If the vertical slope exceeds a certain threshold, the corresponding profile (raw resolution of 30 s) is flagged as cloudy. In our study, we used version 4.0 of the PollyNET processing chain , in which the threshold for the slope of the photon count rate has been set to 0.7×105 MHzm-1. To obtain vertical profiles of the optical properties, the raw signal is averaged over 1 h if cloud conditions allow. However, a minimum of 15 min of contiguous cloud-free raw profiles is required. Thus, the automatically-retrieved profiles of the aerosol optical properties are averaged over time periods between 15 min–1 h, depending on the cloud conditions. Version 4.0 of the processing chain also provides outputs of the lidar-based target categorization version 2, which, in addition to the elastic backscatter signal, uses the Raman one as well . From the target categorization, a cloud information product is derived, including, e.g. the cloud base height. Furthermore, the one-step polarization-lidar photometer networking algorithm POLIPHON; is implemented in the processing chain so that the profiles of the aerosol optical properties are separated into dust and non-dust components following the methodology described in . Like in , the PollyNET processing chain uses by default 0.31±0.03 and 0.23±0.03 as thresholds for the dust particle linear depolarization ratio at 355 and 532 nm, respectively, and 0.05±0.01 as threshold for the non-dust particle linear depolarization ratio at both wavelengths. Thus, the dust fraction can be calculated from the ratio of the dust backscatter to the total particle backscatter coefficient. The uncertainty of the dust fraction was calculated via the analytic Gaussian error propagation using a relative error of 15 % for each, for the total particle backscatter coefficient and the dust backscatter coefficient. The profiles of the aerosol optical properties are vertically smoothed using a moving average filter. The resulting smoothing length is 382.5 m (51 bins) for the NR-measurements and 742.5 m (99 bins) for the FR-measurements, which are the standard values used in the processing chain.

For all of the 73 cases, aerosol layers were defined, using a visual inspection. Although automatic retrievals for the detection of the PBL top height or the aerosol layer top height exist, the application of an automatic algorithm to the PollyNET processing chain to detect the lower and upper boundaries of (multiple) lofted aerosol layers is still under development. The algorithm of can be used only for the detection of the uppermost height where aerosol is present and, thus, is not convenient for defining vertically-homogeneous aerosol layers as well as for cases in which more than one lofted aerosol layer exists. This algorithm also cannot detect the lower boundaries of the lofted aerosol layers. Furthermore, the inhomogeneity of our data set (missing profiles at different wavelengths for different cases) was not a good precondition for applying an automatic algorithm. Therefore, the visual inspection was considered to be the best approach for this study to obtain an optimal data set, including as much profiles of the different aerosol optical properties as possible, while the automatized detection of the aerosol layer top height according to was used additionally for checking the consistency with the manually-defined aerosol layer top heights and, thus, the applicability of an automatic retrieval to this data set.

For the definition of the PBL, the NR-measurements of the particle backscatter coefficient at the available wavelengths were used. The PBL top height was manually defined in the middle of the first significant gradient of the backscatter coefficient. For the lofted aerosol layers, the FR-measurements mainly of the particle backscatter coefficient and the particle linear depolarization ratio were used in combination with the backscatter-related Ångström exponent. In four cases, only the extinction coefficients and the extinction-related Ångström exponent were used as all the other optical properties are not available. Also for the lofted aerosol layers, no specific wavelength but all available wavelengths were used for the definition of the layer boundaries. Similarly to the definition of the PBL top, the gradient of the backscatter coefficient gives a first impression for the location of the layer boundaries. As this gradient is usually less pronounced for the lofted layers than for the PBL, the intensive optical properties were used as an additional metric. The layer boundaries were set such that the depolarization ratio and the Ångström exponent remained almost constant within one layer. Usually, there were also transition zones for these two properties at the bottom and the top of the layer. As they are more an effect of vertical smoothing than of aerosol mixing at the layer edges, they were excluded by the quite restrictive layer definition. If indicated by local minima in the backscatter coefficient or the particle depolarization ratio or by noticeable vertical differences in the intensive aerosol optical properties (i.e. differences of at least 0.05 and 0.5 in the depolarization ratio or the Ångström exponent, respectively), the lofted aerosol layer was split into sub-layers. Transition zones of around 100 m were usually neglected between the sub-layers in favor of a better visibility in the plots. However, excluding or including these few meters should affect the results only in a negligible way as the gradient of the above mentioned properties is much weaker between the sub-layers than at the bottom and top of the entire layer. The layer boundaries within this visual inspection were defined as full hundred meters, i.e. an uncertainty of ±50 m was assumed for the derived heights.

For the defined aerosol layers, layer mean values were calculated for the intensive properties (lidar ratio, Ångström exponents, particle linear depolarization ratio, and dust fraction), while the extensive optical properties (total and dust backscatter coefficients and particle extinction coefficient) were integrated vertically. Because of unreasonable values below, profiles of the backscatter and extinction coefficients were cut-off at 100 m and between 400–500 m above the lidar, respectively. For the integration within the PBL, the profiles were interpolated to the ground, assuming the values at the cut-off heights to be constant down to the ground. In case of the extinction coefficient, the integrated value is equivalent to the layer-AOD. The sum of the layer-AODs at 355 and 532 nm was compared with the columnar AERONET AODs at 340 and 532 nm, respectively. Because the directly measured AERONET AOD at 500 nm was not available for the complete study period, we calculated the AODs at 500 nm (for comparison with the long-term climatology) and at 532 nm (for comparison with the lidar AOD) from the AERONET AOD at 440 nm, using the AERONET Ångström exponent between 440–870 nm. As the sun photometer measurements exist only for daytime and the lidar-based AOD was derived during the night, the latest AERONET hourly mean from the day before was averaged with the first hourly mean of the day after. If only one of both measurements was available, this single hourly mean was taken as reference. For the uncertainty of the layer mean optical properties, layer means of the errors of the lidar and depolarization ratio, the Ångström exponents and the dust fraction were calculated, while for the integrated values the errors of the backscatter and extinction coefficients were used as input for the analytic Gaussian error propagation.

As mentioned above, the algorithm of was applied and the results were compared with the visually-defined top height of the uppermost lofted aerosol layer. This comparison is shown in Fig. . In the automatic algorithm, the layer top of the uppermost aerosol layer is set at the height where the FR-particle backscatter coefficient at 532 nm falls below a certain threshold, i.e. 0.1 Mm-1sr-1 for the first time. For the estimation of the uncertainty of the derived height, the algorithm was applied to a slightly varied threshold of the particle backscatter coefficient (±15 %, which is in agreement with the uncertainty of the particle backscatter coefficient) and the differences between these heights and the previously derived layer top height were calculated. The results are used as asymmetric error bars for the automatically-retrieved aerosol layer top heights in Fig. . As both methods, the visual inspection and the algorithm of , strongly differ in their working principle, differences in the retrieved aerosol layer top heights are expected. Generally, the automatically-retrieved heights are around 200 m larger than the visually-defined ones. Furthermore, the algorithm of was developed for Tajikistan where the aerosol conditions are different and no low and well-defined marine PBL with a well-defined dust layer above exist but a mixture of dust and pollution from ground to several kilometers of altitude with often geometrically complex lofted layers. The aim behind the algorithm was to detect the top of the first significant layer of aerosol over Tajikistan without considering the layers above. Thus, in some cases, the automatic algorithm failed and did not detect the lofted aerosol layers if the backscatter coefficient dropped below the threshold already between the PBL and the lofted layers. The two outliers of the error bars in Fig. are from 9 December 2022 and from 2 June 2023. In the December case, the backscatter coefficient at 532 nm was smaller than the threshold value already in the altitude range between the PBL and the lofted aerosol layer. Thus, the algorithm stopped and detected the PBL top as overall aerosol layer top height. With the 15 % smaller threshold in the error calculation, the correct overall layer top height was detected (upper error bar). In the June case, the algorithm detected the correct aerosol layer top height but in the error calculation (backscatter threshold increased by 15 % for the lower error bar), the PBL top was detected. In all the other cases, the algorithm detected altitudes closely around the layer top height also with the increased and decreased threshold values. Due to the explained limitations of the algorithm of for our dataset, the automatically-retrieved aerosol layer top heights are not considered in the following analysis and discussion.

An overview of the complete measurement period in terms of the vertically-resolved temporal development of (a) the calibrated attenuated backscatter coefficient at 1064 nm and (b) the volume depolarization ratio at 532 nm is provided in Fig. . The lidar measurements provide a good basis for studying the annual cycle of the aerosol in the atmosphere above Mindelo. The local PBL, characterized by strong backscattering signal usually up to about 1 km height (red to white colors in Fig. a) shows no pronounced seasonal variation. During all the time, clouds could be identified at the top of the PBL, represented by the white color, i.e. very strong backscattering signal. The volume depolarization ratio in the PBL usually was low (blue colors in the lowermost kilometer of Fig. b), indicating the presence of spherical particles (marine aerosol and no dust). Very rarely, the volume depolarization ratio exceeded 0.1, e.g. in February 2022, which means the presence of some non-spherical dust particles. Above the PBL, the SAL was visible most of the time. The top height of the aerosol layer strongly varied throughout the two years. Beginning at around 6 km height on 1 July 2021, the aerosol layer top height decreased to around 3 km height in the time of November 2021–February 2022. Afterwards, an increase in the aerosol layer top reaching a maximum of around 6–7 km height in June 2022, followed by a similar pattern as described for the previous period was observed. The backscatter strength in the lofted aerosol layer was much smaller than in the PBL while the volume depolarization ratio varied from values up to 0.3 during the months of northern hemispheric (NH) summer and fall (June–September) to values around 0.1 or even less during the months of NH winter and spring (November–March). The higher values of the depolarization during the summer months indicate the presence of non-spherical particles, i.e. desert dust during that time, while the lower depolarization during NH winter point to a contribution of spherical particles, i.e. smoke as it is known from the previous campaigns.

The temporal development of the manually-defined top height of the uppermost available aerosol layer of the Fri/Sat cases (could be also the PBL if no lofted layer was present) is also illustrated in Fig. . Seasonal mean values and their standard deviations for December–February, March–May, June–August, and September–November are added as numbers. Concerning the temporal evolution of the layer top height, from May to December, clear trends with low short-term variability were observed. Until August, the aerosol layer top heights increased and aerosol was present up to heights of 7 km. Afterwards a decrease of the layer top height was observed. In November and December, aerosol usually was present only below 4 km or even below. From January to April in both, 2022 and 2023, there was slightly more variability in the layer top height, which is in agreement with the observations from SAMUM–2a, showing more diverse aerosol structures during January and February 2008.

A more detailed and layer-resolved overview of the geometrical properties of the Fri/Sat cases is given in Fig. a in terms of time series of the PBL top height and the vertical extent of the lofted aerosol layers. PBL top heights ranged from values as low as 400 m up to 2 km, with the highest values being observed between October–April. The highest observed PBL top heights (between 1–2 km) are often associated with the absence of lofted aerosol layers. However, during October and April, the variation of the PBL top height was large and also low PBL top heights of approximately 400 m occurred. During NH summer, the top height of the PBL was mainly in the range of 700 m up to around 1 km. The top height of the lofted layer distinctly increased during NH spring and summer and was usually lower in NH fall and winter as discussed already in Sect. . Generally, the lofted layer was geometrically thicker during NH summer having an extent of up to 4–5 km. During NH fall and winter, the occurrence of lofted aerosol layers was more diverse. Shallow layers of around 1 km extent as well as layers of around 4 km extent were observed. Sometimes also no lofted layer was present. Mostly, only one lofted aerosol layer was occurring but in 27 cases, two or even three sublayers were identified. These cases are distributed all over the complete year. Multiple Saharan air layers which were distinctively separated from each other were observed in only two further cases, namely on 13 and 27 January 2023.

Besides the geometrical thickness of the layers, the layer-resolved AOD at 532 nm was calculated from the lidar-derived extinction profiles and is shown in Fig. b together with the columnar AERONET AOD at 532 nm. In most cases, the AERONET AOD agreed well with the sum of the layer AODs, proofing the concept of this work. The first five cases are not representative as the NR-observation capabilities were not available and, thus, no AOD for the PBL could be retrieved leading automatically to an underestimation of the total AOD derived by the lidar compared to the one of AERONET. The general agreement of both methods to derive the AOD emphasizes the validity and representativeness of the selected profiles while the lidar measurements allow a layer-resolved evaluation of the AOD.

Strong annual variation can be identified in the total and the layer-AOD. Large total AODs up to values of around 0.5 and 0.6 were observed in June–July 2022 and in July–August 2023, respectively. In these periods, the contribution of the lofted layers to the total AOD was usually at least 50 %. Furthermore, a correlation with the layer thickness was found, i.e. during these summer months, the observed lofted layers were, both, geometrically and optically thick. The opposite was observed in the spring season, both, in 2022 and 2023, when geometrically thick lofted layers (extending up to 4 km) contributed to only a small fraction of the AOD (AOD values usually <0.1). During November–February, the total AOD of <0.1 up to around 0.3 was generally lower compared with June–August. Furthermore, the contribution of the lofted layer was usually smaller than the one of the PBL, i.e. the PBL contributed most to the total AOD during November–February. However, single outliers concerning the contribution of the layer-AODs to the total AOD were observed, e.g. on 24 February 2023, when the total AOD was around 0.55 and a lofted aerosol layer, caused by smoke transport from Africa, with an extent of around 4 km and an AOD of around 0.33 was present. Another outstanding pattern was observed in September/October 2021, when the total AOD reached comparably high values of almost 0.9 driven by a large contribution (60 %–80 %) of the PBL. This anomaly was caused by the volcanic eruption at La Palma, Canary Islands, causing the transport of sulfate aerosol in the PBL towards Cabo Verde as described in . The evening of 24 September represents the largest total AOD of all analyzed Fri/Sat cases with a value of 0.88 at 532 nm. Additionally to that case, the volcanic influence was evident for at least two more dates: 1 and 15 October 2021.

The observed annual variations of the total AOD can be also seen in the long-term climatology (Fig. a). The monthly mean AERONET AOD at 500 nm from Sal, including the years 1999–2024, is lowest in November and December with values slightly above 0.2, starts to increase from January to May (mean values of 0.3–0.4) to a maximum in June (around 0.55) and decreases again from July to October with values between 0.5–0.35. Using these long-term observations, the selected Fri/Sat cases and the period 2021–2023 in general can be set into a larger context. The monthly mean AOD at 500 nm from Mindelo from the years 2021, 2022, and 2023 (star symbols in Fig. a) were often lower than the climatological monthly mean values which points to a generally lower aerosol load over Cabo Verde in our study period. More specifically, from January to June in 2022 and 2023 and in June 2021, the monthly mean AOD was around 0.1–0.2 lower than the climatological mean values. Only the monthly mean AOD of April 2021 was exceptionally high compared with the climatological mean but does not need to be further discussed here because it was before our study period. The months July–September 2021 showed a similar aerosol load like in the climatological long-term, whereas in 2023, the mean AODs of these months were also around 0.1 lower than the long-term monthly means. For 2022, the monthly mean AOD agreed with the climatological one in July and August, but was also smaller in September. The best agreement with the climatological mean values was found for October–December in all three years (2021, 2022, 2023), except for December 2022. These deviations of the study period from the climatological long-term statistics lead also to differences between the monthly mean AODs (500 nm) of the Fri/Sat cases and the long-term monthly means if the selected cases are on average representative for the Mindelo monthly mean of the complete period 2021–2023 (e.g. May, cf. Fig. b). In other months (e.g. February and June), the selected cases overestimate the average aerosol load of 2021–2023 at Mindelo and, thus, agree with the climatological monthly means (cf. Fig. b).

In Fig. c, the layer mean dust fraction at 532 nm is illustrated, which is an indicator for the contributions of dust and smoke/pollution components to the observed aerosol mixtures. The corresponding integrated dust backscatter coefficient at 532 nm for the lofted aerosol layer is shown in Fig. d. Both were derived with the POLIPHON algorithm, assuming a dust particle linear depolarization ratio of 0.31 as typically particle depolarization values between 0.25–0.35 have been observed for pure Saharan dust conditions . However, due to this assumption that pure dust conditions correspond to a particle linear depolarization ratio of at least 0.31, the layer mean dust fraction hardly exceeded 0.9 even in the lofted layers of the summer months, i.e. June–August, when largest dust fractions were observed. Usually, they were in the range of 0.7–0.9. A value of 0.82 was found as the 90th percentile (horizontal line in Fig. c). Thus, there was always a non-dust contribution in the lofted layers, which might have been smoke or pollution. For the PBL, the observed layer mean dust fraction was negligibly low for most of the cases and, thus, we conclude that dust was not present in the PBL for most parts of the evaluated period. A higher dust fraction in the PBL with layer mean values up to 0.5 was found between 31 December 2021–18 March 2022 as well as during September 2022 and on 23 December 2022. During the period from December 2021 to March 2022, the dust fraction of the lofted layer was comparably small and in a similar range as for the PBL. These findings indicate a low dust content and a high smoke fraction in the lofted layers during winter and spring and the mixing of the dust down into the PBL as already discussed in . In contrast, during summer, the dust contribution was clearly separated between the PBL and the lofted layer with almost no dust in the PBL and dust fractions up to 0.9 in the lofted layer.

Time series of the layer-resolved lidar ratios, Ångström exponents, and particle linear depolarization ratios for the different wavelengths are shown in Fig. . Slight temporal variations are visible, but the weekly fluctuations and the large error bars of the lidar ratio and the Ångström exponent make it difficult to identify a clear seasonal cycle. The seasonal cycle, especially for the lidar ratio and the depolarization ratio of the lofted aerosol layers, is more visible when monthly averages of the Fri/Sat cases are considered (Fig. ). Between November and April, the lidar ratio exhibited largest values up to 75 sr (monthly averages, Fig. d) and 80 sr (single Fri/Sat cases, Fig. a) at 355 nm. During that time of the year, the differences between the lidar ratio values at 355 and 532 nm were largest, especially for the winter period 2022–2023. At 532 nm, usually less than 60 sr were observed. These high values and the strong wavelength dependence indicate a mixture of dust and smoke, which is supported by the large extinction-related Ångström exponent values around 1 and the slightly positive values of the backscatter-related Ångström exponent between 355–532 nm shown in Figs. c and f . Furthermore, the particle linear depolarization ratio was usually <0.2 at all wavelengths (Figs. e and h). From April to September, the lidar ratio of the lofted layers was more similar at both wavelengths with values between 30–60 sr, but still slightly larger at 355 nm than at 532 nm. During these months, the extinction- and backscatter-related Ångström exponents mainly varied between -0.5–0.5 with the extinction-related Ångström exponent being always positive and larger than the backscatter-related Ångström exponent. The backscatter-related Ångström exponent in the wavelength range 355–532 nm can be used as an indication for the hematite mass fraction in the dust particles which varies with source region . The particle linear depolarization ratio reached values between 0.2–0.3 and was mostly largest at 532 nm, followed by 1064 nm and was smallest at 355 nm. The wavelength dependence for, both, the particle depolarization ratio and the lidar ratio has already been reported by for slightly polluted Saharan dust from similar source regions.

In the PBL, the seasonal cycle was less pronounced. The lidar ratio (Figs. b and e) was usually below 30 sr at 355 and 532 nm. However, from September to December 2021, values up to almost 70 sr were observed, caused by volcanic sulfate as described in . The Ångström exponents between 355–532 nm showed almost no temporal development during the complete time series from July 2021 to August 2023. In almost all cases, the backscatter-related Ångström exponent was, with values between 0–1, larger than the extinction-related one, which varied around 0. Also the particle depolarization ratio was mostly close to 0 at all three wavelengths, except between October 2021–April 2022 and at the beginning of October 2022, when the values increased up to more than 0.1. The observed values of the lidar ratio point to slightly polluted marine conditions as for pure marine conditions lower values between 16–23 sr were reported in previous studies for the vicinity of Cabo Verde and the Canary Islands . In the months of NH fall and winter, the higher particle linear depolarization ratio indicates a contribution of non-spherical particles, i.e. Saharan dust, which has been also observed by , , and between November–January. As mentioned above, the data of the described time series were also averaged per month for each specific year as shown in Fig. . It is important to note that the shown data points are not statistically significant monthly mean values. Each data point shown in Fig. is the average of at most four/eight (PBL/lofted) single values (four weeks per month and year with maximum two lofted sublayers, which were considered separately). However, the temporal trends in the optical properties are visible more clearly. These monthly mean data points were also used for creating a 2D space of the lidar ratio and the particle linear depolarization ratio as shown for 532 nm in Fig. , allowing a more concrete aerosol typing for the different layers (unfilled circles for the PBL and filled circles for the lofted layers) and different times of the year (indicated by the different colors).

Figure  was created accordingly to Fig. 2 of . This combination of the lidar ratio and the particle linear depolarization ratio is commonly used in aerosol classification as in , or . Using the existing knowledge about the typical values of these two parameters for different aerosol types, we defined some clusters of aerosol types in Fig. . A cluster of data points of the PBL with a particle linear depolarization ratio below 0.05 and a lidar ratio between 20–40 sr can be attributed to polluted marine aerosol (cluster 1) due to the slightly enhanced lidar ratio around 30 sr compared with 20 sr for clean marine aerosol . Most of the data points of the PBL are in cluster 1, but there are also three more data points (March and September 2022 and August 2023) with similar values of the lidar ratio but larger values of the particle linear depolarization ratio between 0.05–0.1, which represents a mixture of marine aerosol with dust (cluster 3). Furthermore, the influence of the volcanic sulfate (cluster 2) can be seen for the three data points of the PBL of September, October, and November 2021, which have particle linear depolarization ratios below 0.05 but enhanced lidar ratios between 40–50 sr. Concerning the lofted aerosol layers, the figure shows quite complex aerosol conditions with lidar ratios between 30–70 sr and a wide range of values of the particle linear depolarization ratio, indicating different contributions of non-spherical and spherical particles. According to , data points with a particle linear depolarization ratio larger than 0.25 point to the occurrence of nearly pure dust in the lofted layers (cluster 6). Data points with a particle linear depolarization ratio larger than 0.2 indicate dust-dominated lofted layers (cluster 5), which is used as a threshold for non-spherical particles in the target categorization of and by for the identification of aerosol layers with a major dust contribution. All remaining data points were considered to indicate dust-smoke-mixtures (cluster 4) with different contributions of dust and smoke components as the influence of transported biomass burning aerosol has been typically observed in the region over West Africa and Cabo Verde . Furthermore, fires in semi-arid areas might emit additional soil dust together with the biomass burning smoke e.g.. In cluster 4 (dust-smoke-mixtures), a few data points of the PBL are included, namely December 2021 and February 2022. Besides them, only data points of the lofted layers can be found in cluster 4 and 5 (dust-dominated). For the data points of the lofted layers, a strong monthly dependence can be observed. The data points in the dust-dominated cluster 5, including the cluster 6 of nearly pure dust, belong to the months June–September of all years and April, May, and October 2022. April, May, and October of the remaining years are located in the more dust-dominated part of the dust-smoke cluster 4 (particle linear depolarization ratio between 0.15–0.2), while the data points of November–January and March are in the more smoke-dominated part of cluster 4 with values of the particle linear depolarization ratio smaller than 0.15. Only the data point of the lofted layer of February 2022 is outstanding with a particle linear depolarization ratio close to 0.18 (lidar ratio around 60 sr). The reason for this anomaly is that the monthly mean data point includes measurements of nearly pure lofted dust layers as well as lofted layers with a large smoke contribution.

The described pattern can be partly explained with the seasonality of the fire activity in 2021–2023 based on assimilated MODIS observations provided by the Global Fire Assimilation System GFAS;, shown in Fig. together with their anomaly compared with 2003–2025. In addition, a seasonal cluster analysis of backward trajectories from the hybrid single-particle Lagrangian integrated trajectories HYSPLIT; model is shown in Fig. . Fire activity in the Sahel zone and south of it was present from September to November with mostly positive anomalies across the entire Sahel zone and especially from December to February with a positive anomaly in the central Sahel and a negative anomaly in large parts of the southern Sahel and south of it. The trajectory analyses indicate air mass transport from Senegal, Mauritania, and Mali, a region with little fire activity but a positive anomaly, with a probability of around 50 % in the time from September to February. Similarly, in Fig. , the data points with the lowest particle depolarization ratio, i.e. smaller than 0.15, belong to the months November, December, January, and March. From March to May, there was a strong fire activity with a positive anomaly in Senegal and Guinea and a strongly negative anomaly of the fire radiative power in the countries south of them. Air masses were advected from Senegal and Guinea Bissau with a 43 % probability, which might have caused the large smoke contribution in March 2023 and a certain smoke contribution in April and May 2023. Very little fire activity in Senegal and Guinea Bissau and in further single spots in the Sahel zone was still remaining in the period June–August indicating a slightly positive anomaly across the complete Sahel region. As all of the cluster mean trajectories for June–August pass the Sahel zone and partly Senegal and Guinea Bissau also a low smoke contribution in the lofted layers of Saharan dust might be assumed as even for the dust-dominated cases lower particle linear depolarization ratio values than in the previous studies of pure dust were observed . Furthermore, a pollution component cannot be excluded for the months in the dust-dominated and even the nearly pure dust clusters (5 and 6, respectively).