Spatial and temporal scales of new particle formation events in eastern North America

Atmospheric Environment 75 (2013) 257e264

Contents lists available at SciVerse ScienceDirect

Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv

Spatial and temporal scales of new particle formation events in eastern North America P. Crippa*, S.C. Pryor Atmospheric Science Program, Department of Geological Sciences, Indiana University, Bloomington, IN 47405, USA

h i g h l i g h t s
New particle formation (NPF) over eastern North America is regionally coherent.
NPF frequently occurs on two sequential days.
The mean spatial scale of NPF is 120e850 km based on the season and assumptions.
Local scale NPF variability is linked to variations in boundary-layer dynamics.

a r t i c l e i n f o

a b s t r a c t

Article history: Received 7 February 2013 Received in revised form 15 April 2013 Accepted 16 April 2013

New particle formation (NPF) events have been observed in numerous locations. However, questions remain as to the scale of the events and their importance to regional and global particle number concentrations, size distributions and climate forcing. This study presents measured particle size distributions (PSD) at multiple sites across eastern North America and evaluates the degree of coherence on large (hundreds of kilometer) scales and the site-to-site variability across scales of tens of kilometers. Longterm data from sites separated by 1500 km demonstrate frequent and synchronous NPF, that over 80% of event days at both sites are followed by another event day and that event sequences are best described by a Markov Chain of order 1. Estimates of the mean spatial scale of NPF from a site in southern Indiana range from at least 120e850 km depending on the season and the precise assumptions applied. Despite the evidence for regional coherence in NPF, detailed measurements along an 80 km transect in southern Indiana also indicate some important sub-regional variability. While PSD from individual days typically indicate NPF at all three sites or at none of the sites, PSD measured in two urban environments show greater coherence than those from a centrally located site in a forest, and both the number of ultrafine particles and their growth rates are typically (but not uniformly) higher at the forested site. Some of the site-to-site variability appears to be causally linked to planetary boundary layer dynamics and variations in land cover. Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: Nucleation Spatial scale Markov chain Boundary-layer dynamics

1. Introduction and objectives New particle formation (NPF) events have been observed across a wide array of ground-based stations (Kulmala et al., 2011), in the near-surface planetary boundary layer (PBL) (e.g. Spracklen et al., 2006) and in the free troposphere (e.g. Weber et al., 1999). Questions pertaining to the temporal and spatial scales of NPF events, subsequent growth and removal rates for the resulting ultra-fine particles (UFP) are critical to determining the global and regional importance of NPF in terms of dictating

* Corresponding author. Tel.: þ1 812 855 5582; fax: þ1 812 855 7899. E-mail address: [email protected] (P. Crippa). 1352-2310/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.atmosenv.2013.04.051

particle size distributions (PSD) and thus their relevance to direct and indirect climate forcing (Spracklen et al., 2006; Riipinen et al., 2011). Observations at regional and sub-regional scales have indicated NPF is observed over tens to hundreds of kilometers, but local variability is manifest at smaller scales. For example, while large spatial scale NPF events are frequently simultaneously observed at five locations over Scandinavia, event characteristics are rarely identical (Hussein et al., 2009). Further, the limited observational data currently available indicate notable variation in nucleation intensity, timing and growth rates. Measurements in northern Germany indicated simultaneous NPF at urban and upwind rural sites, but NPF was uniformly initiated earlier at the rural sites, and there were also cases of horizontally inhomogeneous NPF (Wehner et al., 2007).

258

P. Crippa, S.C. Pryor / Atmospheric Environment 75 (2013) 257e264

PSD in and near Pittsburgh, Pennsylvania, showed frequent simultaneous NPF on scales of tens of kilometers, but number concentrations at the rural site were 2e3 times lower than urban values (Stanier et al., 2004). Observations at five stations on a 500 km transect in southern Ontario also found clear cases of regional scale NPF events, but documented site-to-site variability that was attributed to local availability of sulfuric acid and variations in the background particle population (Jeong et al., 2010). Modeling studies have also indicated NPF is observed at regional scales, but the intensity of NPF and resulting ultra-fine particle concentrations indicate substantial spatial gradients over the eastern USA (Luo and Yu, 2011). There is also evidence that NPF is not uniform through the PBL, and may be concentrated in the residual layer (RL), with subsequent entrainment into the PBL (Wehner et al., 2010; Pryor et al., 2011; Olofson et al., 2009). Thus, there is clear evidence for substantial geographic variability in the temporal and spatial scales of NPF, and for subsequent growth. Here we examine the coherence of nucleation events in eastern North America using PSD measurements from sites separated by a distance of hundreds of kilometers, and detailed transects across three different land-cover types coupled with numerical simulations using the Weather Research and Forecasting (WRF) model to examine variability of NPF on smaller spatial scales and links to boundary-layer dynamics. Thus, we present analyses designed to address the following research objectives:
Analyze the degree to which PSD observed over distances of hundreds of kilometers indicate simultaneous NPF and therefore potentially regional events.
Analyze seasonal and spatial differences in nucleation duration and spatial scales of NPF, and investigate possible causes of the observed variability.
Investigate local variability in NPF and the characteristics there-of and specifically examine the degree to which site-tosite variability is attributable to variations in PBL dynamics. 2. Assessing the regional coherence and scales of NPF 2.1. Methods and approach Long-term PSD measurements analyzed herein were collected at the Morgan Monroe State Forest (MMSF) in southern Indiana (39.317 N, 86.417 W) during 1/1/2007e07/31/2009, and Egbert in a semi-rural location around 70 km north of Toronto (44.23 N, 79.78 W) during 05/01/2007e05/31/2008. The measurements were conducted using Scanning Mobility Particle Spectrometers (SMPS) with inlets at a height of 46 m at MMSF (above a canopy of 26e28 m) and 5.5 m a.g.l. at Egbert (see Pryor et al. (2010) and Riipinen et al. (2011)). Given this study focuses on investigating the role of background conditions and boundary layer dynamics in initiating NPF and controlling the appearance of freshly nucleated particles in the near-surface layer, the small difference in the sampling heights at the two measurement sites is not expected to affect our conclusions. Data from both sites were used to classify NPF events following the approach of Boy and Kulmala (2002) where class A events exhibit a sudden appearance of a new particle mode with number geometric mean diameter <25 nm and consistent growth for at least 1 h. If this sudden increase in number concentration was not followed by a consistent growth profile a C event was identified. When a sudden increase of ultrafine particles was observed, but not from the smallest measured diameters, the day was classified as B event. A day not conforming to these criteria was defined as non-event.

2.2. Spatial and temporal coherence of NPF occurrence Nucleation rates are dictated by the availability of nucleation chemical precursors, condensable species, background particle concentrations and meteorological conditions which determine the radiation flux, degree of stagnation and transport of nucleation precursors (Boy and Kulmala, 2002; Pryor et al., 2010). For these reasons the sequence of event and non-event days at a location may not be independent of air mass history (Sogacheva et al., 2007). We analyzed the importance of atmospheric preconditioning in dictating NPF using a Markov-chain approach:


  Xt
Xt1 ; .Xtk w Be pXt1 ; .pXtk

(1)

where pXt1 ¼ probability of a nucleation event X occurring at time (t1) and Be ¼ Bernoulli distribution. The zero order log-likelihood L0 of the model is defined as:

L0 ¼ n0  ln p0 þ n1  ln p1

(2)

where n0 and n1 ¼ number of non-event and event days and p0and p1 ¼ respective probabilities. A similar approach can be used to derive the log-likelihood of higher orders:

L1 ¼ n01  ln p01 þ n11  ln p11

(3)

To determine the most parsimonious Markov Chain order we use the Bayesian information criterion (BIC) (Schoof and Pryor, 2008) computed as:

BICðmÞ ¼ 2Lm þ 2m ½lnðnÞ

(4)

where Lm ¼ log-likelihood for a model of order m and n ¼ sample size. Data from MMSF indicate 87% of event days are followed by another NPF event day, while for Egbert p11 is 83%. Further, at both sites a first order Markov chain model is the proper model to describe the occurrence of NPF (i.e. BIC is minimized for L1) (Table 1). This implies that the conditions associated with NPF persist for multiple days. Further, consistent with prior research which has indicated weak association between the condensational sink (CS) and occurrence of NPF (Pryor et al., 2010), the occurrence of consecutive event days is not prevented. A test of independence applied to event occurrence (i.e. a binary index of the occurrence (1) or not (0)) of NPF indicates that events are not randomly distributed at the two sites, but rather tend to occur simultaneously (Table 2). This suggests regionally coherent NPF, or at least regional coherence in the conditions associated with NPF, and thus regional-scale forcing. A statistically significant dependence is also evident for lag-1 (i.e. when MMSF events are shifted one day back relative to Egbert), but not for the converse (i.e. when MMSF events are shifted one day forward from Egbert) (Table 2). This result may indicate transport of conditions

Table 1 The probability of NPF on any given day (p1) and two sequential days (p11) during 05/ 01/2007e05/31/2008 at the MMSF and Egbert sites. The Bayesian information criterion (BIC) for a Markov Chain of up to the third order is also reported.

Classifiable days p1 p11 BIC(0) BIC(1) BIC(2) BIC(3)

MMSF

Egbert

292 0.47 0.41 453.02 232.98 414.71 417.27

331 0.41 0.34 405.06 171.64 311.30 294.73

P. Crippa, S.C. Pryor / Atmospheric Environment 75 (2013) 257e264

259

Table 2 Test of independence of NPF events at MMSF and Egbert based on PSD measurements during 05/01/2007e05/31/2008. The c2, p-value and sample size n are reported for simultaneous dates (Dt ¼ 0), for dates shifted one day forward (Dt ¼ þ1) and one day back (Dt ¼ 1) for the MMSF site relative to Egbert.

Dt ¼ 0 Dt ¼ þ1 Dt ¼ 1

c2

p-value

n

7.11 0.61 19.61

0.008 0.434 9.51  106

250 249 249

conducive to NPF from the southwest (e.g. high emissions of nucleation precursors in the southern Ohio Valley (Pryor et al., 2010)), or that the conditions likely to be conducive to NPF are first manifest at the southern end of the transect (i.e. at MMSF). It is possible that sequencing of synoptic scale phenomena may explain both the occurrence of subsequent events and the delay of NPF at the Egbert site. Previous analyses have indicated 75% of class-A events at MMSF were preceded by a cold front passage within the prior 36 h (Pryor et al., 2010). Thus one can envisage the following scenario: anticyclonic conditions cause stable and stagnant air, but dissipate with the passage of mid-latitude cyclones and associated cold fronts from the west which may first encounter MMSF and then Egbert. 2.3. Estimating the spatial extent of regional nucleation events from temporal duration The horizontal scale of NPF was estimated by multiplying the time during which measurements at a site indicate a distinct nucleation mode by the prevailing wind speed (Birmili et al., 2003). For all class A events not followed by rain, we computed the number geometric mean diameter ðDgN Þ for particles with diameters between 10 and 100 nm ðDgN10100 Þ and from 30 to 100 nm ðDgN30100 Þ. The start of an event is when maxðDgN10100  DgN30100 Þ and the event ends when they differ by < 15% (see Fig. 2). The results indicate that nucleation duration (in hours) at both sites appears to be longest during winter and shortest during summer (Table 3). At MMSF this seasonality may reflect faster growth rates (GR) during periods of more intense solar radiation and forest activity leading to the larger abundance of condensable vapors (Pierce

Fig. 2. An example of the calculation of NPF event duration based on the evolution of the number geometric mean diameter ðDgN Þ of the nucleation and Aitken mode in data from MMSF. Details of the procedure applied are given in Section 2.3.

et al., 2012; Riipinen et al., 2011). We might expect longer time scales during the fall too, but 90% of class A events at MMSF occurred before the end of October, prior to senescence of the deciduous forest (Jeong et al., 2011), when the forest was still acting as a significant source of biogenic VOCs. GR derived from the mode merging approach used to obtain the nucleation length scale (Table 3) indicate that, on average, particles were able to grow up to 60 nm during event class A days at MMSF within 18 h of detection at 10 nm assuming that the GR of particles in the kinetic regime is independent of size (i.e. the GR is linear) (Lehtinen and Kulmala, 2003). These GR were used to compute the minimum spatial scale for nucleated particles to attain climate relevance as follows. Assuming the GR for mode merging (GR10e 30 nm) is an analog for GR30e60 nm and the region is characterized by homogeneous meteorological conditions, we can calculate the minimum horizontal scale of NPF (Birmili et al., 2003) as:

 distance ¼ nucleation lengthobs;1030

nm

þ nucleation lengthest;3060

 ¼ nucleation lengthobs;1030



nm

nm þ

u

DDp3060 GR3060

nm nm

 u (5)

Fig. 1. Sampling stations from which data are presented. MMSF and Egbert are the sites used for the long-term data sampling, while MMSF, Indianapolis and Bloomington were the focus for the NIFTy experiment. The underlying colors show the land use categories as used in the WRF simulations over the parent and nested domain which are shown by the two black boxes (i.e. the parent domain extends over most of eastern North America, while the inner nested domain includes only the area encompassed by the smaller, inner box). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

260

P. Crippa, S.C. Pryor / Atmospheric Environment 75 (2013) 257e264

Table 3 Seasonal spatial and temporal scales of NPF estimated from particle GR and mean wind speeds (u) at MMSF on event class A days during 01/01/2007e07/31/2009. Two estimates of the horizontal scale for nucleated particles to attain climate relevance are provided following Equation (5). The upper bound estimate assumes a constant GR10e30 nm and GR30e60 nm derived from analysis of the time required for the nucleation mode to merge into the Aitken mode, whereas the lower bound estimate uses the mean seasonal GR10e30 nm derived from observations as an analog for GR30e60 nm. For comparison the mean seasonal nucleation length and the standard deviation (s) at Egbert on A events during 05/01/2007e05/31/2008 is also reported (note that events during winter were insufficiently numerous to allow stable estimation of the nucleation length scale). It should be noted that the GR reported in this table are computed for growth of particles with Dp > 10 nm, thus they are not directly comparable to GR computed for observations of sub-10 nm particle populations. Variable

Season Winter (DJF)

Egbert

MMSF

Upper bound (MMSF)

Lower bound (MMSF)

Number of days Mean measured nucleation length for mode merging [h] s of mode merging [h] Number of days (u) [m s1] Mean measured nucleation length for mode merging [h] s of mode merging [h] Mean estimated GR for mode merging [nm h1] Spatial scale computed based on the time required for a freshly nucleated particle to merge into the Aitken mode [km] Estimated nucleation length DDp ¼ 30e60 nm [h] Estimated nucleation length DDp ¼ 10e60 nm [h] Minimum spatial scale DDp ¼ 10e60 nm [km] Mean measured GR10e30 nm [nm h1] Estimated nucleation length DDp ¼ 30e60 nm [h] Estimated nucleation length DDp ¼ 10e60 nm [h] Minimum spatial scale DDp ¼ 10e60 nm [km]

Fig. 3. Estimated NPF horizontal spatial extent of event class A days estimated from the back-trajectories analysis during (a) the full year and (b) spring at MMSF based on data from 05/31/2007e07/31/2009. The polar plot was generated discretizing the analyzed domain into grid cells with r ¼ 150 km and q¼p/6. Colors represent the probability distribution of start locations for each sector whereas the black numbers on the edge represent the frequency of the air mass originating within each sector on daily basis. The black circle centered at MMSF indicates the average spatial scale estimated from the GR approach (i.e. 530 and 570 km for the full year and spring, respectively). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Spring (MAM)

Summer (JJA)

Fall (SON)

6 3.65 25.8 14.4 0.77 339

5 10.8 3.3 49 4.15 17.4 12.8 1.15 260

6 9.0 6.1 16 3.42 9.4 3.2 2.13 116

7 14.0 13.9 18 3.33 13.6 11.9 1.47 163

39.0 64.8 851 1.01 29.7 55.5 729

26.1 43.5 650 1.96 15.3 32.7 489

22.6 32.0 393 1.64 18.3 27.7 341

21.0 34.6 414 1.65 18.2 31.8 381

2 e e

where u is the mean wind speed measured at the MMSF site during event class A days. A lower bound on the estimated horizontal scale derived from (5) is 340e730 km, while the upper bound is 390e850 km (Table 3), thus they both indicate large-scale NPF events. The estimate of nucleation mode lifetime derived from the mode-merging is 120e 340 km (Table 3). All three estimates are comparable to the estimate of a few hundreds of kilometers from northern Europe (Birmili et al., 2003; Crumeyrolle et al., 2010). The spatial extent of nucleation events was also quantified following an independent approach based on back-trajectory analysis (Hussein et al., 2009). On each A event day at MMSF a back-trajectory was started at a height of 500 m at the time a nucleation mode was first distinguishable from the Aitken mode and computed for each subsequent hour until the two modes merged. The back-trajectories were computed using the Internet based Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT-WEB) (Draxler and Rolph, 2013) initialized with meteorological conditions from the North American Mesoscale Model at 12 km resolution (NAM12). The mean spatial extent of event class A days averaged over the entire year is several hundreds of kilometers and extends from Missouri and Arkansas in the southwest, and to the northeast toward Egbert and Lake Michigan (Fig. 3a). During spring, when the majority of A-class events occur, the back-trajectory probability density plot is similar but exhibits a higher frequency of starting in the northeast sectors (Fig. 3b). The spatial extent of events from the north is limited to distance of within a few hundreds of kilometers of MMSF, indicating those NPF events originate closer to the site and are likely associated with atmospheric stagnation. Conversely, air masses coming from the southwesterly sectors show both the largest frequency of occurrence and highest probability of distant start locations which is indicative of rapid long-range transport to MMSF and potentially from there on Egbert, located to the northeast. These results are thus consistent with results from the Markov Chain analysis that show the MMSF and Egbert sites are not independent and that NPF events may either occur simultaneously in the presence of atmospheric stagnation, or be delayed in Egbert under conditions of southwesterly flow. These horizontal scales are also consistent with

P. Crippa, S.C. Pryor / Atmospheric Environment 75 (2013) 257e264

evaluated relative to measurements at MMSF. The value added by running high resolution WRF simulations relative to the information provided by NAM12 was also quantified based on Brier skill scores (BSS):

Table 4 Evaluation of hourly WRF simulated values relative to observations at MMSF during 11e26 May 2008. The regression fits are for a forced intercept of 0, and Brier Skill Scores (BSS) are calculated from Equation (6). Number of Estimated 95% CI on slope R2 observations slope

Variable

Friction velocity (u*) 289 Incoming short wave 334 radiation (SW) Air temperature (T) 382

261

BSS (%)

0.874 0.978

0.839, 0.909 0.941, 1.016

0.47 e 0.83 19.58

0.901

0.888, 0.914

0.99 38.34

BSS ¼

MSEðc; xÞ  MSEðf ; xÞ MSEðf ; xÞ  0

(6)

where MSE(c, x)and MSE(f, x) are the mean squared error of the reference (NAM12) and forecast (WRF) respectively, relative to measurements at the MMSF site (x). Generally, the WRF simulations exhibit a high degree of association with in situ observations (R2  0.47, for a regression with 0 intercept) and improved ‘prediction’ of these parameters relative to NAM12 (BSS > 0) (Table 4). The results thus imply WRF is able to generate an accurate description of some key meteorological and boundary-layer properties at the local scale, therefore allowing us to use WRF output to diagnose dynamically causes of variations in NPF events across the three measurement sites.

the average spatial extent of nucleation derived from the GR approach, as indicated by the black circle centered at MMSF in Fig. 3. 3. Analysis of spatial variability at sub-regional scales 3.1. Methods and approach The above analyses are consistent with research in other locations that has indicated simultaneous nucleation events across large distances. This, of course, does not indicate that NPF is observed at all intervening locations, but rather that the conditions associated with NPF are manifest at those spatial scales. However, as also described above, sub-regional variability in NPF has likewise been observed, and thus the degree of coherence in the occurrence and intensity of NPF and particle growth was investigated using data from the Nucleation In Forests (NIFTy) experiment which was conducted in southern Indiana (5e31 May 2008). During NIFTy PSD measurements were collected along an 80 km transect at three locations: downwind of the urban core of Indianapolis (39.811 N, 86.114 W), MMSF, and in the small college town of Bloomington (39.171 N, 86.506 W) (Fig. 1) (Pryor et al., 2011). In support of this analysis, meteorological conditions over eastern North America were simulated for 11e26 May 2008 using Weather Research and Forecasting model Version 3 (WRFV3 e ARW) applied over an outer domain at a spatial resolution of 9 km, and an inner domain at 3 km (Fig. 1) (see Crippa et al. (2012) for details of the simulations settings). Lateral boundary conditions were provided by NAM12 (as in the back-trajectory analysis) and the high-resolution land cover was input from the USGS 24-category Land Use classification (Fig. 1) (Wang et al., 2012). Because we focus on understanding subregional variability in NPF and specifically links to PBL processes, key parameters linked to PBL dynamics as simulated by WRF were

3.2. Site to site variability of nucleation intensity and growth rates during NIFTy Mean nucleation intensity, computed for particle diameters (Dp) ¼ 10e30 nm in the two hours of highest particle number concentrations of that size, is quite similar at the three sites, although the relative intensity on individual event days and the hour of the day at which the increase in UFP concentrations is first observed (i.e. nucleation start time) show marked site-to-site variability (Table 5). The coefficient of divergence (COD) computed for seven size channels with Dp: 10e20 nm, 20e30 nm, 30e40 nm, 40e50 nm, 50e60 nm, 60e70 nm and 70e100 nm is used to investigate the spatial variability of particle number concentrations among the three sites:

CODjk

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi # u n " u1 X N  N 2 ij ik ¼ t n i ¼ 1 Nij þ Nik

(7)

where n is the sample size, j and k represent two different sites and N is the ith hourly averaged particle number concentration. As in prior research in southern Ontario (Jeong et al., 2010), the COD between MMSF and both Indianapolis and Bloomington exhibit highest values for the smallest particles (indicating high spatial variability), and declining values as the particle size increases

Table 5 Observed start hour (LST), nucleation intensities (measured as the total particle number concentration for Dp ¼ 10e30 nm) [cm3] and growth rates (GR) [nm h1] on event days during NIFTy. GR are computed applying a linear fit to the number geometric mean diameter ðDgN Þ in the nucleation mode (10e30 nm) in the three hours subsequent to the minimum ðDgN Þ. Day in May

12 13 16 17 18 19 20 21 22 24 25 Mean

Indianapolis

MMSF

Start hour (LST)

Observed nucleation intensities [cm3]

10 9 10 9 10 9 10 10 12 9 9 9.7

1.61 3.55 1.28 3.81 1.64 5.30 2.63 1.96 1.35 3.74 5.12 2.91

           

104 104 104 104 104 104 104 104 104 104 104 104

GR [nm h

3.13 2.55 1.64 1.96 1.65 1.00 2.01 1.96 1.18 1.50 2.12 1.79

1

]

Bloomington

Start hour (LST)

Observed nucleation intensities [cm3]

10 11 9 9 10 9 13 8 9 10 9 9.7

8.48 2.90 3.58 1.06 5.17 6.95 1.18 2.47 4.90 2.22 5.37 4.90

           

104 104 104 105 104 104 104 104 104 104 104 104

GR [nm h

2.45 2.45 1.34 2.88 2.02 1.36 3.28 2.68 1.92 2.85 2.80 2.42

1

]

Start hour (LST)

Observed nucleation intensities [cm3]

GR [nm h1]

11 11 10 10 11 10 e 8 11 12 12 10.6

5.54  104 3.42  104 2.64  104 2.48  105 4.91  104 7.33  104 Non-event 6.34  104 2.58  105 1.56  104 2.36  104 8.46  104

1.55 1.79 0.44 1.94 1.19 0.64 e 1.12 2.53 1.92 1.60 1.81

262

P. Crippa, S.C. Pryor / Atmospheric Environment 75 (2013) 257e264

(indicating lower spatial heterogeneity) (Fig. 4). High COD for UFP concentrations between MMSF and both Indianapolis and Bloomington and the relatively low COD for UFP at Indianapolis and Bloomington imply that although NPF events were observed on the same days (Table 5), the intensity of NPF varied across the region. We postulate that while Bloomington and Indianapolis are responding to the regional background NPF and PSD, the MMSF site exhibits more intense nucleation events or more rapid growth to 10e30 nm due to the different chemical environment at the forest site (see Table 5 which indicates GR at MMSF were an average of 33% higher at MMSF than either of the urban sites). This inference is consistent with prior work showing that the oxidation of biogenic VOCs contributes to increase UFP number concentrations and subsequent growth in remote areas, where major anthropogenic sources are limited (Tunved et al., 2006; Spracklen et al., 2008; Pierce et al., 2012). 3.3. Diagnosing causes of site-to-site variability Partitioning of semi-volatiles between condensation onto existing particles versus nucleation would suggest the magnitude of the in situ surface area might be negatively correlated with the occurrence or intensity of nucleation (Lehtinen and Kulmala, 2003). However, the observed surface area for Dp ¼ 10e100 nm is significantly higher prior to NPF on event than on non-event days at MMSF (Fig. 5a). Further, after the sunrise when the nocturnal RL starts to be eroded and a fully mixed PBL develops, the particle surface area does not decrease as a consequence of dilution, but is almost constant, possibly indicating the eroding RL contains a higher concentration of 6e30 nm particles than the surface layer. Additionally, the nucleation mode particle surface area starts increasing around 1 h before the maximum rate change of 10 nm particle concentrations is observed (Fig. 5b). Thus we speculate that 6e10 nm particles are formed at higher atmospheric levels (likely close to or within the RL) and are then transported downward with the erosion of the RL and development of a fully mixed PBL (Crippa et al., 2012). Further evidence in support of this postulate may be drawn from a crosscorrelation analysis of hourly average particle concentrations for Dp ¼ 6e30 nm at MMSF and WRF simulated values of turbulent kinetic energy (TKE) on event days (Fig. 6). The highest co-variability between observed UFP concentrations and TKE is observed aloft from the surface (i.e. 1000 m a.g.l.) and approximately 2 h prior to the observed maximum rate change of 10 nm particle concentrations thus supporting our speculation of nucleation being initiated aloft (Fig. 6). Similar results and statistically significant correlations

Fig. 4. Coefficient of divergence (COD) (see Equation (7)) presented as a function of particle diameter (Dp) based on hourly average measurements at Bloomington, MMSF and Indianapolis during NIFTy.

Fig. 5. (a) Average diurnal evolution of particle surface area during event days (solid lines) and non-event days (dashed lines) at the NIFTy measurement sites. Particle surface area is normalized relative to the maximum average value at each site during event days. The bigger dots represent the hours when the mean surface area during class A event days is statistically significantly higher than on non-event days at a confidence level of 90%. (b) Average surface area of particles in the nucleation and Aitken mode, normalized relative to the maximum average value at MMSF during event days. In both panels the x-axis shows the time normalized relative to the event start time which is the hour of maximum rate change of 10 nm particle number concentrations (or 9 am LST for non-event days).

were also found for the Indianapolis and Bloomington sites (not shown). Based on evidence of a strong link between NPF and PBL dynamics, output from WRF was used to examine whether differences

Fig. 6. Cross-correlation between measured 6e30 nm particle number concentrations [cm3] and simulated TKE [m2 s2] values from WRF at multiple levels on event days at MMSF. The time coordinate has been normalized relative to the event start time as defined in the caption to Fig. 5. The threshold for significant correlations at a confipffiffiffi dence level of 90% (0  1.64  1= n ¼ 0.3977) is shown by the horizontal bars.

P. Crippa, S.C. Pryor / Atmospheric Environment 75 (2013) 257e264

in PBL dynamics (and land cover and thus surface energy balance partitioning) explain the site-to-site variability evident in data from NIFTy. Key to this analysis is thus the high spatial resolution (0.9 km) land cover classification which is Urban and Built-up Land for the grid cell containing Indianapolis, Deciduous Broadleaf Forest for MMSF and Cropland/Woodland Mosaic for Bloomington (Fig. 1). In Fig. 7 we present a case study of PBL dynamics as simulated by WRF for May 20 to investigate the factors likely responsible for the different event classification at the three sites (i.e. Table 5 shows a class C in Indianapolis, class B at MMSF and non-event in Bloomington) (Fig. 7). At all three sites the nocturnal temperature inversion was completely eroded by 9 am (LST) but while strong enhancement of TKE was simulated at both Indianapolis and MMSF (Fig. 7aeb) where a NPF event was observed, over Bloomington the

263

maximum gradient of TKE occurs later in the afternoon and shows more limited vertical extent (Fig. 7c). This is consistent with crosscorrelation results which indicate a possible UFP formation aloft with subsequent entrainment into the PBL. The WRF simulation also indicates expansive coverage of low-level and mid-level clouds during nighttime at the three sites which started to dissipate a few hours after the sunrise (Fig. 7gei). Cloud cover decreased earlier and a deeper PBL developed earlier over Indianapolis (Fig. 7g) than MMSF (Fig. 7h) which may explain the earlier nucleation start at Indianapolis (around 10 am LST) (Fig. 7j) relative to MMSF (around 1 pm LST) (Fig. 7k). The fraction of low- and mid-level clouds did not significantly decrease at the Bloomington site during the morning hours (Fig. 7i), which may have contributed to suppression of NPF at this site.

Fig. 7. Case study of the relationship between boundary-layer properties and NPF event on May 20, 2008. The first column (a-d-g-j) refers to the Indianapolis site (C event), the second column (b-e-h-k) to MMSF (B event) and the third column (c-f-i-l) to Bloomington (non-event). Simulated vertical profiles of Log10(TKE) are shown in the upper row, mixed layer depth (MLD) is given in the second row along with incoming short wave radiation, cloud cover fraction in three cloud classes is given in the third row, while the final row shows the measured PSD [cm3] (expressed as dN/dLog10Dp).

264

P. Crippa, S.C. Pryor / Atmospheric Environment 75 (2013) 257e264

4. Concluding remarks Eastern North America is characterized by high concentrations of climate-relevant particles (Luo and Yu, 2011). Long-term data sets from MMSF and Egbert (which are separated by 1500 km) demonstrate frequent and synchronous NPF, that over 80% of event days at both sites are followed by another event day and that event sequences are best described by a Markov Chain of order 1. The time sequencing of NPF events and back-trajectory analyses indicate evidence that conditions associated with NPF may be preferentially advected from the southwest. Depending on the precise assumptions applied, the mean spatial scale of NPF determined both from time-scale analyses and back-trajectory calculations ranges from 340 to over 850 km. There is evidence that the spatial scales of these events are largest in winter and smallest in summer. While there is strong evidence that the conditions associated with NPF extend over large spatial scales, analyses of data collected along an 80 km transect in southern Indiana also reveal local variability in NPF features (event type, nucleation intensity and growth rates). Cross-correlation analysis of UFP concentrations and profiles of TKE exhibit clear evidence for enhancement of TKE aloft prior to detection of NPF in the surface layer. This and other analyses thus provide evidence of an elevated source for NPF, and that site-to-site variability in NPF occurrence and intensity may be dictated by subtle differences in boundary-layer dynamics. Acknowledgments This research was funded by grants to SCP from NSF (0544745 and 1102309), and supplemental funding from the IU Pervasive Technology Institute. Richard Leaitch is acknowledged for providing the PSD data for Egbert and for useful discussions. Thanks also for useful discussions and research support to R.J. Barthelmie, B. Plale and G. S. El Afandi (IU), and P. Hopke (Clarkson). This manuscript was substantially improved by the comments and suggestions of three anonymous reviewers. References Birmili, W., Berresheim, H., Plass-Dulmer, C., Elste, T., Gilge, S., Wiedensohler, A., Uhrner, U., 2003. The Hohenpeissenberg aerosol formation experiment (HAFEX): a long-term study including size-resolved aerosol, H2SO4, OH, and monoterpenes measurements. Atmospheric Chemistry and Physics 3, 361e376. Boy, M., Kulmala, M., 2002. Nucleation events in the continental boundary layer: Influence of physical and meteorological parameters. Atmospheric Chemistry and Physics 2, 1e16. Crippa, P., Petäjä, T., Korhonen, H., El Afandi, G.S., Pryor, S.C., 2012. Evidence of an elevated source of nucleation based on model simulations and data from the NIFTy experiment. Atmospheric Chemistry and Physics 12, 8021e8036. Crumeyrolle, S., Manninen, H.E., Sellegri, K., Roberts, G., Gomes, L., Kulmala, M., Weigel, R., Laj, P., Schwarzenboeck, A., 2010. New particle formation events measured on board the ATR-42 aircraft during the EUCAARI campaign. Atmospheric Chemistry and Physics 10, 6721e6735. Draxler, R.R., Rolph, G.D., 2013. HYSPLIT (HYbrid Single-particle Lagrangian Integrated Trajectory) Model Access via NOAA ARL READY Website. NOAA Air Resources Laboratory, Silver Spring, MD. http://ready.arl.noaa.gov/HYSPLIT.php. Hussein, T., Junninen, H., Tunved, P., Kristensson, A., Dal Maso, M., Riipinen, I., Aalto, P.P., Hansson, H.C., Swietlicki, E., Kulmala, M., 2009. Time span and spatial scale of regional new particle formation events over Finland and Southern Sweden. Atmospheric Chemistry and Physics 9, 4699e4716. Jeong, C.H., Evans, G.J., McGuire, M.L., Chang, R.Y.W., Abbatt, J.P.D., Zeromskiene, K., Mozurkewich, M., Li, S.M., Leaitch, W.R., 2010. Particle formation and growth at five rural and urban sites. Atmospheric Chemistry and Physics 10, 7979e7995. Jeong, S.-J., Ho, C.-H., Gim, H.-J., Brown, M.E., 2011. Phenology shifts at start vs. end of growing season in temperate vegetation over the Northern Hemisphere for the period 1982e2008. Global Change Biology 17, 2385e2399. Kulmala, M., Asmi, A., Lappalainen, H.K., Baltensperger, U., Brenguier, J.L., Facchini, M.C., Hansson, H.C., Hov, O., O’Dowd, C.D., Poschl, U., Wiedensohler, A., Boers, R., Boucher, O., de Leeuw, G., van der Gon, H., Feichter, J., Krejci, R., Laj, P., Lihavainen, H., Lohmann, U., McFiggans, G., Mentel, T., Pilinis, C., Riipinen, I., Schulz, M., Stohl, A., Swietlicki, E., Vignati, E., Alves, C., Amann, M., Ammann, M., Arabas, S., Artaxo, P., Baars, H., Beddows, D.C.S., Bergstrom, R., Beukes, J.P.,

Bilde, M., Burkhart, J.F., Canonaco, F., Clegg, S.L., Coe, H., Crumeyrolle, S., D’Anna, B., Decesari, S., Gilardoni, S., Fischer, M., Fjaeraa, A.M., Fountoukis, C., George, C., Gomes, L., Halloran, P., Hamburger, T., Harrison, R.M., Herrmann, H., Hoffmann, T., Hoose, C., Hu, M., Hyvarinen, A., Horrak, U., Iinuma, Y., Iversen, T., Josipovic, M., Kanakidou, M., Kiendler-Scharr, A., Kirkevag, A., Kiss, G., Klimont, Z., Kolmonen, P., Komppula, M., Kristjansson, J.E., Laakso, L., Laaksonen, A., Labonnote, L., Lanz, V.A., Lehtinen, K.E.J., Rizzo, L.V., Makkonen, R., Manninen, H.E., McMeeking, G., Merikanto, J., Minikin, A., Mirme, S., Morgan, W.T., Nemitz, E., O’Donnell, D., Panwar, T.S., Pawlowska, H., Petzold, A., Pienaar, J.J., Pio, C., Plass-Duelmer, C., Prevot, A.S.H., Pryor, S., Reddington, C.L., Roberts, G., Rosenfeld, D., Schwarz, J., Seland, O., Sellegri, K., Shen, X.J., Shiraiwa, M., Siebert, H., Sierau, B., Simpson, D., Sun, J.Y., Topping, D., Tunved, P., Vaattovaara, P., Vakkari, V., Veefkind, J.P., Visschedijk, A., Vuollekoski, H., Vuolo, R., Wehner, B., Wildt, J., Woodward, S., Worsnop, D.R., van Zadelhoff, G.J., Zardini, A.A., Zhang, K., van Zyl, P.G., Kerminen, V.M., Carslaw, K.S., Pandis, S.N., 2011. General overview: European Integrated project on aerosol cloud climate and air quality interactions (EUCAARI) e integrating aerosol research from nano to global scales. Atmospheric Chemistry and Physics 11, 13061e13143. Lehtinen, K.E.J., Kulmala, M., 2003. A model for particle formation and growth in the atmosphere with molecular resolution in size. Atmospheric Chemistry and Physics 3, 251e257. Luo, G., Yu, F., 2011. Simulation of particle formation and number concentration over the Eastern United States with the WRF-Chem plus APM model. Atmospheric Chemistry and Physics 11, 11521e11533. Olofson, K.F.G., Andersson, P.U., Hallquist, M., Ljungstrom, E., Tang, L., Chen, D.L., Pettersson, J.B.C., 2009. Urban aerosol evolution and particle formation during wintertime temperature inversions. Atmospheric Environment 43, 340e346. Pierce, J.R., Leaitch, W.R., Liggio, J., Westervelt, D.M., Wainwright, C.D., Abbatt, J.P.D., Ahlm, L., Al-Basheer, W., Cziczo, D.J., Hayden, K.L., Lee, A.K.Y., Li, S.-M., Russell, L.M., Sjostedt, S.J., Strawbridge, K.B., Travis, M., Vlasenko, A., Wentzell, J.J.B., Wiebe, H.A., Wong, J.P.S., Macdonald, A.M., 2012. Nucleation and condensational growth to CCN sizes during a sustained pristine biogenic SOA event in a forested mountain valley. Atmospheric Chemistry and Physics 12, 3147e3163. Pryor, S.C., Barthelmie, R.J., Soerensen, L.L., McGrath, J.G., Hopke, P., Petaja, T., 2011. Spatial and vertical extent of nucleation events in the Midwestern USA: insights from the Nucleation In ForesTs (NIFTy) experiment. Atmospheric Chemistry and Physics 11, 1641e1657. Pryor, S.C., Spaulding, A.M., Barthelmie, R.J., 2010. New particle formation in the midwestern USA: event characteristics, meteorological context and vertical profiles. Atmospheric Environment 44, 4413e4425. Riipinen, I., Pierce, J.R., Yli-Juuti, T., Nieminen, T., Hakkinen, S., Ehn, M., Junninen, H., Lehtipalo, K., Petaja, T., Slowik, J., Chang, R., Shantz, N.C., Abbatt, J., Leaitch, W.R., Kerminen, V.M., Worsnop, D.R., Pandis, S.N., Donahue, N.M., Kulmala, M., 2011. Organic condensation: a vital link connecting aerosol formation to cloud condensation nuclei (CCN) concentrations. Atmospheric Chemistry and Physics 11, 3865e3878. Schoof, J.T., Pryor, S.C., 2008. On the proper order of Markov chain model for daily precipitation occurrence in the contiguous United States. Journal of Applied Meteorology and Climatology 47, 2477e2486. Sogacheva, L., Hamed, A., Facchini, M.C., Kulmala, M., Laaksonen, A., 2007. Relation of air mass history to nucleation events in Po Valley, Italy, using back trajectories analysis. Atmospheric Chemistry and Physics 7, 839e853. Spracklen, D.V., Bonn, B., Carslaw, K.S., 2008. Boreal forests, aerosols and the impacts on clouds and climate. Philosophical Transactions of the Royal Society A: Mathematical Physical and Engineering Sciences 366, 4613e4626. Spracklen, D.V., Carslaw, K.S., Kulmala, M., Kerminen, V.M., Mann, G.W., Sihto, S.L., 2006. The contribution of boundary layer nucleation events to total particle concentrations on regional and global scales. Atmospheric Chemistry and Physics 6, 5631e5648. Stanier, C., Khlystov, A., Pandis, S., 2004. Ambient aerosol size distributions and number concentrations measured during the Pittsburgh Air Quality Study (PAQS). Atmospheric Environment 38, 3275e3284. Tunved, P., Hansson, H.C., Kerminen, V.M., Strom, J., Dal Maso, M., Lihavainen, H., Viisanen, Y., Aalto, P.P., Komppula, M., Kulmala, M., 2006. High natural aerosol loading over boreal forests. Science 312, 261e263. Wang, W., Bruyère, C., Duda, M., Dudhia, J., Gill, D., Kavulich, M., Keene, K., Lin, H.C., Michalakes, J., Rizvi, S., Zhang, X., 2012. ARW e Version 3 Modeling System User’s Guide. National Center for Atmospheric Research e Mesoscale & Microscale Meteorology Division. Weber, R.J., McMurry, P.H., Mauldin, R.L., Tanner, D.J., Eisele, F.L., Clarke, A.D., Kapustin, V.N., 1999. New particle formation in the remote troposphere: a comparison of observations at various sites. Geophysical Research Letters 26, 307e310. Wehner, B., Siebert, H., Ansmann, A., Ditas, F., Seifert, P., Stratmann, F., Wiedensohler, A., Apituley, A., Shaw, R.A., Manninen, H.E., Kulmala, M., 2010. Observations of turbulence-induced new particle formation in the residual layer. Atmospheric Chemistry and Physics 10, 4319e4330. Wehner, B., Siebert, H., Stratmann, F., Tuch, T., Wiedensohler, A., Petäjä, T., Dal Maso, M., Kulmala, M., 2007. Horizontal homogeneity and vertical extent of new particle formation events. Tellus Series B-Chemical and Physical Meteorology 59, 362e371.