Short Sampling Periods: A New Setup to Evaluate the Change in the Chemical Composition of Fog at High Time Resolution

Site description

A fog collector system was set up at two different mountain study sites in Taiwan, namely the Lulin Atmospheric Background Station (Lulin; 23°28’06”N, 120°52’25”E; 2862 m asl) and the Xitou Flux Tower in the Experimental Forest of National Taiwan University (Xitou; 23°39’52”N, 120°47’43”E, 1150 m asl) (see Figure 1).

Figure 1

Fog collection

The instrumentation includes a present weather detector that controls an active fog collector, a sample volume gauge, a magnetic valve, and an auto sampler (see Figure 2). The present weather detector (model PWD11, Vaisala OY, Finland) measured the visibility. Whenever the visibility dropped below 1000 m on a 10 minute mean basis, the fog collector was activated and the fog sampler started to collect water. The collection unit of the fog sampler is of a modified Caltech design (Degefie et al., 2015). It is completely made of inert materials: The unit itself is made of polyether ether ketone (PEEK), and the collection strings are made of polytetrafluoroethylene, better known as Teflon (PTFE), which is used to prevent any metal or ion contamination from the construction materials. All tubes and other surfaces that come into contact with fog water are also made of Teflon. The water level in the volume gauge is detected with a photoelectric sensor: the light transmission in air and water is different and therefore indicates whenever the sampling volume of 10 ml is reached. A magnetic valve opens and the sample is released into a standard laboratory auto-sampler (Model PS62, MLE GmbH Dresden, Germany), which is equipped with up to 109 centrifuge tubes (each 15 ml, HDPE). The system is operated with an Arduino microcontroller.

Figure 2

The setting is unique in that it is the first fog collection unit that collects subsequent samples based on a fixed volume of fog water. We decided on a volume of 10 ml per sample, as this volume was sufficient for analyses and allowed us to collect fog water on a rather short sampling period, sometimes less than 1 minute, depending on the liquid water content (LWC) of the fog. During the field campaigns, all samples were filtered as soon as possible after collection with a 0.2 µm pore size string Teflon filter. Electric conductivity and pH were measured in sample aliquots of 3 ml. This subsample volume was then thrown away and not used for further analysis. The other 7 ml were stored in a freezer (15 ml, HDPE) until further laboratory analysis with ion chromatography.

A strict cleaning routine was performed during the field campaigns. The fog collector, including the magnetic valve, all tubes and the auto-sampler, were cleaned with deionized water at least every other day and after each fog event. We expected that neighbouring samples would be more similar in conductivity and pH values than samples further away from each other, as shown by other studies (Simon et al., 2016; Nieberding et al., 2018). In addition, the compositions of neighbouring samples are likely to be more similar to each other than to those of other samples or to that of de-ionized water. Accepting the very low risk of cross-contamination or cross-dilution, a filter was used for several subsequent samples. The electrodes (conductivity and pH) were not cleaned between the samples but were fully drained. Following the reasoning described above, any cleaning with pure water could introduce a dilution effect, which we considered to be more serious than the risk of the carry-over effect. Therefore, we decided to clean the equipment between samples of different events, but not between samples of the same event.

The first sample normally has a longer sampling period than following samples, because all materials need to be wetted before the water drops down into the sampling bottles. Thus, the LWC is slightly underestimated in the first samples. Whenever a sample is taken, the amount of water leftover inside the collector and tubes is around 10%, which creates a carry-over effect between neighbouring samples. Note that the carry-over effect should not reach farther than one subsequent sample. After termination of a fog event, all laboratory material was cleaned with deionized water, dried and put into storage.

We took blank samples of the deionized water for quality assessment and quality control (QA/QC). For laboratory blanks, deionized water was filtered with previously used filters and with new filters. Field blanks were taken after the fog collector was cleaned, blanks were collected by spraying deionized water into the fog collector and collecting it in the same way that real fog water samples were collected. All blanks showed low conductivity, and the pH values also showed no indication of contamination.

Chemical analysis

The ion chromatographic analysis was conducted at the Department of Atmospheric Science of the National Taiwan University after each of the three IOPs. The LWC was calculated for each sample with the fog water collection rate (FWCR, in unit mg min^–1) divided by the product of fog collection efficiency (88%) and flow rate (17.23 m³ min^–1) (Simon et al., 2016). The ion loads of ions i per volume of air (IL(i), in units μeq m^–3) were calculated from the measured ion concentrations (IC(i)) in units μeq L^–1) and LWC using:

where (i) represents the individual ion species and ρ is the density of water (1000 kg m^–3).

Frequency analysis

A frequency analysis was applied to the meteorological data in order to quantify the spectral ranges of the predominant turbulent motion. A Fourier-transform decomposed the time series into a large number of sine and cosine functions. The frequencies and amplitudes of these functions describe the recurring changes of the meteorological parameters with time. This yields a frequency domain of meteorological parameter changes under study and determines the time domain within which the spectral density of turbulent motion was highest. In micrometeorological applications, this approach is used – among others – for data quality assurance (Burba, 2013). Here, we will compare the most relevant frequency domain with the time resolution of fogwater sampling. In order to capture the full spectrum of changes of LWC and ion concentrations, the sample collection should be made at least on the same frequency as the frequency of highest spectral density of turbulent motion.

Backward trajectories

We calculated air mass trajectories with archive meteorological data in order to develop an understanding about the origin of the air masses and the possible sources of pollutants. Backward trajectories (48 hours) were calculated with the Hybrid Single-Particle Lagrangian Integrated Trajectory model (Draxler and Hess, 1998; Stein et al., 2015; Rolph, Stein and Stunder, 2017) provided by the Air Resources Laboratory of the National Oceanic and Atmospheric Administration (NOOA) at https://www.ready.noaa.gov/HYSPLIT_traj.php. We used the Global Data Assimilation System (GDAS) with a 0.5 degree resolution for Lulin 2018 experiment. For the Xitou 2019 and Lulin 2020 experiments, we used the Global Forecast System (GFS) with a resolution of 0.25 degrees. This improved resolution was not available for the Lulin experiment in 2018.

Fog frequency and sampling period

An overview of all IOPs is given in Table 1. During the summer campaigns (Lulin 2018 and Xitou 2019) we had large data losses due to heavy rain that caused power outages and precautionary shutdowns. We presume that the losses of fog-related data are minor. The fog frequency was, in general, somewhat higher at Lulin than at Xitou. The percentage of fog days (minimum 1 hour of fog per day) was 80% – 90% at Lulin and 70% at Xitou, which is within the range of earlier studies at the same sites (Simon et al., 2016; Breuer et al., 2021). The main technical goal of all three IOPs was to collect as many back-to-back samples as possible within single fog events. Therefore, we aimed for long events with a minimum of 8 subsequent samples (Table 1). These events lasted between 1.5 and 13 hours (median 4 hours).

During the 24 long events (Table 1), we sampled 837 samples at a median collection period of 3 minutes, 45 seconds (Figure 3). Compared to other studies (Simon et al., 2016; Nieberding et al., 2018), this is an order of magnitude shorter sampling period than previously achieved sampling frequencies. Overall, 90% of the samples were collected in under 12 minutes, which is less than half of the collection period than that achieved with the traditional sampling routine.

Figure 3

Different event types

We describe four different events of our IOPs that represent four different meteorological conditions.

Advective fog event during the wet season

The wet season at Mount Lulin is dominated by frequent rain and fog, sometimes in combination with each other. Consequently, the atmosphere was rather clean and the electric conductivity of fog samples (median 18.5 µS cm^–1) was low compared to that of earlier studies at the same study site (median 103 µS cm^–1, Simon et al., 2016). Nevertheless, the first event started at a relatively high conductivity of 99.5 µS cm^–1 and a low pH of 3.85 at 01:00 LT. Dense fog formed and the visibility remained low for the next two hours (Figure 4a). As early as during the first hour of the event, the conductivity decreased and the pH increased (Figure 4b). For the next hour (02:00 LT to 03:00 LT), there was little change in conductivity, and the pH remained relatively constant at about 4.5. The LWC increased and decreased later, with a well-pronounced peak just after 02:00 LT (local time, which is UTC + 8 hours) (Figure 4c). The concentrations of inorganic ions dropped synchronously with the conductivity (Figure 4b & 4c) and also increased towards the end of the event. The chemical composition was clearly dominated by H⁺ (about 25%), NH₄⁺ (about 15%), SO₄^2– (about 25%) and NO₃^– (about 10%), which account for around 75% of the total ion concentration during the whole event. Former studies at the study site indicated that these four ions make up more than 85% of the total ion composition (Simon et al., 2016).

Figure 4

The median sampling period for this event was 3 minutes and 8 seconds; the shortest sampling period was only 1 minute (Figure 3). According to spectral analysis (Fourier transform), the highest turbulence intensity was associated with eddies lasting between 5 seconds and 2 minutes. Therefore, our sampling method was not quite able to fully capture the turbulent changes. Related to the wind speed (median 1.45 m s^–1), a sample within the median sample period could cover a cloud cross section of 270 m, the shortest sampling period showed a spatial resolution of around 90 m.

During this specific fog event, we presumably observed the passage of an advected cloud. At the edges of the cloud, which correspond to the beginning and the termination of the fog event at the fixed sampling site, we observed low LWC and thus higher concentrations of ions, high conductivity and low pH. During the centre of the event, which corresponds to the centre of the moving cloud, the foggy air mass was rather homogenous. The phase-out period of the event was less pronounced than the initial phase, which showed rapid changes of all measured chemical parameters.

Advective fog during the dry season

An advective fog during the dry season, showed quite different dynamics. Although it occurred during the dry season, it started with heavy rain and low visibility (Figure 5a). We had refrained from sampling fog during rain, because during such conditions rain enters the fog collector, hindering a clear discrimination between rain and fog droplets. Therefore, the fog sampling started in the middle of a cloud. At the sampling site, the cloudy (foggy) conditions lasted another 6 hours after termination of the rain. The conductivity was rather low in the beginning of the sampling period, well below 100 µS cm^–1 (Figure 5b). A large portion of the aerosol particles carrying the inorganic ions had been washed out by the rain before the onset of the sampling, so only a low level of air pollution remained. The conductivity increased during the event, and later on the fog was less homogeneous (Figure 5a), and the conductivity was more variable during this period (15:00 LT to 18:00 LT). However, the pH was lower (below 4 throughout) than in the event described above and it decreased during the event (Figure 5b). The total ion concentration followed the shape of the conductivity curve (Figure 5c). The dominating ions again were H⁺ (about 20%), NH₄⁺ (about 30%), SO₄^2– (about 20%) and NO₃^– (about 15%), summing up to about 85% of the total ion equivalent concentration, which matches well the results of former studies during the dry season (Simon et al., 2016). The LWC was above 200 mg m^–3 in the beginning and decreased more or less continuously with time down to around 15 mg m^–3. The turbulent energy for this event peaked at somewhat longer timescales (10 seconds to 5 minutes). The median sampling period was 5 minutes and 38 seconds (Figure 3), and the shortest sampling period was taken in 2 minutes and 32 seconds. As such, this median sampling period was again close to catching the turbulent changes. The wind speed during this event was quite high, with a median of 4.36 m s^–1, so that the median sampling period corresponded to a cross section of about 1.5 km within the cloud (around 600 m for the shortest sampling period). Accordingly, these samples still cover a wide range of changing meteorological conditions. We assume that we probed an advective cloud in this case, with edge effects evident at the end of the fog event.

Figure 5

Long event during wet season

The third event of this study site represents a very long fog event during the wet season. Heavy rain occurred before the event, then very dense fog with a visibility below 200 m was measured for around 8 hours onwards, followed by intermittent fog for around 4 hours (Figure 6a). A total of 151 samples were taken during this event. The conductivity (Figure 6b) was very low throughout the event (<25 µS cm^–1); slightly higher conductivity only occurred in the middle of the dense fog (13:00 LT to 16:00 LT, >15 µS cm^–1), and from 17:00 LT on it was very low (<10 µS cm^–1). The pH was more variable through the event (Figure 6b). The first sample was rather high with a pH of 5.29, whereas the next sample (pH 4.56) and the following one were more acidic. One higher pH value (5.1) was measured around 12:20 LT. All other samples until 18:00 LT were below 5. Later on, the pH increased and fluctuated between pH 5.0 and 5.5, except for three samples that jumped up by at least 0.45 units compared to the neighbouring samples (19:40 LT, 19:46 LT and 22:40 LT). Compared with former studies during the same season at the site (Simon et al., 2016), in our samples the pH was high and the conductivity very low. The LWC (Figure 6c) can be categorized into three stages: 12:00 LT to 17:45 LT (median LWC 147 mg m^–3), 17:45 LT to 19:53 LT (median LWC 357 mg m^–3), and 19:53 LT to 23:11 LT (median LWC 83 mg m^–3).

Figure 6

The turbulent energy peak during this event was similar to that of the last event (10 seconds to 5 minutes). The median sampling period was around 3 minutes and 30 seconds, with the shortest sampling period being 30 seconds. Along with the median wind speed (0.78 m s^–1), the air mass moved about 160 m during the median sampling period, 23 m for the shortest sampling period. Our sampling routine was, thus, within the same temporal domain as the turbulent advective processes of the fog. Note that this event also represents orographic fog. The sampling started in the middle of the foggy (cloudy) air mass with very low visibility. Around 20:00 LT, the fog was less homogeneous, and towards the end of the fog event, the edge effects became evident. The ion composition was not measured for this event, as the number of samples would have exceeded the laboratory capacity.

Shorter event during wet season

The shortest event discussed in this paper occurred at the study site Xitou during the wet season. It lasted only 1.5 hours but still included 22 samples. In the upper panel of Figure 7, the data of the automatic sampling routine is shown (Figure 7a & 7b). During this event we had some data storage issues, so, the visibility data is not shown here. This location, at a lower elevation than the Lulin site, is more affected by regional air pollution (Liang et al., 2009; Simon et al., 2016). Therefore, the conductivity and ion concentrations were higher compared with the other events. The conductivity started at 649 µS cm^–1 during the onset of fog and decreased afterwards (Figure 7a). After the 10^th sample (69 minutes), the conductivity slightly increased. The pH was relatively high (4.89) and even increased during the event up to pH 6.21. The ion concentrations followed the shape of the conductivity curve (Figure 7b). The chemical composition was somewhat different compared with that of the other study site: Here H⁺ only played a minor role. The dominating ions were NH₄⁺ (about 45%), SO₄^2– (about 20%) and NO₃^– (about 20%), summarizing to 85% contribution to the total ion composition. In former studies, H⁺ ions also had a less dominant effect on the total ion composition for this study site (Simon et al., 2016), yet our values are really low. The turbulent energy peak during this event was in the same range as the other events (between 5 seconds and 5 minutes). The median sampling period for this event was around 4 minutes; thus, this resolution only covers a small portion of the turbulence elements. The median wind speed during this event was around 0.6 m s^–1, such that the air mass moved 150 m during the time a sample was taken. This sampling period was not short enough to show all changes of meteorological processes. Other studies estimated that the droplet growth time was around 23 seconds (from diameter 0.25 µm to 17.33 µm) for the study site (Baumberger et al., 2021); therefore, the sampling period was rather long compared to the timing of other relevant processes.

Figure 7

Comparison of short sampling period and traditional sampling routines

So far, we have shown that the chemical composition of fog water changes quickly during various fog events, and we were able to determine this by using our newly developed sampling routine, which takes a sample whenever 10 ml of fog water has been collected. We observed rapid changes, especially during the very early and the very late stages of fog events. In this section, we show the difference between the traditional sampling procedure and the new routine.

The traditional sampling procedure is based on fixed time intervals: At the fixed time intervals (typically 30 min), an operator manually checks, whether the required water volume (typically 30 ml, due to historical laboratory capacities) is reached. When the required sampling volume is reached, the sampling bottle is changed; otherwise, the sampling continues for another 30 minutes (Sträter, Westbeld and Klemm, 2010; Degefie et al., 2015; Simon et al., 2016; Nieberding et al., 2018). For the comparison presented here, we merged our short sampling period to simulate a traditional sampling schedule with 30 minute intervals. We computed volume weighted averages of the ion concentrations of samples within 30-min sampling intervals. The respective pH’s were calculated from a volume weighted means of the H⁺ concentrations. This routine was applied for the event on 28 July 2019 at Xitou.

The upper panels of the Figure 7 was described before: It shows the data of the new automatic sampling routine (Figure 7a & 7b), while the (computationally averaged) results of the traditional sampling procedure are displayed in the lower panels (Figure 7c & 7d). For this analysis, the volume-weighted average concentrations were calculated while the pH was calculated from a volume-weighted mean of the H⁺ concentration. The original 22 samples were averaged into only 3 samples (Figure 7c and 7d).

We still see the decrease of the conductivity with time and an increase of the pH (Figure 7c), but the steepness of both the decrease of the conductivity and the increase of the pH are lost. When viewed using the traditional sampling routine, the increase of the conductivity towards the end of the event (Figure 7a) is no longer apparent (Figure 7d). Further, the LWC seems homogeneous during the fog event (Figure 7d), and the decrease of the LWC towards the end of the event, which is shown in the original samples (upper panels), is no longer visible either. Also, the ion concentrations are flattened to a more homogenous level, in which only the first averaged sample is somewhat higher than the subsequent ones (Figure 7d).

To summarise, the traditional sampling procedure shows less variability, fewer changes and fewer peaks, in which all parameters became flatted and more homogenous and information was lost. The new sampling method shows more rapid changes, and the temporal resolution is sufficient to capture many of the quick changes that accompany meteorological conditions.

Trajectories and sources

The chemical composition of fog water is related to the origins of air masses and pollutants along their trajectories. Therefore, backward trajectories were calculated for all presented events (Figure 8). The trajectories for all samples of the event 2 of Lulin during the wet season 2018 (Figure 8a), are similar to each other, and the transport distance was around 1000 km within the preceding 48 hours (around 5 m s^–1). The air masses only moved above the ocean, which led to considerable concentrations of Na⁺ and Cl^– (Figure 4). For the following event of Lulin 2018 (Event 7, Figure 8b), the air movement was much slower (400 km in 48 hours, about 2.3 m s^–1). This event was also dominated by sea salt (Figure 6). The event of Xitou 2019 (Figure 8c) shows trajectories arriving from an eastern direction. The air masses moved first above the ocean and afterwards above Taitung city (south east of Taiwan) and took up air pollutants, which led to higher ion concentrations and conductivity than at the other events (Figure 7). The air masses passed over the high mountains located east of Xitou: They underwent adiabatic lifting, and it is likely that some components had been washed out by rain before the air masses arrived at Xitou. The travel distance over the previous 48 hours was similar to that of the event presented before (around 400 km, about 2.3 m s^–1). The last presented event of Lulin 2020 was the only event affected by air masses moving above mainland China, Taipei and the industrialized western lowlands of Taiwan (Figure 8d). The conductivity (Figure 5b) was much higher than during the other events at the same study site. The samples also might have been affected by seasonal biomass burning (Lin et al., 2014).

Figure 8