The concentration of atmospheric ions affects the total
aerosol particle number concentrations in the atmosphere as well as
atmospheric new particle formation via ion-induced nucleation, ion–ion
recombination, and effects on condensational growth. In this study, we
investigate the concentrations and long-term trends of atmospheric ions in a
boreal forest environment using 16 years of cluster ion (0.8–2 nm) and
intermediate ion (2–7 nm) measurements and characterize the most
important factors that explain those trends. We found that the median
concentration of cluster ions in a boreal forest was 710 cm
Atmospheric ions are produced via the ionization of air molecules (Rutherford, 1897), and primary ions are typically formed from nitrogen and oxygen due to their abundance in the atmosphere (Israël, 1970). The three main atmospheric ionization sources are radon decay, terrestrial gamma radiation, and cosmic rays (Eisenbud and Gesell, 1997; Mironova et al., 2015; Bazilevskaya et al., 2008). Anthropogenic sources of ionizing radiation come mostly from radiation leaks from nuclear power plants and nuclear weapon detonations (WHO, 1968). The charge can then be neutralized by ion–ion recombination, lost to a foreign surface, or transferred to other charged or neutral molecules, clusters, or aerosol particles (Seinfeld and Pandis, 1998; Tammet et al., 2006). Of the charged particles, cluster ions (0.8–2 nm) exist in the atmosphere all the time (Venzac et al., 2007, 2008; Mirme et al., 2010). Intermediate ions (2–7 nm) are typically formed by the propagation of charge onto initially neutral particles of the same size (Tammet et al., 2013) or through ion-mediated nucleation (Hirsikko et al., 2011). All ion diameters mentioned here and further in the text are mobility diameters. Ions of this size are typically generated by atmospheric new particle formation (NPF), snowfall, or rain (Manninen et al., 2010; Kerminen et al., 2018; Leino et al., 2016).
Atmospheric ions are the carriers of electric charge in the atmosphere. The ability for air to conduct electricity was first discovered by Richmann (1751) and Coulomb (1785), and Faraday (1834) proposed an explanation for this by suggesting that neutral molecules are ionized. Early results focused on understanding the charging of air molecules and aerosols in specific charge events, such as thunderstorms and electrification of clouds (Canton, 1753; Franklin, 1751). It was not until after the discovery of radioactivity by Wilhelm Röntgen, Henri Becquerel, and Marie and Pierre Curie that the ionization of neutral air ions in fair weather conditions was successfully explained (Carlson and De Angelis, 2011; De Angelis, 2014; Wilson, 1895, 1899). Air ions were then subsequently used to measure radioactivity as well as air quality (Misaki et al., 1972a, b, 1975; Tuomi, 1989; Israelsson and Knudsen, 1986; Retalis and Pitta, 1989).
The phenomenon of atmospheric ions was historically studied in the field of atmospheric electricity (Israël, 1970) because measuring atmospheric ions and their flow allows for the measurement of the conduction current in the atmosphere (Harrison and Carslaw, 2003; Wilson, 1921). However, in the past 30 years, the relevance of atmospheric ions and aerosols has been recognized, and the role of charged aerosols in both cloud processes and NPF has been debated (Dickinson, 1975; Arnold, 1980; Nadytko et al., 2003, Kulmala and Tammet, 2007; Enghoff and Svensmark, 2008). The current view is that ions can contribute to both the formation (Kirkby et al., 2011, 2016) and growth of small particles (Lehtipalo et al., 2016; Leppä et al., 2011; Stolzenburg et al., 2020; Svensmark et al., 2017), but the overall significance of ion-induced nucleation in the present-day atmosphere is rather limited, at least in the continental boundary layer (Manninen et al., 2010; Kulmala et al., 2013; Dunne et al., 2016).
Despite the increased interest in atmospheric ion concentration in the past decades, there has been relatively little research on how atmospheric ion concentrations have changed over time. Hirsikko et al. (2011) compared ion concentration measurements around the world in different environments. Their study included measurements from several different time periods, but no analysis of the long-term changes in concentrations was done. Other studies have investigated the diurnal and seasonal cycles of atmospheric ions based on a few years of data at best (Chen et al., 2016, Manninen et al., 2009) or have been focused on characterizing the charged aerosol size distribution (Hõrrak, 2001). Long-term trends of atmospheric ions and their characteristic properties have not been properly assessed.
In this study, we investigate the long-term trends of atmospheric cluster (0.8–2 nm) and intermediate (2–7 nm) ions in the boreal forest and attempt to characterize the main drivers behind these trends. We use 16 years of cluster and intermediate ion concentration data from the SMEAR II station in Hyytiälä, Finland, as well as ionization rates calculated from gamma and radon measurements and cosmic ray models.
The data used in this study are from the SMEAR II station (Hari and Kulmala,
2005) in Hyytiälä, Finland (61
The locations of the ion size distribution measurement at site A (BSMA) and the radiation measurements at site C (at site B prior to 2019) and their distances from one another (© Google Maps 2022).
Time periods when data are available for variables we used in our
analysis.
The Balanced Scanning Mobility Analyzer (BSMA; Tammet, 2006) is a
differential aspiration condenser with two parallel aspiration condensers
connected as a balanced capacitance bridge. Because this eliminates the
electrostatically induced current, the BSMA can continuously scan through a
mobility range of 0.023–3.2 cm
We have calculated the ionization rates from gamma radiation and radon decay
using the method prescribed by Chen et al. (2016) and used the modeled
cosmic ray ionization rate from
The main sink for cluster and intermediate ions is their coagulation with
preexisting particles. The concentration of atmospheric cluster ions can be
expressed as a balance equation:
For sub-10 nm particles, the condensation sink (CS) can be used as a good
estimate for the changes in the coagulation sink (CoagS) (Dal Maso et al.,
2002). It is not necessary to know the absolute value of the coagulation
sink because we are only attempting to gauge how the coagulation sink
changes over time. As Lehtinen et al. (2007) show, the coagulation sink and
CS are directly linked, and it can therefore be used to evaluate changes in
the coagulation sink. The CS was calculated in this study by integrating
over the aerosol size distribution as presented in Kulmala et al. (2001)
from a combined particle number size distribution between 3 nm and 20
Temperature and relative humidity (RH) data used in this study were acquired
from the smartSMEAR portal (
The variables used to explain the variance in cluster and intermediate ion concentrations. Coverage is defined as the number of data points available in the time period given.
Prior to trend analysis, time series need to be verified for consistency and breakpoints. Instrument upgrades, relocations, inlet changes, and such can cause breakpoints in which the time series has a discontinuity due to changes in instrument calibration, detection limit, activation probability, or sizing. Therefore, it is important to detect these breakpoints and, if possible, homogenize the time series before attempting trend analysis. The most common way to determine whether a time series has a breakpoint is visually (Collaud Coen et al., 2020), but care must be taken when looking for breakpoints in ion concentrations that include multichannel instruments and/or are interpolated to a specific size range from the initial inverted number size distribution. This is because the summing or interpolation can hide possible breakpoints. We investigated our measured ion number size distributions for the breakpoints in two ways, visually and by making sure the signal variance of the instruments remained steady over the measurement period.
The visual inspection of the BSMA time series revealed no notable breakpoints in the time series. This was further supported by the channel-by-channel signal of the time series, in which the variance for each channel remained steady over the entire measurement period (see Appendix, Fig. A1). The variance increased slightly over time due to the instrument getting older, but this was not a notable change. However, the variance of the signal increased significantly as RH increased towards 100 (Fig. A2), and we therefore excluded data points where the RH was over 80.
The sparsity of the in situ radiation measurements makes it difficult to detect breakpoints, but visual inspection revealed no breakpoints in the data. However, missing data periods in the radiation data can possibly hide breakpoints, and changes due to the change in the measurement location cannot be conclusively dismissed. The condensation sink and meteorological data time series did not exhibit any visually noticeable breakpoints.
Ions are traditionally split according to their size into small, intermediate, and large ions (Hirsikko et al., 2011). Cluster ions, also called small ions, are typically smaller than 2 nm, while intermediate ions are between 2 and 7 nm, and large ions are larger than that (Hirsikko et al., 2011; Hõrrak et al., 2000, 2003). We chose to use three size ranges in our analysis: one for the cluster ions (0.8–2 nm) and two for the intermediate ions (2–4 and 4–7 nm). We split the intermediate ions into two size ranges because earlier studies have shown that the 2–4 nm ion concentration is typically the most sensitive to NPF (Leino et al., 2016), while depending on growth rate, it can take several hours for the newly formed particles to reach 4 nm.
When investigating trends from atmospheric time series, it is important to account for autocorrelation which exists in most atmospheric variables. For time series that we could reasonably assume to be monotonic in trend, we used the 3PW algorithm described in Collaud Coen et al. (2020) to remove the autocorrelation before applying the Mann–Kendall (MK) test with seasonality correction (Hirsch et al., 1982) on the prewhitened time series. The 3PW method removes the autocorrelation typically present in atmospheric time series and then uses the MK test on the prewhitened data to determine the statistical significance of the trend. The trend-free prewhitening (TFPW) method data are used to test for significance due to their high test power. The yearly trend is then calculated using Sen's slope (Gilbert,1987) on the variance-corrected trend-free prewhitened data (VCTFPW) and presented as an annual change. The VCTFPW method produces an unbiased slope estimate, making it appropriate for this analysis (Wang et al., 2015).
We also calculated the dynamic regression for each time series, including those that could not be assumed to be monotonic, using a dynamic linear model (Laine, 2020). The model used assumes a trend and years as well as half-year seasonal variation and monthly medians, which were calculated from the data. The structural parameters of the model were calculated using Markov chain–Monte Carlo methods.
Finally, we tested the effect of various factors on the concentration of cluster and intermediate ions in the atmosphere using multiple linear regression from the MATLAB Statistics and Machine Learning toolbox.
In this section we present and analyze the time series of cluster and intermediate ion concentrations in the boreal forest.
The median concentrations and variability of ions measured with BSMA are presented in Table 2 and as violin plots in Fig. 3. Negative and positive ion concentrations are very similar to one another, with the exception that the 4–7 nm negative ion concentrations are more spread out, with higher concentrations being more common (Fig. 3). This might suggest that more negative particles are created by ion-induced nucleation than positive particles or that the negative ions grow faster to the 4–7 nm size bin before being neutralized or lost.
Our measured concentrations agree well with concentrations reported by
previous studies. Komppula et al. (2007) reported mean cluster ion
concentrations in Hyytiälä to be typically around 840 cm
Measured ion concentrations. Negative polarity is on the left in
blue, and positive polarity is on the right in orange. The cluster ion (0.8–2 nm) concentrations are in
Descriptive statistics of the ion measurements dataset between 2005 and 2021.
The long-term ion concentration trends are shown in Fig. 4 and Table 3. We
observe a slight positive trend of 1 % annual change for the cluster ion
concentration in both polarities, corresponding to a median increase of 7.1 cm
The relative change in ion concentration for BSMA for 2004–2020. The relative changes per year have been calculated from Sen's slopes. The values in brackets are the confidence intervals for the trend at a 90 % level.
The time series of the cluster and intermediate ion concentrations. The relative changes per year have been calculated from Sen's slopes. The values in brackets are the confidence intervals for the trend at a 90 % level.
Out of the three main ionization sources (radon, gamma, and cosmic ray ionization), radon and gamma ionization rates exhibit generally decreasing trends before 2015 (Fig. 5, Table 3), which would suggest that cluster ion concentrations should be decreasing during this period. However, our observations do not follow this. The main sink for cluster and intermediate ions, the sink from larger aerosol particles, which is described here by the CS (see Sect. 2.1.3.), also exhibits a decreasing trend. Therefore, the competing influence of the decreasing trends will be further analyzed in Sect. 3.2.3. to identify the most important factor to the ion concentrations in the boreal forest. Unfortunately, there is a large fraction of missing ionization rate data in the later part of the analyzed time period. Radon and cosmic ray ionization rates exhibit multi-year sinusoidal behavior, and we do not list a trend for them here. This is expected for the cosmic ray ionization rate, but the reason for the changes in radon ionization rate is unclear. Temperature and relative humidity both exhibit no significant trends (Fig. 5).
We investigated the annual variability in the ion concentrations by
calculating the median seasonal cycle of the ions in both polarities (Fig. 6) and ion sources and sinks as well as
CS has an annual maximum in the summer (Fig. 7a), which has been suggested to suppress NPF during the summer (Buenrostro Mazon et al., 2016; Nieminen et al., 2014). The annual minimum is in December, with a local minimum in March. This roughly follows the seasonal behavior of CS reported previously (Lyubovtseva et al. (2005); Vana et al., 2016). Temperature has a maximum in the summer (Fig. 7b) as expected, while RH has a minimum in the summer (Fig. 7c; Lyubovtseva et al., 2005; Hirsikko et al., 2005). The radon ionization rate has an annual maximum in February and a second local maximum in September (Fig. 7d). This has been connected to changes in the mixing layer depth in Chen et al. (2016). The gamma ionization on the other hand has a local minimum in February and March (Fig. 7e), which coincides with the typical maximum snow depth at the station, which suppresses gamma radiation from the soil (Chen et al., 2016). Finally, the cosmic radiation ionization rate remains practically constant (Fig. 7f), as its changes are known to occur in an 11-year period, and a median annual value should appear fairly constant.
The time series of main sources and sinks as well as temperature and relative humidity. The relative changes per year have been calculated from Sen's slopes. The values in brackets are the confidence intervals for the trend at a 90 % level.
We investigated the reasons behind the variations in ion concentrations by fitting a simple multivariate linear regression model. Our assumption was that cluster and intermediate ion concentration variation is caused by the interplay of their sources and sinks. Additionally, concentrations can be modulated by the prevailing meteorological conditions, which are connected to atmospheric air composition and ion processes. The degrees of freedom in the model were adjusted to reflect the number of explanatory variables included. The explanatory variables used are listed in Table 1.
We defined our linear model as
The Belsley collinearity test revealed that
The annual variance of ion concentrations in the size bins 0.8–2, 2–4, and 4–7 nm. The left-hand-side box plots are in the
negative polarity, and the right-hand-side box plots are in the positive polarity. The
numbers on the
The annual variation of CS,
The model assumes that the inputs are normally distributed and independent and
that the variance of the factors is roughly proportional. To account for
this, each factor was normalized by using the formula
The percentage of variability explained by each explanatory variable is presented in Fig. 8. Our model explains between 34 % and 52 % of the variance in the ion concentrations. About 25 % of the cluster ion concentration variability is explained by changes in CS, which corroborates the assumption in the previous chapter that the decrease in CS could be the main reason for the observed negative trend in ion concentrations. The second most important factor was temperature. Intermediate ion concentrations, both 2–4 and 4–7 nm, are best explained by a combination of sources and meteorological variability. However, the combination is different for negative and positive ions. The chemical composition of the negative and positive ions is known to be different (Luts et al., 2011; Ehn et al., 2011), and it is possible that humidity as well as chemical differences in condensable vapors can cause differences in the dynamics of the negative and positive ions. It is clear that we are not capturing all of the long-term variance with the model.
The amount of variation explained by our chosen explanatory
variables for each size range and polarity for BSMA. The symbol (
However, because of strong seasonal effects in ion processes, the variation of ion concentrations could be dominated by different factors in different seasons. For this reason, we also investigated how these factors explained the variation in ion concentrations in different seasons. In Figs. A3–A6 (See Appendix) we show the amount of variability explained by our selected variables when isolating the different seasons.
The different seasons each have some characteristics that are not directly visible in the variation of the entire time series. In spring, the variation in cluster ion concentrations is most strongly explained by CS, about 25%. Radon ionization rate explains a further 15 %–18 % of the variance in spring. For summer, radon ionization rate explains about 40% of the variation, while other factors in our model fail to explain much of the variation. In autumn, RH explains 13 %–40 % of the variation in cluster ion concentrations, while ion sources and sinks contribute less to the variation. Radon ionization rate, however, does contribute roughly 34 % to the negative cluster ion concentration variance. Our chosen variables fail to explain the changes in cluster ion concentrations in winter well. Gamma and radon ionization rates explain about 20 %–30 % of the variation cluster ion concentrations in winter. The fraction of variance explained in autumn and winter suggests that our chosen variables are not ideally suited to explaining the variance in those seasons. Other factors such as the abundance of relevant chemical vapors to grow the ions into the size ranges considered here may be more important during these seasons.
The intermediate ion variation in spring is in general explained best by the radon ionization rate and RH. Temperature has the least effect in spring and summer. Summer and autumn intermediate ion concentration variation is explained by meteorology. RH explains 7 %–47 % of the variance in intermediate ion concentration in the summer, while in the autumn RH explains 62 %–68 % of the variation, but we fail to capture a large portion of the variation. In winter, the variation in intermediate ion concentration is explained by a combination of radon and gamma ionization rates and temperature, likely linked to varying snow depth and its suppressive effect on the ionization rates (Chen et al., 2016). However, the wintertime values are not statistically significant.
We also analyzed the seasonal variation using median concentrations
calculated for each month. By calculating the median monthly concentrations
of our data, we have a median yearly cycle for each factor from which we can
investigate how much of the seasonal variation in ion concentration is
explained by our model (Fig. 9). The variance in cluster ion concentration
in the negative polarity is explained by a combination of CS (35 %) and
temperature (36 %). Meanwhile, in the positive polarity the variation is
best explained by CS (28 %) and
All in all, the effect of ionization rates on the trends and the variability of ion concentrations were surprisingly low. It could be that although the ionization rates control the production of primary ions, the measured concentrations, even in the smallest size bin studied here, depend largely on the factors affecting ion dynamics and growth to large enough sizes to be detected by our instruments, which in turn depend on multiple other factors such as air composition (Chen et al., 2016).
The amount of seasonal variation explained by our chosen
explanatory variables for each size range and polarity. The symbol (
In this work we have presented a 16-year-long time series of ion concentrations at SMEAR II station in southern Finland. Using this unique dataset, we investigated long-term trends of cluster and intermediate ion concentrations in the atmosphere at a boreal forest. A connection between the steadily increasing ambient cluster ion concentrations and decreasing CoagS, estimated in this analysis with CS, appears clear. The decreasing CoagS is likely due to an overall decreasing level of anthropogenic aerosols in boreal forest (Luoma et al., 2019), but possible changes in the air mass origins cannot be easily dismissed and were not investigated here. Although some trends were evident in the ambient ionization sources as well, they did not explain the overall trend of the ambient cluster ion concentrations.
The reasons for the rise of ambient intermediate ion concentration are more complex. This is to be expected, as the dynamics of 2–7 nm aerosols are much more complicated, not only dependent on formation and loss, but also ion–ion recombination, ion–aerosol attachment, and ion growth (Kulmala et al., 2013). And because growth is dependent on many factors, such as the availability of condensable vapors and favorable meteorological conditions, the dependencies will not be as simple (Kerminen et al., 2018). The connection to meteorological conditions is evident in our analysis as well, whereas the season-specific analysis of variation suggests that while our chosen variables appear to be the limiting factors during the summer and spring months, other factors may be more important to the intermediate ion concentration in autumn and especially winter. These factors include boundary layer dynamics and the abundance of condensable vapors (Hao et al., 2018, 2021; Kirkby et al., 2011; Almeida et al., 2013; Vakkari et al., 2015).
Our analysis has several sources of uncertainty worth noting. As mentioned previously, because the BSMA data with RH above 80 % were excluded, our analysis does not include any possible dynamics related to precipitation, which also is known to produce intermediate ions. Therefore, NPF events are likely overrepresented in the data. Additionally, we are missing 3 years of radon ionization rate data and 5 years of gamma ionization data. This makes comparing trends over the entire time period difficult and introduces uncertainty in our interpretation of the results. Also, for detailed studies on the ion balance and transfer of charge from primary ions to cluster ions, the measurements of primary ion size distribution below 0.8 nm would be needed (Chen et al., 2016). Increase in ion concentrations can mostly be attributed to a decreasing CS. The changes in the ionization rates explain the observed trend to a lesser degree. It is possible that this is simply because the ion sink is the dominant limiting factor to the formation of cluster and intermediate ions. However, several open questions remain. Although studies indicate that the neutral pathway dominates NPF in the boreal forest, the role of ions is not totally clear (Wagner et al., 2017). Ion-induced nucleation can be significant in certain conditions, e.g., when the sulfuric acid concentration is low (Rose et al., 2018; Yan et al., 2018; Gagné et al., 2010). Additionally, ion–ion recombination can create neutral aerosols. It is possible that ions are especially important in the initial stages of NPF, which is difficult to measure with current instrumentation. It will have to be determined in the future whether the number of aerosols produced by ion–ion recombination (Kontkanen et al., 2013; Franchin et al., 2015) is affected by the changes in ion concentrations. It can be speculated that the increasing ion concentrations, decreasing concentrations of sulfuric acid in boreal forest (Nieminen et al., 2014), and increasing temperatures will make the ion-induced or ion-mediated NPF pathways more important in future. A more comprehensive analysis of this is needed.
Moving variance of the concentration time series of the BSMA size channels. The symbol n indicates the number of closest data points used for the calculation of the variance at each point. The time resolution is the time resolution of the instrument, which is 12 min.
The standard deviation of ion concentrations measured by BSMA in different RH conditions.
The coefficients of the multiple linear regression model used in the study.
The amount of variation explained by our chosen explanatory variables
for each size range and polarity. The data points selected are only between
March and May and represent changes in springtime concentrations. The symbol
(
The amount of variation explained by our chosen explanatory variables
for each size range and polarity. The data points selected are only between
June and August and represent changes in summertime concentrations. The
symbol (
The amount of variation explained by our chosen explanatory variables
for each size range and polarity. The data points selected are only between
September and November and represent changes in autumn concentrations.
The symbol (
The amount of variation explained by our chosen explanatory variables
for each size range and polarity. The data points selected are only between
December and February and represent changes in wintertime concentrations.
The symbol (
The meteorological data from the SMEAR II station can be accessed from the
smartSMEAR website:
JS performed the data analysis and wrote the paper. JL was responsible for the ion measurements, JL and JS handled the ion data inversion, XC calculated the radon and gamma ionization rates, and TN calculated the cosmic ray ionization rate. JK, VMK, and KL designed the study and provided supervision. All authors contributed to the discussion of the results and provided comments on the paper.
At least one of the (co-)authors is a member of the editorial board of
Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Hanna E. Manninen, Stephany Buenrostro Mazon, Jussi Paatero, the SMEAR II technical and scientific staff, and everyone else contributing to the ion and radiation measurements are gratefully acknowledged.
This research has been supported by the Academy of Finland (grant nos. 337549, 302958, 1325656, 316114, and 325647) and the European Research Council, H2020 European Research Council (grant no. ATM-GTP (742206)).Open-access funding was provided by the Helsinki University Library.
This paper was edited by Andreas Petzold and reviewed by Santtu Mikkonen and one anonymous referee.