Ob

KCa

b

c

a

ARRA

KNCOQ

1

it(tkaot2c

iF

v

h0

Agricultural and Forest Meteorology 213 (2015) 193–202

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology

journa l homepage: www.e lsev ier .com/ locate /agr formet

pen-path eddy covariance measurements of ammonia fluxes from aeef cattle feedlot

ang Sun a,b,1, Lei Tao a,b,1, David J. Miller a,b,2, Mark A. Zondlo a,b,∗, Kira B. Shonkwiler c,hristina Nash c, Jay M. Ham c

Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ 08544, USACenter for Mid-Infrared Technologies for Health and the Environment, NSF-ERC, Princeton, NJ 08544, USADepartment of Soil and Crop Sciences, Colorado State University, Fort Collins, CO 80523, USA

r t i c l e i n f o

rticle history:eceived 13 October 2014eceived in revised form 26 May 2015ccepted 10 June 2015

eywords:H3 fluxattle feedlotpen-path eddy covarianceuantum cascade laser

a b s t r a c t

Eddy covariance (EC) measurements of NH3 fluxes from a cattle feedlot were made with a high-precision,fast-response (20 Hz) open-path laser-based sensor. The sensor employed a continuous wave, quantumcascade (QC) laser and targeted an isolated absorption feature of NH3 at 9.06 �m. It was deployed ona 5-m tall flux tower beside a 22,000-animal cattle feedlot in Colorado, USA for two weeks. Sensibleheat, latent heat, CO2, and CH4 EC fluxes were measured concurrently on the tower. The open-path NH3

sensor showed a comparable time response to well-established commercial open-path sensors for CO2

and H2O. The average high-frequency flux loss over the measurement period was 6.6%, mainly resultingfrom sample path averaging. The sensor showed significant improvement over NH3 EC fluxes measuredby closed-path sensors. The measured NH3 EC fluxes were well-correlated with latent heat EC fluxes.

−1 −1

During the measurement period, the average daily NH3 EC flux was 31.7 kg ha d . The flux-variancerelationship was used to further validate the performance of the NH3 EC flux measurement. A 1� detectionlimit of 1.3 ± 0.5 ng m−2 s−1 for NH3 fluxes measured in 30-min intervals was achieved in this field test.This suite of measurements enabled the evaluation of livestock NH3 emissions at unprecedented temporalresolution and accuracy in the context of other important agricultural trace gases.

© 2015 Elsevier B.V. All rights reserved.

. Introduction

Atmospheric ammonia (NH3) is the dominant alkaline speciesn the atmosphere. It neutralizes gaseous nitric and sulfuric acidso form ammoniated aerosols, which cause human health hazardsPaulot and Jacob, 2014), degrade visibility, and modify the radia-ive forcing of the global climate (IPCC, 2013). Ammonia is also aey compound in the global nitrogen cycle. The deposition of NH3nd ammoniated aerosols contributes to the critical exceedancef nitrogen loading in ecosystems in regions of intense agricul-

ural NH3 sources and threatens those ecosystems’ health (Krupa,003). However, understanding of atmospheric NH3 is limited, andurrent NH3 emission inventories have high uncertainties due to

∗ Corresponding author at: Department of Civil and Environmental Engineer-ng, EQuad E209A, 59 Olden Street, Princeton University, Princeton, NJ 08544, USA.ax: +1 609 258 2799.

E-mail address: mzondlo@princeton.edu (M.A. Zondlo).1 These authors contributed equally to this work.2 Present address: Institute at Brown for Environment and Society, Brown Uni-

ersity, Providence, RI 02912, USA.

ttp://dx.doi.org/10.1016/j.agrformet.2015.06.007168-1923/© 2015 Elsevier B.V. All rights reserved.

a lack of observational constraints (Clarisse et al., 2009; Shephardet al., 2011). Livestock production is thought to be the largest sourceof NH3 emissions globally. For example, in the National EmissionInventory (NEI), livestock production accounts for 54% of total U.S.NH3 emissions (EPA, 2013). Comparisons between atmosphericmodeling and satellite observations imply that NH3 emissions arewidely underestimated in U.S. agricultural regions (Clarisse et al.,2009; Heald et al., 2012; Schiferl et al., 2014; Zhu et al., 2013). Basedon aircraft observations, NH3 emissions from Southern Californialivestock production are estimated to be 3–20 times larger thanemission inventories (Nowak et al., 2012).

Micrometeorological methods (e.g., aerodynamic gradientmethod (AGM), relaxed eddy accumulation (REA), path-integratedtechniques coupled with backward dispersion models, and eddycovariance (EC)) are often thought as the most desirable methodsto measure NH3 emission fluxes from livestock farming activities,because they measure fluxes over large integrated footprint areas

without disturbing the surface or the animals (Baum and Ham,2009; Todd et al., 2011; Whitehead et al., 2008). When applyingthese methods, stationarity, adequate fetch, and chemical reac-tions that occur within the turbulent transport time scale need to

1 rest Meteorology 213 (2015) 193–202

berbesWeoamSdawec

sfmWtaQsu2uf

atfpnmftcvHsN(2apEapN

ubtf2ififlmwlws

200

400

600

800

1000North

South

West East

[m]

Fetch filtered by footprintFetch from all other 30−min segmentsFeedlot boundary

Fig. 1. Map of the feedlot boundaries (red line) and the upwind distances thatrecover 70% of the flux during the measurement period (points). The measurementtower was located at the origin. Red points denote periods that sampled within the

94 K. Sun et al. / Agricultural and Fo

e carefully considered. The EC method is the most direct, leastmpirical and least error-prone approach, but it requires that theesponse time and sampling frequency of the NH3 sensor shoulde on the order of 10 Hz to resolve the fluxes carried by smallddies (Kroon and Hensen, 2007). Recently, high-resolution NH3ensors including tunable laser spectroscopy (Ferrara et al., 2012;

hitehead et al., 2008) and chemical ionization mass spectrom-try (Sintermann et al., 2011) have demonstrated some successf NH3 EC flux in the field. However, due to the strong surfaceffinity of NH3 molecules, no existing closed-path NH3 sensor caneet the optimal response time for EC flux measurements (∼0.1 s).

intermann et al. (2011) reported a short time constant of 0.77 sue to dynamic gas exchange within the closed-path system and

long time constant of 23.06 s due to adsorption/desorption evenhen heating the drift tube of a PTR-MS instrument to 170 ◦ C. Ellis

t al. (2010) reported a short time constant of 0.4 s and a long timeonstant of 15 s for a QC-TDLAS instrument.

The adsorption and desorption of NH3 to the instrumenturfaces and sample tubing introduce significant damping of high-requency signals and significant underestimation of fluxes in

ost potential measurement situations (Brodeur et al., 2009).hitehead et al. (2008) compared NH3 EC measurements from

wo 10 Hz laser spectrometers (QCLAS and TDLAS) and tested thesegainst the aerodynamic gradient method. They found that theCLAS underestimated the NH3 flux by 47% for unexplained rea-

ons. Ferrara et al. (2012) investigated methods to correct the fluxnderestimation, and they calculated that the flux loss ranged from3% to 43% depending upon the correction method. Moreover, thisnderestimation could be dependent on flux magnitude, which

urther complicated the flux corrections (Sintermann et al., 2011).In contrast to closed-path configurations, an open-path design

voids significant adsorption/desorption effects between NH3 andhe instrument surfaces and the consequent damping of high-requency fluctuations, even when sampling high (100 s ppbv topmv) concentrations directly downwind of sources. There is noeed to use a heavy and power-hungry sampling pump, therebyaking the open-path configuration more portable and adaptable

or continuous measurements at remote EC sites. Furthermore,he open-path techniques avoid filters and heated inlets thatause ambiguity between gas phase NH3 and that derived fromolatilization of ammoniated aerosols. Open-path EC fluxes of2O, CO2 and CH4 are routinely measured with commercial sen-

ors from LICOR (McDermitt et al., 2011). Path-integrated, remoteH3 measurements have been demonstrated by open-path FTIRs

Bjorneberg et al., 2009), DOAS (Mount et al., 2002; Volten et al.,012), and TDLAS (Todd et al., 2011), but these sensing systemsll require long path lengths rather than compact sensor foot-rints needed for EC flux measurements. In the field, open-pathC measurements also face the challenges of dust, precipitation,nd spectroscopic/density influences from water vapor and tem-erature. Thus far, no field-based, open-path EC measurements ofH3 have been demonstrated in the literature.

In this study, we demonstrate EC measurements of NH3 fluxessing an open-path, quantum cascade (QC) laser-based sensor at aeef cattle feedlot. The sensor has recently been demonstrated inhe field in non-EC applications and is capable of high-resolution,ast-response, and high-sensitivity measurements (Miller et al.,014; Sun et al., 2014). The sensor of Miller et al., 2014 was mod-

fied with significant improvements to ensure its performance foreld EC measurements. Sensible heat, latent heat, CO2, and CH4 ECuxes were measured concurrently at the same site. The perfor-ance of the open-path NH3 sensor was evaluated and compared

ith the other commercial sensors. The NH3 EC flux was corre-

ated with the other EC fluxes, and the flux-variance relationshipas used to further validate the measurements. This suite of mea-

urements enabled the evaluation of livestock NH3 emissions at

feedlot boundary, while blue points show the fetches from all other 30-min sam-pling periods. (For interpretation of the references to color in this figure legend, thereader is referred to the web version of this article.)

unprecedented temporal resolution and accuracy in the context ofother important tracers.

2. Materials and methods

2.1. Field site and flux footprint

Measurements were performed at a commercial cattle feedlot innorthern Colorado with a cattle population of 22,000 head. The cat-tle were stocked at about 20 m2 per animal and weighed between300 and 600 kg depending on their age and days on feed. The crudeprotein in the animal diet was between 12.7 and 14.3%. The areareceives about 360 mm of precipitation annually, and the terrainis flat with slopes less than 5%. The EC measurements lasted fromNovember 12 to November 26, 2013. No significant precipitationwas experienced during the measurement period except for a snowevent on November 20–21, 2013. The measurement system wasdeployed on the east edge of the pens, providing a fetch in excessof 700 m when winds were from the prevailing westerly direction.Fig. 1 shows the outline of cattle pen area using geographic coor-dinates taken from a georeferenced digital map (ArcMap 10.2, Esri,Redlands, CA) with the sensors located at the origin of the plot. Thesampling area consisted mainly of feedlot pens holding beef cat-tle. An analytical footprint model was used to calculate the sourcearea contributing to the flux measurements (Hsieh et al., 2000).The displacement height and roughness length were determinedfollowing the methods reported in Baum et al. (2008), where thecattle density and average weight were very similar to this study.Only the one-dimensional flux density distribution was calculatedas a function of upwind distance. The crosswind distribution wasneglected because the average standard deviation of wind direc-tion within each 30-min interval was found to be small (13 ± 6◦).An upwind distance that recovered 70% of the flux was used as

the fetch requirement, and only cases with fetch inside the feed-lot boundaries were used in the following analyses (Baum et al.,2008). Roads and alleys were included in the footprint filtering butthese areas accounted for <10% of the total area within the feedlot

rest Meteorology 213 (2015) 193–202 195

beoa

2

wtdpatplrmlfiAtdtmt

a(l(t1r2Lshsslw(trpi

(tNT(ttSafct7aws

s

et al., 2001). The sonic temperature was also converted to ambienttemperature following Liu et al. (2001). The 20 Hz raw data weresynchronized to the sonic anemometer using the cross-covariance

Table 1Sensors’ separations from the sonic anemometer using the midpoint of the opticalpath.

K. Sun et al. / Agricultural and Fo

oundary. Therefore, the fluxes represent the emissions from thentire feedlot, weighted averaged by the footprint density, insteadf emissions from only the pen areas. The fetch distribution beforend after the footprint filtering is also shown in Fig. 1.

.2. Instrumentation

An open-path, quantum cascade (QC) laser-based NH3 sensoras used to measure NH3 concentrations at 20 Hz. The spec-

roscopy and optomechanics of the sensor have been described inetail previously by Miller et al. (2014). Only a brief summary isresented here. The QC laser scanned an NH3 absorption featuret 9.062 �m. The feature was selected to minimize the broad spec-ral interferences from neighboring trace gas species at ambientressure while also optimizing the sensitivity. Wavelength modu-

ation spectroscopy (WMS) was used to enhance the signal-to-noiseatio and resolve the air-broadened NH3 absorption lines. A Herriot

ulti-pass cell achieved a total path length of 46 m with a cell baseength of 0.7 m and a sampling volume of 3 L. A virtual lock-in ampli-er retrieved NH3 concentrations from the averaged waveforms.n inline ethylene (C2H4) reference cell provided a constant signal

hat was continuously probed and can be used to account for sensorrift (Sun et al., 2013). The sensor was calibrated by direct absorp-ion spectroscopy before and after the field deployment using the

ethod described in Miller et al. (2014). The total uncertainty ofhe NH3 measurements was 0.15 �g m−3 ±10%.

The Miller et al. (2014) sensor was modified to ensure its suit-bility for EC measurements. A new QC laser from Corning Inc.Corning, NY) was used in the EC sensor to provide five timesarger optical power and hence to enhance the signal-to-noise ratioSNR) than in the original design. The current sensor has achievedhe same SNR at 20 Hz as Miller et al. (2014) demonstrated at0 Hz. The waveform acquisition and spectral fitting were sepa-ated into parallel routines to enable real-time data processing at0 Hz. Molybdenum (Mo) mirrors (Rocky Mountain Instrument Co.,afayette, CO) were used in the multi-pass cell instead of protected-ilver coated glass mirrors to enhance system performance in theighly dusty feedlot environment. Uncoated Mo mirrors have highurface quality and reflectivity for the wavelength used in NH3 mea-urement. Mo mirrors are capable of guiding high power mid-IRasers for welding/cutting in debris-spattering environments and

ithstand frequent mechanical cleaning with organic detergentsTao et al., 2015). The sensor was horizontally mounted to minimizehe accumulation of precipitation and dust on the mirrors. The mir-ors were cleaned five times during the two-week measurementeriod, and the cleaned mirror showed no noticeable deterioration

n reflectivity or surface quality during the campaign.A high-precision closed-path cavity ring down spectrometer

CRDS, G2103; Picarro Inc., Sunnyvale, CA) was also operated athe site from November 19 to November 26, 2013 to measureH3 with a sampling inlet 0.3 m from the open-path NH3 sensor.he CRDS has a sensitivity of 0.75 �g m−3 at 5 s, excellent stability72 h drift < ±0.15 ppbv), and a claimed response time <30 s (fallime from 90% to 10% of initial concentration). The air inlet forhe CRDS was fitted with a Chemcomb 3500 cartridge (Thermocientific, Electron Corp., East Greenbush, NY) that included bothn impactor plate for course particulate and a 1-�m PTFE filterollowed by a cellulose filter to remove fine particulate. The honey-omb denuders normally used in the Chemcomb were removed sohe cartridge only provided filtration. Air was routed to the CRDS at

LPM through 5 m of Teflon tubing that was heated to reduce thedsorption/desorption effects of NH3 on the surface. The CRDS data

ere used to intercompare concentrations with the open-path NH3

ensor as shown in Section 3.1.Wind vector and high frequency sonic temperature were mea-

ured by a 3D sonic anemometer (CSAT3, Campbell Scientifc, Inc.,

Fig. 2. Experimental setup of the open-path NH3 sensor, sonic anemometer, andLI-7500A.

Logan, UT). Carbon dioxide and H2O were measured by an open-path non-dispersive infrared (NDIR) gas sensor (LI-7500A, LI-CORBiosciences, Lincoln, NE). Methane was measured by an open-path near infrared gas sensor (LI-7700, McDermitt et al., 2011).The LI-7500A and LI-7700 were calibrated by a NOAA ESRL stan-dard before and after the EC measurements (393.444 ± 0.003 ppmvCO2, 0.07 ppmv uncertainty; 1871.26 ± 0.3 ppbv CH4, 3 ppbv uncer-tainty). These data were logged at 20 Hz using the same laptopcomputer for the NH3 sensor. The LI-7500A also measured ambi-ent pressure. The layout of the EC sensing system is shown in Fig. 2.The dimensions, relative positions, heights, and orientations of allthe sensors were precisely measured and used in the flux correc-tions in the following section. The sensors’ separations from theanemometer are summarized in Table 1.

All the open-path sensors were mounted on a motorized tramattached to a tower, and the NH3 measurement height was set to4.92 m above ground. The measurement height gave up to 700 mof fetch when the winds were from the prevailing west direction(Fig. 1).

2.3. Data processing

Eddy covariance fluxes and some other micrometeorologicalparameters were calculated using EddyPro software (Advanced5.0.0, LI-COR Biosciences, Inc., Lincoln, NE) over 30 min durationsfrom the 20 Hz raw data. The results from EddyPro were comparedto the results from MATLAB (Mathworks, Inc., Natick, MA) pro-cessing codes. Both computational programs gave similar resultsin all cases reported here. The general flux calculation procedurefollowed the standard FluxNet methodology (McDermitt et al.,2011). Data pre-processing included raw data despiking, doublecoordinate rotation, and sensible heat correction due to humidityeffects on sonic temperature measurements (Baum et al., 2008; Liu

NH3 sensor LI-7500A LI-7700

Northward separation (cm) −15 −31 −156Eastward separation (cm) 25 24 62Vertical separation (cm) −28 −4 −12

1 rest M

mwiev(t

wBHtcobcawecd

eccldssOct7sea

tMlctflttmafCsus

2

squlflvcvTt

96 K. Sun et al. / Agricultural and Fo

ethod. The spectroscopic effects of pressure, temperature, andater vapor fluctuations on NH3 flux were corrected using a sim-

lar method as the open-path CH4 flux measurements (McDermittt al., 2011). The absorption line broadening effect caused by waterapor was quantified using line width data reported by Owen et al.2013), where spectroscopic parameters of the same NH3 absorp-ion feature at 9.062 �m were measured experimentally.

When calculating the EC fluxes, the WPL density correctionsere applied for NH3, CO2, H2O and CH4 (Webb et al., 1980).

ecause the NH3 sensor did not measure H2O simultaneously, the2O measurements from LI-7500A were used in the WPL correc-

ions. Because NH3 fluxes were large from the feedlot and the WPLorrection term was independent of NH3 fluxes, the WPL correctionnly changed the NH3 fluxes by less than ±2% and was negligi-le for large fluxes (changed a flux of 100 �g m−2 s−1 by 0.2%). Theorrections for the spectroscopic effects of pressure, temperature,nd water vapor were larger (within ±6% of NH3 fluxes) because aide range of ambient temperature and humidity conditions were

ncountered during the measurement period. The spectroscopicorrections are proportional to fluxes at given meteorological con-itions, so they should be considered even for large fluxes.

Sensor self-heating effect specific to LI-7500A and spectroscopicffects specific to LI-7700 were corrected as extra terms of the WPLorrection (Burba et al., 2008; McDermitt et al., 2011). The poweronsumption of the NH3 sensor head was dominated by the QCaser (∼6 W) and the thermoelectric cooler of the laser, which wasependent on ambient temperature and solar radiation on the heatink. During the measurement period in November, the power con-umption of the thermoelectric cooler was usually less than 5 W.verall, the total power consumption of the NH3 sensor head wasomparable to LI-7500A (12 W during normal operation). However,he sampling volume of the NH3 sensor was much larger than LI-500A and more comparable to LI-7700. Because the LI-7700 washown previously to not be impacted by self-heating (McDermittt al., 2011), the self-heating effects for the NH3 sensor were alsossumed to be negligible.

Other calculations within each 30 min block included fric-ion velocity, u∗ (m s−1), standard deviation of the scalars, and

onin–Obukhov stability parameter (z/L), where L is the Obukhovength (m). The random errors of EC flux measurements were cal-ulated following Salesky et al. (2012) by fitting a power law tohe standard deviations of the local fluxes and extrapolating to theux averaging time of 30 min. The random errors resulted fromime averaging over an insufficient period for the temporal meano converge to the ensemble mean, and they are considered as the

ajor sources of uncertainties in EC flux measurements (Saleskynd Chamecki, 2012; Salesky et al., 2012). The diagnostic outputsrom the LI-7500A and LI-7700 were used to exclude questionableO2, H2O and CH4 flux data due to low signal intensity or sen-or malfunctioning. Detector signal intensity of NH3 sensor wassed to exclude NH3 flux data when the laser beam was blocked bynow/dust.

.4. Quality control methods for EC flux measurements

Cospectral analysis is a powerful tool to assess the capability of aensor to detect gas concentration fluctuations over a range of fre-uencies. This is the first time that the open-path NH3 sensor wassed in an EC flux configuration, and its relatively long optical path

ength (0.7 m multi-pass cell base length) could result in additionalux loss by averaging eddies that were smaller than the samplingolume (Moore, 1986). Therefore, it was important to compare the

ospectra of NH3 flux and the cospectra of other scalar fluxes toalidate the efficiency of the NH3 sensor for EC flux measurements.he power spectra for all the scalars, and the cospectra betweenhe vertical wind speed and all other scalars were calculated from

eteorology 213 (2015) 193–202

20 Hz time series, similar to Baum et al. (2008). Frequency responsecorrection functions (Moore, 1986) were calculated to quantify thehigh-frequency flux attenuation due to sample path averaging andsensor-anemometer separation. The sensors’ dimensions, relativepositions, and orientations were taken from the field measure-ments described above.

Another way to quantify the attenuation of high frequencyfluxes is using the ogive functions, which are defined as the flux-normalized integrated cospectra from low to high frequencies.With the hypotheses that the cospectra of all the scalars in theinertial sub-range are similar (Kaimal et al., 1972), and that thecospectra of sensible heat are undamped, the flux attenuations athigh frequencies were calculated by comparing the ogive functionof trace gas fluxes and the sensible heat flux. The high frequencyflux losses were quantified following the same method proposedby Ferrara et al. (2012). The NH3 flux ogive was fit against the ogiveof the sensible heat in the middle of the inertial sub-range, wherethe scalar cospectra showed the closest similarity. The attenuationof high frequency fluxes was then indicated by the right-end of thefitted NH3 flux ogive.

Periods with non-stationary conditions were excluded by usingquality control filters proposed by Foken and Wichura (1996). Devi-ation from stationarity was examined by comparing the 30-mincovariance between the vertical wind speed (w) and the horizontalwind speed (u), or the scalars (temperature, H2O, CO2, and NH3)with the 5-min blocks of data for the same 30-min period:

�ST = | 〈w′S′

5〉 − w′S′30

w′S′30

|

here S represents u or any scalars, and the subscripts 5 and 30denote the 5-min and the 30-min covariances. A filtering crite-rion of �ST < 30% was applied for u and all the scalars to excludenon-stationary runs in the flux quantification.

The flux-variance relationship was used to validate the ECmeasurements by testing the development of turbulence in themeasurement conditions and examining the similarity betweenNH3 EC measurements against other scalars. According to theMonin–Obukhov similarity theory (MOST), in an ideal station-ary, homogeneous turbulent flow, flux-variance relationships havewell-established asymptotic behaviors under very unstable condi-tions (z/L → − ∞) because the friction velocity must disappear fromthe equation due to the insignificant role of shear-generated tur-bulence. As a result, under very unstable conditions, flux-variancerelationships should follow a ‘−1/3’ power law as follows (Hsiehet al., 2008):

�x|x∗| = Cx

(− zL

)− 13

where �x is the standard deviation of the scalar quantity, x; x∗ = w′x′u∗

is the EC flux normalized by friction velocity; and Cx is a similarityconstant. �x

|x∗| is also called the dimensionless standard deviationof scalar x. This asymptotic behavior has been verified extensivelyin experimental studies (Monin and Yaglom, 1975) and is used asa quality control for EC measurements (Ferrara et al., 2012; Zhaoet al., 2013).

The chemical reaction of NH3 was considered negligible in thisstudy. There were no significant sources of nitric or sulfuric acid inthe feedlot surface layer, so rapid gas-to-particle conversion wasnot expected. Previous modeling and airborne observations alsoshow that photochemical reactions and chemical reactions withgaseous acids are unlikely to affect dispersion in the local boundary

layer above NH3 hotspots (Loubet et al., 2009; Staebler et al., 2009).Baum and Ham (2009) measured aerosol fluxes in a similar feedlotenvironment using relaxed eddy accumulation (REA) method andfound that the ammonium fluxes were negligible.

K. Sun et al. / Agricultural and Forest Meteorology 213 (2015) 193–202 197

13

13.5

14

14.5

15

15.5

16

16.5

CO

2 [mm

ol/m

3 ]

130

140

150

160

170

180

190

200H

2O [m

mol

/m3 ]

0 0.2 0.4 0.6 0.820

40

60

80

100

120

140

160

180

Time [s]

NH

3 [μm

ol/m

3 ]

NH3

CO2

H2O

Fig. 3. Synchronized 20 Hz measurements of NH3 (blue), CO2 (black), and H2O (red)dsi

3

3

hiligsotmtmtis7o2swCemLt

sicso5cftsbivCtd

5 5.5 6 6.5 7 7.5 80

2

4

6

8

Time [hour]

NH

3 mix

ing

ratio

[ppm

v]

(a) Open−path QCLPicarro

0 5 100

2

4

6

8

10

Picarro [ppmv]

Ope

n−pa

th Q

CL

[ppm

v]

(b)

3s data

1:1 line

Slope: 1.144±0.007Intercept: −0.19±0.01

0 0.5 1 1.5 20

0.5

1

1.5

2

Picarro [ppmv]

Ope

n−pa

th Q

CL

[ppm

v]

(c)

Intercept: −0.099±0.07Slope: 1.076±0.0730 min data1:1 line

Fig. 4. (a) Measurements by the open-path NH3 sensor and Picarro CRDS over threehours out of the 24-h intercomparison period. (b) Correlation between open-pathNH3 sensor and Picarro CRDS during the 24 h, where the open-path NH3 sensor data

uring a 0.8 s period. The concentrations dropped rapidly due to a sudden windhift. (For interpretation of the references to color in this figure legend, the readers referred to the web version of this article.)

. Results and discussion

.1. Sensor performance and intercomparison

Sensor response time has been a major challenge for currentigh-resolution NH3 sensors to measure EC flux. The response time

s also a crucial parameter to compensate the high-frequency fluxoss. Traditionally, sensor response time for closed-path sensorss characterized by examining rapid changes in inflow samplingas concentration. This is, however, not feasible for open-path sen-ors. Miller et al. (2014) demonstrated the time response of thepen-path NH3 measurements at 5 Hz resolution and concludedhat the sensor response time was less than 1 s. In the current

anuscript, however, the measurement resolution was improvedo 20 Hz, so the response time needed to be no more than 0.05 s to

ake each measurement independent. To more precisely charac-erize the response time of open-path NH3 sensor, we comparedts response to a transient natural concentration change with theimultaneous response of the LI-7500A. The response time of the LI-500A is thought to be negligible in EC flux measurements due to itspen-path configuration and short path length (12.5 cm) (LI-COR,012). Fig. 3 shows the responses of 20 Hz NH3, CO2, and H2O mea-urements to an ambient concentration change caused by a suddenind shift. The NH3 measurements were strongly correlated with

O2 and H2O measurements, indicating that these are real ambi-nt concentration changes. There is no discernible delay of the NH3easurements compared to the CO2 and H2O measurements of the

I-7500A. Therefore, no additional flux loss due to sensor responseime needed to be considered for the NH3 flux.

The accuracy and long-term stability of the open-path sen-or in the field condition were assessed by a co-located, 24-hntercomparison with a Picarro CRDS, a state-of-the-art commer-ial closed-path sensor. The advantage of fast response is clearlyhown in Fig. 4a, where the concentration time series for thepen-path NH3 sensor and the Picarro CRDS are presented from:00 to 8:00 local time on November 20. The measured NH3 con-entrations were characterized by a large dynamic range due torequently changing wind speed and direction. The 20 Hz data fromhe open-path NH3 sensor were first averaged to the Picarro CRDSampling time interval of ∼3 s and then shifted to offset the time lagetween the two sensors. Near the feedlot, NH3 concentrations var-

ed rapidly from tens of ppbv to near 10 ppmv. Large concentration

ariations led to a significant hysteresis effect in the closed-pathRDS system. Lower concentrations were biased high compared tohe open-path sensor after high concentration spikes occurred. Theamping of high frequency concentration changes was also signif-

were averaged to the time resolution of Picarro. (c) Correlation of the two sensorswhen both datasets were averaged to 30 min.

icant when comparing the open-path and closed-path time series(e.g., hour 6.8–7.3 in Fig. 4a). Large scatter between the two mea-surements was observed at the initial 3 s resolution of the Picarro(Fig. 4b). The open-path sensor measured many high concentrationspikes, which were not captured by the Picarro CRDS. The regres-sion analysis yielded a slope of 1.144 ± 0.007 and a R2 of 0.838.A better correlation was seen in the 30-min averaged time seriesin Fig. 4c (R2 = 0.962; slope = 1.076 ± 0.07). This indicated that bothsensors could resolve ambient concentration change at a 30-mintime scale. Although the sampling conditions and time responsesof the two sensors were different, their measurements agreed towithin 10%, which is well within the combined uncertainties ofboth sensors.

3.2. Performance of EC flux measurements

The cross-covariance function method can be used to charac-terize the time lag between two time series. Unlike the EC systemsusing closed-path sensors, where the time lag can vary with pumpspeed or water vapor concentration (Sintermann et al., 2011), theopen-path NH3 time lag varied less than 0.2 s throughout the mea-surement period, and can be easily corrected using the same routinedesigned for LI-7500A and LI-7700. The precision and detectionlimit of EC flux was calculated as the standard deviation of the cross-covariance function within windows far outside the true time lag,usually 70–120 s away (Kroon and Hensen, 2007; Wienhold et al.,1994). The detection limit against a zero flux was estimated as thisstandard deviation in each of the five consecutive 30-min intervalswith minimal fluxes over the measurement period. Fig. 5 shows thecross-covariance function from the lowest flux period at 5–7:30 AMon November 23, 2013. The analysis yielded a 1� detection limitof 1.3 ± 0.5 ng m−2 s−1 for five 30-min intervals during this period.Also note that the peak of open-path NH3 cross-covariance functionwas essentially symmetric, in contrast to the similar plot shown by

Ferrara et al. (2012), where the cross-covariance function was notsymmetric with respect to positive and negative time lags due tosensor hysteresis effects.

198 K. Sun et al. / Agricultural and Forest M

Fa

w|bcadct1cnntiocccc(ic

flt

validate the NH3 EC measurements and shows that MOST assump-

Fa

ig. 5. Cross-covariance between 20 Hz NH3 concentration and vertical wind speeds a function of time shift. Data collected at 5–7:30 AM on November 23, 2013.

Ensemble averages of cospectra were calculated for periodsith sufficient turbulence and surface sensible heat flux (u∗ > 0.15,

H| > 15 W m−2) and when fluxes originated from the feedlot filteredy the footprint model as discussed in Section 2.1. The sensible heatospectra are usually thought to be undamped and represent thectual conditions of turbulent flux distribution in the frequencyomain (McDermitt et al., 2011). At a sufficient height above theanopy with well-developed turbulence, the sensible heat cospec-ra should conform closely to the Kaimal model (Kaimal et al.,972). The cases for unstable conditions (z/L < −0.05) and stableonditions (z/L > 0.05) are illustrated in Fig. 6. The cospectra wereormalized by the correspondent fluxes and plotted versus theon-dimensional frequency. Under both unstable and stable condi-ions, the sensible heat cospectra followed the Kaimal model closelyn the inertial sub-range with the −4/3 slope, which verified the usef sensible heat cospectra as a reference for other gas species. Theospectra in stable conditions peaked at higher frequency (Fig. 6b)ompared to the unstable conditions (Fig. 6a). Therefore, at stableonditions, significant fluxes were transported by smaller eddiesompared to unstable conditions. The cospectra of NH3, CO2, H2Overy similar to CO2, not shown) and CH4 fluxes showed close sim-larity to the sensible heat cospectra in both stable and unstableonditions.

All the gas sensors showed some attenuation of high frequencyuxes. However, the attenuation of NH3 flux was comparable tohose of the CO2 and CH4 fluxes. The cospectra of NH3 flux started

10−3

10−2

10−1

100

10 1

10−4

10−3

10−2

10−1

100

Normalized frequency (fZ/U)

fCw

x(f)/

Fw

x

(a)

w’NH3’

w’CO2’

w’T’w’CH

4’

−4/3 slope

ig. 6. Ensemble averaged cospectra at (a) unstable conditions (z/L < −0.05) and (b) stablso shown for reference.

eteorology 213 (2015) 193–202

to deviate from the sensible heat cospectra significantly at thedimensionless frequency of about 2, whereas the cospectra ofclosed-path NH3 measurements deviated from the sensible heatcospectra at about 0.1 (Whitehead et al., 2008). Compared to thecospectra calculated from closed-path sensor measurements, theopen-path NH3 sensor resolved eddies at frequencies up to 20times higher. High-frequency flux loss was calculated using thefrequency response correction functions derived by Moore (1986)and the sensible heat cospectra as a reference for each 30-mininterval. The average NH3 flux loss over the measurement periodwas 6.6%, which is substantially smaller than those of closed-pathEC measurements and within the instrument uncertainties. Theogive functions of NH3 fluxes also gave a similar high frequencyflux loss value of 5% in unstable conditions. Overall, the frequencyresponse of the NH3 sensor was comparable to the well establishedcommercial open-path sensor LI-7500A at the same measurementconditions and comparable to the results reported in other studiesof open-path EC flux (McDermitt et al., 2011). The CH4 cospec-tra showed the largest deviation from the sensible heat cospectra,mainly caused by the large separation between the CH4 sensor andthe anemometer (1.68 m) due to spatial constraints in the exper-imental setup. Because of the large uncertainties associated withhigh-frequency CH4 flux loss, we only examined the temporal vari-ation of CH4 fluxes and the correlation between CO2 and CH4 fluxesin the following sections, without drawing any quantitative conclu-sions on CH4 emissions.

3.3. EC flux quality control using flux-variance relationships

Under unstable conditions, the flux-variance relationships oftemperature, water vapor, CO2, and NH3 showed strong similar-ity and followed nicely the theoretical ‘−1/3’ power law derivedfrom MOST (Fig. 7a–d). The similarity constant, Cx, was calculatedwith a bisquare robust fitting routine under unstable conditions(z/L < −0.1) and also shown in the plot. The similarity constantsfor temperature, water vapor, NH3, and CO2 were indistinguish-able from one another within the uncertainties. These similarityconstant values were also consistent with the range reported inthe literature (0.9–1.4) (Asanuma and Brutsaert, 1999; Assoulineet al., 2008; Hsieh et al., 2008; Monin and Yaglom, 1975). Moredetailed analyses on the flux-variance relationships derived fromthis feedlot dataset can be found at Sun et al., 2015. This helps to

tions were generally satisfied. However, the similarity constantfor CH4, as shown in Fig. 7e, was significantly larger than thoseof the other scalars. As discussed in the cospectra analyses, the

10−3

10−2

10−1

100

10 1

10−4

10−3

10−2

10−1

100

Normalized frequency (fZ/U)

fCw

x(f)/

Fw

x

(b)

w’NH3’

w’CO2’

w’T’w’CH

4’

−4/3 slope

le conditions (z/L > 0.05). The −4/3 slopes from the Kaimal model (black lines) are

K. Sun et al. / Agricultural and Forest Meteorology 213 (2015) 193–202 199

10−2

10−1

100

101

100

101

(a)

−z/L

σ T/|T

*|

CT = 1.04±0.05

10−2

10−1

100

101

100

101

−z/L

σ q/|q*|

(b)

Cq = 1.01±0.04

10−2

10−1

100

10 1

100

101

−z/L

σ NH

3/|NH

3*| (c)

CNH

3

= 1.00±0.04

10−2

10−1

100

10 1

100

101

−z/L

|σC

O2/C

O2*

| (d)

CCO

2

= 1.04±0.05

10−2

10−1

100

10 1

100

101

−z/L

σ CH

4/|CH

4*| (e)

CCH

4

= 1.25 ±0.09

F O2 (d), and CH4 (e) under unstable conditions. The similarity constant Cx was calculatedf

asdsiuoof

3

tTtscslwca3sroWflT

14 16 18 20 22 24 26−10

0

10

20

30

Air

T [ °

C]

(a)

0

100

200

300

Win

d di

rect

ion

[deg

ree]

14 16 18 20 22 24 260

5

10

15

Win

d sp

eed

[m/s

]

(b)

14 16 18 20 22 24 260

50

100

150

NH

3 flux

[μg

m−

2 s−

1 ]

Date in Nov 2013

(c)All measurements

Filtered measurements

Snow

Fig. 8. Time series of ambient temperature (a), wind speed and direction (b), andNH3 EC flux (c) during the measurement period. The major data missing was due tothe snow event on November 20–21, 2013. (c) shows the flux data when the concen-

ig. 7. Flux-variance relationships for temperature (a), water vapor (b), NH3 (c), Crom a fitting routine using flux-variance relationships of each scalar.

ttenuation of CH4 flux was large at high frequency due to largeensor-anemometer separation, but this did not cause similaramping effects on the variance of CH4. Therefore, the dimen-ionless standard deviation of CH4 was likely to be overestimated,nducing a biased value of CCH4 . The results of this quality controlsing flux-variance relationship were consistent with the previ-us cospectra analyses. Because of the significant loss of CH4 flux,nly qualitative analyses were performed using the CH4 data in theollowing sections.

.4. Flux measurements

Data with concurrent valid NH3, CO2, H2O, and CH4 concen-ration measurements covered 76% of the measurement period.he largest data gap (7%) occurred on November 20–21, 2013 dueo a snow event. The rejected data due to precipitation were theame for NH3, CH4, and CO2/H2O sensors because the sensors wereleaned of accumulated snow at the same time after the snowtopped. Because the NH3 sensor mirror surfaces were perpendicu-ar to the ground, accumulation of water on the mirrors after rainfall

as minimized. The other shorter gaps were caused by logisti-al reasons like power outages and instrument problems. Averagembient temperature, wind speed and wind direction over each0-min period were derived from the 20 Hz anemometer mea-urements (Fig. 8a and b). Fig. 8c presents NH3 EC fluxes withandom errors. Additional data filtering was then applied to include

nly data with u∗ > 0.15 m s−1, |H| > 5 W m−2, �ST < 30% (Foken andichura, 1996; Li et al., 2012; Todd et al., 2011), and when the

uxes came from the feedlot area according to the footprint model.he u∗ threshold was chosen arbitrarily and in a conservative way.

tration measurements were available (black dots) and after all the meteorologicaland footprint filtering were applied (red dots). Error bars in (c) denote the randomerrors associated with NH3 EC flux. (For interpretation of the references to color inthis figure legend, the reader is referred to the web version of this article.)

200 K. Sun et al. / Agricultural and Forest Meteorology 213 (2015) 193–202

0 4 8 12 16 20−50

0

50

100

Local time [h]

Sen

sibl

e he

at [W

/m2 ]

(a)

0 4 8 12 16 200

50

100

Local time [h]

Late

nt h

eat f

lux

[W/m

2 ]

(b)

0 4 8 12 16 200

50

100

Local time [h]

NH

3 EC

flux

[μg/

m2 /s

]

(c)

0 4 8 12 16 200

2

4

6

8

Local time [h]

CO

2 EC

flux

[mg/

m2 /s

]

(d)

0 4 8 12 16 200

50

100

Local time [h]

CH

4 EC

flux

[μg/

m2 /s

]

(e)PercentilesMean

Fig. 9. Composite diurnal flux variation curves were calculated for sensible heat (a), latent heat (b), NH3 EC flux (c), CO2 EC flux (d), and CH4 EC flux (e) based on the localt centilo on of to

WlTc

sbfi5esmm(tntalStmhst

NfNt

ime of day. The boxes and whiskers denote the 5th, 25th, 50th, 75th, and 95th perriginated within the feedlot boundaries were included in the plot. (For interpretatif this article.)

ith these filters, there were valid EC flux data from the feed-ot over 20% of the measurement period, as also shown in Fig. 8c.he average relative random errors of NH3 EC fluxes were 10% andomparable to those of CO2 (11%), H2O (12%), and CH4 (15%).

Composite diurnal flux variation curves were calculated for sen-ible heat, latent heat, NH3 EC flux, CO2 EC flux, and CH4 EC fluxased on the local time of day using only 30-min intervals afterootprint and meteorological filtering. A total of 124 runs werencluded (Fig. 9). Besides the mean values, percentiles (5th, 25th,0th, 75th, and 95th) outlining the distributions of all data withinach bin are also illustrated in Fig. 9. The composite fluxes for sen-ible and latent heat followed a diurnal pattern with peaks duringidday. These fluxes agreed very well with the sensible/latent heateasurements in November at a feedlot in Kansas by Baum et al.

2008). Some positive values of sensible heat flux appeared duringhe night on November 21–24, 2013 during the snow storm. Theighttime ambient temperature dropped to below −10 ◦C, whereashe surface temperature stayed around 0 ◦C. The positive temper-ture gradient between the surface and the atmospheric surfaceayer was likely the cause of nighttime positive sensible heat flux.ignificant positive latent heat fluxes were found at night whenhere was a strong wind, likely due to strong forced convection. The

ajor water vapor source at a cattle feedlot was urination, whichappened continuously. Animal breathing was another constantource of water vapor, the magnitude of which was roughly 10% ofhe urinated water.

The composite curve for NH3 followed a similar pattern. Both

H3 and water fluxes originated from the ground surface of the

eedlot. The equilibrium constant governing the concentrations ofH3/NH4

+ in the soil solution and the Henry’s constant governinghe evaporation of NH3 from liquid to gas phase are both tempera-

es, and the red circles denote the means. Only periods with at least 70% of the fluxhe references to color in this figure legend, the reader is referred to the web version

ture dependent (Baum and Ham, 2009). Hence higher temperaturesduring the midday favored the emissions of both water vaporand NH3. The diurnally-averaged NH3 EC flux was 36.7 �g m−2 s−1

(31.7 kg ha−1 d−1), which is comparable to the results of Todd et al.(2011) from two cattle feedlots in the High Plains of Texas duringwinter time at comparable ambient temperature. In contrast, thediurnal patterns for CO2 and CH4 EC fluxes were different from NH3with only weak enhancements in the late afternoon and evening.CO2 fluxes from feedlots mainly originated from cattle respirationand ruminant digestion (Baum et al., 2008). CH4 fluxes originatedfrom enteric fermentation and associated with feeding practices(Johnson et al., 1994). Detailed discussion on CO2 and CH4 EC fluxesis beyond the scope of this work, though the results are broadlyconsistent with the temporal evolution of the expected sources.For example, Kinsman et al. (1995) also observed significant corre-lation between CO2 and CH4 fluxes. Baum et al. (2008) also showedthat diurnal variation in CO2 flux was small above a large cattlefeedlot with weak enhancement in the late afternoon and evening.Although the feeding patterns may be different between thesestudies, the significant correlation and relatively small diurnal vari-ations of CO2 and CH4 fluxes were commonly observed for differentfeeding facilities.

As indicated by Fig. 9, the composite fluxes of both water vaporand NH3 were strongly temperature-dependent and showed dis-tinct diurnal patterns overall. In contrast, CO2 and CH4 fluxeswere not linked to temperature. To further examine the relation-ships between fluxes, water vapor and NH3 fluxes are correlated

in Fig. 10a, and CO2 and CH4 fluxes are correlated in Fig. 10b.The correlation between water vapor and NH3 EC fluxes (Fig. 10a)was significant (Pearson correlation coefficient r = 0.88) and arosebecause both were driven by temperature and originated from

K. Sun et al. / Agricultural and Forest Meteorology 213 (2015) 193–202 201

0 10 20 30 40 500

20

40

60

80

100

H2O flux [mg/m2/s]

NH

3 flux

[μg/

m2 /s

]

(a)

r = 0.88

0 2 4 6 80

20

40

60

80

100

120

CO2 flux [mg/m2/s]

CH

4 flux

[μg/

m2 /s

]

(b)

r = 0.93

F earsor this a

tpNTNlwbiieeioauwr

4

uaisTiaswpcm3rahmadaspfls

fl

ig. 10. Correlations between the EC fluxes of water vapor, CO2, NH3, and CH4. Peferences to color in this figure legend, the reader is referred to the web version of

he surface. The feedlot surface was generally wet, whereas theatches of urine were not uniformly distributed. A snow event onovember 20–21 also contributed to extra water on the surface.hese may have caused some differences between water vapor andH3 source regions and hence degraded the correlation. The corre-

ation between CO2 and CH4 EC fluxes (Fig. 10b) was very significantith r = 0.93. Unlike NH3, both CO2 and CH4 from cattle feedlot are

iogenic with much weaker diurnal variations. There was no signif-cant temperature dependence of these CO2 and CH4 sources. Thismplies that it is possible to use NH3:CO2 or NH3:CH4 concentrationnhancement ratios to characterize the temporal variability of NH3missions. Because there are much less relative diurnal variationsn CO2 and CH4 fluxes than in NH3 fluxes, the diurnal variationsf these enhancement ratios generally represent the diurnal vari-tions of NH3 fluxes. Moreover, these enhancement ratios can besed to estimate NH3 fluxes if only CO2 or CH4 fluxes are available,here NH3 fluxes are estimated as the product of the enhancement

atios and CO2 or CH4 fluxes.

. Conclusion

An open-path, quantum cascade (QC) laser based sensor wassed for eddy covariance studies to characterize NH3 fluxes from

cattle feedlot. The key improvements to the original sensorncluded a more efficient light source, a higher frequency mea-urement of 20 Hz, and Mo mirrors for the optical multi-pass cell.he new Mo mirrors did not show any significant degradation dur-

ng two-week measurements in the dusty feedlot environmentfter routine manual cleaning. The mean concentrations mea-ured by the open-path sensor were in excellent agreement (8%)ith a state-of-the-art commercial NH3 sensor, whereas the open-

ath sensor demonstrated much better time response, the majorhallenge for NH3 EC measurements. The open-path NH3 EC fluxeasurements had a 1� detection limit of 1.3 ± 0.5 ng m−2 s−1 for

0-min intervals. The shapes of NH3 EC flux cospectra closelyesembled those of CO2 from the LI-7500A, the sonic sensible heat,nd the theoretical curves in the inertial sub-range. The averageigh-frequency flux loss over the measurement period was 6.6%,ainly resulting from sampling path averaging. Ogive analyses

lso indicated that the high-frequency NH3 flux loss due to variousamping effects was comparable to those of the CO2 flux. The aver-ge high-frequency flux loss over the measurement period showedignificant improvement over NH3 EC fluxes measured by closed-ath sensors. Flux-variance relationships of multiple trace gas

uxes validate NH3 EC flux measurements against Monin–Obukhovimilarity theory.

The average daily NH3 EC flux was 3.17 g m−2 d−1. Ammonia ECux showed a distinct diurnal pattern with high peak in the midday

n correlation coefficients are also shown in the figure. (For interpretation of therticle.)

and significantly correlated with latent heat flux. However, neitherCO2 nor CH4 EC fluxes correlated significantly with NH3 flux. In thefuture, an automated mirror-cleaning technique with momentarybursts of air will be used in the sensor, as described in Ham et al.(2012). This would make the sensor more suitable for long termdeployment at a feedlot and maintenance by inexperienced users.The open-path sensor enables the most direct NH3 flux measure-ments and represents a robust approach for future investigation ofNH3 flux dynamics in agricultural regions.

Acknowledgements

The authors acknowledge Prathap Ramamurthy, Dan Li, Qi Li,Josh DiGangi, Anthony O’Brien, Minghui Diao, Da Pan, Levi Golston,and Elie Bou-Zeid at Princeton University for helpful discussions.Special thanks to the research group of Azer Yalin at Colorado StateUniversity for the use of laboratory space. The feedlot work wassupported by the Center for Mid-Infrared Technologies for Healthand the Environment (MIRTHE) under National Science FoundationGrant No. EEC-0540832. Kang Sun acknowledges support by a NASAEarth and Space Science Fellowship (NN12AN64H). Additional sup-port was provided by USEPA grant R834551. Its contents are solelythe responsibility of the grantee and do not necessarily representthe official views of the USEPA. Further, USEPA does not endorsethe purchase of any commercial products or services mentionedin the publication. This work was partially supported by the USDANational Institute of Food and Agriculture project 2012-03407. Thiswork was also partially supported by grant number 2012-67021-19978 from the USDA/NSF National Robotics Initiative. We thanktwo anonymous reviewers for very helpful feedback and commentson the manuscript.

References

Asanuma, J., Brutsaert, W., 1999. Turbulence variance characteristics oftemperature and humidity in the unstable atmospheric surface layer above avariable pine forest. Water Resources Res. 35, 515–521, http://dx.doi.org/10.1029/1998wr900051

Assouline, S., Tyler, S.W., Tanny, J., Cohen, S., Bou-Zeid, E., Parlange, M.B., Katul,G.G., 2008. Evaporation from three water bodies of different sizes andclimates: measurements and scaling analysis. Adv. Water Resources 31,160–172, http://dx.doi.org/10.1016/j.advwatres.2007.07.003

Baum, K.A., Ham, J.M., 2009. Adaptation of a speciation sampling cartridge formeasuring ammonia flux from cattle feedlots using relaxed eddyaccumulation. Atmos. Environ. 43, 1753–1759, http://dx.doi.org/10.1016/j.atmosenv.2008.12.021

Baum, K.A., Ham, J.M., Brunsell, N.A., Coyne, P.I., 2008. Surface boundary layer ofcattle feedlots: Implications for air emissions measurement. Agric. For.Meteorol. 148, 1882–1893, http://dx.doi.org/10.1016/j.agrformet.2008.06.017

Bjorneberg, D.L., Leytem, A.B., Westermann, D.T., Griffiths, P.R., Shao, L., Pollard,M.J., 2009. Measurement of atmospheric ammonia, methane, and nitrous oxide

2 rest M

B

B

C

E

E

F

F

H

H

H

H

I

J

K

K

K

K

L

L

L

L

M

M

M

M

Meteorol. 148, 495–516, http://dx.doi.org/10.1007/s10546-013-9829-7Zhu, L., Henze, D.K., Cady-Pereira, K.E., Shephard, M.W., Luo, M., Pinder, R.W., Bash,

02 K. Sun et al. / Agricultural and Fo

at a concentrated dairy production facility in southern Idaho using open-pathFTIR spectrometry. Trans. ASABE 52, 1749–1756.

rodeur, J.J., Warland, J.S., Staebler, R.M., Wagner-Riddle, C., 2009. Technical note:laboratory evaluation of a tunable diode laser system for eddy covariancemeasurements of ammonia flux. Agric. For. Meteorol. 149, 385–391, http://dx.doi.org/10.1016/J.Agrformet.2008.08.009

urba, G.G., Mcdermitt, D.K., Grelle, A., Anderson, D.J., Xu, L., 2008. Addressing theinfluence of instrument surface heat exchange on the measurements of CO2

flux from open-path gas analyzers. Global Change Biol. 14, 1854–1876, http://dx.doi.org/10.1111/j.1365-2486.2008.01606.x

larisse, L., Clerbaux, C., Dentener, F., Hurtmans, D., Coheur, P.F., 2009. Globalammonia distribution derived from infrared satellite observations. Nat. Geosci.2, 479–483, http://dx.doi.org/10.1038/Ngeo551

llis, R.A., Murphy, J.G., Pattey, E., van Haarlem, R., O’Brien, J.M., Herndon, S.C.,2010. Characterizing a quantum cascade tunable infrared laser differentialabsorption spectrometer (QC-TILDAS) for measurements of atmosphericammonia. Atmos. Meas. Tech. 3, 397–406.

PA, 2013. The National Emission Inventory 2011 [WWW Document]. URL http://www.epa.gov/ttnchie1/net/2011inventory.html

errara, R.M., Loubet, B., Di Tommasi, P., Bertolini, T., Magliulo, V., Cellier, P.,Eugster, W., Rana, G., 2012. Eddy covariance measurement of ammonia fluxes:comparison of high frequency correction methodologies. Agric. For. Meteorol.158–159, 30–42, http://dx.doi.org/10.1016/j.agrformet.2012.02.001

oken, T., Wichura, B., 1996. Tools for quality assessment of surface-based fluxmeasurements. Agric. For. Meteorol. 78, 83–105, http://dx.doi.org/10.1016/0168-1923(95) 2,248-1

am, J., Williams, C., Shonkwiler, K., 2012. Automated Dust Blow-off System forthe LI-7700 Methane Analyzer.

eald, C.L., Lee, T., Benedict, K.B., Schwandner, F.M., Li, Y., Clarisse, L., Hurtmans,D.R., Van Damme, M., Clerbaux, C., Coheur, P.-F., Philip, S., Martin, R.V., Pye,H.O.T., 2012. Atmospheric ammonia and particulate inorganic nitrogen overthe United States. Atmos. Chem. Phys. 12, 10295–10312, http://dx.doi.org/10.5194/acp-12-10295-2012

sieh, C., Katul, G., Chi, T., 2000. An approximate analytical model for footprintestimation of scalar fluxes in thermally stratified atmospheric flows. Adv.Water Resources 23, 765–772.

sieh, C.-I., Lai, M.-C., Hsia, Y.-J., Chang, T.-J., 2008. Estimation of sensible heat,water vapor, and CO2 fluxes using the flux-variance method. Int. J.Biometeorol. 52, 521–533, http://dx.doi.org/10.1007/s00484-008-0149-4

PCC, 2013. Intergovernmental Panel on Climate Change. Cambridge UniversityPress.

ohnson, K., Huyler, M., Westberg, H., Lamb, B., Zimmerman, P., 1994.Measurement of methane emissions from ruminant livestock using a sulfurhexafluoride tracer technique. Environ. Sci. Technol. 28, 359–362, http://dx.doi.org/10.1021/es00051a025

aimal, J.C., Wyngaard, J.C., Izumi, Y., Coté, O.R., 1972. Spectral characteristics ofsurface layer turbulence. Q. J. R. Meteorol. Soc. 98, 563–589.

insman, R., Sauer, F.D., Jackson, H.A., Wolynetz, M.S., 1995. Methane and carbondioxide emissions from dairy cows in full lactation monitored over asix-month period. J. Dairy Sci. 78, 2760–2766.

roon, P., Hensen, A., 2007. Suitability of quantum cascade laser spectroscopy forCH and N2O eddy covariance flux measurements. Biogeosciences 4, 715–728.

rupa, S.V., 2003. Effects of atmospheric ammonia (NH3) on terrestrial vegetation:a review. Environ. Pollut. 124, 179–221, http://dx.doi.org/10.1016/S0269-7491(02)00434-7

i, D., Bou-Zeid, E., De Bruin, H.A.R., 2012. Monin–Obukhov similarity functions forthe structure parameters of temperature and humidity. Boundary-LayerMeteorol. 145, 45–67, http://dx.doi.org/10.1007/s10546-011-9660-y

I-COR, 2012. LI-7500A Open Path CO2/H2O Gas Analyzer Instruction Manual[WWW Document]. LI-COR Biosci. Inc., Lincoln, NE http://envsupport.licor.com

iu, H., Peters, G., Foken, T., 2001. New equations for sonic temperature varianceand buoyancy heat flux with an omnidirectional sonic anemometer.Boundary-Layer Meteorol. 100, 459–468.

oubet, B., Asman, W.A.H., Theobald, M.R., Hertel, O., Tang, Y.S., Robin, P.,Hassouna, M., Dämmgen, U., Genermont, S., Cellier, P., 2009. Ammoniadeposition near hot spots: processes, models and monitoring methods. In:Atmospheric Ammonia. Springer, pp. 205–267.

cDermitt, D., Burba, G., Xu, L., Anderson, T., Komissarov, A., Riensche, B.,Schedlbauer, J., Starr, G., Zona, D., Oechel, W., 2011. A new low-power,open-path instrument for measuring methane flux by eddy covariance. Appl.Phys. B 102, 391–405.

iller, D.J., Sun, K., Tao, L., Khan, M.A., Zondlo, M.A., 2014. Open-path, quantumcascade-laser-based sensor for high-resolution atmospheric ammoniameasurements. Atmos. Meas. Tech., 81–93, http://dx.doi.org/10.5194/amt-7-

81-2014

onin, A.S., Yaglom, A.M., 1975. Statistical Fluid Mechanics. MIT press, Cambridge,Massachusetts.

oore, C.J., 1986. Frequency response corrections for eddy correlation systems.Boundary-Layer Meteorol 37, 17–35, http://dx.doi.org/10.1007/BF00122754

eteorology 213 (2015) 193–202

Mount, G., Rumburg, B., Havig, J., Lamb, B., Westberg, H., Yonge, D., Johnson, K.,Kincaid, R., 2002. Measurement of atmospheric ammonia at a dairy usingdifferential optical absorption spectroscopy in the mid-ultraviolet. Atmos.Envir. 36, 1799–1810.

Nowak, J.B., Neuman, J.A., Bahreini, R., Middlebrook, A.M., Holloway, J.S., McKeen,S.A., Parrish, D.D., Ryerson, T.B., Trainer, M., 2012. Ammonia sources in theCalifornia South Coast Air Basin and their impact on ammonium nitrateformation. Geophys. Res. Lett. 39, http://dx.doi.org/10.1029/2012GL051197

Owen, K., Es-sebbar, E., Farooq, A., 2013. Measurements of NH3 linestrengths andcollisional broadening coefficients in N2, O2, CO2, and H2O near 1103.46 cm−1.J. Quant. Spectrosc. Radiat. Transfer 121, 56–68, http://dx.doi.org/10.1016/j.jqsrt.2013.02.001

Paulot, F., Jacob, D.J., 2014. Hidden cost of U. S. agricultural exports: particulatematter from ammonia emissions. Environ. Sci. Technol. 48, 903–908, http://dx.doi.org/10.1021/es4034793

Salesky, S.T., Chamecki, M., 2012. Random errors in turbulence measurements inthe atmospheric surface layer: implications for Monin–Obukhov similaritytheory. J. Atmos. Sci. 69, 3700–3714, http://dx.doi.org/10.1175/JAS-D-12-096.1

Salesky, S.T., Chamecki, M., Dias, N.L., 2012. Estimating the random error ineddy-covariance based fluxes and other turbulence statistics: the filteringmethod. Boundary-Layer Meteorol. 144, 113–135, http://dx.doi.org/10.1007/s10546-012-9710-0

Schiferl, L.D., Heald, C.L., Nowak, J.B., Holloway, J.S., Neuman, J.A., Bahreini, R.,Pollack, I.B., Ryerson, T.B., Wiedinmyer, C., Murphy, J.G., 2014. An investigationof ammonia and inorganic particulate matter in California during the CalNexcampaign. J. Geophys. Res. Atmos. 119, 1883–1902, http://dx.doi.org/10.1002/2013JD020765

Shephard, M.W., Cady-Pereira, K.E., Luo, M., Henze, D.K., Pinder, R.W., Walker, J.T.,Rinsland, C.P., Bash, J.O., Zhu, L., Payne, V.H., Clarisse, L., 2011. TES ammoniaretrieval strategy and global observations of the spatial and seasonalvariability of ammonia. Atmos. Chem. Phys 11, 10743–10763.

Sintermann, J., Spirig, C., Jordan, A., Kuhn, U., Ammann, C., Neftel, A., 2011. Eddycovariance flux measurements of ammonia by high temperature chemicalionisation mass spectrometry. Atmos. Meas. Tech. 4, 599–616, http://dx.doi.org/10.5194/amt-4-599-2011

Staebler, R.M., McGinn, S.M., Crenna, B.P., Flesch, T.K., Hayden, K.L., Li, S.-M., 2009.Three-dimensional characterization of the ammonia plume from a beef cattlefeedlot. Atmos. Environ. 43, 6091–6099.

Sun, K., Tao, L., Miller, D., Khan, M.A., Zondlo, M., 2013. Inline multi-harmoniccalibration method for open-path atmospheric ammonia measurements. Appl.Phys. B 110, 213–222, http://dx.doi.org/10.1007/s00340-012-5231-2

Sun, K., Tao, L., Miller, D.J., Khan, M.A., Zondlo, M.A., 2014. On-road ammoniaemissions characterized by mobile, open-path measurements. Environ. Sci.Technol. 48, 3943–3950, http://dx.doi.org/10.1021/es4047704

Sun, K., Li, D., Tao, L., Zhao, Z., Zondlo, M.A., 2015. Quantifying the influence ofrandom errors in turbulence measurements on scalar similarity in theatmospheric surface layer. Boundary-Layer Meteorol., http://dx.doi.org/10.1007/s10546-015-0047-3

Tao, L., Sun, K., Miller, D.J., Pan, D., Golston, L.M., Zondlo, M.A., 2015. Low-power,open-path mobile sensing platform for high-resolution measurements ofgreenhouse gases and air pollutants. Appl. Phys. B, http://dx.doi.org/10.1007/s00340-015-6069-1

Todd, R.W., Cole, N.A., Rhoades, M.B., Parker, D.B., Casey, K.D., 2011. Daily, monthly,seasonal, and annual ammonia emissions from Southern High Plains cattlefeedyards. J. Environ. Qual. 40, 1–6, http://dx.doi.org/10.2134/jeq2010.0307

Volten, H., Bergwerff, J.B., Haaima, M., Lolkema, D.E., Berkhout, A.J.C., van der Hoff,G.R., Potma, C.J.M., Kruit, R.J.W., van Pul, W.A.J., Swart, D.P.J., 2012. Twoinstruments based on differential optical absorption spectroscopy (DOAS) tomeasure accurate ammonia concentrations in the atmosphere. Atmos. Meas.Tech. 5, 413–427, http://dx.doi.org/10.5194/Amt-5-413-2012

Webb, E.K., Pearman, G.I., Leuning, R., 1980. Correction of flux measurements fordensity effects due to heat and water vapour transfer. Q. J. R. Meteorol. Soc.106, 85–100.

Whitehead, J.D., Twigg, M., Famulari, D., Nemitz, E., Sutton, M.A., Gallagher, M.W.,Fowler, D., 2008. Evaluation of laser absorption spectroscopic techniques foreddy covariance flux measurements of ammonia. Environ. Sci. Technol. 42,2041–2046, http://dx.doi.org/10.1021/Es071596u

Wienhold, F.G., Frahm, H., Harris, G.W., 1994. Measurements of N2O fluxes fromfertilized grassland using a fast response tunable diode laser spectrometer. J.Geophys. Res. Atmos. 99.

Zhao, Z., Gao, Z., Li, D., Bi, X., Liu, C., Liao, F., 2013. Scalar Flux–Gradient relationshipsunder unstable conditions over water in coastal regions. Boundary-Layer

J.O., Jeong, G.R., 2013. Constraining U. S. ammonia emissions using TES remotesensing observations and the GEOS-Chem adjoint model. J. Geophys. Res.Atmos. 118, 3355–3368, http://dx.doi.org/10.1002/jgrd.50166.