2.1 ENA central facility (C1) and aerosol supplementary site (S1)

The ENA central facility (C1) is located on Graciosa Island within the Azores archipelago at 39^∘5^′28^′′ N, 28^∘1^′36^′′ W. C1 is located on the northern part of the island as the area is flat, has access to local power, and is mostly unpopulated (Fig. 1). High-temporal-resolution measurements (seconds to minutes) of aerosol properties at C1 are made with the ENA AOS (McComiskey and Ferrare, 2016; Uin et al., 2019). The AOS at ENA C1 includes instruments for measuring aerosol optical, physical, and chemical properties; trace gases; and meteorological parameters. The AOS is comprised of one container that samples aerosols using instrumentations connected to a central inlet located approximately 10 m above ground level (Bullard et al., 2017; Uin et al., 2019).

The aerosol supplementary site (S1) was deployed at 39^∘5^′43^′′ N, 28^∘02^′02^′′ W, ∼0.75 km from C1 (Fig. 1), in July 2017. S1 was sited within 1 km of C1 to maintain the relevance of S1 data to the AMF measurements at C1. S1 was located at ∼0.2 km from the shore (closer than C1) at ∼50 m a.s.l. Data were collected at S1 until the site was decommissioned in April 2018 after the conclusion of ACE-ENA.

Figure 1Satellite image of ENA C1 and S1 on Graciosa Island, Azores, Portugal (© Google Earth).

Three instruments, duplicate models of those used within the AOS at C1, were deployed at S1. Two aerosol instruments were selected for their ability to measure submicron aerosol concentrations in high time resolution. The third instrument was included to associate the measurements with meteorological parameters as is done in the AOS. The aerosol instruments were powered and located inside a converted garage in an unoccupied house, with the computer for data acquisition. The meteorology sensor was mounted above the inlet at ∼3 m above the roofline. Measurements were designed to duplicate those made within the AOS as best possible without the use of an AOS inlet at S1. Prior to the deployment at S1, the instruments were calibrated at C1 alongside the AOS instruments. Ambient data from the three instruments were compared over a period of 1 week at S1. The S1 inlet flow rate was optimized to minimize submicron particle loss (Bullard et al., 2017).

Briefly, the fine-mode condensation particle counter (CPC) (TSI, Inc., Shoreview, MN, USA; model 3772) measures the submicron number concentration (N_tot) of aerosols from ∼7 nm to 1 µm in particle diameter (D_p). Particles are grown by condensing butanol vapor onto the particles before they are optically counted by illuminating them with a laser beam to count the number of light pulses that are scattered (Kuang, 2016). The Ultra-High Sensitivity Aerosol Spectrometer (UHSAS) (Droplet Measurement Technologies, Inc., Longmont, CO, USA) is an optically scattering, laser-based aerosol spectrometer for sizing particles from ∼60 to 1000 nm D_p. Aerosols scatter the laser light as a function of their optical D_p. The UHSAS detection efficiency is ∼100 % for particles > 100 nm and for concentrations < 3000 cm⁻³ (Cai et al., 2008). Concentration measurement errors occur for smaller particles that have low scattered light intensities and during periods of higher N_tot due to particle coincidence. Sizing of spherical and irregular particles by the UHSAS is within 10 % of the mobility diameters measured by the scanning mobility particle sizer (SMPS) for particles with D_p > 70 nm (Cai et al., 2008). Therefore, in this study, we use the UHSAS submicron data for particles > 70 nm (Uin, 2016). CPC and UHSAS sample flows are dried using a shared Nafion dryer that reduces the relative humidity of the samples to ≤30 % (Uin et al., 2019). Since submicron data were collected at S1 and compared with the submicron data collected at C1, we make no inferences on supermicron particles. The meteorology sensor (Met) (Vaisala, Finland; WXT520) provides ambient air temperature, relativity humidity, atmospheric pressure, wind speed and direction relative to true north, and precipitation data (rain amount, duration, and intensity) (Kyrouac, 2016).

2.2 ARM Aerial Facility (AAF)

The ARM Aerial Facility (AAF) Gulfstream-159 (G-1) research aircraft flew from Terceira Island (∼90 km from the ENA C1 site) during two IOPs in early summer 2017 (June to July) and winter 2018 (January to February). Flight patterns included spirals to obtain vertical profiles of aerosol and clouds and ascendant and descendent legs at multiple altitudes to provide characterization of the boundary layer and lower-free-troposphere structure. Data were collected up to an altitude of ∼4970 m.

N_tot data collected by the AAF with the CPC (TSI, Inc., Shoreview, MN, USA; model 3772) during the summer were compared to CPC data collected at C1 and S1. The CPC was installed behind an isokinetic inlet to minimize particle loss in aircraft sampling and was operated on the AAF G-1 (Schmid et al., 2014). During the first ACE-ENA IOP there were 20 research flights (RFs). We analyzed data from the seven flights that collected data over C1 at low altitudes between 54 and 500 m.

3.1 C1 and S1 intercomparison

We present and evaluate different strategies to identify periods when the AOS data are impacted by high submicron aerosol concentrations and associate them with nearby potential aerosol sources. The impacts of local aerosol sources at ENA C1 are evaluated by comparing data collected at C1 and S1. We analyzed two 1-month time periods that represent two seasons: summer (22 July–20 August 2017) and winter (1–30 December 2017).

Measurements from the UHSAS and CPC are combined to describe the submicron aerosol size distribution by dividing the data into three optical size modes. Zheng et al. (2018) used lognormal fitting of the submicron aerosol size distributions from the UHSAS to define three modes to study aerosol–cloud interactions at ENA. The lognormal fittings gave three parameters: mode diameter, mode number concentration, and mode σ (standard deviation; Table 2 in Zheng et al., 2018). Number concentrations (N) of the fitted modes were classified by the mode diameter into three bins: (1) N_At, number concentration of Aitken (At) mode aerosol (D_p≤100 nm); (2) N_Ac, number concentration of accumulation (Ac) mode aerosol (D_p=100–300 nm); and (3) N_LA, number concentration of large accumulation (LA) mode aerosol (D_p=300–1000 nm). The N_Ac- and N_LA-mode number concentrations reported here are directly measured by the UHSAS. Since there is not a direct measurement of the full range of At-mode particles, N_At is determined by combining the measurements from the CPC and the UHSAS. N_At is calculated as the difference between the N_tot, as measured by the CPC, and the sum of the UHSAS number concentrations from the two larger modes: $N_{\mathrm{At}}=N_{\mathrm{tot}}-$ (N_Ac+ N_LA). All D_p values referenced in the text refer to aerosol optical diameter unless otherwise stated.

One way to determine statistical outliers in the data is by comparing the difference between the median and the mean. Time periods when the median and mean N_tot differ significantly are used to indicate periods when the data are affected by outlying events, such as high-number-concentration aerosol events. Median values represent the midpoints in the data, which are minimally affected by outlying events. Mean values describe the central tendency of the data and are affected by outlying events. As such, comparison between the two values provides information about the variability within the overall dataset. Erroneous data and their ARM quality assurance/quality control (QA/QC) flags (e.g., negative values and −9999) have been removed prior to the analysis presented here. Significant deviations between the mean and median concentrations, where the mean is biased high, are used to indicate when aerosol N_tot have a statistically relevant higher variability due to the presence of high-concentration aerosol events.

3.2 ENA Aerosol Mask (ENA-AM)

ENA Aerosol Mask (ENA-AM) is a standard deviation algorithm that was parameterized for the ENA N_tot data collected by the CPC. Application of the algorithm requires the statistical differences between adjacent data points to distinguish periods of short-duration high aerosol number concentrations from the baseline measurements. The time resolution of the data has to be shorter than the typical time period of the high-concentration events. The variation within the clean baseline periods also has to be smaller than the variation of N_tot during the high-concentration events. Therefore, the algorithm works best with high-time-resolution data, as is collected by the AOS at time intervals on the order of seconds to minutes, and for identifying local sources that have high temporal variability. An additional requirement is that at least half of the total data points have to be representative of the baseline conditions; otherwise the algorithm is not able to identify the high-N_tot events properly. The flow chart in Fig. 2 describes the requirements and recommended procedures to apply the algorithm to data affected by local aerosol events. A perfect algorithm would identify only the noise and retain all of the natural variability. Since data may include periods when the local sources are less variable than the natural baseline no separation will be perfect. Here, we test and develop an algorithm optimized to balance the separation of the noise from the baseline. The 1 min N_tot data collected at C1 and S1 fulfill these requirements, and we, therefore, developed ENA-AM as described below using two 1-month periods of data collected at ENA (Aiken and Gallo, 2020; Gallo and Aiken, 2020a, b).

Figure 2Flow chart to apply the standard deviation algorithm to high-time-resolution aerosol data.

Download

We determined the standard deviation of the data below the median (σ_b) of N_tot for each of the two 1-month periods. Any data point that differs by more than α times the σ_b from the preceding data points is identified and masked as a high-concentration aerosol event. The retained data points are defined as the baseline. The variable α is used to set the threshold, and its value is defined as a function of the specific dataset and time series variability. The utilization of 1-month time periods was chosen to limit biases in the characterization of the regional baseline after testing a range of periods from 2 weeks to 2 months. At ENA, we observed that when using longer periods of time (> 6 weeks), σ_b removed the long-term variability associated with seasonal changes. Simultaneously, considering shorter time periods (< 2 weeks), σ_b was unable to retain periods when ENA was affected by episodes of long-range-transported continental aerosols. An alternative parameterization would be to use the standard deviation between the first and the third quartiles. This approach has been shown to be effective for masking continuous time series of greenhouse gas measurements that present daily and monthly natural fluctuations and positive short-term spikes (seconds to minutes) due to local emissions (El Yazidi et al., 2018). We tested this alternative at ENA and observed similar results for both methods. The data filtered using σ_b agreed with the data filtered between the first and third quartiles 98.6 % of the time.

Whenever a data point is identified above the threshold, the next point in the time series is evaluated using a random walk method (threshold = (${\mathit{\sigma }}_{\mathrm{b}}+\sqrt{\left(n\right)}$) ×α), where n is the number of data points since the last data point that was within the standard variability. In this way, the threshold is slightly increased to account for normal temporal development of the baseline. If the density of the high-concentration particle events is high, the algorithm is not able to identify the baseline variability properly. In such cases, α should be set to a lower value, and the random walk method threshold might be better substituted with a two-point thresholding method. With two-point thresholding, the two data points after each masked point are considered to be part of the event. Thus, the value of α and the thresholding method are dependent on the time series variability as is the selection of the time period over which to apply the algorithm. Selection of both must be optimized for the specific dataset.

We tested six different parameterizations of the algorithm, which included two α values and the two thresholding methods as well as different time lengths. Table 1 presents the combination of α values and thresholding methods used. A sensitivity analysis was conducted to determine the best parameterization of the algorithm for N_tot measurements at ENA. After the parameters for ENA-AM were determined, we compare masked C1 with AAF N_tot measurements.

Table 1Standard deviation algorithm input parameters tested at C1 and S1 in the summer.

Download Print Version | Download XLSX

4.1 High-concentration aerosol events

Wind directions can be used with aerosol measurements to determine aerosol sources (Zhou et al., 2016; Cirino et al., 2018). To understand the frequency and direction from which local aerosols originate at ENA, we present mean aerosol N_tot and N_UHSAS as a function of wind direction. N_tot and N_UHSAS are used to understand the directional and temporal influence of observed high aerosol concentrations at C1 and S1 and to evaluate the use of wind direction data to create an aerosol mask at ENA.

In Fig. 3, 1 min N_tot,N_UHSAS, and wind measurements were averaged as a function of wind degree direction in the summer and winter. A detailed analysis of wind speeds and wind directions at ENA during summer and winter is presented in the Supplement (Sect. S1). When plotted by wind degree direction, we observed N_tot > 1000 cm⁻³ at C1 and S1. Mean N_tot values for all directions in the summer were 710 (C1) and 490 cm⁻³ (S1). N_UHSAS mean concentrations were less than half of N_tot during the same time periods: 342 (C1) and 210 cm⁻³ (S1). The higher N_tot is due to a significant fraction of aerosol below the UHSAS lower detection size limit of 70 nm since the instruments have similar upper limits for counting particles. Without the N_tot that counts particles < 70 nm D_p, the high-concentration aerosol would be harder to identify by wind direction alone due to the lower variability in N_UHSAS. For this reason, we continue our analysis by wind direction focusing on N_tot.

The largest mean N_tot plotted by wind degree direction that was observed at C1 was ≥3000 cm⁻³ (Fig. 3a, c) during summer and winter when the winds were from the west to northwest, wind directions that are associated with the airport (see Fig. 1; Table S1 in the Supplement). These directions were attributed to the utilization of the runway and the airplane parking lot with AOS camera visual validations of aircraft. The next highest N_tot values were observed from the south to southeast at C1. Values of N_tot≥1000 cm⁻³ were observed in the summer and N_tot≥1600 cm⁻³ in the winter. These directions are associated with the direction of the road that leads from the airport to the town of Santa Cruz (Figs. 1 and S1 in the Supplement).

Figure 3Polar graphs of the mean N_tot and N_UHSAS as a function of wind direction during summer (a, b) and winter (c, d) at C1 and S1. The 1 min N_tot data from the CPC, in orange (data not available at S1 in the winter), and N_UHSAS from the UHSAS, in blue, were averaged as a function of wind degree direction. The frequency of wind direction is in gray.

While mean N_tot values were lower at S1 than C1, S1 also had N_tot > 1000 cm⁻³ (Fig. 3b). The three highest N_tot values at S1 that exceeded 1000 cm⁻³ were observed from the south-southeast, east-southeast, and east. The wind directions of the maxima N_tot were associated with the airport runway, rural road, and pasture at S1. Wind directions with N_tot∼1000 cm⁻³, observed from the northeast, were likely due to the rural road along the shore. Values of N_tot∼500 cm⁻³ from the southwest were from the direction of the decommissioned landfill that still has active vents as well as the airport runway. N_tot was not available during the winter at S1 to make a comparison with summer.

The results of the wind direction analysis indicate that the main sources of N_tot≥1000 cm⁻³ at C1 and S1 are most likely associated with airport activities and road traffic due to the proximity and direction of the sources. However, at ENA, other unattributed local sources, not related to airport operations, that are not identified here could also be present. One example of an aerosol source that we could not verify was a potential brick production facility ∼1 km to the south-southeast of C1. Complex meteorological conditions known to exist in the region might also be responsible for high N_tot at C1 and S1 that we were not able attribute to local sources based on wind direction.

4.2 Submicron aerosol modes

Number concentrations from three aerosol modes that we defined in Sect. 3.1 are presented in Fig. 4 from C1 and S1 in the summer and winter. The smallest mode number concentration, N_At, represents the size range most likely impacted by nearby combustion sources (aircrafts and gasoline and diesel vehicles) as discussed in the Sect. S3. N_Ac is expected to include some of these particles as well, especially for the less efficient combustion sources and operational modes, such as those produced by diesel engines and wood burning sources that may or may not be significant at ENA and are not discussed here due to their unconfirmed use on the island. The third and largest mode number concentration, N_LA, is not expected to be significantly impacted by nearby combustion aerosol sources. However, N_LA is presented since it includes natural aerosol sources such as sea spray (Burrows et al., 2014; Quinn et al., 2015) and secondary organic aerosol (Jimenez et al., 2009; Shrivastava et al., 2019) that can also be formed in association with combustion sources.

For the three submicron size modes analyzed at C1 and S1, N_At had the largest median and mean number concentrations, equating to 44 % of the median and 31 % of the mean for the total submicron aerosol concentrations, N_tot, when averaged from the different sites and seasons. N_At also had the highest deviation between the mean and median of the three size modes during the summer and winter. Median N_At values at C1 were relatively constant at 245 cm⁻³ in the summer and 258 cm⁻³ in the winter. Median N_At at S1 was 78 % of C1 with 190 cm⁻³ in the summer. While median N_At values were relatively constant for the data shown here at both seasons and sites, mean N_At varied with site and season. Mean N_At values were 540 (C1) and 330 cm⁻³ (S1) in the summer (Fig. 4a). In the winter at C1, the mean N_At was 800 cm⁻³, which was 48 % higher than what was observed in the summer (Fig. 4b).

The higher observed mean N_At at C1 during the winter indicated that the influence of nearby aerosol sources was likely to be larger in the winter than in the summer. This result is supported by the earlier results from N_tot (Sect. 4.1: Fig. 3a, c) and the submicron size distributions (Fig. S3b). The reason for the higher fraction of N_At observed in the winter at C1 could have been due to additional seasonal sources that were not attributed here, such as the burning of wood or other fuels to heat homes. Different meteorological conditions experienced in the winter versus the summer could also have contributed to the seasonal differences. For example, higher N_At from the known sources discussed in Sect. S3 might also be due to different winter meteorological conditions, e.g., lower boundary height and higher wind speeds.

Figure 4Box and whisker plot of At-, Ac- and LA-mode aerosol number concentrations at C1 (orange) and S1 (pink) in the (a) summer and (b) winter. Mean (x) and median (red line). Box bottom at 25 %, box top at 75 %, whisker bottom at 10 %, and whisker top at 90 %. No At-mode data were available at S1 during the winter.

Download

Mean and median N_Ac were lower than N_At during summer and winter at C1 and S1, yet still represented a significant fraction of N_tot. In the summer, C1 and S1 N_Ac had similar mean and median values, indicating low variability in the data. Mean N_Ac values observed were 160 (C1) and 151 cm⁻³ (S1). Median N_Ac values were 156 (C1) and 147 cm⁻³ (S1). The similar values between the mean and median N_Ac at both sites indicated that the mode was not largely affected by high-concentration aerosol events. In the winter, mean N_Ac values were 13 % (C1) and 22 % (S1) lower than mean N_Ac in the summer. Mean N_Ac values were 140 (C1) and 118 cm⁻³ (S1). Median N_Ac values were 91 (C1) and 89 cm⁻³ (S1). Overall, N_Ac values at C1 and S1 were more similar than N_At in summer and winter. However, there was a higher variability between the mean and median N_Ac observed during the winter that was not observed in the summer (see Sects. S3 and S4 for discussion).

N_LA did not represent a significant fraction to N_tot at ENA for the data presented here. Mean N_LA values during the summer were 6 cm⁻³ at C1 and S1. Similar N_LA values were observed in the winter at 8 cm⁻³ at C1 and 10 cm⁻³ at S1. While N_LA is important in regard to mass concentrations, scattering properties, and cloud condensation nuclei, all properties measured by the AOS (Uin et al., 2019), N_LA values are not generally attributed to local combustion aerosol sources, which is the focus here. Contributions and impacts to N_LA due to sea spray aerosol were beyond the scope of this work yet were not considered to be a large contribution at C1 or S1 based on the low N_LA observed here.

4.3 High-time-resolution data

Time series of N_tot at C1 and S1 indicated that both locations periodically sample high concentrations over time periods < 4 min. High N_tot such as these are typically the result of local sources due to their high concentrations and short durations which would become less evident at greater distances from the source. Since aircraft idling, taxiing, takeoff, and landing are all potential times when high N_tot could be sampled at C1 and S1, we used the Graciosa Airport flight logs and the AOS camera observations to validate high-time-resolution N_tot data at ENA.

In Fig. 5, we present two 1 d periods sampled at C1 and S1 during the summer. N_tot > 25 000 particles cm⁻³ were observed on a daily basis at C1 in the raw 1 s data (Fig. 5a). Lower N_tot daily maximum concentrations > 11 000 cm⁻³ were observed at S1. Winter N_tot daily maximums at C1 were > 20 000 cm⁻³, with maximum concentrations occasionally ∼80 000 cm⁻³. Figure 5b shows a time period when the overall trend is the reverse of Fig. 5a when higher N_tot values were observed at S1 in comparison to C1. While this period did not represent the overall trend in N_tot between C1 and S1, it is included to show that C1 and S1 both observed N_tot maximums at different times and that both sites were impacted by high-concentration aerosol events in high time resolution.

Figure 5The 1 s N_tot data during two 1 d periods at C1 (orange) and S1 (pink) raw in the summer. (a) Typical day when C1 sampled higher N_tot than S1; (b) atypical day when S1 sampled higher N_tot than C1. To highlight the smaller peaks, the four highest peaks are off-scale by the factors indicated in the figure.

Download

Graciosa Airport on average hosts two flights a day, the first typically in the late morning/early afternoon and the second in the late afternoon. The airport time tables for 2017 and 2018 reported that planes landed and took off from Graciosa Island during three distinct time periods throughout the day: ∼17 % of the planes arrive at Graciosa Airport between 08:30 and 11:00 UTC, ∼26 % between 13:00 and 15:00 UTC, and ∼56 % between 17:00 and 20:00 UTC. Taking into account the wind direction, planes typically land from the east and take off from the west. We confirmed that, during the summer, 97 % of the flights occurred in this direction by analyzing the daily video from the AOS cameras at C1. However, due to the runway's limited length, planes often utilize the full length of the runway, which was observed in N_tot at C1 and S1. Such occurrences affected N_tot at C1 the most when the wind direction was between northeast and west, as well as S1 when the wind came from the east to southwest.

To further understand the potential influence of the airport operations on N_tot at C1 and S1, we examined a 1 d time period in detail. In Fig. 6, we present C1 and S1 1 min time resolution N_tot on 3 August 2017. N_tot at S1 was largely unaffected by the short-duration high-concentration aerosol events as N_tot was never > 1000 cm⁻³. While this is only a 1 d time period and was by no means representative of daily N_tot, we show it as an example of the complexity within N_tot at ENA.

Throughout the day, abrupt changes in wind direction were observed. Winds from the south, southwest, and west dominated until 17:58 UTC. Starting at 18:00 UTC, the dominant wind directions were northwest, north, and east. Analysis of the video from the AOS camera at C1 showed that diesel trucks were on the runway from 09:07 to 09:27 UTC for daily maintenance. At two times during the afternoon, 13:42 to 15:02 and 18:46 to 19:51 UTC, the aircraft was idling near the airport terminal (Fig. 6). During the first part of the day, when the wind directions were from the south and west, N_tot values were greater than 1000 cm⁻³ at C1 at numerous times. In the Supplement we identify these directions at C1 with the airport terminal, parking lot, and the road to the airport. Later in the day, when the winds were coming from the northwest to east, in the direction of the runway at C1, values of N_tot < 1000 cm⁻³ at C1 were similar to N_tot at S1.

The high-N_tot events at C1 and S1 were associated with the airport activities and increased road traffic that generally occurred before the arrival and after the departure of the aircraft, based here on visual observations, airport flight logs, and wind degree direction analysis. The aircraft and vehicle impacts were observed by sharp peaks occurring on timescales on the order of minutes when N_tot was an order of magnitude above the baseline signal. In contrast, the airport operations tended to cause periods of elevated N_tot that occurred over longer timescales on the order of hours. Therefore, the impact of the airport, its operation, and associated traffic on the AOS data at ENA could not be constrained to the arrival and departure times of the aircraft since it was also impacted by airport operations that occurred throughout the day and the wind direction in relation to C1 and S1.

Figure 6N_tot and wind direction at C1 (orange) and S1 (pink) on 3 August 2017. Yellow and gray periods indicate when the AOS cameras observed trucks on the runway (yellow) and planes near the terminal building (gray).

Download

While the influence of the airport operations may not be readily apparent from the short-duration high concentrations observed at C1 and S1 (Fig. 5), further information constraining this influence was obtained by looking at the diurnal cycle of mean and median N_tot at C1 (Fig. 7). The three hourly periods with highest mean N_tot were observed during 09:00 to 10:00 UTC at 916 cm⁻³, 13:00 to 14:00 UTC at 860 cm⁻³, and 17:00 to 18:00 UTC at 1595 cm⁻³. These three elevated mean N_tot periods occurred during the three time periods when the airport flight logs on average observed flights. These periods were identified using the airport flight logs and are shown as the black boxes in Fig. 7. Mean N_tot from 07:00 to 08:00 UTC reached a value of 615 cm⁻³. The highest mean N_tot was observed during the third time period identified by the airport to host on average half of the daily flights, while the two earlier time periods were only associated with ∼25 % of the flights each. Two other high-mean-N_tot periods were observed during the diurnal profile at C1. Mean N_tot values were 615 cm⁻³ from 07:00 to 08:00 UTC, which occurred during a similar time that the AOS cameras observed the daily maintenance of the runway with diesel trucks from 07:45 and 08:30 UTC. The second period from 20:00 to 21:00 with mean N_tot > 800 cm⁻³ was attributed to unknown potential aerosol sources at this time.

The diurnal variation observed in the mean N_tot at C1 in the summer was not observed in the median N_tot. Hourly averaged medians exhibited low variability throughout the day with a minimum of 380 cm⁻³ during the night between 23:00 and 24:00 UTC. A maximum of 506 cm⁻³ was observed in the late afternoon between 17:00 and 18:00 UTC.

At ENA, the periods with the largest deviation between the median and mean N_tot were the three periods when most of the flights occurred at the airport. A diurnal variation was observed in the mean N_tot yet was not statistically relevant for the median N_tot of the same data at C1 and S1. While not shown here, S1 had a similar trend in the diurnal profile to what was observed at C1 in the summer. The main difference was that the mean N_tot values were all < 1000 cm⁻³. Winter data at C1 also had the highest mean N_tot values and deviations from the medians during the hours of airport operations. We use the information from the diurnal profiles at ENA to validate the statement that the airport operations and associated activities were the largest sources of high-concentration N_tot observed at ENA.

Figure 7Box and whisker diurnal profile of N_tot at C1 during summer. N_tot mean (orange x) and median (red line). Black boxes from 08:00–10:00, 13:00–15:00, and 17:00–20:00 UTC indicate the three daily time periods when aircraft were present at the Graciosa Airport.

Download

4.4 High-number-concentration aerosol event mask

4.4.1 Algorithm parameterization and validation

To apply a mathematical algorithm to mask high-N_tot events at C1 and S1, we first calculated the standard deviation of the data below the median (σ_b). We found σ_b values of 298 and 264 cm⁻³ for C1, respectively in the summer and winter, and σ_b values of 234 cm⁻³ for S1 in the summer. Then, we conducted a sensitivity test to select the optimal parameterization of the algorithm to apply to the 1 min resolution N_tot data at ENA using the combination of the α parameter and thresholding method shown in Table 1, Sect. 3.2. First, we analyzed the efficiency of six parameterizations to detect high-N_tot aerosol events that were independently validated using additional collocated measurements at C1 (AOS camera and airport flight logs). Subsequently, we assessed the percentage of data removed and the R² generated between masked N_tot at C1 and S1. Finally, we evaluated the ability of the best parameterizations to discriminate short-lived high-N_tot events from periods when ENA was affected by long-range-transported continental aerosol. In our analysis, we use 1-month time periods as the utilization of longer periods was found to bias the characterization of the regional baseline due to seasonality, which could accentuate long-term variabilities and confuse the high signal of local events (El Yazidi et al., 2018). Previous studies use the random walk (RW) threshold for aerosol data. Drewnick et al. (2012) proposed using α=3 to remove sharp and short peaks lasting a few seconds in N_tot from CPC and gas-phase CO measurements from a mobile aerosol research laboratory. The authors found that the application of the α3-RW parameterization worked well when the density of the high-concentration events was low. Similarly, El Yazidi et al. (2018) used α1-RW with gas-phase CO₂ and CH₄ data at four different stations in Europe that were affected by sharp events over time periods of a few minutes. The α1-RW parameterization was able to detect ∼96 % of the events that were visually identified by the station manager. Therefore, we began at ENA by testing three α values with the RW threshold that were used in these two studies.

We present, in Fig. 8, the results from the application of the algorithm over the same 24 h period that we analyzed in Sect. 4.3, Fig. 6. The first three parameterizations selected, α05-RW, α1-RW, and α3-RW, were able to identify the first data points during a high-N_tot event but were not able to identify events that occurred for extended periods of time on the order of hours, as is shown in Fig. 8a for the application of α3-RW at C1. While α05-RW and α1-RW are not included in the figure for simplicity, similar results were produced from these parameterizations. Next, we constrained the threshold more by applying the TP method with the same α values. For both C1 and S1 sites and seasons, the α05-TP (not shown for simplicity) and the α1-TP parameterization were the only parameterizations able to identify longer-duration events that lasted from minutes to hours, as were experienced due to airport operations as shown in Fig. 8b. Results from α3-TP were not included in the figure as the combination of the relaxed α and constrained two-point threshold parameters, α3-TP, yielded similar results to the RW threshold parameterizations tested previously. The α3-TP parameterization was not able to identify the longer-duration high-N_tot events. In conclusion, when high-N_tot events had durations on the order of hours, the difference in the signal between the adjacent points was not high enough to be identified by either the RW threshold or the higher α=3 parameter combinations tested at C1.

Figure 8Original (orange points) and masked N_tot at C1 using (a) α3-RW (blue points) and (b) α1-TP (green points) parameterizations over a 24 h period on 3 August 2017. Yellow and gray boxes indicate periods when the AOS cameras at C1 detected trucks and planes, respectively, on the runway.

Download

Similar results were obtained when we tested the six parameterizations of the algorithm on N_tot at S1 (not shown). As we observed at C1, the tightest combinations of parameters, α05-RW and α1-TP, were able to most accurately identify all of the high-N_tot events of all the parameterizations tested here. The higher α values and the random walk threshold relaxed the algorithm such that the number of data points identified was likely to underestimate the number and duration of high-N_tot events observed at ENA.

Due to the diverse high-N_tot events and local sources at ENA, the R² value between N_tot measurements collected at C1 and S1 in the summer was minimal (Fig. 9, R²=0.03, slope $=\mathrm{0.05}\pm \mathrm{0.001}$). In Table 2, we show a comparison of the percentage of data that were masked and the R² values with corresponding slopes after applying the parameterizations of the algorithm.

Table 2R² values and percentage of masked N_tot data during summer at C1 and S1 using six different combinations of the α parameter and thresholding methods.

Download Print Version | Download XLSX

Application of the RW threshold generated R² values ≤0.8 between C1 and S1 independent of α. The highest α value (α=3) with the TP threshold generated similarly low regressions between the two ENA sites (R²=0.79, slope $=\mathrm{0.79}\pm \mathrm{0.001}$) confirming that the α3-TP parameterization was not able to detect all of the local aerosol events. After applying the α0.5-TP and the α1-TP parameterizations, the linear regressions and slopes were closer to unity. α0.5-TP generated a R²=0.88 with a slope $=\mathrm{0.86}\pm \mathrm{0.001}$ and α1-TP a R²=0.87 with a slope $=\mathrm{0.84}\pm \mathrm{0.001}$ (data fit through zero by orthogonal distance regression). The percentages of masked data were 35 % (α0.5-TP) and 26 % (α1-TP) at C1 and 23 % (α0.5-TP) and 15 % (α1-TP) at S1. C1 retained a higher N_tot, likely due to incomplete removal of sources discussed in Sects. 4.1 and S4. The variability in the original N_tot due to high-concentration aerosol events was removed, and a regional baseline was able to be identified based on the agreement between the two locations within measurement and mask uncertainties.

Figure 9Scatter plot of N_tot at C1 and S1 in the summer. Original 1 min data (black) are shown with ENA-AM masked data (green).

Download

Furthermore, we evaluated the ability the α0.5-TP and the α1-TP parameterizations to mask short-lived high-N_tot events during periods when ENA was sampling long-range-transported aerosol. Periods with elevated aerosol concentrations due to long-range-transported continental sources have been observed to occur at ENA for durations on the order of days to weeks (Zheng et al., 2018). Through the analysis of back trajectories and aerosol optical properties, here we identify and present an episode of transported aerosol from Central Africa and the Canary Islands from 7 to 12 January 2017 (Fig. 11a). During this time, N_tot at ENA remained above 700 cm⁻³. The transported aerosol was likely due to a mixture of mineral dust and carbonaceous aerosol species from biomass burning sources (Logan et al., 2014), as have been observed from other continental sources at ENA. After applying the two parameterizations of the algorithm, we observed that during the long-range-transported aerosol event, at C1, the α05-TP approach removed 47 % versus 29 % of the data with α1-TP. The α05-TP algorithm was not able to discriminate between variations in the baseline due to regional processes, e.g., entrainment of particles from the free troposphere due to long-range-transport events shown in Fig. 10, from weak local aerosol events from unattributed local sources. After applying the α1-TP parameterization to N_tot at C1 we observed that the majority of the data associated with the multiday event were retained with the baseline N_tot. Simultaneously, the short-duration high-N_tot events, attributed to local sources, were also removed (Fig. 10b). We use the results from this case study to further validate the application of ENA-AM using α1-TP with 2 min N_tot data during periods when multiday entrained long-range-transported aerosol was sampled at ENA.

Figure 10An episode of long-range-transported continental aerosol at C1 determined (a) with an 8 d back trajectory arriving at 10 m a.g.l. and (b) elevated N_tot with original (black) and masked (green) data using ENA-AM.

Therefore, we used the α1-TP parameterization to create an aerosol mask at ENA, heretofore referred to as the ENA Aerosol Mask (ENA-AM), using 1 min resolution N_tot. At ENA the 1 min N_tot had sufficient time resolution to mask the high-N_tot events. Application to the higher-time-resolution 1 s N_tot data was not necessary based on the validation of ENA-AM presented here and saves computational time when analyzing continuous data.

4.4.2 Identification of a regional baseline and impact of ENA-AM on N_tot and N_Ac

After the application of ENA-AM, we observed that mean, deviation between mean and median, and standard deviation N_tot all experienced reductions. In Table 3 we show mean, median, and standard deviations for the original and ENA-AM masked N_tot and N_Ac measurements at C1 and S1 in the summer and winter. After applying ENA-AM to C1 N_tot, mean and standard deviation values dropped from 707±2780 to 428±228 cm⁻³ in the summer and from 537±630 to 347±223 cm⁻³ in the winter. At S1, the decrease was lower yet still significant, from 489±370 to 384±355 cm⁻³. In the summer, ENA-AM mean N_tot was 9.1 % higher at C1 than at S1. Satellite images and analysis of local aerosol sources (see Supplement) show that C1 is located ∼1 km closer to urbanized areas and to the town of Santa Cruz than S1. The N_tot generated from these more distant and diffuse sources is likely too weak to be completely masked by ENA-AM as discussed in Sect. 4.5.1.

Contrarily to N_tot, in the summer, N_Ac mean, median, and the deviation between them remained largely unchanged after the application of ENA-AM. This is in agreement with Sect. 4.3.2, where we showed summer N_Ac was only minimally affected by local aerosol events. However, in the winter, mean, deviation between mean and median, and standard deviation N_Ac at C1 experienced a higher reduction when masked with ENA-AM (25 % for the mean, 51 % for the deviation between mean and median, and 73 % for the standard deviation). These results are likely related to the presence of additional sources in the winter (e.g., burning of wood for home heating) which might affect N_Ac in a way that was not masked completely by ENA-AM.

Table 3Mean, median, and standard deviations (σ) of original and ENA-AM masked 1 min N_tot at C1 and S1 during the summer and winter.

Download Print Version | Download XLSX

To estimate the influence of local aerosol events on daily N_tot and N_Ac, we investigated the deviation between the original and ENA-AM masked N_tot and N_Ac daily means at C1 in the summer and winter in Fig. 11. We observed that after applying ENA-AM, depending on the day, N_tot daily means experienced reductions varying between 7 % and 81 % in the summer and between 2 % and 67 % in the winter. N_Ac reductions were lower than 27 % in the summer and 40 % in the winter (with the exception of two days, 16 and 22 December, when N_Ac daily means experienced reductions of 61 % and 80 %). As already observed in Sect. 4.2, at ENA, especially in the summer, local sources are mainly responsible for the emission of small particles in the At mode, while the Ac mode is generally only minimally impacted. Thus, in general, the reduction in N_Ac after masking the data does not impact the daily mean values as much as it does for N_tot. The higher daily mean reductions observed for N_tot in comparison to N_Ac after the utilization of ENA-AM demonstrated that the algorithm was able to selectively detect and isolate periods impacted by local aerosol events without having to use size distribution data. The high original N_tot and N_Ac daily means and the large deviation after application of ENA-AM (80 %) observed on 22 December were exceptions likely related to a poor-efficiency combustion source, a bulldozer, not normally present at C1 that was observed by the AOS cameras. The time series plots highlight N_Ac up to 11 000 cm⁻³ between 14:50 and 18:30 UTC. Thus, while N_tot and N_Ac measurements were impacted, ENA-AM was able to mask the data.

Figure 11Original (black) and ENA-AM masked (green) N_tot and N_Ac daily means at C1 and corresponding reduction (%) (blue) in the (a) summer and (b) winter.

Download

4.4.4 Masked AOS data and AAF overflights

After determining the regional baseline for N_tot from the ground AOS measurements at ENA, we compare ENA-AM masked C1 N_tot with AAF N_tot data in Fig. 12. We restricted our comparison of N_tot from the AAF to an area within a 10 km diameter box centered at C1 at altitudes ≤500 m. Before applying ENA-AM, the R² value obtained from comparing the original C1 N_tot and AAF measurements from seven overflights (data not shown) was poor (R²=0.26). After applying ENA-AM at C1, we obtained an R² of 0.71 and a slope of 1.04±0.01, which indicated a good agreement between the AOS and AAF data. The largest deviations from the 1:1 line occurred during two flights, on 21 June (RF1) and 29 June (RF6). On 21 June the AAF flew over Graciosa Island at two distinct times (12:00 and 13:30) during the day that we represent as two different periods in Fig. 12. While the AAF and C1 N_tot data fell on the 1:1 line within measurement uncertainties for the first period during RF1, C1 sampled an average of 43 % more N_tot than the AAF during the second flyover. The second flyover might have coincided with a period of time when C1 was affected by local events that ENA-AM was not able to identify and mask, as discussed in the paragraph above. Unfortunately, due to a lack of data at S1 during this time, this could not be verified. The largest deviation from the 1:1 line of all flights was observed on 29 June when higher N_totwas observed by the AAF. The mean N_tot was 659±17 at C1 and 1141±828 cm⁻³ at AAF. The flight trajectory indicated that the AAF flew south of C1 over the center of the island and around the town of Santa Cruz and two smaller towns, Guadalupe and Vitoria, on an ascending path. The AAF N_tot measurements might be affected by local aerosol when the AAF flew over these urbanized areas. AAF N_tot measurements might also have been biased by AAF emissions sampled through the aerosol inlet while the aircraft was gaining altitude. The standard deviation of the AAF data was also significantly greater than what was observed at C1, indicating the AAF likely intercepted plumes not observed at C1 during this time. Further analysis of the AAF data from RF6 would be required to determine the source of the discrepancy with the AOS data.

The high R² obtained from the masked C1 and AAF N_tot demonstrated that aerosol in the summer was well mixed within the first 500 m of the marine boundary layer. This is likely due to the high-sea-level-pressure system and advection in the summer at ENA which might enhance submicron aerosol mixing within the MBL (Davis et al., 1997). Since our focus here was on summer data at ENA due to the deployment of S1 to constrain C1 measurements, the winter season and vertical characterization of the MBL was beyond the scope here. During the winter, aerosol in the MBL is expected to be less well mixed due to a strong polar front activity and low-pressure system (Barbosa et al., 2017). Therefore, we expect less correlation between the AOS data and the AAF in the winter than what was observed in the summer.

Figure 12Scatter plot of ENA-AM masked C1 N_tot and AAF N_tot collected within 10 km from C1 and at altitudes ≤500 m.

Download