Comparison of eddy covariance and modified Bowen ratio methods for measuring gas fluxes and implications for measuring fluxes of persistent organic pollutants

Semi-volatile persistent organic pollutants (POPs) cycle between the atmosphere and terrestrial surfaces; however measuring fluxes of POPs between the atmosphere and other media is challenging. Sampling times of hours to days are required to accurately measure trace concentrations of POPs in the atmosphere, which rules out the use of eddy covariance techniques that are used to measure gas fluxes of major air pollutants. An alternative, the modified Bowen ratio (MBR) method, has been used instead. In this study we used data from FLUXNET for CO2 and water vapor (H2O) to compare fluxes measured by eddy covariance to fluxes measured with the MBR method using vertical concentration gradients in air derived from averaged data that simulate the long sampling times typically required to measure POPs. When concentration gradients are strong and fluxes are unidirectional, the MBR method and the eddy covariance method agree within a factor of 3 for CO2, and within a factor of 10 for H2O. To remain within the range of applicability of the MBR method, field studies should be carried out under conditions such that the direction of net flux does not change during the sampling period. If that condition is met, then the performance of the MBR method is neither strongly affected by the length of sample duration nor the use of a fixed value for the transfer coefficient.


Introduction 1
Despite the more than decade-old global ban on the production and use of persistent organic 2 pollutants (POPs) such as polychlorinated biphenyls (PCBs), hexachlorobenzene and several 3 organochlorine pesticides, these chemicals are still present in the environment and continue to 4 raise concerns due to their persistence, bioaccumulation, toxicity and potential for long-range 5 atmospheric transport (The Secretariat of the Stockholm Convention, 2010). As the 6 production and use of POPs continues to decline, large cities, old stocks and re-volatilization 7 from soil are expected to become more important sources to the atmosphere (Nizzetto et al., 8 2010). Studying the sources and fate of organic pollutants in the environment is an important 9 prerequisite to exposure and risk assessment, and environmental fate models that calculate The preferred approach to measure the flux of major air pollutants between the earth's surface 17 and the atmosphere is the eddy covariance (EC) technique (Baldocchi et al., 1988). It is based 18 on measuring the covariance of the concentration of a pollutant and the vertical wind velocity, 19 using data from very fast measurements (e.g. 5-10 Hz). This approach works well for 20 compounds such as CO 2 , methane, ozone and more recently also for mercury (Pierce et al., 21 2015), since concentrations can be measured at a high temporal resolution. However, it cannot 22 be applied directly when studying trace-level organic micropollutants that require sampling 23 times of at minimum several hours when using active high-volume sampling, or even several 24 weeks when using passive samplers (Hung et al., 2013) to result in reliable, quantifiable data. 25 26 One way to estimate chemical fluxes from measurements based on sampling times of hours to 27 days is to use the modified Bowen ratio (MBR) method (Businger, 1986). The MBR method 28 is based on the assumption that turbulent atmospheric transport occurs indiscriminately for 29 chemicals, heat and other scalar quantities that can be described entirely by their magnitude 30 without reference to direction. It can be used to measure the flux of a chemical pollutant (x) 31 from measurements of its concentration at two heights and the measured transfer coefficient 1 of another scalar such as heat (y) over the same height interval (Meyers et al., 1996, eq. 1): 2 x = − y * ∆ x ∆ eq. 1 3 Where F x (ng m -2 h -1 ) is the flux of the chemical x of interest, K y (m 2 h -1 ) is the measured 4 eddy diffusion coefficient for a scalar y over the height interval Z (m) and C x (ng m -3 ) is 5 the measured concentration gradient of x over the height interval. The negative sign on the 6 right hand side of the equation enforces the convention that downward fluxes have a negative 7 sign, and upward fluxes a positive sign. 8 9 Among other applications, the MBR method has been used to measure volatilization fluxes of 10 pesticides applied to agricultural fields (Majewski, 1999), to estimate PCB fluxes from Lake 11 Superior to the overlying air phase (Rowe and Perlinger, 2012) and fluxes of polycyclic 12 aromatic hydrocarbons (PAHs) above a forest canopy in Canada (S.-D. . In 13 the study of Choi et al., air was sampled for 24 h every 3 days at different heights for a period 14 of one month while leaves in the forest canopy were developing. The samples were analyzed 15 for PAHs and the data was combined with the eddy diffusivity of heat (K heat ) determined from 16 eddy covariance measurements from the FLUXNET network to estimate vertical PAH fluxes 17 using the MBR method. 18 19 Relaxed eddy accumulation (REA) is another method that is used to measure fluxes of 20 chemicals for which the EC approach is not feasible. Unlike the MBR method, the REA only 21 samples air at one height but uses fast switching valves in combination with high frequency 22 measurements of the wind speed and direction to split the incoming airflow according to the 23 prevailing vertical wind direction (Businger and Oncley, 1990). The air can then be collected 24 in bags or other reservoirs (Pattey et al., 1993) for further analyses or be passed through 25 denuders or sorbents such as polyurethane foam (PUF) (Majewski et al., 1993) as is done with 26 conventional high volume sampling to accumulate the levels needed for analysis. To our 27 knowledge, the REA method has not seen any recent uses to measure the fluxes of POPs or 28 POP-like pollutants, unlike the MBR method which has seen an increasing number of 29 applications in recent years. The likely reason is that it is technically more demanding to set-30 up the REA, and it requires specialized equipment. 31 Our goal in this study was to test the limits of applicability of the MBR method and to 1 evaluate its accuracy relative to the 'standard' EC technique. We used data from the 2 FLUXNET network to calculate fluxes of CO 2 and water vapor (H 2 O) with the MBR method  3   under different sampling duration scenarios and different assumptions about data availability  4 for the eddy diffusion coefficient K y . Thus, we took advantage of the high-frequency 5 measurement data for CO 2 and H 2 O, and used them as proxies for organic micropollutants in 6 order to analyze the performance of the MBR method compared to the EC method. By 7 averaging the FLUXNET data over periods that ranged from 1 hour to 1 week, we simulated 8 sampling times that are typically required to measure POPs and other organic micropollutants 9 in air. Our approach is similar to the one used by Majewski (see (Majewski, 1999a), who 10 simulated 24 h sampling periods from higher frequency data to characterize the potential for 11 long sampling times to introduce error to the aerodynamic profiling method. Prior to any calculations, we filtered the data to remove about 25% of the observations that 24 were flagged as unreliable for CO 2 or H 2 O. Details on the criteria for the flags can be found 25 on the FLUXNET homepage. Common reasons for flagging are instrument malfunctions, 26 calibration problems and outliers. All flagged data was filtered out simultaneously, such that 27 our analysis only includes data points collected at times when data were not flagged for either 28 CO 2 or H 2 O. 29 30 On inspection of the distribution of the CO 2 and H 2 O concentration gradients, it was apparent 1 that a few outliers that had not been flagged could significantly alter the average gradient 2 when pooling the data to simulate long sampling times. These outliers in some cases led to net 3 flux estimations based on the MBR method that were in the opposite direction compared to 4 the EC method. To exclude such outliers and reduce the influence of extreme values of 5 measured parameters, the highest and lowest 2.5% of values of the CO 2 gradient and the H 2 O 6 gradient were removed from the dataset prior to further calculations. 7

Modified Bowen Ratio 8
We used K heat derived from EC flux measurements and measurements of the scalar 9 temperature at two heights, as described in the paper by Choi et al. (S.-D.  to 10 specify the eddy diffusivity in the MBR method (K y in eq. 1). Specifically, K heat was inferred 11 from the dataset by first calculating the vertical turbulent flux of the sonic anemometer 12 temperature W'T' (K m s -1 , eq. 2) from the turbulent sensible heat flux (Q in W m -2 or J s -1 m -13 2 ), using the air density (σ air in kg m -3 ) and the gravimetric heat capacity (c p ) of air measured 14 at 33 m height (1005.7 J kg -1 K -1 ). Spurious K heat values less than or equal to zero were 15 removed from the dataset as these would indicate a heat flux against the measured 16 temperature gradient. represent sampling times typical for organic micropollutants, as described below. Fluxes 28 calculated with the MBR method were compared with those measured by the EC method 1 available from the FLUXNET dataset. 2

Data-analysis 3
To simulate sampling times typical for organic micropollutants, concentrations of CO 2 and 4 H 2 O reported as 30 minute averages in the database were pooled and averaged over periods of November fall. 14 15 We tested two approaches to specify K heat in the MBR method calculations. In the first 16 approach hourly average K heat values were calculated from 30 min averages of temperature 17 measurements reported in the database. In the second approach a geometric mean of K heat was 18 calculated for all time points across the entire period corresponding to the simulated sampling 19 time. The first approach takes advantage of the availability of high temporal resolution 20 information about K heat at the FLUXNET site, but the second approach is likely to be common 21 when applying the MBR since high frequency meteorological data is not always available. 22

23
The direction of flux for CO 2 can change on a diurnal basis (See Fig. 1

and Fig. S.M. 3). 24
During the day the flux of CO 2 is often negative (i.e., downward) due to photosynthesis while 25 during the night, plant respiration produces CO 2 and fluxes are positive (i.e. to the 26 atmosphere). In addition, atmospheric conditions during the night are typically much more 27 stable than during the day, resulting in a lack of large turbulent eddies and a higher 28 contribution of additional transport mechanisms, such as horizontal advection, to the total 29 flux. The result is that fluxes measured using EC during the night are often underestimated 30 (Aubinet, 2008). 31 1 To understand the impact of changing directions of flux and to investigate potential 2 underestimation of flux at night, fluxes during the day and during the night calculated with the 3 MBR method were evaluated against EC measurements separately. Nighttime data was set to 4 cover from 9 p.m. to 5 a.m. local time across all seasons, and daytime data was set from 9 5 a.m. to 5 p.m. local time. The nighttime / daytime divisions were selected based on the 6 shortest interval between sunrise and sunset at the site. The 8 hour periods representing day 7 and night allowed us to construct 24 hour, 3 day and 1 week sampling periods by averaging a 8 whole number of 8 hour periods taken at the same time of day over multiple days. In addition 9 to the nighttime/daytime split data, we also examined the performance of the MBR method 10 relative to the EC method when using continuous data that ignored day/night differences.   The use of either hourly-resolved data for K heat or a fixed value did not significantly affect the 8 the MBR method. A student t-test comparing the similarity of the 2 datasets resulted in a P 9 value below 0.0001. 10 11 Results using a fixed value for Kheat are shown in Table 1; those using hourly-resolved data 12 for Kheat are given in the supplementary material (See Table S.M. 2) 13

Performance of the MBR method on day/night split data 14
Fluxes of CO 2 during the nighttime only, measured using the MBR method in combination 15 with simulated sampling times ranging from 1 hour to 1 week are on average a factor of 1.7 16 and up to a factor of 2.1 larger than those derived with the EC method (see Table 1). Fluxes of 17 CO 2 during the daytime only, measured using the MBR method have, in some cases, the 18 opposite sign of the fluxes reported using the EC technique. Specifically, the MBR method 19 produces daytime fluxes with the opposite sign compared to the EC method during daytime in 20 the spring and for the 1-week duration simulated sampling scenario in the winter (values 21 marked with (!) in Table 1). In those cases where the direction of flux calculated with the 22 MBR does not agree with the EC method, the disagreement is attributable to the median value 23 of CO 2 (between 41.5 m and 33m) selected to represent the sampling period having a sign 24 that implies a flux in the opposite direction of the flux measured with the EC method. 25 In cases where the direction of daytime flux measured using the MBR method agreed with the 1 EC method, the ratio of the two fluxes ranged from 0.32 to 1.4, implying that the two methods 2 differed by factors that range from 1.4 to 3.0 and that the MBR method may either 3 underestimate or overestimate fluxes relative to the EC method. 4  it is very likely that steady-state conditions during the sampling period are not achieved. Our 29 results indicate however that when fluxes were unidirectional, measurements using the MBR 30 method were usually within 1 order of magnitude of those from the EC method, and that in 31 most cases the difference was less than a factor of 4. The summer period, in which K heat is low 32 and the gradients of CO 2 and H 2 O are large due to stable atmospheric conditions, showed the 1 best agreement between the MBR and EC methods. In the study of Choi  was an order of magnitude, which is within the range of agreement we obtained between the 4 MBR method and the EC method. 5 6 Fluxes measured with the MBR method corresponded better with the EC data for CO 2 than 7 for H 2 O, the reason of which is unknown. However our investigation of the discrepancy in 8 direction of flux between the two methods for H 2 O in the winter suggests that the 9 concentration gradient of H 2 O is more variable over height than that of CO 2 . Therefore the 10 gradient that we selected between 41.5 m and 33m may be more representative of the flux 11 measured at 33.3 m with the EC method for CO 2 than for H 2 O. 12 13 In general, the variation between fluxes determined using the MBR method for different 14 sampling frequencies was small, suggesting that longer sampling times did not introduce a 15 higher bias relative to EC measurements. Using a single value for K heat , in this case the 16 geometric mean across the sampling period, was found to be a good substitute for hourly K heat 17 values, indicating that it is possible to use the MBR method when there is no high frequency 18 data available for the transfer coefficient. 19 20 Some studies have shown that additional transport mechanisms aside from eddy diffusion, 21 such as advection, can become more important during night, thereby violating the conditions 22 needed for EC measurements to take place, and resulting in underestimations of the night time 23 fluxes (Aubinet, 2008). In this study, there is no clear difference in the performance of the 24 two methods relative to one another during either day or night. We note however that the 25 MBR method relies on K heat determined under the assumption that only turbulent atmospheric 26 processes occur, so the effect of additional transport mechanisms might not be evident in our 27 analysis. 28 Based on the findings in this study it is clear that field studies that use the MBR method to 29 measure gas fluxes of POPs and other organic micropollutants should be designed such that 30 the direction of the flux does not change during the sampling period. If this condition is 31 fulfilled and the concentration gradients are large enough to be measured accurately, then we 1 find no strong evidence that the duration of sample collection affects the performance of the 2 MBR method. Furthermore, using a fixed value for the transfer coefficient instead of hourly 3 data is a good alternative that should not introduce a significant bias when there is no high 4 frequency data available.