2.1 Storm Peak Laboratory

Storm Peak Laboratory (SPL) is a remote, mountaintop observatory (3210 m a.s.l., 40.455^∘ N, 106.745^∘ W) located in Steamboat Springs, CO. SPL is one of the only sites in North America with long-term measurements of aerosol size distributions and CCN number concentrations (Lowenthal et al., 2002; Borys and Wetzel, 1997; Hallar et al., 2017). The laboratory is commonly in-cloud during storms and sees frequent NPF events (Hallar et al., 2011; Borys and Wetzel, 1997; Lowenthal et al., 2019). The primary wind direction at the laboratory is westerly, which allows for the potential transport of SO₂ and the formation of sulfuric acid (H₂SO₄), an NPF precursor, from multiple powerplants 50–250 km upwind of SPL (Hallar et al., 2016; Obrist et al., 2008). SPL is located above a mixed forest allowing for the emission of a variety of different biogenic volatile organic compounds (BVOCs) that can impact aerosol formation and growth (Amin et al., 2012). Given the remote, mountaintop location of SPL, clean atmospheric conditions are common at the laboratory (Obrist et al., 2008).

To measure aerosols at SPL, we use a TSI Inc. (Shoreview, MN) scanning mobility particle sizer (SMPS) 3936 (with a TSI 3010 condensation particle counter (CPC)) for particles with diameters between 8 and 340 nm that scans every 5 min. Data are collected on a log-normal scale with particle diameter on a log scale and time on a normal scale. The instrument is periodically shipped back to TSI Inc. for routine maintenance and calibrations. The sheath and sample flow rates are 10 and 1 L min⁻¹, respectively, for the SMPS. Multiple charge corrections and diffusion corrections are applied to all SMPS data used in the analysis. SMPS data from SPL are now available on the European Monitoring and Evaluation Programme (EBAS) database, including level 1 data, which maintain 5 min time resolution while removing invalid values and calibrations, as well as level 2 data which present hourly averages and quantify atmospheric variability. Level 1 SMPS data are used in this study. The goal of EBAS data is to store long-term atmospheric science datasets and provide standards for quality assurance; thus rigorous standards for data quality are implemented to any data admitted to EBAS (Norwegian Institute for Air Research). SPL consistently runs a single-column droplet measurement technology (DMT; Boulder, CO) CCNC that collects number concentrations of CCN every second (Lance et al., 2006; Roberts and Nenes, 2005). We consider instrument measurements at supersaturation levels between 0.2 % and 0.4 % in our study.

2.2 An automatic method to classify new particle formation

A crux of studying atmospheric NPF is the identification of NPF events. The identification process historically utilized three-dimensional plots of log-normal size distributions and visual inspection aimed at identifying a burst of particles below 20 nm, followed by growth over the course of multiple hours that forms a new nucleation mode (Dal Maso et al., 2005; Kulmala et al., 2012). By visually inspecting these plots, the viewer sorts days into the following broad categories based on the observed growth patterns: event, non-event, or undefined. In an effort to improve the visual classification process proposed by Dal Maso et al. (2005), studies split events into subcategories to provide more specific classifications detailing whether particle growth is sustained during a given day or if the given day exhibits a burst of particles (Hirsikko et al., 2007; Kulmala et al., 2012; Boy et al., 2008; Svenningsson et al., 2008; Dal Maso et al., 2005). The visual classification of NPF can present problems, since human biases can influence classification leading to issues with the reproducibility and comparability of studies (Joutsensaari et al., 2018). To minimize the potential biases that influence visual classification, we present a statistically based, automatic sorting technique that evaluates particle burst and growth patterns to classify days into one of the following categories: type 1a event, type 1b event, class II event, undefined, or non-event (Hirsikko et al., 2007; Tröstl et al., 2016). The logic of the automatic classification technique is shown in a flowchart (Fig. 1) and described below.

Figure 1Flowchart illustrating the step-by-step process of the automatic NPF classification method. PGR: particle growth rate.

Download

The first step of the automatic classification method is to ensure the availability of SMPS level 1 data. Although NPF events can span multiple days, we consider daily data (00:00–23:59 MST) as well as the first 12 h (00:00–12:00 MST) of the next day to ensure the consideration of an NPF event does not prematurely end if growth continues overnight. The 5 min SMPS data are only considered if the first 24 h period meets the following conditions: there are at least 16 h of data present and the period between 10:00–23:00 MST (the times in which NPF is most common at SPL) has less than 1 h of data missing.

The days that successfully undergo quality control are then considered by the automatic classification method. For data to be classified as an event, two general conditions must be met: a burst of particles in the nucleation mode and growth that spans multiple hours contributing to the formation of a new nucleation mode. To first address the presence of a burst and identify days that are non-events, we compute the percentiles of all particle concentrations in our dataset below 25 nm from 10:00–23:00 MST. Days below the 10th percentile were automatically categorized as non-events, since they are automatically assumed to not have high enough nucleation mode number concentrations for an NPF event to have occurred. For days where the average particle concentration below 25 nm is above the 10th percentile of all data considered, the maximum of the Gaussians is calculated at each size bin. The normal distributions were fit by solving for the non-linear least-squares estimates using the R programming language (Eq. 1), which considers the particle size distribution at each diameter to return the time that corresponds to the maximum concentration at that given diameter (Bates and Watts, 1988). In the equation, “k” is the maximum aerosol number concentration, “t” is the time index where the normalized maximum at D_p occurs, “μ” is the mean aerosol concentration, and “σ” is the corresponding standard deviation. This equation is used for the calculation of individual maximum Gaussians at each size bin:

The derived time index represents the time at which the maximum of the peak-fitted particle size distributions occurs for each value of D_p. For data where at least five different Gaussian maximum points are calculated, a linear regression is fit to these maxima allowing for further analysis of growth over the course of an event (Lehtinen and Kulmala, 2003). R² values for the linear regression (one below 20 nm and another from 20 to about 70 nm), as well as the time differences between the maxima, are also considered to ensure growth. For days to be defined as an event, the time difference between bin maxima must be positive and non-zero for at least 40 % of occurrences, the largest r² value must be at least 0.6, there must be at least five maxima considered in the fit, and the largest size bin with a calculated Gaussian maximum must be above 25 nm for type 1a event classification and 15 nm for type 1b event classification. While 15 nm may seem like a low threshold for NPF growth, the growth of a given event often reaches sizes exceeding the diameter where the last Gaussian is calculated. Figure 2 is an example of a day that is calculated as an event because the threshold is lowered to 15 nm. Days that do not meet the above statistic-based criteria are initially classified as undefined but can be classified as a class II event later in the method.

Figure 2An example of a day classified as a type 1b event. Setting 15 nm as the diameter that the growth Gaussian maxima must reach allows for this day to be classified as an event demonstrating why the threshold is set at 15 nm. Gaussian maxima (black points) are outlined by the first-order derivative of the fitted distribution at each size (black line). The vertical red lines denote the initiation and end times of a given event as assigned by the automated methodology.

Download

For days that are defined as an event, the growth rate and event start and stop times are calculated. To find the growth rate, a linear regression is fit to the maximum Gaussians which are time dependent. The growth rate is determined by the following equation:

Because the slope of the linear regression fit of the maximum Gaussians represents particle growth over time during NPF events, this value is used when determining the growth rate. This method is most similar to the log-normal function fitting method of calculating growth rate but finds the growth rate by fitting a linear regression to the maximum Gaussians. Derivatives of the linear regressions are used to determine the start and end time of events, where the start time of the event is defined by the time of the first maximum of the first-order derivative and the end time of the event is defined by the time of the last first-order derivative minimum. Figure 3 illustrates an example of an NPF event and a day classified as a non-event.

Figure 3Strong NPF event (a) with midpoint size bin maxima (black points) outlined by the first-order derivative of the fitted distribution at each size (black sloped lines). The vertical red lines denote the initiation and end times of a given event as assigned by the automatic methodology. A non-event (b) is added for comparison. The vertical black lines represent the time period when CCN is considered (CCN_start through CCN_end) which is determined for each individual event day, while the seasonal average of this period is used for comparing CCN during non-event days.

Download

Days that are not classified as a type 1a or type 1b event are further considered to determine whether the given day is a class II event or undefined. Class II events are different than type 1a and type 1b events due to the presence of a particle burst which resembles an “apple” shape rather than persistent growth (Dal Maso et al., 2005; Junninen et al., 2008). Because the methodology detailed thus far considers growth patterns, significant class II particle bursts are initially classified as undefined due to weak growth (undefined stats fail) or Gaussian stacking in which greater than 75 % of the calculated Gaussian maxima occur at the same time (undefined burst). To address class II events, we apply an additional set of threshold tests to determine if days initially classified as undefined should be classified as a class II event.

Days that were classified as “undefined burst” or “undefined stats fail” are eligible for reclassification as class II events based on multiple thresholds. Class II events exhibit Gaussian maxima that occur at elevated number concentrations, exhibit particles bursting to a larger size, and originate in the nucleation mode; however, there is often a minimal difference in the temporal location of the calculated Gaussian maxima. Thus, the focus of this analysis is the identification of a significant particle burst. To confirm that the burst originates with smaller particles and exceeds the sizes required for class II events, the lowest size bin of a calculated Gaussian for a given day must be below 15 nm and the highest size bin of a calculated Gaussian must be above 15 nm. To ensure that the given day exhibits a strong enough burst for consideration as a class II event, at least 50 % of calculated Gaussians must have $\mathrm{d}n/\mathrm{d}\log D_{\mathrm{p}}$ values above the 95th percentile of all values in a given day. In addition, the diameters of consecutively calculated Gaussian maxima for days initially classified as an undefined burst cannot differ by more than 20 nm. The reason this threshold is not applied to days initially defined as undefined stats fail is because there is some growth observed, thus removing days with large Gaussian maxima differences could lead to the accidental removal of a class II NPF event that exhibits weak growth in addition to a significant burst.

2.3 Formation rates (J₈) and condensation sink (CS) values

The aerosol formation rate (J₈) and condensation sink (CS) values are calculated as part of the automatic classification method. J₈ values are calculated for type 1a and type 1b events. CS values are calculated for the comparison of values between type 1a and type 1b events and non-events.

The J₈ value for an event is defined by the formation rate equation (Kulmala et al., 2004, 2012):

where $\mathrm{\Delta }{N}_{\mathrm{8},D_{max}}$ is the change in the number concentration of particles across the considered size distribution from about 8 to 25 nm during Δt which is the time difference from the defined start of an event to the defined end of an event. When calculating the initial and final number concentrations, we utilize the average number concentration observed between 4 and 1 h prior to NPF initiation as the initial number concentration. The final number concentration is the average number concentration from all 5 min scans taken during an event. Doing so allows for the comparison of the initial conditions of an NPF event and aerosol formation across the entirety of a given event. The additional loss terms in the equation represent loss to the coagulation sink and loss due to growth out of the size range (Kulmala et al., 2012). The entire size distribution measured by the SMPS is used when calculating the coagulation sink loss term (Casquero-Vera et al., 2020).

CS values are calculated for the entire size distribution using the following equation (Pirjola et al., 1999; Kulmala et al., 2001):

In the equation, r_i is the radius of a given size bin (cm), and N_i is the number concentration (cm⁻³) of the given size bin. D is the diffusion coefficient of vapor, which is assumed to be 0.13 cm² s⁻¹ for H₂SO₄ at SPL based on calculations using representative pressure and temperature at the site (Hanson and Eisele, 2000; Welty et al., 2020; Tuovinen et al., 2021). β_m is calculated following the protocols of Kulmala et al. (2012) and Tuovinen et al. (2020) where the mass accommodation coefficient in these calculations is assumed to be unity (Kulmala et al., 2001; Nishita et al., 2008; Hallar et al., 2011; Kulmala et al., 2012; Tuovinen et al., 2020).

2.4 Determining when to consider cloud condensation nuclei concentrations

Determining whether an NPF event is impacting CCN concentrations is crucial in understanding the exact contribution of aerosols to cloud formation and, thus, understanding the potential climatic impacts. While environmental supersaturation and particle hygroscopicity are both crucial factors for CCN activation, aerosols from NPF must grow to CCN-relevant sizes before activating as CCN. Therefore, it is important to consider CCN enhancements due to NPF at times when particles reach CCN sizes. In this study, we propose and apply a statistical method to determine the time in which to consider the contribution of NPF to CCN concentrations. Our method sets a start and end time for CCN concentrations based only on aerosol concentration measurements that consider growth patterns of aerosols over and around the time of NPF.

For days classified as type 1a events and type 1b events, the start time of CCN consideration (CCN_start) is the first time after the start of an NPF event that 25 % of all particles in a given scan (ranging from 8 to about 340 nm) are above 40 nm. Utilizing a percentile-based threshold method to determine CCN_start allows for newly formed particles to grow to CCN sizes and is an effective metric when dealing with multi-year datasets. CCN_start for non-event days is calculated by using the average of CCN_start calculated for each season during events. We consider CCN concentrations during non-events to determine if NPF events result in an enhancement of CCN. Sunlight is generally necessary for NPF and growth; therefore, it is important to consider the variations in the seasonal diurnal cycle and obtain one unique value of CCN_start for each season that accurately represents the time that NPF impacts the site during each season (Hallar et al., 2011).

The end time of CCN consideration (CCN_end) is determined by finding the time at which particle growth from an event tapers off. To do so, we estimate the bin that corresponds to the normalized maximum aerosol concentration at each timestamp from the start of the NPF event to 17:00 MST the next day. This ensures that consecutive events are not erroneously considered. The maximum bin diameter at each timestamp is determined in a similar way to the NPF classification method (Eq. 1), but when considering CCN, we find the maximum of fitted Gaussians at each timestamp. Because the formation of CCN from nucleated particles can exceed the time period of NPF, especially in remote environments, our method allows for the evolution of particle growth over a time period long enough to ensure particles originating from NPF can grow to CCN sizes.

Once each time has a corresponding diameter maximum, we evaluate the overall growth pattern by fitting a polynomial curve to the Gaussian bin maxima over time. Once the curve is fit, the time at which the last inflection point occurs (in which the fitted line transitions from positive slopes to negative slopes) is selected as CCN_end. The last inflection point of the curve serves as an indicator of growth tapering off, and, therefore, we assume that the enhancement of CCN from NPF has concluded. For non-events, CCN_end is determined by adding the average duration of CCN consideration (CCN_end − CCN_start) to the previously averaged CCN_start. Four different values of CCN_end, one for each season, are determined when finding CCN_end values for non-events. An example NPF day including labels of CCN_start and CCN_end illustrating the time at which we assume CCN is enhanced by NPF is included in Fig. 3. To compare the impact NPF events have on CCN, CCN number concentrations directly measured are considered during the time period spanning from CCN_start to CCN_end during valid events and non-events. An average CCN number concentration for supersaturation levels between 0.2 % and 0.4 % is calculated for each individual time period. These values are then averaged separately each season between events and non-events. The goal is to determine whether CCN concentrations are enhanced by NPF events. During long-term studies, especially at clean, remote locations like SPL, directly comparing events and non-events will result in the relative enhancement of CCN due to events at a given location. By removing the subjectivity of selecting idealized cases, we provide a more robust methodology to evaluate long-term datasets. The methodology within this paper carefully considers similar timeframes within the diel pattern with and without NPF, to look at the relative change induced by NPF. At other high-altitude mountaintop sites around the globe, this approach could have sources of error, since NPF can be associated with the transport of both condensable vapors and pre-existing aerosol that could become CCN (Sellegri et al., 2019). However, SPL seems to be an exception to this rule, since previous observations of NPF show an association with lower existing particle surface areas which allows for a more direct comparison of events and non-events (Hallar et al., 2011; Sellegri et al., 2019). By further comparing events to non-events through a seasonal lens, we ensure that days with similar meteorological conditions are compared.

3.1 Fifteen years of new particle formation at Storm Peak Laboratory

Over the course of 15 years (2006–2021), we consider 835 d that pass basic quality control protocols and have both aerosol and CCN data available for analysis. The automatic method to determine NPF classification splits the data into one of the following five categories: type 1a event, type 1b event, class II event, undefined, or non-event (Hirsikko et al., 2007; Tröstl et al., 2016). Of the 835 d considered, 95 d are classified as a type 1a event, 80 d are classified as a type 1b event, 244 d are classified as a class II (burst) event, 269 d are classified as undefined, and 147 d are classified as a non-event. When considering the overall NPF event frequency, which includes type 1a events, type 1b events, or class II events, the overall event frequency calculated by the automatic method is 50 %, which compares well to the 52 % overall event frequency observed at SPL by Hallar et al. (2011).

Evaluating NPF from a seasonal lens at SPL creates a better understanding of how important variables, such as temperature, SO₂ concentrations, and the presence of organics, affect NPF (Hallar et al., 2016, 2013). Table 1 details the frequency of NPF events across all seasons. The summer and fall display the highest frequency of events with either a type 1a event, type 1b event, or class II event occurring on 56 % of days in the summer and 59 % of days in the fall. The spring (53 %) and winter (41 %) display similar but slightly lower event frequencies than the summer and fall at SPL. An analysis focusing on the frequency of different event types is conducted to determine which seasons may be conducive to the occurrence of type 1a and type 1b events in which persistent growth occurs. We find that type 1a events and type 1b events are more likely to occur in the winter (62 % of all NPF events) and spring (51 % of all NPF events) than in the summer (17 % of all NPF events) and fall (32 % of all NPF events) where burst events are more common partially due to higher temperatures (Yu et al., 2015).

Table 1A summary of variables related to the frequency of NPF split by season. The total event frequency considers the percentage of days that a type 1a event, type 1b event, or class II event occurs compared to an undefined or non-event day. The “Frequency of type 1a/1b events” row considers the percentage of all events in a given season that are persistent growth events (type 1a event or type 1b event).

Download Print Version | Download XLSX

When analyzing the impact of NPF on CCN concentrations, it is important to focus on days that exhibit a prolonged period of consistent particle growth, allowing for aerosols from NPF to reach CCN-relevant sizes. While type 1a events, type 1b events, and class II events are all considered NPF events, class II events do not exhibit strong, consistent growth, making it difficult to calculate growth statistics (Dal Maso et al., 2005). From this point forward, we focus on comparing type 1a events and type 1b events against non-events to better understand how aerosols from NPF affect CCN. Figure 4 compares the average number of particles of a given size produced during type 1a and type 1b events against non-events. We find that NPF days at SPL produce significantly more particles than non-event days up to diameters of 82.0 nm, which is larger than the critical diameter, theorized to be as low as 30 nm at SPL, required for aerosols to activate as CCN (Lowenthal et al., 2002). The significant increase in particles between 30.0 and 82.0 nm during type 1a and type 1b NPF events, providing an average enhancement of 2.78 (cm⁻³) times more particles during events, demonstrates an important influx of particles from NPF that reach sizes relevant to CCN formation at SPL. Above 82 nm, days with NPF events do not indicate more particles than non-events, which suggests any enhancements in CCN due to NPF events are likely due to particles below 82 nm. Previous work at SPL has shown that during NPF events, particles as low as 5 nm are observed alongside events demonstrating that particles observed during NPF originate from nucleation (Hallar et al., 2016).

Figure 4Average number of particles produced at each particle diameter for NPF events (blue) and non-events (red). NPF events produce significantly more particles at aerosol diameters below the vertical line (82.0 nm) as determined by a two-sample t test (p<0.05 indicates significance).

Download

3.2 Enhancements of cloud condensation nuclei due to new particle formation

To better understand the extent that aerosols from NPF events affect CCN concentrations, additional quality control is conducted to determine days when NPF events grow to CCN-relevant sizes and days with available CCN data taken at supersaturation levels between 0.2 % and 0.4 %. If there are errors in the CCN data or the difference between CCN_start and CCN_end is less than an hour, the day is discarded from CCN consideration. We compare 139 type 1a and 1b events that exhibit growth to CCN sizes against 111 non-events.

Figure 5 illustrates comparative CCN number concentrations following type 1a and 1b events and non-events. We find that NPF enhances CCN concentrations by a factor of 1.54 in the spring and 1.36 in the winter. Higher CCN concentrations during NPF events than non-events are statistically significant in both the winter (p=0.020) and spring (p=0.025). However, CCN concentrations between events and non-events during the summer (p=0.889) and fall (p=0.432) are not statistically significant. Average number concentrations of CCN are higher during NPF events in the spring (event: 146.47 cm⁻³, non-event: 94.92 cm⁻³), winter (event: 98.60 cm⁻³, non-event: 72.67 cm⁻³), and fall (event: 258.84 cm⁻³, non-event: 245.61 cm⁻³) but lower during NPF events in the summer (event: 306.63 cm⁻³, non-event: 388.05 cm⁻³).

Figure 5Seasonal comparisons of average CCN number concentrations (cm⁻³) during NPF events (blue) and non-events (red). CCN concentrations during events are considered starting at the CCN_start time and ending at the CCN_end time of a given day. CCN is considered during non-events starting at the seasonal average of CCN_start and ending at the seasonal average of CCN_end. Displayed p values represent the results of a two-sample t test with a one-sided hypothesis that NPF days would exhibit greater CCN concentrations than non-events. We interpret values below p=0.05 to be statistically significant. p values show that the spring and winter display statistically significant enhancements of CCN due to NPF, a trend that is absent during the summer and fall.

Download