Agricultural and Forest Meteorology 96 (1999) 19±38
Nitrous oxide ¯uxes from a fertilised maize crop using micrometeorological and chamber methods Patricia Lavillea,*, Corinne Jambertb, Pierre Celliera, Robert Delmasb b
a INRA, Unite de Bioclimatologie, Grignon-Thiverval, France Laboratoire d'AeÂrologie, Universite Paul Sabatier, Toulouse, France
Received 4 August 1998; received in revised form 14 May 1999; accepted 1 June 1999
Abstract Emissions of N2O from maize ®elds irrigated and fertilised with 200 kg N haÿ1 were measured using several chamber and micrometeorological techniques at a single site in the Landes de Gascogne (France). The methods were compared over similar periods, but the sampling areas covered range from 0.18 m2 for the chamber methods to 104±105 m2 for the micrometeorological methods. Nitrous oxide concentration was measured using electron capture gas chromatography (GC) for the chamber technique. Micrometeorological ¯uxes were determined by eddy covariance and gradient methods with N2O mixing ratios measured using fast-response tunable diode laser absorption spectroscopy. The spatial heterogeneity of N2O emissions was studied with a set of 30 closed chambers. This study showed an anisotropic distribution of N2O emissions which was attributed to the pattern of fertiliser application. A sensitivity analysis of the micrometeorological techniques indicated that uncertainties on ¯ux assessment are mainly due to spatial heterogeneity of N2O emissions. Fluxes measured using chamber and micrometeorological techniques were compared over a 2 h period. Depending on the precision of the techniques, good agreement between the magnitude of the ¯uxes measured by chamber and micrometeorological methods was observed. For micrometeorological methods, the ¯ux ranged from 20 to 400 ng N-N2O mÿ2 sÿ1 and from 25 to 275 ng N-N2O mÿ2 sÿ1 for chamber method. Average relative variations of the ¯ux were similar in both methods, and were mainly dependent on the soil water content. Using micrometeorological measurements, large variations of the ¯ux were observed from 1 h to the next related to variations in the wind direction combined with spatial heterogeneity. # 1999 Elsevier Science B.V. All rights reserved. Keywords: Nitrous oxide emissions; Micrometeorological technique; Chamber technique; Footprint analysis; Tunable diode laser spectrometer
1. Introduction Despite its low abundance in the atmosphere (310 ppbv), N2O plays a major role in global warm*
Corresponding author. Tel.: +1-30-81-55-30; fax: +1-30-81-5563 E-mail address:
[email protected] (P. Laville)
ing due to its long lifetime (130 years) and the radiative forcing of the N2O molecule (about 200 times higher than that of CO2) (Houghton et al., 1992). N2O contributes to global warming potential (GWP) of 5% (Bouwman, 1990). In the stratosphere, the principal mechanisms for N2O destruction are photolysis and the reaction with excited oxygen atoms. The latter reaction is the largest source of
0168-1923/99/$ ± see front matter # 1999 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 8 - 1 9 2 3 ( 9 9 ) 0 0 0 5 4 - 4
20
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
stratospheric NO which is a reactant involved in ozone depletion (Warneck, 1988). In the atmosphere N2O concentration increase is estimated at about 0.3 ppbv yearÿ1. Nevertheless, the net annual emissions, estimated by source and sink strengths, does not balance the observed rate of N2O increase in the atmosphere. It is generally accepted that the most important sources are tropical and agricultural soils. However, there are signi®cant uncertainties on the estimated magnitude of these sources which ranges from 2.7 to 5.7 and 1.8 to 5.3 Tg-N yÿ1 (IPCC, 1994), respectively. There is evidence that the ¯uxes of N2O from agricultural land (fertilised ®elds and grasslands) contributes to the global emission of N2O. For these reasons, improving the assessment of N2O ¯uxes from fertilised land is a priority for research. The global N2O emission estimates are based on limited data sets, which cannot account for the large spatial and temporal variability of N2O emission from soils. During the last decade, emissions at the ®eld scale have been principally measured by chamber methods. Such devices are simple and especially adapted to study the relationships between the ¯uxes and soil chemical or microbiological factors. However, they are less effective for quantifying ¯uxes at the landscape level. Indeed, many studies have demonstrated a large spatial and temporal variability of N2O sources at the ®eld scale (e.g., Ambus and Christensen, 1994). Because the area enclosed in a chamber is typically smaller than 1 m2, it is necessary to use a large number of chambers to get a representative estimate of the ¯uxes at ®eld scale. An alternative is to continuously monitor the N2O ¯ux at the ®eld scale by using micrometeorological methods (Hargreaves et al., 1994; Smith et al., 1994; Wiehnold et al., 1994; Wagner-Riddle et al., 1996). These methods require fast gas analysis with a sensitivity better than 1 ppbv. Recent developments in analytical techniques have lead to the possible application of micrometeorological methods to measure N2O emissions. Depending on the height of the sampling tower, micrometeorological methods lead to an integrated ¯ux over areas of 0.01±1 km2. An experiment was carried out near Bordeaux (Landes de Gascogne, France) in a maize ®eld to compare chamber and micrometeorological methods for measurement of N2O emissions at the ®eld scale.
The theory, performance and drawbacks of the chamber and micrometeorological methods for N2O emission estimates are reported there. Major characteristics of each method will be illustrated with the results obtained during this ®eld campaign. 2. Experimental site, climatic conditions This experiment was conducted in a 60 ha fertilised maize ®eld in south western France at 25 km southwest of Bordeaux (458450 N, 08450 W and altitude of 60 m). This site is characteristic of the humid areas of the region converted to agriculture. The soil is a sandy soil (acid humic, podzosoil) with an acidic pH (6, after successive limings). Two hundred kilogram of N haÿ1 were injected into the soil under pressure as anhydrous ammonia. Similar sites in the region were previously studied (Jambert, 1995; Jambert et al., 1997a, b). In these sites, the emissions of nitrogen compounds were measured throughout the year, and more intensively after fertiliser application. The plot received nitrogen on 12 June 1996 while the experiment took place from 11 to 24th June. The cultivated ®eld was bordered by a path in the SE direction (Fig. 1). The distance between rows was 0.8 m and the rows were curvilinear. All treatments (fertilisation, irrigation) were applied according to this geometry. Mean meteorological variables (global radiation, wind direction, temperature, air humidity and rain) were recorded hourly throughout the campaign. Air temperature was high with a mean daily value of about 208C and a maximum of about 308C at midday. Wind was blowing mostly from NE (508N) to SE (1508N) from 12 to 17 June, and then shifted to SW±W direction (210±2758N). Rain occurred on 11, 20 and 21 June. The ®eld was irrigated (20 mm) with a pivot system on 12 and 16 June. Each day, soil water content was measured on six replicates randomly sampled over the ®eld by a gravimetric method, in the 0± 10 cm and 10±20 cm layer. Soil moisture was expressed as volumetric water content and water ®lled pore space (WFPS) considering apparent and bulk soil densities of 1.4 and 2.6 g cmÿ3. The WFPS is currently used, and allows to compare the level of anoxia in different soils, which is one of the main factor controlling N2O emissions (Groffman and Tiedje, 1991).
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
21
Fig. 1. Location of the micrometeorological mast and the chambers in the experimental field. Ec mast: eddy correlation; Gr mast: gradient mast; manual chambers (the filled squares correspond to the chambers numbered `2', `8', and `14' which were used every day throughout the experiment period). For the micrometeorological methods N2O concentrations were measured using tunable diode laser absorption spectrometry (TDLAS).
3. Methodology and instrumentation 3.1. Chamber method A set of 30 static chambers were deployed on three lines of 100 m long joining at the micrometeorological gradient mast (Fig. 1). The median line (Y-axis in Fig. 1) was in the direction 2908N and the angle difference between this axis and the two others was 278. Along the three lines, three chambers were placed every 12 rows. Flux measurements were made on the 30 chambers on 17, 20 and 24 June. Measurements were made twice a day throughout the experiment only on three numbers (2, 8 and 14, see Fig. 1) out of the 30 chambers. Static chambers were made of open PVC cylinders of 44 cm diameter (0.18 m2, in area) and 25 cm height. Cylinders were inserted into the soil to a depth of over 5 cm the day before the experiment. At the start of each measurement period, the cylinders were covered with a removable air tight aluminium lid equipped with a septum. Four air samples of 50 ml were taken over a 2 h period (0±30±60±120 min after closing the
chamber) to estimate the increase in concentration in the chamber. Samples were injected into prevacuumed glass tubes closed with Te¯on septa and crimped with aluminium caps. Over the ®rst hour of integration the N2O concentration increase in the chamber was linear. Therefore, the ¯ux was calculated as the slope of the linear regression of concentration with time over this period. Accumulation kinetics were considered to be signi®cant when the correlation coef®cient was higher than 0.95. Corrections for air density variations were made by measuring the mean air temperature during the 1 h integrating time. The precision of the ¯ux measurement by this method was 10% (Laville et al., 1998) with a detection limit of 2 ng N-N2O mÿ2 sÿ1. Analyses of the samples were performed once a week. The nitrous oxide concentration was measured in the laboratory using gas phase chromatography. The chromatograph is a Girdel GC, equipped with a 63 Ni electron capture detector, two Porapak Q ®lled columns, and a back¯ush system allowing a high frequency analysis (once every 3 min). The precision of the analysis was of 5% in N2O concentration.
22
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
3.2. Micrometeorological methods N2O ¯uxes were measured by the eddy correlation method during the ®rst part of the experiment, from 11 to 17 June. The gradient method was then used, from 18 to 24 June. The location of the masts for each method are shown in Fig. 1. In the eddy correlation method, the N2O ¯ux density was calculated as the mean product of instantaneous vertical wind velocity (w) and concentration (c): F wc. To take into account each elementary eddy contributing to the ¯ux, the wind speed and the N2O concentration were sampled at 10 Hz and the ¯uxes were averaged over periods of 15 min (N = 9000). The product wc may be considered as a sum of the two components, ®rst being the product of the mean vertical wind speed ( w) and mean trace gas density (c), and the second the mean product of the ¯uctuations from the average for the same components (w0 c0 and F wc w0 c0 ). The three components of the wind turbulence structure were rotated in such a way that the mean vertical wind speed was 0. Therefore, the ¯ux is equal to the mean product between w0 and c0 which is the covariance between w and c: F w0 c0
N N 1X 1X wprimec i0
wi ÿ w
ci ÿc i N i1 N i1
R2 w c
(1)
where N is the number of measurements over the averaging period, ci and wi are the instantaneous values of N2O and wind speed for the index of time i; R2, w and c denote the correlation coefficient between wind speed and N2O concentration and the standard deviation (SD) of vertical wind speed and N2O concentration. Wind speed was measured using a three-dimensional sonic anemometer (Gill Instrument, Hampshire, UK, Model 1012R2), which allowed the measurements of u, v and w, the three components of wind speed in an orthogonal system. N2O concentrations were measured with a tunable diode laser absorption spectrometer (TDLAS) developed by Aerodyne Research, Billerica, USA (Kolb et al., 1995). The TDLAS has a multipass cell with an optical path of 36 m for a volume of 0.27 l. The effect of the air density variations due to the air temperature and humidity were eliminated by passing sampled air
previously through a Perma Pure dryer and by monitoring air temperature using a thermocouple (Omega, Stanford, Model TAC80B-AC). Under ®eld conditions, this spectrometer allowed to measure N2O concentrations at 10 and 1 Hz with a precision of 1 and 0.1%, respectively. The wind speed and the nitrous oxide concentration were measured simultaneously at 0.8 m above crop height. Air was driven to the N2O analyser cell located 10 m away through Te¯on tubing. Data were recorded with a Campbell Scienti®c datalogger (Model 21X) that performed analogical-digital conversion data, synchronisation and sent data to a computer at a frequency of 10 Hz. The time lag between air sampling and analysis of the N2O concentration was determined as that giving the maximum correlation coef®cient between w and c. The gradient method, consists in calculating ¯uxes from vertical pro®les of mean wind speed (u) and concentration (c) in the ®rst few meters above the surface. In the constant ¯ux layer, the ¯ux is proportional to the product of u* and c* calculated from the slope of u and c versus a logarithmic vertical height scale, and corrected by atmospheric stability functions, according to the equations: F ÿu c u zÿd zÿd ln u
z ÿcm z0 L k c zÿd zÿd ln ÿch c
z z0 L k
(2) (3) (4)
The terms cm and ch are empirical nondimensional stability correction functions for momentum and heat or gas concentration (Dyer and Hicks, 1970; Brutsaert, 1982), z0 is the roughness length, d is the zero plane deplacement height, k is von Karman constant (k = 0.41) and L is the Monin±Obukhov length which gives an estimate of the stability of the atmosphere. Wind speed, temperature and concentration pro®les were measured at ®ve different heights: 0.1, 0.3, 0.7, 1.2, and 2.0 m above the mean crop height. The zero plane displacement and the roughness length were taken as 2/3 and 0.1 of the canopy height (Hicks et al., 1979), respectively. The wind speed measured with cup anemometers, air temperature with thermocouples, and N2O concentration with TDLAS were measured every second. Concentration pro®les were
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
calculated over a 3 min cycle. Air pumping was switched from one height to the other using solenoid valves. Air samples at each height lasted 36 s, but data were averaged only over the last 26 s to reduce contamination from the previous height. For each height, mean concentration and mean SD were calculated and recorded over 15 min periods (26 5 = 130 data per height). 4. Treatment of micrometeorological data: flux uncertainties Both eddy correlation and gradient methods require a uniform distribution of the sources in the ®eld, and assumes the stationarity of the properties of the air ¯ow. However, many studies demonstrated a large spatial variability of N2O sources at the ®eld scale with mean coef®cient of variations (CVs) between 100 and 200% and sometimes higher (Matthias et al., 1980; Duxbury et al., 1982; Denmead and Raupach, 1993). Using a micrometeorological method in such cases requires to process the data and to estimate ¯ux uncertainties due to the `nonhomogeneous' conditions. In the gradient method, departures from a linear pro®le in a logarithmic vertical scale could be a symptom of a divergence of the experimental conditions from ideal conditions for the method. In the eddy correlation method, the presence of an additional low frequency component in the N2O concentration signal could be a symptom of heterogeneity of the source distribution in the ®eld. In this chapter, the sensitivity analysis of these methods and the estimation of the ¯ux uncertainties are described. 4.1. Gradient method The relative error on the ¯ux is expressed from the partial derivation of the ¯ux (Eq. (2)) as follows: @F 2 @F 2
du
dc 2 dF @u dc s DF Du 2 Dc 2 F u c 2
(5)
where u* and c* are the uncertainties on the friction velocity and the concentration scales deter-
23
mined from the uncertainties of the linear vertical slopes. To compare micrometeorological ¯uxes with the chamber method or to investigate the process of N2O production in relation to environmental variables, such as nitrate or ammonium soil content, soil humidity and temperature, it is necessary to minimise the variation and uncertainty of ¯ux estimates. To overcome this variability, the hourly mean ¯uxes were calculated from the 15 min ¯uxes weighted by the inverse of their mean errors as follows: Fp
4 X
Pi F i
i1
with
Pi
DFi
P4
1
j1
1=DFj
(6) where Fp is the hourly weighted mean, Fi the 15 min flux and Fi the 15 min flux error deriving from Eq. (5), Pi is the weight assigned for each 15 min flux. The resultant hourly mean error was then: v u 4 uX (7) DFp t P2i
DFi 2 i1
Fig. 2(a) shows, for 22 June the mean 15 min difference of N2O concentrations at heights 0.1 and 2 m above the canopy, the higher and lower inlets in the gradient method. The SD of the concentration difference presented as error bars were calculated as the square root of the sum of squares of SDs of mean concentrations over the 5 26 s. On 22 June, mean 15 min differences varied between ÿ0.5 and 50 ppbv. Lower differences were observed during midday and higher at night. The concentration difference and the SD were similar, with an average SD of 5.7 ppbv and an averaged difference of 6.2 ppbv. The large SD could be related to spatial variability of N2O sources and modi®cations of the `footprint' due to wind direction and speed changes during the integrating time of 5 26 s. Compared to the measured SD, the instrumental resolution was high with a mean of about 0.2 ppbv. Uncertainties in the ¯ux measurements on 22 June were estimated from Eq. (5). During this day the averaged relative error (F/F) for a 15 min integrating time was 51%. The study of all data sets gave a mean relative error of 69% for 15 min integrating time. Fig. 2(b) shows the contributions of the error linked
24
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
Fig. 2. (a) Mean difference of atmospheric concentration on 22 June between 0.1 and 2 m above the canopy. Error bars are the SD of the differences. (b) Relative contribution of the error linked with N2O concentration gradient and friction velocity on 22 June. Data are averaged over 15 min.
to the N2O concentration gradient (Dc =c ) and to the wind speed gradient (Du =u ). On average, the uncertainty in N2O concentration pro®le was large (35 and 57% for the whole data set) compared to that on the wind speed pro®le (19 and 14% for the whole data set). However, errors related to the wind gradient were larger at night because of the low wind speed and low turbulence close to the cup anemometer stalling speed (0.3 m sÿ1). For the N2O concentration, main uncertainties occurred at midday with large unstable conditions, with rapid diffusion of the pollutant, and thus weak differences in N2O concentration gradients which could be close to the TDLAS instrumental resolution. Fig. 3(a) shows the mean ¯uxes on 22 June measured with the gradient method at 15 min intervals.
Fluxes varied between ÿ6 and 900 ng NN2O mÿ2 sÿ1. At the 15 min time scale, large ¯ux ¯uctuations, were observed. They could not be imputed to temporal variability in the process of N2O production. They were attributed mainly to modi®cation in measurement conditions. At any time, a change in wind speed and wind direction modi®ed the apparent N2O source pattern.To overcome the uncertainties and the ¯uctuations of the 15 min ¯uxes the hourly mean ¯ux were calculated from the weighted 15 min ¯uxes according to Eq. (6). This treatment reduced the mean ¯ux errors (Eq. (7)) by about a half depending on the relative values of Fi. Fig. 3(b) shows on 22 June the weighted hourly mean ¯uxes and the error bars of these ¯uxes. This treatment considerably smoothed the variability of the ¯uxes and reduced
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
25
Fig. 3. (a) Five min fluxes and error bars obtained with the gradient method on 22 June. (b) Mean hourly weighted fluxes obtained from the 15 min fluxes in (a). Error bars are the mean hourly errors associated to the weighted fluxes.
the methodological uncertainty. On 22 June the mean hourly error bar was equal to 20% of the mean daily ¯ux.
order to limit the edge effect (Kunt, 1988). The analytical expression of the ®ltered N2O concentration (c0 ) was:
4.2. Eddy correlation method
k cos c
i A c
iÿk n kÿn=2
4.2.1. High frequency filtering The N2O time series were affected by high frequency noise due to the time resolution of the instrument. These high frequency ¯uctuations were independent of the boundary layer turbulence (`Aliasing' process described by Kaimal and Finningan, 1994) and could have induced an error into the estimate of the ¯ux. This noise modi®ed the covariance between the vertical wind speed and N2O concentration and could have induced an overestimation of the ¯ux. To minimise TDLAS high frequency noise a lowpass ®lter was applied. A cosine window was used in
0
n=2 X
with
n=2 1 X k cos A kÿn=2 n
(8)
Where c(i) is the measured N2O concentration, c0 (i0 ) the ®ltered signal, and i, k the indexes of time for a given time sequence. The integer n de®nes the value of the cut-off frequency: the higher n, the lower the cutoff frequency. The best value of cut-off frequency (i.e., n = 3) was chosen as to maximise the correlation between the vertical wind speed and the concentration. The study of all data sets showed that using low-pass
26
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
®ltering of the N2O signal improved the correlation coef®cient between vertical wind and N2O concentration by 48% (R0 2 =R2 1:48: ratio of the correlation coef®cients after and before ®ltering) and also reduced the SD of the N2O signal by 58% (0c =c 0:58: ratio of the standard deviations after and before ®ltering). This ®ltering also reduced by about 16% the covar-
iances between N2O concentration and vertical wind and likewise the N2O ¯uxes (Ffiltering =Fraw
R0 2 =R2
0 c =c 1=1:16). Fig. 4 is an example of raw data (16 June at 9 : 00 UT) on a 3 min period. Vertical wind speed, air temperature and nitrous oxide concentration are shown. The corresponding mean ¯ux was 130 ng
Fig. 4. Raw data of eddy correlation measurements at 10 Hz: w (vertical wind speed); T (air temperature); raw N2O (mixing ratio of N2O). The bottom figure corresponds to the raw N2O data after using low-pass filtering.
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
N-N2O mÿ2 sÿ1. After using low-pass ®ltering (bottom ®gure) there was a reasonable correlation of N2O measurements with temperature (R2 = 0.4) and with vertical wind speed (R2 = 0.3). 4.2.2. Low frequency filtering The heterogeneity of the N2O emissions in relation with changes in wind direction could cause non stationarity of the N2O signal. The consequence would be the presence of an additional low frequency component on N2O concentration signal which is not related to the local ¯ux. The magnitude of this additional trend could be of the same order as the ¯uctuations of higher frequency components. To remove this trend, high-pass ®ltering was used. This treatment consisted in removing a smoothed mean on the N2O concentration signal. The ¯ux was then written as follows: F
N 1X
wi ÿ w
ci ÿcsi N i1
(9)
where csi the smoothed mean written as: n=2 1 X csi ciÿj n jÿn=2
(10)
n corresponds to the time interval over which the smoothed mean was calculated and defines the lower cut-off frequency. n was selected to maximise the correlation coefficient between vertical wind and N2O concentration. The relative error on the ¯uxes was estimated from the derivatives of the ¯ux as follows: " # 2 N 1 X @Fi 2 0 2 @Fi 2 0 2
dci
dwi dF 2 N i1 @c0i @w0i using Eq. (1), the following equations were obtained: N N 1X @Fi 2 1X @Fi 2 2 w and 2c N i1 @c0i N i1 @w0i Introducing these equations in the preceding one, we obtained: q 1 (11) F p 2w
Dc2 2c
Dw2 N where 2w and 2c are the vertical wind and N2O
27
concentration variances, w and c are the N2O concentration and vertical wind uncertainties. These uncertainties are both instrumental and methodological. For the instrumental uncertainty, w and c were taken as the resolutions of the sonic anemometer and of the gas analyser. For the methodological uncertainty, c was linked to N2O source heterogeneities. By assuming that the removed low-frequency component of high-pass filtering was mainly due to the field heterogeneity, we estimated an error in the flux. Assuming w = 0 in Eq. (11), the expression of the error on the flux was then written as: q 1 2w Dc2 (12) DF N c was taken as the average SD deviation between the mean and the smoothed mean of N2O concentration: Dc2
N 1X
cÿcsi 2 N i1
(13)
Flux errors due to instrumentation were determined from Eq. (11). Errors linked to the TDLAS analyser were about of 1.2 ppbv for concentration and about of 3 cm sÿ1 for vertical wind. Considering the experimental conditions, i.e., values of the SD of the vertical wind and the N2O mixing ratio, ¯ux errors due to instrumentation were estimated on average at about 3%. The high pass ®ltering previously described was tested on all N2O eddy correlation data sets. The best correlations between N2O concentration and vertical wind speed were observed with a cut-off frequency of about 0.03 Hz corresponding to n = 600. This treatment improved the correlation coef®cient between N2O concentration and wind speed by about 6% (slope = 1.06 in Fig. 5(a)) but reduced the SD of N2O concentration by 33% (slope = 0.67 in Fig. 5(b)). The ®ltering of the N2O signal almost removed the negative correlations between N2O concentration and the vertical wind speed. Most of the data appeared positively correlated after ®ltering. The major effect of the high-pass ®ltering was to smooth the N2O ¯ux over 15 min periods. Fig. 6 shows 15 min ¯uxes on 17 June calculated with and without ®ltering. The main sources of error came from the experimental conditions and resulted from the heterogeneity
28
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
Fig. 5. (a) Effect of high-pass filtering on the correlation coefficient between N2O and vertical wind speed. (b) Effect of high-pass filtering on the N2O SD.
of the N2O emission. Assuming that the low frequencies of the N2O signal corresponded to the source heterogeneity, the relative ¯ux errors were calculated according to Eqs. (12) and (13). For all data sets the relative errors exhibited large variabilities ranging from 2 to 200% with a mean of 25%. The use of high-pass ®ltering could introduce an error by truncating the N2O concentration ¯uctuations. This error was estimated as the relative correction on the heat ¯ux using the same high-pass ®ltering according to Eqs. (12) and (13) (`c' is substituted by `T' in the text). The truncation error was found to be 8% on the heat ¯ux. No relationships between the wind direction and the wind speed were observed. The cospectrum of the vertical wind speed and tempera-
ture, CwT could be used as a surrogate for the cospectrum of the wind speed and N2O concentration according to the Monin±Obukhov similarity theory (Kaimal et al., 1972). According to this theory, the shape of the wc and wT cospectra were expected to be similar, therefore we can expect that the N2O ¯uxes were underestimated by about 8%. However, care must be taken with this approach, because the spectral shape at a given measurement height may be different for different scalars if their source/sink distributions differ within the underlying canopy as was certainly the case for N2O and heat sources. This study only showed that, if the N2O sources were distributed as the heat sources, the use of high-pass ®ltering would reduce the value of the N2O ¯uxes by only 8%.
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
29
Fig. 6. Eddy correlation fluxes on 17 June calculated with and without high-pass filtering.
5. Flux measurements: spatial and temporal variabilities 5.1. Chamber method 5.1.1. Spatial variability Fluxes were measured twice a day with three static chambers from 17 to 24 June, except on 17, 20 and 24 June when the 30 chambers were used in the morning. These measurements allowed us to check the heterogeneity of the N2O emissions. On 17 June, N2O emissions were nearly lognormal distributed with a Shapiro±Wilk coef®cient of 0.983 (Shapiro and Wilk, 1965) which allowed statistical description (Fig. 7). Because of this distribution, the mean N2O emissions and standard deviations were calculated according to the uniformly minimum var-
iance unbiased estimator (UMVUE) method of Finney described by Parkin et al. (1988). Such estimations are considered to be less sensitive to the in¯uence of longtailed distributions and departures from normality. The UMVUE mean and SD values were similar to the arithmetic mean and SD values (Table 1). Therefore, data distributions were considered as almost normal. As arithmetic standard deviations (x) ranged from 48 to 53% of mean values, the 95% con®dence interval (IC95%) for the mean ¯uxes (x) corresponding to in®nite sets of data couldpbe estimated. It is expressed as: IC95% x
1:96= n 100 for 30 samples (n = 30) and is equal to 18±20% of the mean values (Table 1). On 17, 20 and 24 June, the measurements with the three chambers chosen for continuous measurement were included in the SD and the IC95% interval of the whole set of chambers (except for 24
Table 1 Mean and standard deviations of N2O fluxes measured in 30 chambers and 3 chambers on 17, 20 and 24 June. The coefficient of variation, and ratio of 95% IC to the mean are also shown. The uniformly minimum variance unbiased estimator (UMVUE) are given in brackets Date (chamber no)
Mean arithmetic (UMVUE) (ng N-N2O mÿ2 sÿ1)
17 17 20 20 24 24
102.0 122.1 (121.4) 135.1 114.0 (113.5) 211.6 179.6 (178.8)
June June June June June June
(3) (30) (3) (30) (3) (30)
SD arithmetic (UMVUE) (ng N-N2O mÿ2 sÿ1)
CV arithmetic (UMVUE) (%)
IC95%/ mean (%)
65.5 (64.0)
54 (53)
19.2
56.1 (55.1)
49 (48)
17.6
88.1 (86.5)
49 (48)
17.5
30
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
Fig. 7. Frequency distribution of N2O fluxes measured in chambers using linear and logarithmic scale.
June). Thus, we considered that these three chambers gave a good estimate of the 30 chamber set, and thus of the whole ®eld plot if the small value of IC95% intervals is taken into account. Fig. 8 shows for 17, 20 and 24 June, the mean ¯uxes as a function of the distance from the gradient mast along the perpendicular row direction. The emissions were the mean of the three chambers on each row. Large spatial variability was observed along the direction perpendicular to the rows. It is particularly evident on 17 June where N2O emission peaks were observed at 15, 45 and 75 m. This day, the mean of the 10 coef®cients of variation over the three chambers at the same Y location but for different locations on the row was 26% (corresponding to the mean value of the
error bars in Fig. 8). The variability along the Y-axis was larger with a mean CV of about 55% for 10 measurements along each direction. The spatial distribution of N2O sources seemed to have been anisotropic. This was observed between rows and along the perpendicular direction and certainly resulted from agricultural practices: indeed, the fertiliser was injected into the soil with a tine. The amount of fertiliser applied along the tine was probably not constant and was probably a maximum at the locations 15, 45 and 75 m on the Y-axis. For the three dates the maxima were at similar positions (Fig. 8). However, the location of the maximum in ¯uxes was different. On 20 and 24 June, the relative maximum occurred near the gradient mast at 15 m. This was not
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
31
the case for the ®rst date (17 June) with a maximum located at 75 m from the mast. The magnitude of N2O emission depends on soil nitrogen nutrient availability (Weier et al., 1996), but also of local topographic features as suggested by Pennock et al. (1992) and Ball et al. (1997). A small uneven ®eld surface could induce modi®cations of the local soil properties such as the bulk density, soil water or organic matter content. The variation of maxima between dates could result from microtopographic features. Addition of water by rain (19 and 20 June) and irrigation events (13 and 16 June) could have altered the initial distribution of N2O emission, which might explain the difference between days.
Fig. 8. Spatial variation of N2O emissions measurement with chambers as a function of the distance from the gradient mast for the 3 days (17, 20 and 24 June 1996). The error bars show the SD of the emissions over 30 chambers for each row (see Fig. 1).
5.1.2. Temporal variability of emissions ± modelling Fig. 9 shows the ¯uxes measured with the chambers during the whole experimental period. Fluxes increased from 14 June (2 days after fertilisation) to 22 June, with variations following rain and irrigation.
Fig. 9. N2O fluxes measured using chamber and micrometeorological methods along with modelled fluxes. Soil moisture content is also shown. Micrometeorological measurements were averaged over an hour, and chamber measurements over 2 h. Micrometeorological fluxes were estimated by eddy correlation method till 18 June and then by gradient method.
32
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
Apparently, nitrous oxide emission seemed to be correlated with soil moisture content (Fig. 9). To test this, a simple model developed by Jambert et al. (1997b) was applied to the data set, with daily averaged ¯uxes. In this model the daily mean ¯ux is expressed as an exponential function of time and a Gaussian function of the soil water ®lled pore space (WFPS). The maximum value (WFPSmax) was ®tted to 35%. The sandy soil might be a reason for this small value of WFPSmax (Parton et al., 1996): JÿJmax with j WFPS exp Tc "ÿ 2 # WFPSj ÿWFPSmax (14) WFPS max exp where: j: calculated flux N2O for the day j (ng NN2O mÿ2 sÿ1); WFPS: flux expressed as a function of soil water content (percentage of water filled pore space); Jmax: day of year of the maximum flux (24 June) : maximum flux measured on day Jmax; Tc: time interval (max =2ÿmax ) during increasing emission phase (Tc = 10) in days : fitted coefficient ( = 15) WFPSj: water filled pore space in the 10± 20 cm soil layer on day j. The model was made for determining N2O ¯ux after fertiliser application over maize ®elds of this region. It accounted for the dependence of ¯uxes on irrigation and rain events, which modi®ed the soil water content. Good agreement was observed between the modelled and measured ¯uxes, with a correlation coef®cient of 0.95 (Fig. 9). These results were used as a validation of the model which was previously used to estimate nitrous oxide emissions over a 1 year period (Jambert et al., 1997b). 5.2. Micrometeorological methods 5.2.1. Spatial variability: footprint analysis The development of a relationship between the emission pattern and the measured ¯uxes by micrometeorological methods is a key step in the interpretation of these measurements. Schuepp et al. (1990) and Leclerc and Thurtell (1990) developed simple methods to quantify the spatial contributions of the upwind source area to the measured ¯ux. Schuepp et al. (1990) used an analytical solution where the contributions of different upwind sources vary with the distance
between the source and the observation, with the height of the observation, as well as with the characteristics of the wind speed pro®le and atmospheric stability. The cumulative normalised contribution to the ¯ux F0 is then given by: u
zÿd F0 exp ÿ (15) u kx where u is the wind speed (m sÿ1) averaged between the height z0 + d and the measurement height z. The ratio u=u depends on the atmospheric stability functions (see Eq. (3)) and was defined as a function of the Monin±Obukhov length scale. The footprints for the eddy correlation method were quantified for a range of atmospheric stabilities. (z-d)/L ranged from ÿ2 to 0.7 with corresponding u/u* of 2.8 and 9.8, respectively (Fig. 10). Under neutral conditions u=u was equal to 5.4. Fig. 10 shows that under unstable conditions, 75% of the flux was due to sources located within 20 m of the mast. Under stable conditions, only 30% of the flux came from sources within 20 m of the mast and at 100 m, only 75% of the flux was recovered. It is clear that the bulk of the flux measured by micrometeorological systems was emitted from a fetch less than 100 m away. It is particularly true under very unstable atmospheric conditions and should be emphasised with low u=u. The relative corrections F/F corresponding to low frequency ®ltering in the eddy correlation method (Eqs. (12) and (13)) were plotted versus the wind direction and the wind speed (Figs. 11(a) and (b)). The largest errors were observed in the ®rst period of the experiment (14±17 June) when the wind-direction was oriented along the axis of the rows (508N). After 16 June, the mean wind direction changed and was oriented perpendicularly to the row direction. After this date, corrections due to low frequencies ®ltering, were small, and the relative error ¯uxes was only 6% on average. When the wind direction was parallel to the row axis, good correlations were also observed between the wind speed and F/F (R20.53). The higher the wind speed was, the larger was the low frequency corrections to the N2O signal. These results are in agreement with the measurements made with the chambers. Using a segment shaped layout for the chamber locations, it was found that the maximum variability in N2O emission was in
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
33
Fig. 10. Predicted flux footprints at 0.8 m above the maize canopy under three conditions of atmospheric stability: u=u 9:8 (stable); u=u 5:4(neutral); u=u 2:8 (unstable). Bold lines are flux footprint, and thin lines are cumulated flux footprint.
the direction perpendicular to the row. Source strength was more variable along the `cross-row' direction than along the row. Therefore, when the mean wind direction was along the row axis, the measured ¯uxes were more sensitive to slight variations of wind direction. In contrast, when the mean wind direction was perpendicular to the rows, the distribution of the sources contributing to the measured ¯uxes were similar for a slight change in wind direction. From 14 to 16 June, using recorded wind speed at 10 Hz on 15 min period, the SD of the wind direction varied between 17 and 908 with a mean of about 508. Therefore, for a mean SD of 508 and considering the location of the maximum of the footprint function (xmax u
zÿd=u 2k), the explored upwind length at the maximum of the function (Eq. (15)) in the crossrow direction would be about 23 m (2 9.8 Tan(508)) for stable atmospheric conditions (u=u 9.8), and 7 m under unstable conditions ( u=u 2.8). Thus, assuming that u=u varies in the same way that wind speed, the ¯ux footprint would be smaller for low wind speed and would not integrate all the spatial variability of N2O source emission observed with the chambers in the perpendicular row direction. This could explain why low frequency corrections on N2O signal were lower. Sensitivity studies of the two micrometeorological methods indicate that the eddy correlation method was apparently more accurate with a relative uncertainty of
the ¯uxes of 51% for the gradient method and 25% for the eddy correlation method using 15 min period. The eddy correlation method corresponds to a direct estimate of the ¯uxes at measurement height without assumptions about the trace gas dispersion. The gradient method is based on the use of a diffusion model depending on atmospheric stability (Eqs. (3) and (4)). These assumptions might explain, the large uncertainties in ¯ux estimates by this method. In addition, the accuracy of the gradient method would certainly be improved by simultaneous recording the N2O mixing ratio at each height. This would integrate the same temporal variability of the footprint during the 15 min time period. 5.2.2. Temporal variability Fig. 9 shows the ¯uxes measured by the micrometeorological methods and averaged over hourly periods for the whole campaign. As was observed for the chamber measurements, the ¯uxes increase from 17 till 24 June. Apparently the N2O emissions were related to the soil water content. A high variability of the emissions was observed on a hourly scale. Hourly ¯uxes reached a maximum greater than 700 ng N-N2O mÿ2 sÿ1 and 200 ng N-N2O mÿ2 sÿ1 was often exceeded. Fig. 12 shows the mean relative diurnal variations of N2O emissions between the 18 and 24 June. To allow comparison between different conditions the hourly
34
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
Fig. 11. (a) Relative correction to eddy correlation N2O fluxes and wind direction over the experimental period. (b) Relative correction to eddy correlation N2O fluxes and wind speed over the experimental period.
mean ¯ux was normalised by the corresponding daily mean ¯ux. A slight diurnal trend was observed. Fluxes were higher than the mean between 11 : 00 and 19 : 00 UT. However, high emissions were observed early in the morning around 6 : 00 UT. Error bars indicate the SD obtained for the 8-day-period. They were roughly 50% of the mean. They indicate a large variability of ¯uxes between days, of which a large part could be attributed to footprint variability. 6. Intercomparison Fig. 9 shows the comparison between N2O emissions measured by micrometeorological methods and
chambers during the whole experimental period. The chamber ¯uxes were integrated over 2 h and micrometeorological ¯uxes were calculated hourly. Comparison during the common periods of measurements (i.e., micrometeorological ¯uxes measurement at the same time as the chambers) are shown in Fig. 13. The error bars indicate the SD of the whole sets (3 or 30 chambers) for the chamber method. The error bars for the micrometeorological methods are the averages of the correction factors expressed by Eqs. (12) and (13) for the eddy correlation method, and the uncertainties derived from the Eqs. (5) and (7) for the gradient method. A good agreement between the chamber and micrometeorological methods is observed with a mean coef®cient of determination R2 0.6 and a linear
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
35
Fig. 12. Diurnal variation of normalised N2O fluxes determined by the gradient method. Fluxes have been normalised using their mean daily values.
regression equation of Fm = 1.14 Fc + 3.6 (ng NN2O mÿ2 sÿ1),where Fc is the chamber ¯uxes and Fm the micrometeorological ¯uxes. Except in the evening of 19 June corresponding to an abrupt change in the N2O ¯ux magnitude measured by micrometeorological method, the differences in ¯uxes are within the
error bars of each method. The `OBS' column in Table 2 indicates the type of micrometeorological method used during the comparison. The ratios between the chambers included in the wind sector of the micrometeorological mast and the total number of used chambers are indicated in parentheses.
Fig. 13. Mean flux estimates: for 2 h integration time, using 30 or 3 manual chambers, for 2 h integration time, using micrometeorological fluxes. For chamber fluxes, error bars indicate standard deviations from the 30 or 3 chamber data. For micrometeorological fluxes, the error bars indicate methodological uncertainties: uncertainties of the gradient profiles or relative correction for low frequency from N2O data for gradient and eddy correlation methods, respectively.
36
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
Table 2 Mean chamber and micrometeorological fluxes (F) with associated SD (F) in ng N-N2O mÿ2 sÿ1, Coefficient of variation (CV%). In the column `OBS' micrometeorological method used are labelled with `gr' for gradient, `ec' for eddy correlation method and in the brackets the proportion of chambers included within micrometeorological footprint; mean wind direction and SD (, ) associated with the micrometeorological methods Date (June)
UT time
Chamber F
14 15 16 17 18 18 19 19 19 20 20 21 21 22 23 24
3 p.m. 8 a.m. 9 a.m. 9 a.m. 8 a.m. 3 p.m. 8 a.m. 12 a.m. 3 p.m. 8 a.m. 3 p.m. 8 a.m. 1 p.m. 9 p.m. 9 p.m. 10 p.m.
24 22 31 122 115 98 72 86 95 114 116 27 90 228 260 180
Micromet. methods F 10 10 8 65 43 16 27 33 31 56 48 15 39 127 145 88
SD
CV
F
F
CV
OBS
40 46 25 53 37 16 37 38 32 49 42 54 43 56 56 49
33 17 43 131 49 100 62 70 266 82 108 13 196 396 253 195
29 6 10 5 3 6 17 20 18 19 16 1 48 58 33 39
89 38 23 3 6 6 27 28 7 23 14 9 25 15 13 20
ec (1/3) ec (3/3) ec (2/3) ec (12/30) ec (0/3) ec (3/3) gr (3/3) gr (3/3) gr (3/3) gr (30/30) gr (3/3) gr (3/3) gr (3/3) gr (0/3) gr (3/3) gr (0/30)
Depending on the wind direction and the location of the different micrometeorological masts, chambers were sometimes not included in the micrometeorological footprint area. This did not induce any difference in the comparison analysis and indicated that either the 3 or 30 selected location areas for the chambers were representative of the mean N2O emission at the ®eld scale or that the spatial variability of N2O emission at the ®eld scale was not so pronounced. The mean coef®cient of variation was 50% for the 30 chambers and 40% for the 3 chambers. Similar CVs were observed for a 2 h period for eddy correlation estimates (27%) and for gradient estimates (18%). For the gradient method, the mean and the SD were estimated from the 15 min values according to Eq. (7). For the eddy correlation method the mean and the SD correspond to the mean of the 15 min ¯uxes and to the correction ¯uxes (Eqs. (12) and (13)). Similar CVs between the chamber and micrometeorological methods strengthen the assumption that low frequency components of the N2O concentration for the eddy correlation method mainly resulted from N2O source heterogeneities and that uncertainties of the gradient method also re¯ected this heterogeneity. The highest CVs for micrometeorological methods were observed
55 45 88 81 195 261 238 258 260 294 277 279 270 215 296 97
21 38 58 50 48 75 ? ? ? ? ? ? ? ? ? ?
for the eddy correlation method corresponding to a wind direction across the row direction (N-NE) and resulting from the pattern of nitrogen fertiliser application. This anisotropy was supported by studying the chamber ¯ux CVs along and across the row direction. Given the small number of chambers, it was not possible to go further with this comparison. In fact, for the gradient method it would have been useful to record the SD of the wind direction to obtain the area `scanned' by the micrometeorological method during the measurement period. To improve the comparison between the micrometeorological and the chamber methods it would have been interesting to estimate the mean values of chamber ¯uxes, to weight each ¯ux of the different chambers by the footprint function modelled according to Schuepp et al. (1990). This treatment would have been possible if more chambers had been available to obtain a better map of N2O source distribution. 7. Conclusions Given the variability of N2O emission at the ®eld scale good agreement was found between the micro-
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
meteorological and the chamber techniques considering the temporal variability and the ¯ux magnitude of each measurement methods. Comparing chamber and micrometeorological methods demonstrated that the wind direction, wind velocity and atmospheric stability can complicate the comparison of N2O emissions determined by `patch' and `integrative' methods due to the footprint area modi®cations. For the micrometeorological method, large ¯uctuations of ¯ux at 15 min time scale were observed. The interpretation of these variations is dif®cult because the temporal variability between 15 min data re¯ects a temporal N2O emission event as well as the variability of the footprint from one measurement period to the other. To compare micrometeorological methods to chamber methods it is essential to have information on the heterogeneity of the soil and to apply different treatments to the micrometeorological raw data: weighted 15 min ¯ux versus their relative error to smooth hourly mean ¯uxes for the gradient method or using high-pass ®ltering for the eddy correlation N2O data signal. Future improvement of the comparison between micrometeorological and chamber methods would consist in using more chambers allowing a variogram and geostatistical analysis to study the different scales of N2O heterogeneity emission. These scales could be compared to the heterogeneity scales of other physical parameters such as NH3 ÿ , NH4 and soil water contents, to understand the major cause of N2O emission variability.
Acknowledgements This work was supported by the French programme AGRIGES of the MinisteÁre de l'Environnement, the Institut National de la Recherche Agronomique (INRA) and the Agence de l'Environnement et de la MaõÃtrise de l'Energie (ADEME). Dominique Flura, Sylvie Masson and Brigitte Durand (INRA), Laurent Coppola (CNRS) assisted us with the in situ measurements. We are grateful to Claire Gaudout of the Translation Department of INRA for reviewing the English version of the manuscript.
37
References Ambus, P., Christensen, S., 1994. Measurement of N2O emission from a fertilised grassland: an analysis of spatial variability. J. Geophys. Res. 99, 16549±16555. Ball, B.C., Horgan, W.G.W., Clayton, H., Parker, J.P., 1997. Spatial variability of nitrous oxide fluxes and controlling soil and topographic properties. J. Environ. Qual. 26, 1399±1409. Bouwman, A.F., 1990. Exchange of greenhouse gases between terrestrial ecosystems and the atmosphere. In: Bouwman, A.F. (Ed.), Soils and the Greenhouse Effect. Wiley, Chichester, UK, pp. 61±127. Brutsaert, W., 1982. Evaporation into the Atmosphere. Reidel Publishing Company, London, UK, pp. 64±76. Denmead, O.T., Raupach, M.R., 1993. Methods for measuring atmospheric gas transport in agricultural and forest systems. In: Harper, L.A., Mosier, A.R., Duxbury, J.M., Rolston, D.E. (Eds.), Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. Am. Soc. Agron., Madison,WI, pp. 19±43. Duxbury, J.M., Bouldin, D.R., Terry, R.E., Tate, R.L., 1982. Emissions of nitrous oxide from soils. Nature 298, 462±464. Dyer, A.J., Hicks, B.B., 1970. Flux-gradient relationships in the constant flux layer. Q .J. R. Meteorol. Soc. 96, 715±721. Groffman, P.M., Tiedje, J.M., 1991. Relationships between denitrification, CO2 production, CO2 production and air-filled porosity in soils of different texture and drainage. Soil Biol. Biochem. 23, 299±302. Hargreaves, K.J., Skiba, U., Fowler, D., Arah, J., Wienhold, F.G., Klemedtsson, L., Galle, B., 1994. Measurement of nitrous oxide emission from fertilised grassland using micrometeorological techniques. J. Geophys. Res. 99, 16569±16574. Houghton, J.T., Callander, B.A., Varney, S.K., 1992. Climate Change 1992. The supplementary report to the IPECAC scientific assessment. Published for the Intergovernmental Panel on Climate Change. Cambridge University Press, UK. IPCC, 1994. Radiative forcing of climate change and an evaluation of the IPECACIS92 emission scenario. In: Houghton, J.T., Filho, Meira L.G., Bruce, J., Hoesung, L., Callander, B.A., Haites, E., Harris, N., Maskell, K. (Eds.), Climate Change. Cambridge University Press, UK. Hicks, B.B., Hess, G.D., Wesley, M.L., 1979. Analysis of fluxprofile relationships above tall vegetation-an alternative view. Q. J. R. Soc. 105, 1074±1077. Jambert, C., 1995. Emissions de composeÂs azoteÂs dans l'atmospheÁre par les agrosysteÁmes fertiliseÂs: maõÈsiculture dans les Landes de Gascogne, Ph.D. Thesis. Universite de Paul Sabatier, Toulouse (in French). Jambert, C., Serc,a, D., Delmas, R., 1997. Quantification of Nlosses as NH3, NO and N2O-N2, from fertilised maize fields in the southwest of France. Nutr. Cycling Agroecosystems 48, 91± 104. Jambert, C., Delmas, R., Serc,a, D., Thouron and Delprat, L., 1997. N2O and CH4 emissions from fertilised agricultural soils in the southwest of France. Nutr. Cycling Agroecosystems 48, 105±114. Kaimal, J.C., Wyngaard, J.C., Izumi, Y., Cote, O.R., 1972. Spectral characteristics of surface layer turbulence. Q. J. R. Meteorol. Soc. 98, 563±589.
38
P. Laville et al. / Agricultural and Forest Meteorology 96 (1999) 19±38
Kaimal, J.C., Finningan, J.J., 1994. Atmospheric Boundary Layer Flows: Their Structure and Measurement. Oxford University Press, US, pp. 259±264. Kolb, C.E., Wormhoudt, J.C., Zahniser, M.S., 1995. Recent advances in spectroscopic instrumentation for measuring gases in the natural environment. In: Matsou, P.A., Harriss, R.C., (Eds.), Biogenic Trace Gases: Measuring Emission from Soil and Water. Blackwell Science, US, pp. 259±290. Kunt, M., 1988. Traitement numeÂrique des signaux. In: Neirynck, J., (Eds.), Traite d'ElectriciteÂ, d'Electronique. Dunod, F, pp. 107-123. Laville, P., Henault, C., Renault, P., Cellier, P., Oriol, A., Devis, X., Flura, D., Germon, J.C., 1998. Field comparison of nitrous oxide emission measurements using micrometeorological and chamber methods. Agronomie 17, 375±388. Leclerc, M.Y., Thurtell, J.W., 1990. Markovian analysis. BoundaryLayer Meteorol. 52, 247±258. Matthias, A.D., Blackmer, A.M., Bremner, J.M., 1980. A simple chamber technique for field measurement of emission of nitrous oxide from soils. J. Environ. Qual. 9, 251±256. Parkin, T.B., Meisinger, J.J., Chester, S.T., Starr, J.L., Robinson, J.A., 1988. Evaluation of statistical estimation methods for lognormally distributed variable. Soil. Sci. Am. J. 52, 323± 329. Parton, W.J., Mosier, A.R., Ojima, D.S., Valentine, D.W., Schimel, D.S., Weier, K., Kulmala, A.E., 1996. Generalised model for N2 and N2O production from nitrification and denitrification. Global Biogeochem. Cycles 10, 401±412. Pennock, D.J., van Kessel, C., Farrell, R.E., Sutherland, R.A.,
1992. Landscape-scale variation in denitrification. Soil Sci. Soc. Am. J. 56, 770±776. Schuepp, P.H., Leclerc, M.Y., Macpherson, J.I., Desjardins, R.L., 1990. Footprint prediction of scalar fluxes from analytical solutions of the diffusion equation. Boundary-Layer Meteorol. 50, 355±373. Shapiro, S.S., Wilk, M.B., 1965. An analysis of variance test for normality (complete samples). Biometrika 52, 591±611. Smith, K.A., Clayton, H., Arah, J.M.R., Christensen, S., Ambus, P., Fowler, D., Hargreaves, K.J., Skiba, U., Harris, G.W., Wienhold, F.G., Klemedtsson, L., Galle, B., 1994. Micrometeorological and chamber methods for measurement of nitrous oxide fluxes between soil and the atmosphere: overview and conclusions. J. Geophys. Res. 99, 16541±16548. Wagner-Riddle, C., Thurtell, G.W., King, K.M., Kidd, G.E., Beauchamp, E.G., 1996. Nitrous oxide and carbon dioxide fluxes from a bare soil using a micrometeorological approach. J. Environ. Qual. 25, 898±907. Warneck, P., 1988. Chemistry of the natural atmosphere. In: Dmowska, R., Holton, J.R. (Eds.), International Geophysics Series. Academic Press, London, U K, pp. 41±45. Weier, K.L., Doran, J.W., Power, J.F., Walters, D.T., 1996. Denitrification and the dinitrogen/nitrous oxide ratio as affected by soil water, available carbon, and nitrate. Soil Sci. Soc. Am. J. 57, 66±72. Wiehnold, F.G., Frahm, H., Harris, G.W., 1994. Measurements of N2O fluxes from fertilised grassland using a fast response tunable diode laser spectrometer. J. Geophys. Res. 99, 16557± 16567.