The study sites are located in Beijing Olympic Forest Park (40.02∘ N, 116.38∘ E, 51 m above mean sea level, AMSL) and Badaling Tree Farm (40.37∘ N, 115.94∘ E, 535 m a.m.s.l.), Beijing, China. Beijing's air quality is generally poor, with high concentrations of suspended particulate in the atmosphere (Fig. 1) and hazy conditions, at times with visibility < 10 km. Haze is a common problem, particularly during the winter and spring, stemming from a variety of contributing factors including home heating, traffic congestion, industrial activity, stable synoptic conditions and surrounding mountainous topography (e.g., Yang et al., 2015; Zhang et al., 2016; Zheng et al., 2015).

The Olympic Forest Park is the largest urban forest park in Asia, with an area of 680 ha and vegetation coverage of about 90 %. The site is an ecological conservation and restoration area. The site is dominated by Pinus tabulaeformis L. Other species include Platycladus orientalis, Sophora japonica L., Fraxinus chinensis and Ginkgo biloba, with an understory of Iris tectorum and Dianthus chinensis. All trees were tagged and identified by species and trees with a diameter at breast height (DBH) > 3 cm are being assessed annually. Stand density was 210 trees ha-1, with a mean tree height of 7.7 m and a mean DBH of 20 cm. Cover ratio of trees to shrubs was about 7:3. The shrubs were Prunus davidiana, Amygdalus triloba, Swida alba and Syzygium aromaticum, with a mean height of 2.8 m (Xie et al., 2015).

The Badaling Tree Farm is about 60 km from the downtown core of Beijing. Local terrain is generally flat and uniform. The study site is composed of Acer truncatum, Koelreuteria paniculata, Fraxinus bungeana, Ailanthus altissima and Pinus tabuliformis. Stand density was 975 trees ha-1, with a mean tree height of 4 m and a mean DBH of 4.7 cm. The study site has a sparse herbaceous cover with no well-defined understory canopy (Jia et al., 2013).

Data presented in this study were acquired from field deployment of two EC systems; one at each of the two sites described above. Both systems were deployed in 2011 with the original inline filter intakes. Data with the inline filter design were collected from January 2011 to July 2014 at the Olympic Park and from January 2011 to September 2014 at Badaling Farm. Both EC systems were switched to the vortex intake in 2014. Data with vortex intakes were acquired from July 2014 to December 2015 at Olympic Park and from September 2014 to December 2015 at the Badaling Farm.

An EC155 (model EC155, Campbell Scientific, Inc. Logan, UT, USA) is an in situ, closed-path, mid-infrared absorption gas analyzer (IRGA) that measures molar mixing ratios of CO2 and H2O at high frequency. The original EC155 analyzer includes a heated intake tube, inline filter and rain cap (Fig. 2a). The modified EC155 system includes a prototype vortex chamber and rain cap in place of the original filter and rain cap (Fig. 2b).

The vortex intake is a small, lightweight device with no moving parts, and requires no chemicals to clean the sample air. Its simple design makes it essentially maintenance free. The vortex assembly (Fig. 2b) consists of a rain cap and inlet nozzle, a vortex chamber and two outlet ports. Schematics of both systems are shown in Fig. 3. Unlike a filter, the vortex intake design is based on fluid and particle dynamics. Sampled air enters the vortex chamber through a tangent port to induce rotational flow. Entrained dust-particle motion is governed by centrifugal force (inertia), aerodynamic drag and chamber wall impact forces. The high rotational speed of the vortex flow provides the larger and heavier (relative to air) dust particles with greater centrifugal force, keeping them close to the chamber wall and leaving the air in the center of the vortex free of dust. Clean sample air flows from the vortex center through a tube to the EC155 sample cell. The dust particles in the air close to the wall of the vortex chamber are pulled out through the bypass tube. The dirty air passes through a relatively large (12 mL internal volume) filter (9922-05-DQ, Parker Hannifin Corp., Mayfield Heights OH, USA) with moderate porosity (25 µm) that requires only infrequent replacement due to clogging. The filter protects a flow-control orifice that balances the airflow split such that 6 is sampled and 2 L min-1 is bypassed. The two flows rejoin downstream of the analyzer, then go to the single, low-power (5 W) vacuum pump. In contrast, the EC55 with the inline filter has slightly higher flow through the analyzer (7 L min-1). Burgon et al. (2016) measured the frequency response of the EC155 with the inline filter and with the production vortex design using the impulse response method (Sargent, 2012), showing that the cutoff frequency is proportional to flow, as expected: 5.1 Hz for the original design at 7 L min-1 and 4.3 Hz for the vortex design at 6 L min-1.

For flux determination (covariances), high-frequency wind velocity measurements are needed. Wind velocities are acquired with a fast-response, three-dimensional sonic anemometer (CSAT3A; Campbell Scientific, Inc. Logan, UT, USA). Spatial separation between the EC155 intake and the CSAT3A sample volume is 15.6 cm.

At Olympic Forest Park, a 12 m-tall tower is surrounded by uniform forest cover with a homogeneous fetch of about 600 m in all directions. The EC instruments were mounted on the tower at a height of 11.5 m from the ground. The EC instrumentation was installed similarly at the Badaling Tree Farm. All flux-related data were collected at 10 Hz using a CR3000 data logger (Campbell Scientific, Inc. Logan, UT, USA). Half-hourly turbulent fluxes were calculated from the covariance between the fluctuations in the vertical wind speed and the scalar quantities (Aubinet et al., 2000). Sonic temperature was corrected for changes in atmospheric humidity and pressure (Schotanus et al., 1983).

Attenuation of high-frequency fluctuations is one of the systematic errors in EC measurements, and it is especially important to address it when trace-gas concentrations are measured with a closed-path analyzer (Aubinet et al., 2000). The frequency response of a system is defined as a ratio of its output to its input as a function of the signal frequency. In an ideal system, the value would be one at all frequencies. In a real system, fluctuations in CO2 or H2O tend to be damped at higher frequencies due to adding a rain cap, filter and intake tubing to a gas sampling system (Aubinet et al., 2016). Measuring the frequency response quantifies the loss of high-frequency information so that a correction may be applied.

With this in mind, the frequency response of a closed-path EC system with vortex intake is compared to one with an inline filter. In situ field measurement of system frequency response is challenging because the true signal inputs (scalar variables) are not known a priori. However, it can be evaluated by comparing the cospectra of the vertical component of wind with fluctuations of sonic temperature (WTs, where W denotes the vertical component of wind and Ts the sonic temperature) to the cospectra of the vertical component of wind with fluctuations of CO2 (WC, where C denotes the mixing ratio of CO2) and H2O (WH, where H denotes H2O mixing ratio). Analysis of cospectra are based on data from the Olympic Forest Park over 1 h periods (12:00–13:00 and 21:00–22:00 Beijing Standard Time) averaged daily for the months of January (high incidence of haze) and June (low incidence of haze) 2014 for the inline filter-based EC measurements, and January (high incidence of haze) and June (low incidence of haze) 2015 for the vortex-intake-based measurements. A fast Fourier transform was applied for each variable's time series, consisting of about 36 000 data points.

Decreases in sample-cell differential pressure and CO2 optical signal strength indicate when the inline filter needs to be replaced and the optical windows cleaned. Clogged filters can induce substantial pressure drops (Aubinet et al., 2016). Generally, the pressure drop in the original intake assembly is approximately 2.5 kPa at 7 LPM flow without filter. The filter adds approximately 1 kPa pressure drop when it is clean. This pressure drop will increase as the filter clogs. The filter should be replaced before the differential pressure reaches ±7 kPa. Additionally, the windows of the analyzer should be cleaned when the optical signal strength of CO2 drops below 80 % of the original value.

Due to damping of high-frequency signals in closed-path systems, gas cospectra commonly exhibit reduced response at high frequencies, causing flux loss (Burba et al., 2010; Leuning and King, 1992). Brach and Lee (Brach et al., 1981; Lee et al., 2004) found that the cospectrum of vertical wind velocity with sonic temperature (WTs) is often very close to the ideal cospectrum. In field experiments, WTs is often used as a standard to evaluate whether there is a high-frequency loss for other measured scalars. To examine the effect of a vortex intake, spectral analysis was applied to the measurements collected in situ. Ensemble cospectra of vertical wind velocity with CO2 (WC) and H2O vapor (WH) were compared to those for the sonic temperature (WTs) for both the inline filter and vortex intake in different periods (Fig. 4). The normalized cospectra for both systems were consistent at all frequencies, with no significant difference (P > 0.05); thus the frequency response of the EC155 sampling system with either the inline filter or a vortex intake could not be distinguished from that of the sonic anemometer. However, the frequency response of a sonic anemometer depends on horizontal wind speed, and the results in Fig. 4 are based on data with relatively low wind speed (0.3 to 1.5 m s-1). The corresponding cutoff frequencies for the sonic anemometer are 2 to 10 Hz (Massman, 2000), which bracket the EC155 cutoff frequencies measured in the laboratory (5.1 and 4.3 Hz for the inline filter and vortex intake, respectively, Burgon et al., 2016).

Closed-path IRGAs with long inlet tubes often show additional high-frequency damping for H2O due to sorption effects (Goulden et al., 1996; Laubach and Teichmann, 1996; Leuning and Judd, 1996), which is not apparent in Fig. 4. The EC155 has a very short intake tube (0.6 m), which has been shown to reduce this effect (Burba et al., 2010; Clement et al., 2009). Ibrom et al. (2007) showed that this sorption effect increased with higher relative humidity, but Runkle et al. (2012) showed a weaker dependence on relative humidity at higher temperatures. Figure 4 shows results with relative humidity of ≤ 50 % for all cases except summer evenings (Fig. 4d and h), which had higher relative humidity (65 %), but also warm temperatures (27 ∘C). The water sorption effect is expected to be relatively small for these dry and warm conditions.

Air quality at Olympic Park was mostly poor during the measurement periods, being worse in winter than in summer. To verify performance of the vortex intake, we chose periods of high and low incidence of haze to compare the differential pressure and optical signal strength of CO2. Figure 5 shows time series of 3 months for each case. The differential pressure decreased as the inline filter clogged (Fig. 5a and c), quickly exceeding the range of the pressure sensor (±7 kPa), invalidating data until the filter could be replaced. The differential pressure with the vortex intake was much more stable than with the inline filter. Over a period of 3 months, the differential pressure with vortex intake exceeded the pressure sensor range once during very hazy conditions (Fig. 5b). The optical signal strength of CO2 with the vortex intake remained above 90 %, indicating that the optical windows remained free of debris for a substantially longer time than with the inline filter.

At the Badaling Farm site, we also chose periods of high and low incidence of haze to compare differential pressure and optical signal strength of CO2. As shown by the differential pressure in Fig. 6a and c, the inline filter clogged multiple times in a period of 2 months resulting in large data losses. Pressure drop with the vortex intake (Fig. 6b and d) was typically about 3 kPa, remaining within the working range for the entire observation period (2 months). The optical signal strength of CO2 with vortex intake was higher and more stable than that of the system with an inline filter.

In order to further verify the performance of the EC155 system with vortex intake, field maintenance records from the two sites were compared. These maintenance records included the number of maintenance services; downtime due to clogged intake filters; and percentage downtime, defined as a ratio of downtime due to clogged intake to the actual testing-period duration × 100. Also included are the minimum and average time periods for the intake filter to clog.

A summary of field maintenance at Olympic Park and Badaling Farm is shown in Table 1. The vortex design reduced the number of maintenance services from 15 to 4 at Olympic Park and from 7 to 0 at Badaling Farm. The percentage downtime at Badaling Farm was reduced from 26 % with the original inline filter intake to 0 % with the vortex intake. At Olympic Park the percentage downtime was 31 % for the inline filter and 5 % for vortex intake. The percentage downtime reflects not just the number of times the filter was clogged, but also how soon the filter was replaced after it clogged. The minimum and average clog times, shown in the last two columns of Table 1, highlight the reduced maintenance requirement of the vortex intake design. At Olympic Park, the inline filters clogged in as little as 1 day, with an average clog period of just 6 days. The minimum maintenance interval for the vortex intake was 21 days, with an average of 46 days. At Badaling Farm, inline filters clogged within as few as 9 days with an average of 20 days, while the vortex-intake design required no maintenance for an entire period of 122 days.

Independent of high or low haze cover, the maintenance required for the vortex-intake EC system was markedly reduced, thus decreasing overall downtime substantially. Overall, the vortex intake can improve long-term monitoring of CO2 and H2O fluxes in conditions of high particulate concentration.