Evaluating the calculated dry deposition velocities of reactive nitrogen oxides and ozone from two community models over a temperate deciduous forest

the


Introduction
Global atmospheric emissions of nitrogen oxide have increased dramatically during the past 150 years, and the supply of reactive nitrogen to ecosystems has doubled due to anthropogenic activities such as nitrogen fertilization, biomass burning, and fossil fuel combustion (Galloway et al., 2008).Dry deposition is responsible for a significant portion of the total (wet and dry) nitrogen deposition (e.g.34%, Munger et al., 1998;58%, Sparks et al., 2008).Up to 43% of NO x eN emissions over North America have been estimated to be removed from the atmosphere by dry deposition (Shannon and Sisterson, 1992).Reactive nitrogen oxides, called NO y , is a class of oxidized nitrogen compounds including NO, NO 2 , NO 3 , N 2 O 5 , HNO 3 , PAN (peroxyacetyl nitrate), other organic nitrates, and particle nitrate, which supply significant nutrient and acidic quantities to ecosystems.Augmented atmospheric deposition of NO y associated with increased emissions of NO x poses many environmental threats, including acidification of soil and surface water, eutrophication of lake, river and estuary, loss of biodiversity, damage to forests, and global climate change (Galloway et al., 2008).Increased anthropogenic emissions of NO x combined with hydrocarbons have produced high levels of surface O 3 concentration.O 3 can penetrate the tissues of leaves easily through stomatal uptake, causing stomatal occlusion and leaf damage.The direct uptake by vegetation through the stomata is also a major sink of O 3 in the lower troposphere (Turnipseed et al., 2009).
Given the significant impacts of NO y and O 3 deposition on atmospheric chemistry and ecosystem health, it is desirable to quantify the deposition amount and assess the effects.Measuring deposition fluxes for reactive nitrogen compounds and O 3 with the eddy-covariance technique (e.g.Munger et al., 1996;Turnipseed et al., 2006) or the gradient method (e.g.Meyers et al., 1989;Sievering et al., 2001) have formed the basis for deposition models aimed at predicting dry depositions of reactive nitrogen compounds and O 3 .
Models have been developed (e.g.Wesely, 1989;Meyers et al., 1998;Zhang et al., 2002Zhang et al., , 2003;;Niyogi et al., 2009;Wu et al., 2003) to estimate the dry deposition velocity (V d ) by commonly utilizing the resistance approach analogous to Ohm's law in electrical circuits.Accurately parameterizing the complex surfaceatmosphere exchange process remains challenging for V d modeling due to large variability in surface conditions (e.g., vegetation types, and soil contents) at model sub-grid scales.It is difficult to fully describe the physiological processes concerning the vegetation stomatal responses to various environmental conditions, leaf age, injury, and so on.The rapid within-canopy chemical reactions are not often considered in simple single-layer models, neither for the role of horizontal flow to receptor surfaces over non-uniform surfaces and terrains (Wesely and Hicks, 2000).Therefore, large uncertainties still exist in modeling V d .A recent study (Flechard et al., 2010) modeled the V d of inorganic reactive nitrogen species (i.e.NH 3 , NO 2 , HNO 3 , and HONO and aerosol NH 4 þ and NO 3

À
) over 55 monitoring sites throughout Europe, using four existing dry deposition models.Their result revealed that differences between models can reach a factor 2e3 and are even greater than differences between monitoring sites.Hence, there is a continuous need to evaluate modeled V d over different land-cover types and for different chemical compounds.
Observational deposition fluxes of SO 2 and O 3 are often used to evaluate models (Zhang et al., 2002;Wu et al., 2003).However, few studies have evaluated modeled V d for nitrogen species primarily because accurate quantifications of dry deposition fluxes and speciation of the reactive nitrogen species are difficult and expensive to obtain (Horii et al., 2005).Munger et al. (1996) demonstrated that the dry deposition fluxes of NO y can be measured reliably using the eddy-covariance technique and yearround observations have been conducted at the Harvard Forest Environmental Measurement Site (HFEMS) since 1990.In a campaign attempting to estimate NO y concentration and deposition budget, concentrations of individual NO y species (i.e.NO, NO 2 , PAN and HNO 3 ) have been measured at HFEMS.The reactive nitrogen dataset along with the O 3 fluxes/concentrations available at HFEMS are used to evaluate two community dry deposition models here.
One model is the Weather Research and Forecasting-Chemistry model (WRF-Chem) dry deposition module (hereafter WDDM).WRF-Chem is a state-of-the-art, regional atmospheric chemistry model (Grell et al., 2005) and has been successfully applied for regional air quality studies (Wang et al., 2009).Due to lack of observational data, few studies have evaluated the ability of the WDDM for calculating nitrogen V d , even though dry deposition is one of the most important sinks for pollutants.The other model is the Noah land surface model (LSM) (Chen and Dudhia, 2001) coupled with a photosynthesis-based Gas-exchange Evapotranspiration Model (Niyogi et al., 2009) (hereafter Noah-GEM).The Noah LSM has been used to provide surface heat fluxes as boundary conditions for WRF.It is of broad interest to develop capacities of computing V d in Noah LSM (Charusombat et al., 2010).This evaluation effort is part of a broader effort to eventually integrate the balance of hydrosphere, biosphere, and atmosphere with environmental modeling such as atmospheric nitrogen input for the ecosystems in Noah.There are also plans to couple surface deposition and emission information more closely in Noah by linking with biogenic emission models such as MEGAN (Model of Emissions of Gases and Aerosols from Nature; Guenther et al., 2006).So one main purpose of this paper is to document current deficiencies in WDDM and raise the awareness of such problems.Also, because an investigation of nitrogen deposition calculation has not been done for these models, this study takes advantage of recently available nitrogen flux data to investigate nitrogen-deposition algorithms, which can serve well in the deposition models.The objectives are to: 1) assess the performances of WDDM and Noah-GEM in calculating V d (NO y ) and V d (O 3 ) over a temperate deciduous forest, 2) understand the sensitivity of modeled V d (NO y ) and V d (O 3 ) to the key variables/parameters, and 3) improve the models by comparing with the field observations.
We will first describe the measurements used in this study (Section 2) and the modeling framework and formulations of WDDM and Noah-GEM (Section 3).Next, the observation data and model results and discussions are presented in Section 4, which is followed by the conclusions in Section 5.

Site description
The HFEMS is located in a temperate 80e100 year-old mixed deciduous forest in central Massachusetts (42.54 N,72.18 W;elevation,340 m), which consists of red oak (Quercus rubra), red maple (Acer rubrum) with scattered hemlock (Tsuga canadensis), red pine (Pinus resinosa), and white pine (Pinus strobus).The canopy height near the observation tower is approximately 20 m with a peak leaf area index (LAI) of 3.4 m 2 m À2 during summer.The nearest sources of significant pollution are a secondary road about 2 km west of the site and a main highway about 5 km north of it.
A permanent 30-m Rohn 25 G tower has been used at HFEMS to measure eddy-covariance fluxes of CO 2 , NO y , and O 3 , along with vertical profiles of NO, NO 2 , and O 3 since 1990.Measurements of PAN concentrations were added to the tower in 2000.A temporary 23-m steel scaffolding tower, situated about 100 m to the southeast of the Rohn tower, was configured with a Tunable Diode Laser Absorption Spectrometer (TDLAS) to measure concentrations of HNO 3 from JuneeNovember 2000.Due to physical constraints, the second tower did not match the measurement height of HNO 3 (22 m) with the measurement height of O 3 , NO y , NO, NO 2 , and PAN (29 m) on the first tower.However, Horii et al. (2005) confirmed that the two datasets are spatially coherent on the hourly timescale.Details on the site and the instrumental methods can be found in Munger et al. (1996) and Horii et al. (2005).Data used in this study are available online at http://atmos.seas.harvard.edu/lab/data/nigec-data.html.

Calculations of flux and dry deposition velocity
The fluxes (F) of O 3 and NO y were measured using the eddycovariance technique.The ratio of observed heat flux and heat flux mathematically smoothed to simulate the attenuation of high-frequency variations by the instruments was used to account for loss of scalar covariances at high frequencies.Corrections were typically less than 20% (Munger et al., 1996;Horii et al., 2005).Flux data were also omitted during periods of very low turbulence intensity (when the friction velocity, u * < 0.2 m s À1 ), resulting in approximately 18% and 21% of the data being omitted for O 3 and NO y , respectively.In addition, periods with [O 3 ] < [NO y ] (1%) were excluded for O 3 to avoid periods when O 3 chemical reactions may exceed O 3 deposition (Munger et al., 1996).
Assuming a zero concentration on the absorbing surface, the dry deposition velocity (V d ) can be determined as where C(z) is the gas concentration at a reference height, z.

Modeling framework
The resistance method determines V d as the reciprocal of a total resistance (R t ) which consists of a series of resistances to perform gas transport from the atmosphere down to the surface. (2)

Further developments of GEM
The GEM model (Niyogi et al., 2009) was further developed here (see Appendix B), but the parameters were kept the same and not specifically tuned for this study.R s is the primary output of GEM and since direct measurements of R s were not available at HFEMS, examining modeled surface heat fluxes provides an independent assessment of R s .The new results from the Noah-GEM model with modified R s substantially improved calculations of heat fluxes for both summer and autumn (Fig. 1), implying that it produced more reasonable R s and better surface energy partitioning between sensible and latent heat fluxes.In section 4, we discuss the performance of Noah-GEM in calculating V d (O 3 ) and V d (NO y ), based on this modified version.

Model configuration
The WDDM was extracted from the WRF-Chem model V3.1.1 and executed in a 1-D mode, and the Noah LSM V3.1 plus GEM was executed in the same fashion.Hourly tower measurements of air temperature (T a ), relative humidity (RH), wind speed (WS), wind direction (WD), atmospheric pressure (P a ), downward shortwave radiation (R g_in ), downward long-wave radiation (R long_in ), and precipitation rate (P recip ) at the height of 29 m were used to drive Noah-GEM.The u * and L are obtained in Noah via an iterative process, using T a , RH, WS, and P a (Chen et al., 1997).WDDM requires inputs of T a , R g_in , RH and P recip from observations, and u * , and L calculated by Noah.Hourly V d were computed for O 3 , NO, NO 2 , PAN and HNO 3 .

Modeling analysis
Model results are evaluated using descriptive statistics such as the degree of agreement (d) and fractional bias (FB) (e.g., Charusombat et al., 2010): (3) FB ¼ 2 where o i is the observation, m i the model result, and n the number of samples.

Results and discussion
4.1.The observations of O 3 deposition and its environmental drivers showing maxima in summer, associated with the high solar radiation and temperature.The peak values ranged from 40 to 80 ppbv, slightly lower than the observations in 1991e1994 (Munger et al., 1996).F(O 3 ) followed the same seasonal trend with maxima during summer, closely coinciding with high concentrations and canopy growth (Munger et al., 1996).As shown in Fig. 3, V d (O 3 ) augmented with increasing PAR (when <1400 mmol m À2 s À1 ) and decreased with increasing VPD.At moderate temperature (8e24 C), V d (O 3 ) exhibited relatively small variations, but declined at more extreme temperature conditions.The environmental factors are often not independent from each other, e.g., high PAR often accompanying high temperature and VPD.V d (O 3 ) tended to decrease with increasing PAR above 1400 mmol m À2 s À1 , suggesting that temperature and VPD, rather than light, take controls in regulating stomatal openings.V d (O 3 ) increased almost linearly with increasing latent heat flux (LE), consistent with stomatal control of the O 3 uptake and plant evapotranspiration.These trends agree with the analysis by Turnipseed et al. (2009) over a subalpine forest.V d (O 3 ) had a strong diurnal cycle.During summer, mean V d (O 3 ) peaked before 1200 LST at 0.9 cm s À1 and dropped throughout the day to minimum value of 0.2 cm s À1 at night (Fig. 4), as seen in Munger et al. (1996).

Evaluation of modeled V d (O 3 )
Fig. 4 compares the modeled summer V d (O 3 ) by WDDM and Noah-GEM against observations.Table 1 presents the statistical results of the comparison.WDDM and Noah-GEM produced low values of V d (O 3 ) at night (w0.1 cm s À1 ), much smaller than the observations (FB ¼ 0.43e0.82).Zhang et al. (2002) and Charusombat et al. (2010) reported a similar bias, indicating an overestimation of nighttime non-stomatal resistance (R ns ).Wesely (1989) scheme estimates R ns mainly using constant values specified for each season and each land-use category, while a recent R ns scheme developed by Zhang et al. (2003) is a function of u * , RH, LAI and canopy wetness for non-stomatal uptake.As the main purpose of this paper is to compare the performance of different algorithms for stomatal uptake, Noah-GEM deploys the same R ns parameterization as WDDM for convenience of comparison.The performance of Noah-GEM and WDDM can be improved by utilizing the more realistic and accurate R ns parameterization (e.g.Zhang et al., 2003) in the future work.
V d (O 3 ) increased in the morning as canopy photosynthesis became active.In Fig. 4, WDDM and Noah-GEM produced V d (O 3 ) with similar magnitude.However, WDDM did not capture the peak and underestimated morning V d (O 3 ) by w 0.2 cm s À1 .Noah-GEM, on the other hand, was able to capture the V d (O 3 ) decline possibly as a result of stomatal closure at noon.Noah-GEM also produced a second peak in V d (O 3 ) in the late afternoon that was not observed.Zhang et al. (2006) also observed the early morning peak of V d (O 3 ) over two forest sites and proposed a threshold of accumulated O 3 stomatal flux for leafs, above which stomatal uptake of O 3 is slowed down probably due to an increased substomatal CO 2 concentration or non-zero O 3 concentrations inside the stomata.However, those factors are not considered in the current deposition models.
In Fig. 5, the observed V d (O 3 ) showed expected patterns of behavior with respect to the main environmental drivers (PAR, temperature and VPD) and, therefore, exhibited both unimodal (June 03, 16, and 18) and bimodal diurnal patterns (June 04, and 17).Noah-GEM captured the bimodal diurnal pattern on June 04 and 17, while under the unimodal conditions it reproduced the peak before noon but overestimated the afternoon O 3 uptake.WDDM was found to hardly capture the diurnal behaviors of V d (O 3 ), probably due to its Jarvis-type R s (Eq.( A8)).However, it should be pointed out that Wesely (1989) R c scheme was developed for general application, which requires very little data to use and intends to produce average estimate for a long time over large areas, rather than a period of days at a particular site.
WDDM considerably underestimated V d (O 3 ) in autumn (Fig. 6), and the minimum canopy stomatal resistance (R i , which is also broadly denoted as R smin in the atmospheric and plant modeling community) was found to be responsible for this large discrepancy.The R i parameter in WDDM for deciduous broadleaf forests was assigned to be an infinite value (10 25 s m À1 was used) for early autumn (Wesely, 1989).Here we defined the season classification for JuneeAugust as category 1(summer) and SeptembereOctober as seasonal 2 (early autumn) respectively, based on the general climate at HFEMS (see also Munger et al., 1998).The infinite R i implies that there is no air-surface exchange via the stomatal pathway (Wesely, 1989) and is only valid for leafless condition.However, this was not the case for the Harvard Forest during Sep-tembereOctober, as indicated by the observations of net ecosystem exchange of CO 2 and also LAI (Urbanski et al., 2007).The dominant effect of R i on modeled R s has been emphasized (e.g.Cooter and Schwede, 2000) and is well illustrated in Fig. 6.For this particular study, the value of R i for summer (70 s m À1 ) seems appropriate for the early autumn.Fig. 6 shows that neither model captured the rising of V d (O 3 ) in the early morning hours (0300e0600 LST).The early morning rising is possibly due to some factors (e.g.episodic mixing events and transport) which are not adequately represented by the resistance analogy, and also note that the small number of observations available during this period is likely hard to smooth those effects.
As illustrated by Fig. 6, the uncertainty in specifying R i is one main reason for modeled bias in R s and V d of gases that are under stomatal control for WDDM.However, the prescription of R i inherently has significant uncertainty because R i cannot be measured or determined independently in the laboratory (Niyogi et al., 2009) and also the assumption of a constant R i value within a season is inappropriate because of its temporal variations including diurnal cycle (Avissar, 1993).Better approaches have been proposed to solve the issue related with seasonal category classification, such as using continuous LAI without the need of defining different seasonal categories (e.g., Zhang et al., 2003), which could avoid the abrupt change of input parameters (e.g., R i ) from one season to the next.Charusombat et al. (2010) identified LAI as the first-order parameter affecting Noah-GEM estimates of R s .The Noah LSM prescribed a maximum value of 3.3 m 2 m À2 for LAI in this case, slightly lower than the field measurement (3.4 m 2 m À2 ).Model performance can be improved by assimilating more accurate and seasonally-varying LAI data in the future work.
R s is a complex and dynamic variable representing the coupled effects of resistance imposed by plants to vegetation-atmosphere exchange through leaf stomata (Niyogi et al., 2009).The difference between modeled V d (O 3 ) by WDDM and Noah-GEM is mainly caused by the use of different R s schemes.Noah-GEM simulates the response of stomata to various environmental variables (e.g.PAR, canopy temperature, soil moisture, CO 2 concentration and relative humidity at the leaf surface) (Niyogi et al., 2009).A significant feature of Noah-GEM is that it is structured to consider the impacts of physiological   processes including CO 2 assimilation rate on the responses of leaves to environmental parameters, which can predict R s better than the Jarvis-style approach that is based on the minimum canopy stomatal resistance parameter (Niyogi et al., 2009).

The observations of NO y deposition and its environmental drivers
HFEMS experienced minor pollution events during the selected period (Fig. 7).[NO y ] was generally lower than 5 ppbv, occasionally reaching 15 ppbv.F(NO y ) and V d (NO y ) showed large day-to-day variations, with maximum values of 22 mmol m À2 h À1 and 4.5 cm s À1 , respectively.F(NO y ) and V d (NO y ) tended to peak during midday, consistently following similar diurnal behavior to those of turbulence development.Large values of V d (NO y ) on October 02, 04, 07 and 08 accompany large ratios of HNO 3 /NO y , inferring a key role HNO 3 played in the NO y deposition and V d (NO y ) (see also Munger et al., 1996).
The measured V d (NO y ) represent averaged V d of the total NO y species.But, in current gas dry deposition models, V d values are estimated for individual species (see Section 3).Similar to Michou et al. (2005), the concentrations and V d of individual NO y species were used to derive a composite V d (NO y ), which can be more directly compared to observations.Simulated V d (NO y ) can be defined as where x i is the member of the NO y family, and n the number of the members.
Modeling-wise, the unique value of this data set at HFEMS lies in the availability of the simultaneous concentrations of the main NO y species (e.g., NO, NO 2 , PAN, HNO 3 ) and the high temporal resolution (1 h).However, data gaps exist in the concentration measurements especially for HNO 3 , which is very difficult to measure at a short integration time (e.g., at hourly interval) due partially to its tendency to adsorb onto surfaces (Horii et al., 2005).All the gaps in the concentrations will be reflected in the simulated V d (NO y ) (see Eq. ( 5)).To obtain a less patchy simulation of V d (NO y ), a "x/NO y " ratio method was used to fill the gaps.The average diurnal cycles of "x/NO y " were derived from the measurements from JuneeNovember 2000 (not shown here).Along with the NO y concentrations, inferred concentrations of NO, NO 2 , PAN, and HNO 3 were derived as The inferred concentrations were used to fill the gaps in the measured concentrations.Observations at HFEMS suggested that HNO 3 played a critical role in V d (NO y ).To minimize the errors in simulated V d (NO y ) that result from inferred concentrations, the period (October 1e12, shown in Figs.7 and 8) with the fewest gaps in HNO 3 concentrations was selected.Fig. 8 presents the measured concentrations of the NO y species and also the gap-filled data from the "x/NO y " ratio method.A few gaps existed for NO and NO 2 while PAN concentrations were all inferred.Given that PAN usually showed a relatively small V d , and the averaged PAN/NO y ratio had a relatively small deviation, the simulated V d (NO y ) should not be significantly affected by the uncertainties in the inferred PAN concentrations.The relative differences between the concentrations of NO y and the sum of gapfilled NO, NO 2 , PAN and HNO 3 were typically less than 30% (Fig. 8a).

Evaluation of modeled V d (NO y )
The roughness length for momentum (z 0 ) is an essential parameter in calculating R a in LSMs and can be prescribed as a function of land-cover type, as in Noah where the value of 0.5 m is assigned for deciduous broadleaf forest.Alternatively, if information about the vegetation morphology (e.g., canopy height (h c ), and LAI) is known, z 0 can be calculated following Meyers et al. (1998): or simply assumed 0.1 h c (e.g., Chen and Zhang, 2009).In our sensitivity simulations, the model was run with three z 0 values: 1) 0.5 m (Noah default), 2) 1.6 m (Eq.( 7)) and 3) 2 m (0.1 h c ).As z 0 increased from the initial model value of 0.5 to 1.6 and 2 m, modeled u * increased significantly and approached observations (Fig. 9).Increased z 0 and u * can reduce R a and R b (Eqs.( A3) and (A7)), which, in turn, leads to an increase of up to 1.5 cm s À1 in modeled V d (NO y ).This exercise demonstrates that adjusting z 0 can substantially alter the V d of compounds sensitive to R a and R b (e.g.HNO 3 ).Ultimately, a value of 2 m for z 0 seems a reasonable representation of the canopy structure for HFEMS in this scenario.WDDM showed similar response to the parameter z 0 , and the results are not presented here.Hereafter, we assessed the models performance with z 0 set to 2 m.These sensitivity tests highlight the importance of treating the atmospheric surface layer in modeling the biosphere-atmosphere exchange.Indeed, current LSMs (including WDDM and Noah) employ the MonineObukhov Similarity Theory (MOST) to parameterize surface exchange coefficients.While MOST provides a dimensionally-based set of relationships that links the vertical fluxes of scalars to the gradients of the mean profiles within the atmospheric surface layer, it is only valid well above the rough surface (Högström, 1996) and fails in the so-called roughness sublayer, which is above tall canopies and within canopies (e.g., Harman and Finnigan, 2008).Simply adjusting parameters such as z 0 used in MOST may not solve this fundamental problem, and future work to improve the models will involve the use of vertically-varying profiles of mean scalar concentration (e.g., Harman and Finnigan, 2008) or a multi-layer canopy model that explicitly resolves the radiative, dynamical, and thermal transport within vegetation canopies.
Fig. 10a compares the modeled V d (NO y ) by WDDM and Noah-GEM against the observations.Table 1 presents the statistical results of the comparison.As described in section 4.2, WDDM prescribed an infinite value for R i , which results in no air-surface exchange via stomata during autumn.To assess the sensitivity of R s parameterization to V d (NO y ) estimate, we conducted an additional simulation, reducing R i to 70 s m À1 as validated in V d (O 3 ) study.The WDDM modeled daytime V d (NO y ) increased from 1.2 cm s À1 to 1.37 cm s À1 (on average), closer to the observations of 1.41 cm s À1 (also see Table 1, FB decreased from 0.16 to 0.04).This result is consistent with Munger et al. (1996) in that the stomatal influence on NO y dry deposition at HFEMS is relatively small.WDDM with corrected R i presented quite similar results with Noah-GEM (Table 1 and Fig. 10), as WDDM has the same expressions for R b , R m , and R ns with Noah-GEM and the predicted R a values by the two models are generally close.
Although the models are generally in good agreements with the observations (d ¼ 0.88), they seem to underestimate the nighttime value of V d (NO y ) significantly (FB ¼ 1.09e1.18).Overestimation of nighttime R a may be a chief reason for this unsatisfactory model performance as the conventional micrometeorological equations (e.g.Eqs.(A2), (A3)) have been known to have poor R a estimate for nighttime stable regime (Wesely and Hicks, 2000).The underestimation of V d (HNO 3 ) in turn caused the poor performance of V d (NO y ) during nighttime.Part of these model deficiencies can also be attributed to the fact that both models do not have a multiple canopy scheme to represent a realistic wind shear within forest canopies.
Models reproduced the daytime V d (NO y ) with satisfactory statistic results (d ¼ 0.94, FB ¼ 0.02e0.04),and captured most variations in the observation (Fig. 10a), while the bias occurred mainly on October 04, 07 and 08.The underestimation of u * on October 07 and 08 (Fig. 9a) leaded to a too high R a , which could be one reason for the V d (NO y ) underestimation.A large ratio of HNO 3 / NO y was observed (Fig. 7) on October 04, which caused the large V d (NO y ) in the observations.The models appear to substantially underestimate the HNO 3 deposition on this particular day.
At HFEMS, HNO 3 dominated the NO y deposition, so the modeled V d (NO y ) did not show much sensitivity to R s , but mostly depended on R a and R b .However, a recent field study over a coniferous forest (Turnipseed et al., 2006;Sparks et al., 2008) estimated that HNO 3 accounted for only w24% of the NO y flux and PAN exhibited a close portion of w20%.Zhang et al. (2009) investigated the total nitrogen flux budget over eight rural sites across eastern Canada, and estimated that HNO 3 constituted less than half of the NO y flux (46%), and the flux from other measured gaseous nitrogen species (NO 2 þ PAN þ PPN) were also a significant portion (35%).Those new field data suggested that other forms of gaseous nitrogen like NO 2 and PANs instead of HNO 3 can constitute the dominant portion of the NO y deposition at some sites.As V d (NO 2 ) and V d (PAN) are not evaluated directly due to lack of observations, we compared the simulations between models.Table 2 shows the model estimates of R c and V d for NO 2 and PAN during different daytime periods of summer.PAN presented a larger R c (w200 s m À1 ) than NO 2 (w120 s m À1 ).Compared with WDDM, Noah-GEM produced smaller R c for NO 2 and PAN during the morning and afternoon period, while slighter lager ones during the noon.The differences of R c between WDDM and Noah-GEM reached a factor of 1.1e1.5 on average, which in turn leads a factor of 1.1e1.3 for V d .
The NO y species have various physical and chemical natures, leading to very different behaviors in the deposition process.The V d (HNO 3 ) estimated by Noah-GEM is one order of magnitude larger than V d (NO 2 ) and V d (PAN) (Fig. 11), whereas V d (NO) is close to zero.Dry deposition of NO is usually assumed to be negligible as NO is almost inert to the mesophyll, the surface of canopy, or ground.Significant rates of NO 2 uptake by vegetation through stomata have been observed in chamber experiments (Hanson and Lindberg, 1991).The assumption that V d of NO 2 is similar to that of O 3 is usually used in deposition models.However, the deposition processes of NO and NO 2 are difficult to quantify in field measurements as NO x (NO x ¼ NO þ NO 2 ) rapidly interconverts between the surface and the height of flux measurement, which violates the assumption for a constant flux layer.Ideally, the NO x flux above the canopy should be examined as a whole that is conserved chemically on timescales relevant to turbulent exchange (Wesely, 1989).Noah-GEM estimates R c for PAN based on molecular diffusivity, solubility, chemical reactivity and comparison to O 3 or SO 2 deposition, leading to a prediction of daily maximum V d (PAN) on the order of 0.5 cm s À1 , almost half of V d (O 3 ).Turnipseed et al. (2006) reported the first direct measurement of eddy-covariance fluxes of PAN to a coniferous forest and showed a mean daily maximum value of V d (PAN) at w1.2 cm s À1 .The findings of fast deposition of PAN (Turnipseed et al., 2006) imply large uncertainties in parameterizing R c of organic compounds, and also indicate the importance of organic nitrogen in the reactive nitrogen deposition budget.Peak values of V d (HNO 3 ) modeled by Noah-GEM ranged from 3 to 6 cm s À1 , on the same order with the gradient-method measurements by Meyers et al. (1989) over a dense deciduous forest (2.2 to 6.0 cm s À1 ) and Sievering et al. (2001) over a conifer forest (7.6 cm s À1 ).
We estimated the deposition fluxes of individual NO y species by multiplying the predicted V d (using z 0 ¼ 2 m) with the observed concentrations (Eq.( 1)).On average, NO x , PAN and HNO 3 accounted for 19%, 4%, and 70% of the measured NO y fluxes, respectively.In the current models we only considered the unidirectional fluxes (deposition), and this assumption is not valid for gases with emission fluxes from the surface.Emissions of NO from soils at HFEMS are negligible (Horii et al., 2005).But Horii et al. (2004) observed bi-directional fluxes of NO 2 and suggested a compensation point for NO 2 near 1.5 ppbv at HFEMS, the ambient concentration below which NO 2 is emitted from stomata.Given that [NO 2 ] approached this level occasionally during the daytime within the selected period (Fig. 8), the estimate of the contribution of NO x to NO y fluxes should be overestimated at some degree.This uncertainty can be narrowed by incorporating a parameterization of the compensation point within the models in the future work.Because the overestimated nocturnal R a resulted in underestimation of V d (HNO 3 ), the contribution of HNO 3 to NO y fluxes presented here should be considered more of a lower limit.

Summary and conclusions
We evaluated the ability of two models (WDDM and Noah-GEM) to calculate V d (O 3 ) and V d (NO y ) against direct observations at HFEMS, and identified key variables/parameters and uncertainties in the two models.WDDM employs Wesely (1989) parameterization for R c , which uses a simple R s scheme based on the R i parameter prescribed for each season and land-cover category.The uncertainty in prescribed R i dominates the errors in estimating V d for O 3 and other gases that are controlled by the stomatal pathway.An infinite R i value for deciduous forest in autumn in the default WDDM was not appropriate and resulted in too low values of V d (O 3 ), while using R i values originally prescribed for summer (70 s m À1 ) produced better V d (O 3 ).More evaluations of WDDM for R i at different seasons are needed to mitigate the underestimation of V d (O 3 ).Several revisions to the original GEM formulations were justified by comparing the formulations with other literature and also by evaluating modeled surface sensible/latent heat fluxes against observations.Compared with WDDM, Noah-GEM has a more sophisticated R s scheme considering the response of physiological processes to environmental variables (such as soil moisture, vapor pressure deficit, and CO 2 concentration at the leaf surface) and shows a better ability to capture the variations in V d (O 3 ) than WDDM.The models still need to be improved to better represent the nocturnal O 3 dry deposition process.
On the other hand, results showed that V d (NO y ) calculation was not sensitive to R s , as expected, because V d (NO y ) was mainly affected by the rapidly depositing species such as HNO 3 and controlled by the atmospheric resistances (i.e.R a and R b ).The difference in calculated V d (NO y ) by WDDM and Noah-GEM was small as these two models produced similar R a and R b .WDDM and Noah-GEM agreed well with the observed daytime V d (NO y ), but underestimated it under nighttime stable conditions.A modest adjustment in the z 0 values can significantly alter and/or improve the predicted V d (NO y ).Daytime V d (NO y ) was more sensitive to z 0 .These sensitivity tests regarding z 0 and inferior performance of WDDM and Noah-GEM models under stable conditions illustrate the importance and difficulties in modeling the biosphereeatmosphere exchange within the forest canopies, the layer known as roughness sub-layer where the traditional MOST theory is not valid.Therefore, our future model development effort will involve utilizing vertically-varying profiles of mean scalar concentration including chemical species such as PAN, NO and NO 2 or a multi-layer canopy model that explicitly resolves the radiative, dynamical, and thermal transfer within vegetation canopies.Finally, with a combination of the observed concentrations and modeled V d , it was estimated that NO x , PAN, and HNO 3 were 19%, 4%, and 70% of the measured NO y dry deposition fluxes, respectively.Comparison of the simulated R c and V d for NO 2 and PAN shows that differences of R c estimates between WDDM and Noah-GEM were large and would cause differences in V d reach a factor of 1.1e1.3.
Very few studies were done in the past to extensively focus on evaluating nitrogen-deposition models, which are critical to estimating the surface and atmospheric nitrogen budget in atmospheric chemistry models, primarily due to lack of observations.This is particularly true regarding the WRF-Chem model, because its dry deposition calculation has not been systematically evaluated despite its popularity in atmospheric air quality community.This work is a first step and yet a preliminary study to evaluate the effects of two modules with different treatment of canopy resistance on deposition estimation.The implementation of Noah-GEM calculated V d in WRF-Chem is underway.And more comprehensive studies at different seasons and locations (for different forest types) will be done in the future.

Fig. 1 .
Fig. 1.Comparison of averaged diurnal cycles of observed and modeled heat fluxes by Noah-GEM.(a) Latent heat flux for JuneeAugust, (b) sensible heat flux for June-August, (c) latent heat flux for September-October, and (d) sensible heat flux for SeptembereOctober.d and FB were calculated from the original hourly data.

Fig. 2
Fig. 2 shows the time series of hourly-averaged [O 3 ] and F(O 3 ) from JuneeOctober 2000.There was a distinct seasonal cycle of [O 3 ]showing maxima in summer, associated with the high solar radiation and temperature.The peak values ranged from 40 to 80 ppbv, slightly lower than the observations in 1991e1994(Munger et al., 1996).F(O 3 ) followed the same seasonal trend with maxima during summer, closely coinciding with high concentrations and canopy growth(Munger et al., 1996).As shown in Fig.3, V d (O 3 )

Fig. 4 .
Fig. 4. Comparison of averaged diurnal cycles of observed and modeled O 3 deposition velocities by WDDM and Noah-GEM during JuneeAugust.d and FB were calculated from the original hourly data.
6) where D indicates the date, H is the hour of day (H ¼ 0,1,2,.,23),[NO y ] D,H is the measured concentration of NO y at the hour of H, on the date of D, [x i ] inferred is the inferred concentration of x i , and Ratio ([x i ]/[NO y ]) H is the averaged ratio at the hour of H.

Fig. 7 .
Fig. 7. Time series of (a) NO y mixing ratio, (b) NO y fluxes, (c) NO y dry deposition velocities, (d) wind speed, (e) friction velocity, and (f) the ratio of HNO 3 /NO y .

Fig. 9 .
Fig. 9. Sensitivity of Noah-GEM modeled friction velocity and NO y deposition velocity to z 0 parameter: observation (black line, cycle symbol); z 0 ¼ 0.5 m (green line); z 0 ¼ 1.6 m (yellow line); z 0 ¼ 2 m (blue line) (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).

Fig. 10 .
Fig. 10.(a) Comparison of a time series of observed and modeled NO y deposition velocity by WDDM and Noah-GEM, and (b) modeled R a by WDDM and Noah-GEM (using z 0 ¼ 2 m).

Table A .
1 describes each resistance component, and TableA.2compares the formulations between WDDM and Noah-GEM.

Table 1
Statistical results of the observed and modeled V d (O 3 ) and V d (NO y ). a a Note: Daytime is 0900e1700 (LST); Nighttime is 1900e0600 (LST).The sample numbers are 1134, 551 and 430 for V d (O 3 ), and 170, 80 and 70 for V d (NO y ), respectively.

Table 2
Median of the modeled deposition velocity (surface resistance) in summer. a