Spatial heterogeneity of ammonia fluxes in a deciduous forest and adjacent grassland

Abstract

Gas-phase ammonia (NH₃), emitted primarily from agriculture, contributes significantly to reactive nitrogen (N_r) deposition. Excess deposition of N_r to the environment causes acidification, eutrophication, and loss of biodiversity. The exchange of NH₃ between land and atmosphere is bidirectional and can be highly heterogenous when underlying vegetation and soil characteristics differ. Direct measurements that assess the spatial heterogeneity of NH₃ fluxes are lacking. To this end, we developed and deployed two fast-response, quantum cascade laser-based open-path NH₃ sensors to quantify NH₃ fluxes at a deciduous forest and an adjacent grassland separated by 700 m in North Carolina, United States from August to November, 2017. The sensors achieved 10 Hz precisions of 0.17 ppbv and 0.23 ppbv in the field, respectively. Eddy covariance calculations showed net deposition of NH₃ (−7.3 ng NH₃-N m⁻² s⁻¹) to the forest canopy and emission (3.2 ng NH₃-N m⁻² s⁻¹) from the grassland. NH₃ fluxes at both locations displayed diurnal patterns with midday peaks and smaller peaks in the afternoons. Concurrent biogeochemistry data showed over an order of magnitude higher NH₃ emission potentials from green vegetation at the grassland compared to the forest, suggesting a possible explanation for the observed flux differences. Back trajectories originating from the site identified the upwind urban area as the main source region of NH₃. Our work highlights that adjacent natural ecosystems sharing the same airshed but different vegetation and biogeochemical conditions may differ remarkably in NH₃ exchange. Such heterogeneities should be considered when upscaling point measurements, downscaling modeled fluxes, and evaluating N_r deposition for different natural land use types in the same landscape. Additional in-situ flux measurements accompanied by comprehensive biogeochemical and micrometeorological records over longer periods are needed to fully characterize the temporal variabilities and trends of NH₃ fluxes and identify the underlying driving factors.

1. Introduction

Ammonia (NH₃), a major component of reactive nitrogen (N_r) and a precursor to fine particulate matter (PM_2.5), has significant impacts on the environment locally and downwind of its sources. Agricultural activities, particularly synthetic fertilizer application and animal husbandry, contribute 80% of NH₃ emissions worldwide (Paulot et al., 2014). Anthropogenic emissions of NH₃ remain largely unregulated and are projected to increase in coming decades (Prather et al., 2013). Ammonia reacts with nitric acid and sulfuric acid in the atmosphere to form aerosols, which reduce visibility, affect radiative forcing and pose threats to human health (Langridge et al., 2012; Ostro et al., 2015). Ammonia emitted to the atmosphere can also return to the land through dry or wet deposition. Dry deposition occurs when gas-phase NH₃ or its ammoniated form, ammonium (NH₄⁺) directly deposits to surfaces, and wet deposition refers to the scavenging of NH₃ and NH₄⁺ by precipitation (Behera et al., 2013). Deposition of NH₃ has become increasingly important (e.g., accounting for 65% of the total N_r deposition in the U.S., Li et al., 2016) with associated environmental and ecological consequences. For example, Zhan et al. (2017) showed that dry deposition of NH₃ was the largest contributor to the eutrophication of Dianchi, a lake in China that suffers from harmful algal blooms. Ammonia deposition also causes acidification of terrestrial and aquatic ecosystems (Greaver et al., 2012). Eutrophication and soil acidification alter the balance between living species, which can lead to a decline in biodiversity (Fenn et al., 2003; Stevens et al., 2010). On a global scale, the nitrogen cycle has been heavily disrupted due to increased anthropogenic NH₃ emissions (Fowler et al., 2013).

Estimating total N_r deposition to assess whether the critical load has been exceeded in natural ecosystems requires accurate quantification of NH₃ emissions, transport, and deposition. Chemical transport models (CTMs), such as the Community Multiscale Air Quality Model (CMAQ), have incorporated bidirectional NH₃ exchange schemes based on the current-resistance analogy proposed by Nemitz et al. (2001) to improve NH₃ dry deposition estimates (Pleim et al., 2013; Pleim & Ran, 2011). Figure 1 shows the basic schematic of this two-layer bidirectional model, which accounts for the exchange of NH₃ in the canopy layer as well as the ground layer (Pleim et al., 2013). Net NH₃ exchange at the canopy scale (F_t) can be estimated from this model using the compensation points (χ) and resistances (R), usually parameterized as a function of meteorological variables as well as chemical and physical properties of vegetation and soil (Pleim et al., 2019). Many of the recent developments on modeling, however, focused primarily on improving NH₃ emission estimates from croplands. Less attention has been paid to natural ecosystems despite known high uncertainties in quantifying the dry deposition of NH₃ (Dennis et al., 2010). Non-stomatal processes such as exchange with leaf cuticles and litter on the soil surface are poorly constrained by observations (Hansen et al., 2017; Massad et al., 2010). Moreover, sub-grid variability in surface characteristics like leaf area index, surface roughness, soil moisture, and plant stomatal response are often ignored in CTMs, adding to the total uncertainty of deposition estimates at the grid scale (Walker et al., 2019). A few studies have demonstrated the significance of spatial heterogeneity in modeling nitrogen deposition (Paulot et al., 2018; Schwede et al., 2018), yet direct evidence from measurements is still lacking.

Figure 1.

Concept of the two-layer bidirectional model used by Pleim et al. (2013) for CMAQ. χ_a, χ_c, χ_g, and χ_st are the atmospheric NH₃ concentration, canopy compensation point, soil compensation point, and stomatal compensation point. R_a, R_ac, R_bg, R_soil, R_b, R_w, and R_st are the aerodynamic, in-canopy aerodynamic, ground quasi-laminar boundary layer, soil, quasi-laminar boundary layer, cuticular, and stomatal resistances. F_t, F_g, F_cut, and F_st are the total, soil, cuticular, and stomatal NH₃ fluxes.

Flux measurements can be used to evaluate the spatial heterogeneity of NH₃ deposition and constrain model estimates. Deposition velocities calculated directly from measured fluxes can also help validate parameterized resistances. However, long-term NH₃ flux measurements are lacking, in part due to the challenges associated with measuring gas-phase NH₃. Moreover, NH₃ exists at very low levels in natural environments, typically well less than 1 part per billion by volume (ppbv) (Langford et al., 1992). To date, there only exist a few NH₃ flux studies in deciduous forests, where fluxes were quantified using the aerodynamic gradient method (Hayashi et al., 2011; Langford et al., 1992; Pryor et al., 2001) or the relaxed eddy accumulation technique (Hansen et al., 2013, 2015). A summary of past NH₃ flux measurements in a wider category of forest environments is provided in Table 1. Many of these studies reported an overall negative flux, suggesting that forests often act as a sink for NH₃. However, significant variabilities were seen in the magnitude of NH₃ fluxes, sometimes reflecting the forests as a source of NH₃. There also remain open questions about the spatiotemporal representativeness of the measured fluxes, as many studies collected samples over relatively short periods (e.g., a few weeks) using instruments with long integration times (e.g., > 30 min).

Table 1.

Summary of past NH₃ flux measurements in forest environments

Forest type	Instrument	Detection limit (ng NH3-N m−3)	Integration time	Duration	NH3 flux (ng NH3-N m−2 s−1)	Deposition velocity (m s−1)	Reference
Deciduous
	Molybdenum oxide annular denuder	-	30 min	3 weeks	−0.8 - ≤0.8	-	(Langford et al., 1992)
	Wet effluent diffusion denuder	-	2 min	2 months	−17	-	(Pryor et al., 2001)
	Filter pack	30	4 hours	1 week	−108 – 31	~0	(Hayashi et al., 2011)
	Wet effluent diffusion denuder	3	30 min	1 month	−250±300 – 670±280	-	(Hansen et al., 2013)
	Wet effluent diffusion denuder	-	30 min	2 months	−70 – 110	-	(Hansen et al., 2015)
Coniferous
	Thermodenuder	82	1 hour	4 months	−82	3.2	(Wyers et al., 1992)
	Oxalic acid coated denuder	66	1.5 hours	1 week	−159	3.6	(Duyzer et al., 1992)
	Oxalic acid coated denuder	50	3 hours	2 months	−7.5	2.7	(Andersen et al., 1999)
	Annular denuder	238	8 – 12 hours	6 months	−11 – −3	-	(Tarnay et al., 2001)
	Annular denuder	41±33	1.5 hours	1 week	−32 – −19	1.2	(Huber et al., 2002)
	Parallel-plate denuder	82	1 hour	3 weeks	−2 – 9	-	(Hrdina et al., 2019)
Mixed
	Passive diffusion sampler	-	1 week	3 weeks	−159 – −32	-	(Fowler et al., 1998)
	Annular denuder	-	30 min	6 months	−74	3.0	(Neirynck & Ceulemans, 2008)

Note: Detection limits are for the specified integration time only. Durations indicate the total length of the measurements. Some studies had samples collected during multiple campaigns spanning a longer period. Positive and negative fluxes indicate emission and deposition, respectively. If the range of flux was not reported, the mean flux (or the median flux if the mean flux was not available) is shown here instead. A short dash (−) indicates that the parameter was not explicitly reported in the referenced work. Reported numbers were converted to the same unit for consistency. 1 unit of NH₃-N ≈ 0.82 units of NH₃ in weight.

Highly time-resolved measurements of NH₃ at low concentration are notoriously difficult as its affinity to surfaces slows the overall response time of instruments that require inlets and tubing to draw air samples to a detector (Sauren et al., 1989; Sun et al., 2015; Roscioli, 2016). For this reason, NH₃ fluxes have typically been measured using slow-response or discrete sampling methods by gradient or relaxed eddy accumulation as reflected in Table 2. Only more recently have fast response online mass-spectrometry and laser spectroscopy-based techniques been used for eddy covariance (EC) flux measurements (Famulari et al., 2004; Whitehead et al., 2008; Sintermann et al., 2011; Ferrara et al., 2012; Zöll et al., 2016), which requires a sampling frequency of 10 Hz. As these techniques employ a closed detector, spectral correction of the flux (20 – 50%) is needed to account for dampening of the high frequency signal (Moravek et al., 2019). Another approach for measuring NH₃ at high frequency and precision in low-concentration environments is open-path laser spectroscopy. An open-path sensor design allows ambient air to flow freely through the optical cell, thereby minimizing sampling artifacts due to NH₃ adsorption. Open-path sensors also do not require pumps, significantly reducing power consumption and making deployments possible at locations where electrical power is not readily available (Sun et al., 2015; Pan et al., 2021). A few studies have successfully demonstrated the application of quantum cascade laser (QCL)-based open-path sensors for fast NH₃ flux measurements in the field. For example, Sun et al. (2015) measured 20 Hz NH₃ fluxes from a feedlot continuously for two weeks with a detection limit of 1.1 ± 0.4 ng NH₃-N m⁻² s⁻¹ (68% confidence interval, CI). Pan et al. (2021) improved the instrument design and conducted flux measurements in an alpine grassland over two summers with a 10 Hz precision of 0.1 ppbv and a detection limit of 1.6 ng NH₃-N m⁻² s⁻¹ (95% CI).

Table 2.

Instrumentation of the forest and grassland EC systems

	Forest				Grassland
Sensor	NH3	CO2/H2O	Wind		NH3	CO2/H2O	Wind
Model	Zondlo Group	LI-7500	RM Young 81000V		Zondlo Group	LI-7500	METEK uSonic-3 CLASS A
Sensor path length (m)	0.72	0.13	0.15		0.72	0.13	0.17
Sensor height (m)	43.9	44.4	44.5		1.6	1.9	1.8
Horizontal separation (m)	0.45	0.25	-		0.40	0.44	-
Vertical separation (m)	0.50	0.06	-		0.25	0.10	-

Note: Sensor heights and separations were measured to the midpoint of each sensor’s sampling path. The ground level was used as the reference point for sensor height and the wind anemometer’s location was used as the reference point for horizontal and vertical separations.

To enhance the understanding of NH₃ bidirectional exchange and its spatial heterogeneity in forests, we present EC NH₃ fluxes measured concurrently on a tall tower above a deciduous forest and at an adjacent grassland in North Carolina, U.S. using custom-developed QCL-based open-path sensors. Flux measurements are analyzed by ecosystem and diurnal patterns, followed by a discussion on the driving factors using micrometeorological and biogeochemical measurements. We then explore the influence of regional NH₃ emissions on the forest due to atmospheric transport. We close by discussing the importance of considering the spatial variability of NH₃ fluxes originating from different land surfaces and future work to help constrain NH₃ dry deposition estimates.

2. Methods

2.1. Site Description

The State of North Carolina, located in southeastern U.S., is known for its swine industry concentrated in the southeastern part of the state. Agriculture accounts for 94% (191 kt) of NH₃ emissions in North Carolina, with livestock waste alone contributing to 95% of agricultural emissions (National Emissions Inventory, 2021). The study site, Duke Forest, lies ~20 km west of the Raleigh-Durham metropolitan area in central North Carolina. Small patches of agricultural lands exist within a few km of this area with the main crop types being corn and soybeans (United States Department of Agriculture, 2017). The nearest permitted animal facility is a 150-head cattle farm 4 km west of the site; all the other animal facilities are farther than 10 km away (North Carolina Department of Environmental Quality, 2020). At the time of this study, several landfill facilities were located approximately 1 km southeast of the site, and a construction site a few hundred meters further away was operational during the day. While Duke Forest is not directly surrounded by agricultural activities, it is a good candidate for studying the transport of agricultural emissions and the impacts on natural ecosystems due to its proximity (~100 km) to the animal production source region in eastern North Carolina

Duke Forest has a relatively flat and homogeneous terrain, covering an area of at least 1 km² composed of deciduous trees with scattered evergreen trees (United States Department of Agriculture, 2017). Major tree species in the forest stand include hickory (Carya tomentosa), white oak (Quercus alba L.), willow oak (Quercus phellos L.), yellow poplar (Liriodendron tulipifera L.), and sweetgum (Liquidambar styraciflua L.) in the overstory layer, and elm (Ulmus sp.), red maple (Acer rubrum), redbud (Cercis canadensis L.), blue beech (Carpinus caroliniana Walt.), and dogwood (Cornus florida L.) in the understory (Geron et al., 1997; Oishi et al., 2008). A 44-m flux tower is located at the center of the forest (35°58’27”N, 79°06’00”W) at an elevation of 200 m above sea level. The average canopy height within a 450 m × 450 m area around the tower was 29.7 m calculated using Quality Level 2 (QL2) light detection and ranging (LIDAR) data collected in 2015 (https://sdd.nc.gov/). Average leaf area index (N = 5 locations surrounding the tower) measured by optical Plant Canopy Analyzer (model LAI-2200, LI-COR, Inc) was 5.2 and 2.5 m² m⁻² on August 31, 2017 and November 3, 2017. The forest EC system was deployed above the canopy on the top deck of the 44-m flux tower. The measurements therefore represented integrated fluxes over the forest, which did not resolve processes below the canopies such as fluxes directly above the forest floor.

A small (~ 480 m × 305 m) grass field (35°58’16”N, 79°05’36”W) is located approximately 700 m southeast of the forest tower at an elevation of 174 m above sea level. Vegetation in the field is primarily tall fescue (Festuca arundinacea Shreb.) with lesser amounts of C3 and C4 grasses, herbs, and forbs (AmeriFlux, 2021). The grass field is not fertilized or grazed but is managed as a hay crop and mowed once or twice per year. The field was mowed to a height of 10 cm in July, 2017 after which grass height reached ~ 85 cm by early November 2017. Average leaf area index (N = 5 locations within the grass field) was 1.6 and 1.5 m² m⁻² on August 31, 2017 and November 3, 2017. A 2-m tripod was installed at the center of the grassland for EC flux measurements. While the grassland sensor height was lower than recommended by some literature (e.g., 4–5 times the sensor’s path length [~70 cm] according to Burba & Anderson, 2010), it allowed measurements of NH₃ exchange closer to the grass to better understand the surface processes and better accommodation of the field fetch. Human activities were maintained at minimal levels except for regular maintenance of the equipment, and periods potentially under the influence of these activities were removed from the analyses.

Figure 2a shows the satellite map of the study site with the locations of the EC systems highlighted. Figures 2b and 2c show the detailed instrumentation setup, where each EC system consisted of one open-path NH₃ sensor, one 3-dimensional sonic anemometer (forest: R. M. Young Co.; grassland: Metek, GmbH) for wind measurements, and one open-path sensor (LI-COR, Inc.) for measuring water (H₂O) and carbon dioxide (CO₂) needed for the spectroscopic and air density corrections. Data from all three sensors in each system were logged to a single board computer using a custom LabVIEW program. The systems were monitored remotely at least twice a day to ensure normal operation, and the external mirrors were cleaned regularly as well as after each precipitation event throughout the campaign. Information on the instrumentation setup can be found in Table 2, and details on NH₃ measurements are described in Section 2.2.

Figure 2.

a) Google Earth satellite map of the study site with the locations of the two EC systems highlighted. The inset shows the geographical location of Duke Forest in the contiguous U.S.; b) forest EC system on the 44-m tower; c) grassland EC system on the 2-m tripod.

2.2. NH₃ Measurements

Two custom QCL-based open-path NH₃ sensors were developed for 10 Hz flux measurements following the approaches described in Miller et al. (2014), Sun et al. (2015) and Pan et al. (2021). A summary is provided here; interested readers are referred to the publications for further reading. For simplicity, the two NH₃ sensors will be referred to as the FS (forest sensor) and GS (grassland sensor) herein. The FS and GS each used a 72-cm long Herriott multi-pass cell (Herriott et al., 1964) with a pair of 3-in diameter concaved spherical mirrors, achieving a total optical path length of 60 m. The mirrors were fabricated with Molybdenum substrates to achieve a reflectivity of > 98%. The QCLs used for NH₃ detection were integrated with a high heat load (HHL) package (FS: Hamamatsu Photonics K.K.; GS: Corning Inc.). Different QCLs were used because only one Hamamatsu QCL was functional at the time of instrument development. The Hamamatsu QCL also had a higher current requirement (~1 A) compared to the Corning QCL (0.3 A), so we did not acquire another one for the GS. Due to this higher current requirement of the Hamamatsu QCL, a solid-state recirculating chiller (ThermoTek Inc.) was attached to the mounting plate inside the FS to ensure temperature stability. The NH₃ absorption peak at 9.06 μm (1103.8 cm⁻¹) was selected to minimize spectral interferences from other trace gases species. Wavelength modulation was applied to the QCL current for increased sensitivity, and the 2nd harmonic (2f) signal was detected to further minimize interferences from water vapor (H₂O). An in-line reference cell (5 cm) filled with ethylene gas (2% at 50 hPa) was placed in front of the detector inside the FS as a reference for NH₃ concentration. Due to limited availability, the GS was not equipped with such reference cell. Mercury-Cadmium-Telluride (MCT) infrared (IR) detectors made by Teledyne Judson and Intelligent Material Solutions Inc. were used on the FS and GS, respectively.

Because wavelength modulation spectroscopy (WMS) is a relative detection technique, the sensors were calibrated in the laboratory before and after the field deployment by fitting WMS data to direct absorption spectroscopy (DAS) readings in the 0 – 400 ppbv range using a linear model similar to Miller et al. (2014). During the calibration, the sensor was covered with a custom-designed Nylon tube to isolate the optical path from ambient air and minimize NH₃ adsorption. The zero point of NH₃ was obtained by flowing pure nitrogen gas first through an NH₃ scrubber (Perma Pure LLC) that neutralizes NH₃ using phosphoric acid and then to the enclosed optical cell continuously for 2 – 3 hours until readings became stable. We used the same calibration standard to minimize systematic biases between the systems. The FS and GS were also mounted on a tripod for an intercomparison before field deployment and demonstrated similar responses of Allan deviation against averaging time within 30-min intervals.

In addition, hourly concentrations of NH₃ were measured at the height of the FS (44 m) using the Monitor for Aerosols and Gases in Ambient Air (MARGA, Metrohm Applikon B.V., the Netherlands). Details and principles of the MARGA system have been previously described (Chen et al., 2017; Rumsey et al., 2014). Briefly, the MARGA system consists of a sample box positioned on the tower and a detector box located in a climate-controlled building at the base of the tower. The sample box is comprised of an inlet of ½” outer diameter 30-cm long PFA Teflon tubing with no particle size selection, through which air flow is mass controlled at ~ 16.7 LPM, a wet rotating denuder (WRD) for collection of soluble gases and a steam jet aerosol collector (SJAC). Liquid sample from the WRD and SJAC is continuously drawn from the sample boxes down the tower to the analytical box for Ion Chromatography (IC) analysis on an hourly basis. The sample time is adjusted for a delay time of 1 hour required for the sample to travel down the tower to the analytical box. During the course of the experiment, multi-level NH₄⁺ standards were introduced at the WRD, with airflow turned off, to assess the analytical accuracy of the NH₃ measurement. Air flows were routinely checked using a MesaLabs Bios Defender (Brandt Instruments, USA) flow meter. The detection limit is 0.05 μg m⁻³ for gas-phase NH₃, and the accuracy is estimated to be around 20 – 30% (Rumsey et al., 2014).

The MARGA system provided important reference NH₃ concentrations for the forest QCL sensor, particularly at sub-ppbv levels, where the FS had significant long-term drift on the order of ~1 ppbv. To evaluate whether the calculated fluxes were impacted by drift of the FS, we mapped the 10 Hz QCL NH₃ data to the corresponding MARGA half-hourly means downscaled from hourly measurements using the z-score normalization method:

$$C_{QCL}^{\prime }=\frac{{\text{C}}_{QCL}-\overline{C_{QCL}}}{S_{QCL}}{S}_{\mathit{\text{MARGA}}}+\overline{C_{\mathit{\text{MARGA}}}}$$ #(1)

where $C_{QCL}^{\prime }$ is the normalized 10 Hz QCL NH₃ concentration, ${\text{C}}_{QCL}$ is the unnormalized 10 Hz QCL NH₃ concentration, $\overline{C_{QCL}}$ is the unnormalized half-hourly mean QCL NH₃ concentration, $S_{QCL}$ is the standard deviation of unnormalized QCL NH₃ concentration in the same half hour, $S_{\mathit{\text{MARGA}}}$ is the standard deviation of MARGA NH₃ concentration in the same half hour, and $\overline{C_{\mathit{\text{MARGA}}}}$ is the half-hourly mean MARGA NH₃ concentration. Data were processed in 30-min groups to observe the EC conventions. $S_{\mathit{\text{MARGA}}}$ is set to be equal to $S_{QCL}$ based on the assumption that the two sensors were adjacently located and measured largely the same air currents. Therefore, the distribution of readings should be similar if MARGA was able to perform measurements at higher frequencies. Equation (1) is simplified to Equation (2) after the assumption. NH₃ fluxes calculated using unnormalized and normalized concentrations are discussed in Section 3.1.

$$C_{QCL}^{\prime }={\text{C}}_{QCL}-\overline{C_{QCL}}+\overline{C_{\mathit{\text{MARGA}}}}$$ #(2)

At the grassland, additional NH₃ measurements were made on a biweekly frequency under the National Atmospheric Deposition Program’s (NADP) Ammonia Monitoring Network (AMoN, site ID: NC30). Details about AMoN measurements can be found in existing literature (Puchalski et al., 2011). Briefly, the AMoN system consists of a Radiello^® passive sampler (Sigma‐Aldrich, Germany) that takes advantage of the simple diffusion technique. The interior cartridge of the sampler, coated with phosphoric acid, is sent back to NADP’s Central Analytical Laboratory every two weeks for analysis. Flow Injection Analysis is used to determine NH₄⁺ concentrations in the collected samples. Concentrations presented here are blank corrected using the corresponding yearly median travel blank (N = 6) and the corresponding equivalent detection limit of NH₃ is approximately 0.16 μg m⁻³ over the two-week averaging period. The blank corrected concentrations in Class A and B, defined by the AMoN metadata as fully qualified data, and valid data with minor issues, are presented in this work. Another AMoN site (ID: NC98) was established at the forest site a year after the study ended and hence not included in this analysis.

2.3. Biogeochemical Measurements

Live grass, leaves and litter on the soil surfaces were collected within the area surrounding the hardwood forest tower and in the grass field during summer and fall, 2017. In the forest, vegetation was collected from several of the primary tree species within the tower footprint, including white oak (Quercus alba L.), willow oak (Quercus phellos L.), hickory (Carya tomentosa), sweetgum (Liquidambar styraciflua L.), and red maple (Acer rubrum). Sunlit leaves were sampled from duplicates of each species. Litter was collected in five randomly distributed 1 × 1 m plots in the tower footprint. In the grass field, grass and litter were collected in five randomly distributed 1 × 1 m plots.

Approximately 50 g of material was collected during each sampling event from each plot or tree, of which 5 g was subsampled, ground in liquid nitrogen using a mortar and pestle and small coffee grinder then extracted with 25 mL of deionized water. pH was determined directly on the extracts. [NH₄⁺] in the extracts, which reflects the bulk tissue concentration, was determined by colorimetry after separation of the NH₄⁺ from the solution as NH₃ using headspace equilibration. For the headspace method, 5 mL of tissue extract was added to a 250 mL high density polyethylene jar containing two Adapted Low-cost Passive High Absorption (ALPHA) passive samplers (Center for Ecology and Hydrology; Tang et al., 2001), without the diffusion barrier, affixed to the interior of the lid. The jar was sealed and 5 mL of 0.3 N NaOH was added to the extract via septum. NH₃ liberated from the liquid extract into the headspace was collected by the passive diffusion samplers over a period of 48 hours, after which the passive sampler was extracted with 10 mL of deionized water. NH₄⁺ in the extracts was determined by colorimetry using a Salicylate Method – QuickChem^®10–107-06–2-A.

Each batch of vegetation analyses included 5 extracted blanks determined by adding 5 mL of deionized water to the headspace jar along with 5 mL of 0.3 N NaOH and otherwise processed as normal vegetation samples. Extracted blank concentrations (NH₄⁺) were averaged and subtracted from vegetation sample results. Each batch of vegetation analyses also included a set of extracted standards to evaluate headspace recovery of NH₄⁺. Extracted recovery standards consisted of adding 5 mL of a standard containing 5 μg mL⁻¹ NH₄⁺ to the headspace jar along with 5 mL of 0.3 N NaOH and were otherwise processed as normal vegetation samples. Recoveries were > 90% on average.

Emission potentials of the grass (Γ) were estimated directly from measured concentrations of [H⁺] (M) and [NH₄⁺] (μg g⁻¹ tissue fresh weight) in the bulk tissue as:

$$\Gamma =\frac{\left[{NH}_4^{+}\right]\times \left(5.56E^{-5}\right)\times LD}{\left[H^{+}\right]}$$ #(3)

where LD is leaf density (kg L⁻¹ fresh tissue, equivalent to g cm⁻³ fresh tissue). In this case, LD for deciduous tree species is 0.37 (Poorter et al., 2009). For grass species, the median LD (0.156 kg L⁻¹) of the 16 C3 species reported by Sugiyama (2005) is used. Leaf densities for litter assumed the value for live vegetation. The factor of 5.56E-5 in the equation for Γ is necessary to convert [NH₄⁺] from μg NH₄⁺ g⁻¹ tissue to mol NH₄⁺ kg tissue⁻¹.

Extractable ammonium (NH₄⁺_ex) was determined on soil samples collected (0–3 cm depth) in the forest and grass areas in the same five 1 × 1 m plots as the vegetation samples during summer and fall, 2017. Approximately 50 g of soil was collected and composited during each sampling event in each plot. Samples were immediately placed on ice, returned to laboratory, and stored under refrigeration until extraction, typically within 24 hours. Soils were sieved to remove material > 2 mm immediately prior to extraction. For extraction of NH₄⁺, 5 g of field moist soil was added to 25 mL of 0.01 M CaCl₂, rotated for 30 minutes to mix, centrifuged for 5 minutes and approximately 15 mL of supernatant collected and frozen until analysis. As with the vegetation extracts, NH₄⁺ was determined by colorimetry using a Salicylate Method – QuickChem^®10–107-06–2-A. pH was determined directly in a 1:1 soil/water slurry. Gravimetric moisture was determined by drying 10 g of soil at 100 °C for 24 hours. The concentration of NH₄⁺ in the soil pore solution (NH₄⁺_sol) was determined from NH₄⁺_ex using a Langmuir sorption model (Hinz, 2001). The soil emission potential (Γ) was estimated as the molar ratio of NH₄⁺_sol to H⁺ using the maximum sorption capacity (Q_max, 302 mg NH kg⁻¹ soil) and binding coefficient (K_L, 0.014 L mg⁻¹) determined for the Duke Forest grass field (Iredell gravelly loam) (Alnsour, 2020).

2.4. Eddy Covariance Calculations

The EC method calculates the upward and downward turbulent fluxes of a scalar, such as a trace gas in the atmosphere, using the covariance between the perturbations of vertical wind and the scalar of interest (Burba & Anderson, 2010). The NH₃ flux ($F_{{NH}_3}$) at the measurement height is defined by Equation (4):

$$F_{{NH}_3}={\rho }_a\overline{w^{\prime }{c}_{{NH}_3}^{\prime }}$$ #(4)

where ${\rho }_a$ is the density of air and $\overline{w^{\prime }{c}_{{NH}_3}^{\prime }}$ is the covariance between the turbulent vertical wind speed and the turbulent mixing ratio of NH₃ in the air.

We mainly followed the procedures outlined in Mauder et al. (2013) to calculate half-hourly integrated EC fluxes using the 10 Hz raw data, with modifications specific to the custom-developed open-path QCL NH₃ sensors. Table 3 summarizes the main procedures and corresponding methods used during the flux calculation. A brief description is provided here; more details on EC flux processing can be found in Pan et al. (2021). Initial quality control on the raw data included removing low laser signal intensity, unstable laser thermal stability, and half-hourly periods with missing or invalid data greater than 30%. Next, wind data from the anemometer were rotated using the double rotation method. Valid data from the QCL NH₃ sensor and the LI-7500 sensor within the same 30-min period were then synchronized with the winds using the covariance maximization method with rolling averages (Taipale et al., 2010). Calibration coefficients obtained in the laboratory were applied to raw NH₃ readings, which were then spectroscopically corrected using correction factors from a lookup table based on spectroscopic simulations (Sun et al., 2015), an approach similar to the spectroscopic correction of methane as described in McDermitt et al. (2011). Block averaging was used to remove time-dependent correlations, and the time series were de-spiked using the median absolute deviation (MAD) method. These steps were followed by spectral corrections for high frequency losses of fluxes (Moore, 1986), the self-heating correction for LI-7500 (Burba et al., 2008), and the Webb, Pearman, and Leuning (WPL) correction (Webb & Leuning, 1980) for the effect of air density fluctuations on gas densities.

Table 3.

Procedures and methods used in EC flux calculations

Procedure	Method	Reference
Wind rotation	Double rotation	(Wilczak et al., 2001)
Time lag compensation	Covariance maximization with rolling averages	(Taipale et al., 2010)
Spectroscopic correction	Lookup table based on spectroscopic simulations	(McDermitt et al., 2011; Sun et al., 2015)
Detrending	Block average	(Mauder et al., 2013)
De-spiking	Median absolute deviation	(Mauder et al., 2013)
Spectral loss correction	Transfer functions	(Moore, 1986)
Flux quality test	0–1-2 quality flag system	(Mauder et al., 2013)
LI-7500 self-heating correction	Linear regression	(Burba et al., 2008)
Air density fluctuation correction	WPL approach	(Webb & Leuning, 1980)
Footprint estimation	Model parameterizations	(Kljun, 2004)

Note: HITRAN: High-resolution Transmission Molecular Absorption Database; WPL: Webb, Pearman, and Leuning.

Several flux quality indicators were calculated at the end of the process to remove periods that did not satisfy the EC conditions. These included the stationary test (ST) and the Integral Turbulence Characteristics (ITC). Based on the ST and ITC values, quality flags were assigned to the half-hourly fluxes using the 0–1-2 system outlined in Mauder et al. (2013). Footprint sizes contributing to the NH₃ flux were estimated following Kljun (2004). To evaluate sensor performances and flux uncertainties, the instrumental noise (${\sigma }_{ins}$) and stochastic error (${\sigma }_{sto}$) of fluxes were calculated following Mauder et al. (2013). The relative errors were calculated using raw fluxes before spectral corrections and multiplied by the corrected fluxes for more accurate estimates of absolute errors. For each half-hourly period, the detection limit ($LoD$) of NH₃ flux at 95% CI was calculated as

$$LoD=1.96{\sigma }_{sto}\left|F_{{NH}_3}\right|$$ #(5)

which is defined by the relative stochastic error as well as the NH₃ flux magnitude (Langford et al., 2015). In addition, the 10 Hz precisions of the QCL NH₃ sensors were estimated using the Allan deviation method that quantifies the drift of a time series caused solely by instrumental noise. Allan deviations (square roots of Allan variances) were obtained from the lowest nighttime flux period and the adjacent ±30 min periods during the first week of the campaign. The field precision of each sensor was then approximated as the average Allan deviation at 10 Hz over the three half-hourly periods.

2.5. Back Trajectory Analysis

Back trajectory analyses of air parcels are commonly used to identify the source region of atmospheric pollutants for the location of interest. To evaluate the impact of atmospheric transport on the fluxes observed at Duke Forest, we used the HYbrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model version 4 with the PySPLIT package version 0.3.6 and the North American Model (NAM) meteorological data at a resolution of 12 km (Draxler & Hess, 1998; Stein et al., 2015; Warner, 2018). Back trajectories of air parcels originating from the study site were initiated at an altitude of 50 m and the beginning of each hour. A total of 24 hourly back trajectories were generated per day over the entire campaign, and each trajectory was traced back for 24 hours.

3. Results and Discussion

3.1. Sensor Performance

Flux measurements at Duke Forest were carried out between August 12 and November 9, 2017. The two systems were operational for a total of 3414 hours (forest: 1638 hours; grassland: 1776 hours), leading to an up-time of 76% and 82%, respectively. After initial quality control on the raw data, 1386 and 1224 half-hourly NH₃ fluxes were calculated from the forest and grassland datasets. Following the method in Mauder et al. (2013), 29% (18%), 65% (61%), and 6% (21%) of the half-hourly integrated NH₃ fluxes were classified as Class 0, 1, and 2 for the forest (grassland) system. Class 0 is defined as high-quality data for fundamental research; Class 1 is moderate quality data and Class 2 is low-quality data that should be discarded. Despite regular sensor maintenance, the performance of QCL NH₃ sensors degraded over time due to the contamination of mirror surfaces from dust, dew, and rain, as well as long-term drifting caused by optical fringing. To obtain a comprehensive view of NH₃ exchange, fluxes in both Class 0 and 1 data were used for the following analyses.

Additional filters were applied to remove fluxes that did not meet the EC conditions or might be of low quality: (1) the friction velocity must be greater than 0.1 m s⁻¹ to ensure enough turbulence; (2) 70% of the NH₃ flux must be coming from inside the footprint with homogeneous vegetation cover (400 m for the forest and 110 m for the grassland). The final number of 30-min fluxes after all the quality controls was 1019 and 655 for the forest and grassland, respectively. Notably, 68% of the nighttime (1800 – 0600 local time) forest fluxes remained acceptable, compared to only 34% at the grassland, indicating that turbulence was more developed over the forest tower than the grassland tripod at night.

Quality-controlled forest NH₃ fluxes calculated from unnormalized QCL NH₃ concentrations were compared to those calculated from normalized concentrations using collocated MARGA data (Figure 3). Orthogonal regression was used to minimize perpendicular distances from both the independent and dependent variables to the fit. Non-overlapping periods between QCL and MARGA measurements further reduced the availability of final forest fluxes by 30%. The unnormalized fluxes showed a good agreement with the normalized fluxes with a slope of 1.04±0.01, a negligible intercept (0.44±0.18) compared to the scale of the fluxes, and an R² of 0.98. These results demonstrate that the half-hourly fluxes were relatively insensitive to long-term drifts of QCL NH₃ concentrations. Because no reference MARGA dataset was available for the grassland sensor, only unnormalized NH₃ fluxes will be presented onward for consistency between the forest and grassland unless explicitly stated otherwise.

Figure 3.

Orthogonal regression between normalized NH₃ fluxes and unnormalized NH₃ fluxes at the forest. Only quality-controlled fluxes in overlapping periods were used for comparison. The standard deviations of the slope and intercept are included for reference. The uncertainties of fluxes are not shown on the graph for better clarity. Dash line indicates the 1:1 relationship.

The best estimates of 10 Hz precision of the QCL NH₃ sensors, calculated as the square roots of the Allan variances, were 0.17 ppbv (forest) and 0.23 ppbv (grassland). While we selected nighttime periods where turbulence was generally suppressed to assess the precisions in the field, some fluctuations in the data would be inevitably attributed to changes in temperature, pressure, as well as ambient NH₃ concentration. Therefore, these numbers represent the lower bounds of the actual sensor performance, as they were expected to improve under controlled environmental conditions. The median instrumental noises were 11% (forest) and 19% (grassland), and the median stochastic errors of the half-hourly NH₃ flux were 28% (forest) and 27% (grassland). The resulting median NH₃ flux detection limits at 95% CI for the final quality-controlled datasets were 7.6 ng NH₃-N m⁻² s⁻¹ (forest) and 5.7 ng NH₃-N m⁻² s⁻¹ (grassland).

As noted in Section 2.4, several corrections were needed for fluxes calculated using the EC method. The average spectroscopic and WPL correction terms combined for the filtered NH₃ fluxes were less than 1% (forest) and 4% (grassland). The average spectral correction term was 7% (forest) and 39% (grassland), which included high frequency losses due to sensor separation and path averaging of vectors and scalars. The larger spectral loss in the grassland NH₃ fluxes was likely related to the lower deployment height (2 m) compared to the forest (44 m). The self-heating correction for the LI-7500 CO₂/H₂O sensor resulted in an average increase of 3.5% (forest) and 1.4% (grassland) in sensible heat flux, which led to an average increase of 0.002% (forest) and 0.02% (grassland) in NH₃ flux. While the self-heating correction terms had negligible impacts on NH₃ fluxes and their diurnal patterns in our study, the validity of this correction has been questioned by some previous studies. In particular, Deventer et al. (2021) found over-corrections using the semi-empirical models and no improvements in performance when using measured data instead, which implies that the correction may not be universally appliable to all study sites.

Figure 4 shows the normalized composite co-spectra of sensible heat, H₂O, CO₂, and NH₃ fluxes at the forest and grassland during times when NH₃ fluxes were above the sensor’s respective detection limit. In general, all the scalar fluxes started to fall off at similar frequencies, peaking around a natural frequency of 0.02 Hz, and NH₃ co-spectra did not show much deviation from the others. However, some deviations from the ideal −4/3 slope in the Kaimal model (Kaimal et al., 1972) were seen in the inertial subrange, showing again the necessity for spectral corrections of flux attenuation during EC calculations due to path averaging and physical separations between the sonic anemometer and trace gas sensors. Co-spectra of fluxes under different atmospheric conditions were also inspected, and no noticeable shifts of peak position were found between stable and unstable conditions, suggesting a consistency of eddies responsible for energy dissipations in these conditions.

Figure 4.

Composite co-spectra of sensible heat, CO₂, H₂O, and NH₃ fluxes averaged over the quality-controlled periods and when the NH₃ fluxes were larger than the detection limits for (a) forest and (b) grassland. The normalized dimensionless frequency is plotted on the x axis and the normalized co-spectral density is plotted on the y axis. Note the different range of axes on the two panels.

3.2. NH₃ Concentration and Flux Profiles

NH₃ concentrations measured by the forest MARGA system and the grassland AMoN system throughout summer (June) and fall (November) 2017 are compared in Figure 5a. Ambient NH₃ levels were found to be similar across the two locations with a mean of 0.61 and 0.64 μg m⁻³ for forest and grassland, and a median of 0.48 and 0.64 μg m⁻³ for forest and grassland. The composite diurnal profile of MARGA concentration (Figure 5b) during the flux study period (August 12 – November 9) shows higher NH₃ concentrations during daytime, which is most likely driven by higher daytime temperature and its effect on surface compensation points and upwind emissions (Sutton et al., 2013). For reference, the average diurnal profile of air temperature (not shown here) measured by the sonic anemometer on the forest tower peaked around 1600 local time (26 °C), and hourly air temperature showed an overall correlation of 0.31 with MARGA concentrations during overlapping periods. A diurnal profile was not available for the grassland because AMoN measurements have a two week sampling interval.

Figure 5.

(a) Overview of MARGA and AMoN NH₃ measurements between June and November 2017. Mean concentrations are shown as black diamonds. The bottom, center, and top of the boxes denote the 25^th (Q1), 50^th (Q2), and 75^th (Q3) percentiles. Outliers are shown with a “+” symbol and are defined as values greater than Q3+1.5*(Q3-Q1) or smaller than Q1–1.5*(Q3-Q1). Outliers beyond 3σ from the means are not shown. (b) Composite MARGA NH₃ concentration diurnal profile during the flux study period (August 12 – November 9). Top panel: hourly MARGA concentrations averaged using days with more than 70% data coverage. The blue line shows the mean concentration in each hour. Boxes and whiskers show the distribution of concentrations. Bottom panel: number of MARGA measurements available in each hour.

In contrast to the NH₃ concentrations, NH₃ fluxes at the two locations showed large differences: The forest experienced an overall average deposition of 7.3 ng NH₃-N m⁻² s⁻¹ (median: 7.8 ng NH₃-N m⁻² s⁻¹), whereas net emission was observed at the grassland with an average flux of 3.2 ng NH₃-N m⁻² s⁻¹ (median: 1.4 ng NH₃-N m⁻² s⁻¹). While 76% of filtered forest fluxes and 73% of filtered grassland fluxes were above their corresponding median detection, we included those below the detection limits to avoid potential biases so long as the individual flux periods passed the quality controls. Figure 6 shows the composite NH₃ flux diurnal profiles during the quality-controlled periods upscaled from 30-min averages to hourly scales, the number of observations available during each hour, and a side-by-side comparison of forest and grassland fluxes. Clear diurnal variations can be seen at both locations, but they showed opposite patterns (i.e., deposition at forest and emissions at grassland). In general, the largest NH₃ flux peaked around noon and slightly lower peaks were found throughout the afternoons (1400 – 1800 local time). The forest remained a sink for NH₃ for most of the day while the grassland likely turned into a sink in the evening until midnight, but we note again the smaller number of observations available at night due to suppressed turbulence, especially at the grassland. The mean and median deposition velocities calculated from the MARGA-normalized forest dataset were 0.4 and 1.9 cm s⁻¹. These values are in line with those reported in literature for deciduous forest environments (0.3 – 1.9 cm s⁻¹, Fan et al., 2009; Staelens et al., 2012; Zhang et al., 2009; Schrader & Brümmer, 2014).

Figure 6.

Top panels: composite diurnal profiles of NH₃ fluxes at forest (left) and grassland (right) by hour over the study. The red line shows the mean flux in each hour. The box and whisker plot shows the distribution of fluxes, with the bottom, center, and top of the box denoting the 25^th (Q1), 50^th (Q2), and 75^th (Q3) percentiles. Outliers are shown with a “+” symbol and are defined as values greater than Q3+1.5*(Q3-Q1) or smaller than Q1–1.5*(Q3-Q1). Outliers beyond 3σ from the means are not shown. The horizontal dash-dotted line specifies the location of zero flux. Note the different scales on the y axis for better visualization of smaller fluxes. Middle panels: number of observations in each hour that passed the flux quality controls at forest (left) and grassland (right). Bottom panel: side-by-side comparison of forest and grassland NH₃ fluxes during overlapping periods.

Many studies have revealed that NH₃ fluxes in natural environments display diurnal patterns, but the dominant direction of exchange depends on many factors such as location and measurement period, and therefore can vary for the same type of land cover. For example, Horváth et al. (2005) found a large deposition of NH₃ to grassland during daytime of all seasons except spring. Langford et al. (1992) and Hansen et al. (2015) observed high emissions of NH₃ from grassland and deciduous forest in the afternoon. Our results were unique and specific to the Duke Forest site, but the opposite directions of NH₃ exchange found in these two adjacent ecosystems imply that it is important to consider the spatial heterogeneity within the same natural landscape. We note, however, that the availability of high-quality flux data decreased over time due to degraded sensor performance. This could lead to biases as fluxes lower than the detection limits might not be captured.

3.3. Drivers of NH₃ Flux and Spatial Heterogeneity

Because the forest NH₃ concentrations and fluxes showed clear but opposite diurnal patterns compared to the grassland, it is worthwhile to explore the relationship between the two quantities. Figure 7a shows the correlation between MARGA-normalized NH₃ fluxes and MARGA NH₃ concentrations at the forest, sorted into 10 bins of approximately equal size from the smallest to the largest concentration. A negative correlation of −0.92 was found between NH₃ fluxes and concentrations, indicating that deposition to the forest canopy is more likely to happen when ambient NH₃ concentration is higher. Figure 7b and 7c show the relationship between NH₃ fluxes and friction velocities (u*) sorted from the smallest to largest u* at the forest and grassland, respectively. Higher u* generally corresponds to more negative flux (r = −0.95) at the forest and more positive flux (r = 0.20) at the grassland, showing that the magnitude of flux is closely related to turbulent exchange. A side-by-side comparison of u* at these two locations during overlapping periods (Fig. 7d) shows that the forest generally observed higher u* (median: 0.39 m s⁻¹) than the grassland (median: 0.22 m s⁻¹). These statistics suggest that turbulent exchange at these two locations is very different, which is expected given that the underlying surface roughness is substantially different.

Figure 7.

Correlations between NH₃ fluxes and (a) MARGA NH₃ concentration (forest), (b) friction velocity (forest), and (c) friction velocity (grassland). Median values in each bin are shown as squares and error bars denote the interquartile ranges. (d) Friction velocities at the forest and grassland during overlapping periods. The means are shown as blue diamonds. Boxes and whiskers follow the same convention as the other box plots in this work.

We also examined biogeochemical drivers as another possible reason for the differences in fluxes between the forest and grassland. Figure 8 shows the NH₄⁺ concentrations and pH levels from the collected samples at the two locations, as well as the derived dimensionless NH₃ emission potentials. Due to the small sample size (N = 5 in each category), medians were used for comparison instead of means. Soil NH₄⁺ concentrations are larger in the grassland compared to the forest but pH is lightly lower, leading to overall similar soil emission potentials (median Γ_soil ~ 40) between the two ecosystems. Soil compensation points (Γ_soil @ 25 °C; Nemitz et al., 2001) derived from overall median Γ_soil are ~ 0.3 μg NH₃ m⁻³ in both the forest and grassland. Emission potentials of litter on the soil surface (Γ_litter) were larger than for soil in both ecosystems. While some seasonal differences are evident, particularly for pH of the forest litter, generally higher concentrations of NH₄⁺ in the grassland litter translate to overall larger Γ_litter (median = 144) compared to the forest (median = 81). Median Γ_litter correspond to Γ_soil @ 25 °C of 0.6 and 1.0 μg NH₃ m⁻³ in the forest and grassland, respectively. The most notable difference between the forest and grassland is the chemistry of the live vegetation. Vegetation emission potentials (Γ_veg) are comparable to Γ_litter in the forest but significantly higher than Γ_soil and Γ_litter in the grassland. The combination of higher NH₄⁺ content and pH in the grassland lead to much larger Γ_veg (median = 2741) compared to the forest (median = 147). Median Γ_veg corresponds to Γ_veg @ 25 °C of 1.0 and 19.0 μg NH₃ m⁻³ in the forest and grassland, respectively. Due to dampened turbulent exchange at the forest floor (deep canopy) and the potential for recapture of any ground emissions by the overlying canopy (higher leaf area index compared to grass), soil and litter NH₃ exchange likely play a smaller role in the net canopy-scale NH₃ exchange in the forest compared to the grassland. Taken together these factors suggest that the much higher Γ_veg in the grassland may explain the general tendency of the grassland to be a net source of NH₃ where the forest is a net sink.

Figure 8.

Median values of (a) NH₄⁺ concentration, (b) pH, and (c) derived emission potential of NH₃ in soil, ground litter, and live vegetation at the forest and grassland. Values are shown for the August and November sample collection periods as well as the overall medians across the two periods. NH₄⁺ concentration unit: mg NH₄⁺-N per kg of fresh material. Note that y axes in (a) and (c) are in log scale.

3.4. Impact of Atmospheric Transport

To visualize air flows and their influence on NH₃ deposition to Duke Forest, we analyzed HYSPLIT back trajectories using the method described in Section 2.5. Figure 9a shows the wind rose diagram from the forest tower anemometer during the quality-controlled flux periods, the distribution of which was representative of winds over the whole study. Frequencies of occurrence of HYSPLIT back trajectory during the same periods are plotted in Figure 9b, and Figure 9c shows further-filtered frequencies during times when NH₃ deposition was observed by the forest system. The map in Figures 9b and 9c covers a 4°×4° geographical area centered on the study site, corresponding to a typical area along the air parcels’ possible paths within a day under an average wind speed of 2 m s⁻¹ measured during the study. The area was divided into 400 grids with a 0.1° resolution. The occurrence of trajectories within each grid was counted and the frequencies of occurrence were then calculated to visualize the overall pattern.

Figure 9.

(a) Wind rose diagram from the forest tower wind anemometer during the filtered flux periods. (b) Frequencies of the hourly HYSPLIT back trajectories originating from the site during the same periods in (a). (c) Frequencies of the hourly HYSPLIT back trajectories during periods of NH₃ deposition to the forest. The upper limit of the color scale in (b) and (c) was set to 0.2 to improve the clarity of the overall patterns, but the highest frequencies were greater than 0.2.

Figures 9a and 9b reveal that the back trajectory frequencies generally aligned with wind directions from the wind anemometer, indicating consistency between modeled meteorology and measurements. The incoming directions of air parcels during the deposition periods (Figure 9c) did not show a distinguishable difference compared to the overall distribution (Figure 9b), indicating that the deposition was not driven by air flows from a particular direction. While the predominant incoming direction gradually shifted from easterly to westerly during the study, the overall pattern suggests that the Raleigh-Durham urban area located east of the site was likely the primary source of NH₃ for Duke Forest, with possible influences from croplands further upwind. Few animal facilities exist in the northeast direction, but two industrial facilities in the upwind Durham area were identified with NH₃ emissions greater than 1 ton yr⁻¹ among the top 50 NH₃ emitters in the state (National Emissions Inventory, 2021). A smaller fetch was found in the southwesterly direction within 0.5° of the site, which was likely caused by local effects that deflected air flows from easterly directions. Winds from the southeasterly direction, where most swine productions are located, were less frequent during the study, accounting for only 17% of the time. A full-year analysis of hourly winds measured by a wind vane on a 10-m tower at the grassland showed that southeasterly winds were only observed 9% of the time in 2017 and occurred most frequently in spring (MesoWest, 2019), which was not covered by this study. Additional measurements are needed to assess the potential for enhanced deposition during transport of NH₃ from agricultural sources in eastern North Carolina.

4. Summary and Implications

Despite the negative effects of NH₃ on human society and natural environments, NH₃ emissions and deposition are still poorly quantified in natural ecosystems due to uncertainties in modeling and challenges in measuring NH₃. To provide more insight on the bidirectional exchange of NH₃ in forests, we measured EC NH₃ fluxes using open-path QCL-based sensors at a deciduous forest and an adjacent grassland in southeastern U.S. from August to November 2017. The NH₃ sensors achieved 10 Hz precisions of 0.17 ppbv (forest) and 0.23 ppbv (grassland) in the field. The mean fluxes measured were −7.3 ng NH₃-N m⁻² s⁻¹ (forest) and 3.2 ng NH₃-N m⁻² s⁻¹ (grassland) with an overall random error of 28% and 27%, respectively. Our results showed that the forest canopies serve as a sink for NH₃, broadly consistent with the literature. On the other hand, small net emissions of NH₃ were observed at the grassland, likely due to the higher emission potentials of NH₃ derived from biogeochemical measurements. Diurnal variations of NH₃ fluxes were observed at both locations. Hourly HYSPLIT back trajectories originating from the site identified the Raleigh-Durham urban area as the primary source of NH₃ and found no clear distinction in incoming air directions between the deposition periods and overall distribution.

Our results highlight the importance of accounting for the spatial variabilities of NH₃ fluxes between adjacent ecosystems. For investigating the impacts of NH₃ on PM_2.5 formation, spatial heterogeneity should be considered when upscaling fluxes from point measurements or within a grid when using CTMs. Models need to be able to resolve specific vegetation types or at least broad vegetation categories respective to regional differences, and therefore will require accurate parameterizations of the compensation points and resistances. For evaluating N_r deposition and exceedance of critical loads in the ecosystem, the spatial variability of NH₃ exchange processes should also be considered when downscaling modeled NH₃ fluxes, because sub-grid variabilities are often averaged out at the grid scale (Walker et al., 2019). In addition, continuous monitoring of ambient NH₃ levels and surface characteristics in a broader range of forest environments across the world is needed, as CTM parameterizations such as emission potentials currently rely heavily on European studies, causing large uncertainties in NH₃ dry deposition estimates when applied elsewhere (Dennis et al., 2010).

Vertical measurements of NH₃ fluxes above and within the forest canopy, and directly at the forest floor are needed to further understand the exchange processes within canopy, soil, and ground litter. Fast measurements at sub-hourly scales are necessary to capture the dynamics because fluxes vary quickly within a day. However, due to the fully-grown canopies in the summertime, turbulence is often not well developed on the forest floor. The conditions, therefore, may be challenging for EC deployments. Other techniques such as automatic chambers or relaxed eddy accumulation may serve as alternative options for this purpose. Our results may underestimate fluxes if extrapolated to annual scales because winds from the swine production source region were rare during the study. Longer studies, ideally over multiple years, will help to unveil the seasonal, annual, and interannual variabilities of NH₃ fluxes. Similarly, collocated measurements of biogeochemistry such as vegetation and soil properties over finer temporal resolutions (e.g., daily to weekly) should accompany flux measurements to help characterize the underlying mechanisms of the different exchange pathways for NH₃. Finally, while we applied several tests to remove periods with low turbulence that are more susceptible to advection, we were unable to perform a full energy balance closure analysis due to the lack of net radiation and soil heat flux measurements. Upon investigating the relationship between CO₂ flux and friction velocity, we found no evidence of flux underestimation during low-turbulence periods. Nonetheless, energy balance closure analyses should be included in future eddy covariance NH₃ measurements as advection can result in significant underestimation of fluxes, which have been documented in CO₂ flux studies, especially over sloped surfaces (Aubinet et al., 2003; Lee & Hu, 2002).

Highlights.

• NH₃ flux was measured concurrently over a deciduous forest and adjacent grassland

• Net deposition and emission were observed over forest and grassland, respectively

• NH₃ fluxes peaked in middays and afternoons with negligible flux overnight

• Grassland emissions are consistent with soil and vegetation biogeochemical drivers

• Forest canopies act as a sink of NH₃ coming from upwind sources

Data Availability

The 30-min averaged meteorological and flux data used in this study are archived on the Princeton DataSpace (permanent doi to be added in the final version if accepted). The 10 Hz raw data are available upon request.