Evaluation and improvements of two community models in simulating dry deposition velocities for peroxyacetyl nitrate (PAN) over a coniferous forest



[1] Dry deposition velocities (Vd) for peroxyacetyl nitrate (PAN) calculated using two community dry deposition models with different treatments of both stomatal and nonstomatal uptakes were evaluated using measurements of PAN eddy covariance fluxes over a Loblolly pine forest in July 2003. The observed daytime maximum of Vd(PAN) was ∼1.0 cm s−1on average, while the estimates by the WRF-Chem dry deposition module (WDDM) and the Noah land surface model coupled with a photosynthesis-based Gas Exchange Model (Noah-GEM) were only 0.2 cm s−1 and 0.6 cm s−1, respectively. The observations also showed considerable PAN deposition at night with typical Vd values of 0.2–0.6 cm s−1, while the estimated values from both models were less than 0.1 cm s−1. Noah-GEM modeled more realistic stomatal resistance (Rs) than WDDM, as compared with observations of water vapor exchange fluxes. The poor performance of WDDM for stomatal uptake is mainly due to its lack of dependence on leaf area index. Thermal decomposition was found to be relatively unimportant for measured PAN fluxes as shown by the lack of a relationship between measured total surface conductance and temperature. Thus, a large part of the underprediction in Vd from both models should be caused by the underestimation of nonstomatal uptake, in particular, the cuticle uptake. Sensitivity tests on both stomatal and nonstomatal resistances terms were conducted and some recommendations were provided.

1. Introduction

[2] Peroxyacetyl nitrate (PAN, CH3C(O)OONO2) is an abundant secondary pollutant of photochemical oxidation, which is produced in the atmosphere by reactions 1 and 2f:

display math
display math

The reverse reaction (R2r) represents the thermal decomposition of PAN, a process highly sensitive to temperature. The reaction of the peroxyacetyl radical (PA, CH3C(O)OO) with NO is the primary removal mechanism of PAN from the atmosphere:

display math

Other chemical decay mechanisms for PAN, including oxidation by hydroxyl radical (OH) and photolysis, are relatively slow and negligible relative to thermal decomposition in the lower troposphere [Singh, 1987; Talukdar et al., 1995].

[3] PAN acts as an important reservoir of reactive nitrogen and plays an important role in photochemical reactions. PAN is thought to contribute significantly to the global transport and distribution of reactive nitrogen as it can be transported over long distances in the free troposphere where low temperatures prevent its thermal decomposition, and it can return to the warmer, lower troposphere in remote areas where NO2 is released from the thermal decomposition reaction and contributes to ozone formation [Singh and Hanst, 1981; Cox and Roffey, 1977; Moxim et al., 1996]. Deposition of PAN on surfaces provides another removal pathway through which the amount of reactive nitrogen within the surface layer is reduced and a source of atmospheric nitrogen is supplied to the ecosystem.

[4] Sparks et al. [2003], using a PAN concentration of 250 pptv in a chamber study, reported maximum dry deposition velocities (Vd) of PAN between 0.20–0.54 cm s−1 for eight plant species. The PAN uptake was found to be controlled by the stomatal pathway. In earlier field measurements, Vd(PAN) was determined by some indirect techniques (e.g., the boundary layer budget technique, the gradient method). Shepson et al. [1992] measured Vd(PAN) greater than 0.5 cm s−1 at night over a rural/agricultural area and two deciduous/coniferous forest sites. Schrimpf et al. [1996] determined Vd(PAN) to be 0.54 ± 0.94 cm s−1 during nighttime above a corn field. Doskey et al. [2004] reported a small Vd(PAN) of 0.13 ± 0.13 cm s−1 during the daytime at a grassland site and suggested that the stomatal uptake contributed mainly to the PAN flux while the contribution of thermal decomposition was less than 15%. Currently, the eddy covariance (EC) method is being used as a standard technique for flux measurements for energy, water vapor, CO2, as well as air pollutants [Ingwersen et al., 2011]. Farmer et al. [2006] conducted full annual cycle EC measurements of total peroxyacyl and peroxy nitrates (∑PNs) over a ponderosa pine canopy at the University of California Blodgett Forest Research Station. Downward fluxes were observed during wintertime, but the summertime measurements showed upward fluxes, probably due to a chemical flux divergence involving oxidation of NO2 and acetaldehyde below the measurement height by elevated OH concentration within the canopy [Farmer and Cohen, 2008]. Recently, reliable fast response sensors for PAN have been developed and the EC fluxes of PAN to canopy were measured over two U.S. pine forests at Duke Forest, North Carolina [Turnipseed et al., 2006] and at Blodgett Forest, California [Wolfe et al., 2009]. Both data sets showed net downward PAN flux and similar diurnal trends that peaked around midday and remained at smaller values throughout the night. Vd(PAN) at Duke Forest showed a daytime maximum of ∼1 cm s−1 and typical nighttime values of 0.2–0.6 cm s−1, which were much faster than predicted values by the Regional Acid Deposition Model (RADM). The deposition of PAN represented ∼20% of the simultaneously measured daytime NOy flux. At Blodgett Forest, the measured maximum Vd(PAN) showed typical values of 0.4 cm s−1 and nighttime values were generally <0.2 cm s−1. At Blodgett Forest, since there were strong vertical gradients in temperature due to surface heating, thermal decomposition was estimated to account for 31–65% of the measured PAN flux, and the portion varied significantly with temperature. Wolfe and Thornton [2011]developed a vertically resolved 1-D chemical transport model (CAFE) which had a fine vertical resolution of 0.1 m near the surface and described the details of turbulent diffusion, emission, deposition, and chemical reactions extending from the within canopy air space to the mixed layer. Modeled gradients of PAN concentrations were found to be in good agreement with the measurements, however the underprediction in the PAN flux andVdreached above a factor of 2. This model-measurement disagreement was attributed to the underestimation of nonstomatal uptake, or uncertainties in gas phase chemistry and vertical mixing [Wolfe et al., 2011].

[5] To date, the performance of Vd(PAN) estimation by dry deposition models widely used in Chemistry and Transport Models (CTM) has not been systematically evaluated. Although the dry deposition process cannot compete with chemical losses because of thermal decomposition at typical daytime temperatures during the summer season, it could be a primary sink of PAN at night or during cold seasons [Turnipseed et al., 2006]. It may also represent a significant source of atmospheric nitrogen to ecosystems that has thus far been ignored [Sparks et al., 2003]. Therefore, Vd(PAN) in CTMs must be treated as accurately as possible in order to predict PAN deposition and O3 formation, especially for remote areas. It is imperative to assess the ability of dry deposition models to capture the variation in Vd(PAN).

[6] The present study aims to evaluate the performance of two community dry deposition models in calculating Vd(PAN), taking advantage of the newly available data of Turnipseed et al. [2006]. One model is the WRF-Chem dry deposition module [Grell et al., 2005; Wesely, 1989] (hereinafter referred to as WDDM) and the other is the Noah land surface model (LSM) [Chen and Dudhia, 2001; Ek et al., 2003] coupled with a photosynthesis-based Gas Exchange Model [Niyogi et al., 2009; Wu et al., 2011] (hereinafter referred to as Noah-GEM). We previously evaluated the performance of WDDM and Noah-GEM in simulatingVd(O3) and Vd(NOy) [Wu et al., 2011] and found that a large WDDM measurement discrepancy was attributed to the minimum canopy stomatal resistance (Ri) based treatment of the canopy and Noah-GEM shows a better ability to capture the variations of canopy stomatal uptake than WDDM. In this paper, we coupled Noah-GEM with a more sophisticated nonstomatal resistance scheme [Zhang et al., 2003] which is a function of u*, RH, LAI, and canopy wetness for nonstomatal uptake, instead of the constant values used in WDDM. The measurement site, available data, and the two dry deposition models are briefly described in section 2. Model evaluation and sensitivity tests are described in detail in section 3. The summary of major conclusions and some recommendations are provided in section 4.

2. Methodology

2.1. Site Description

[7] The measurements were conducted at the Duke Forest FACE (Free Air CO2 Enrichment) site at one of the control towers (i.e., no enhanced CO2 levels). The site is in a 25 year old forest plantation of Loblolly pine (Pinus taeda L.) with a diversity of understory primarily of sweetgum (Liquidambar styraciflua), but also containing red maple (Acer rubrum), winged elm (Ulmus alata Michx), and flowering dogwood (Cornus florida) in North Carolina (35.98°N, 79.09°W; elevation, 163 m). The vegetation is fairly homogeneous within the distance of ∼1 km in the direction of the dominant wind direction (southwest). The canopy height is on average ∼17 m with a leaf area index (LAI) of 5.6 m2 m−2 [Palmroth et al., 2005].

[8] Measurements of 30 min average PAN fluxes and concentrations were taken at a height of 26 m between 16 and 25 July 2003 (day of year 197–206), accompanied with the associated meteorological data (i.e., air temperature, relative humidity, wind speed/direction, surface pressure, precipitation rate, and solar radiation). Turnipseed et al. [2006]reported that the average relative flux error on single flux measurements was ±41%. Although not reported in the previous paper, daytime values were typically smaller (25–35%). This large measurement uncertainty causes a large degree of scatter in the direct 30 min flux comparisons with the model. Therefore we also provided a comparison of simulated to observed diurnal averages, since this reduces the large run-to-run measurement variability. Further details on the site, instrumental methods, and uncertainty analysis can be found athttp://c-h2oecology.env.duke.edu/site/main.html or in the work of Turnipseed et al. [2006].

2.2. Model Description and Configuration

[9] In existing single-layer (or big leaf type) deposition models, the dry deposition velocity (Vd) for a gaseous compound is determined as the reciprocal of a series of resistances to transport down to the surface [Wesely and Hicks, 2000]:

display math

where Ra is the aerodynamic resistance, Rbthe quasi-laminar sublayer resistance,Rc the surface resistance, and z the reference height.

[10] The conventional micrometeorological approaches based on similarity theory are used to estimate Ra and Rbin WDDM and Noah-GEM, as described byWu et al. [2011], while the fundamental difference between WDDM and Noah-GEM exists in theRc parameterization employed. The deposition may take place both through stomata and onto the exterior surface. Rc can then be generalized from both models as

display math

where Rs is the canopy stomatal resistance, Rm the mesophyll resistance, Rns the resistance for uptakes by leaf cuticles, bark, soil, or ground litter, grouped together as nonstomatal, and Wst the fraction of stomatal blocking under wet conditions [Zhang et al., 2003]. For WDDM, the effect of stomatal blocking is not considered and a value of 0 is given to Wst. Instead, dry and wet canopies were treated separately in this model. WDDM employs the Rc parameterization of Wesely [1989] that estimates Rs using a minimum canopy stomatal resistance (Ri) according to

display math

where G is the solar irradiation and Ts the surface air temperature. In the work of Wesely [1989], Rns is formulated as

display math

where Rlu is the leaf cuticle resistance, Rdc the resistance for buoyant convection in canopies, Rcl the resistance for leaves, twig, bark, or other exposed surfaces in the lower canopy, Rac the in canopy aerodynamic resistance, and Rgs the ground or soil resistance. The Rns components for SO2 and O3are mainly derived from look-up tables, and then scaled for other gases based on their effective Henry's law constant (H*) and reactivity factor (f0). The readers are directed to the work of Wesely [1989]for a detailed description of all of these formulas and their parameters. Noah-GEM estimatesRs by considering the physiological process of the leaf response to net CO2 assimilation/photosynthesis rate (An), the relative humidity fraction at the leaf surface (hs), and CO2 partial pressure at the leaf surface (Cs) [see Ball et al., 1987; Niyogi et al., 2009; Wu et al., 2011]:

display math

where P is the atmospheric pressure and m and b are the slope and intercept obtained by linear regression analysis of data from gas exchange experiments. The Rnscalculated in Noah-GEM is parameterized according toZhang et al. [2003]

display math

where Rac is the in canopy aerodynamic resistance which is common to all gases and Rg and Rcut are the resistances for the uptake by ground/soil and canopy cuticle. Similar to the work of Wesely [1989], Rg and Rcut are parameterized for SO2 and O3 and then scaled for other gases according to

display math

where i represents the particular gas and parameters α and βare two scaling factors based on the chemical species' solubility and half-redox reactivity.Zhang et al. [2003] parameterized the nonstomatal resistances as a function of u*, RH, LAI, and canopy wetness.

[11] Another difference between WDDM and Noah-GEM is the wetness definition. WDDM defines the surfaces to be wet either during precipitation events or whenRHis >95%. Noah-GEM calculates the water content on the canopy surface (CMC) from the intercepted precipitation and dew formation [Chen and Dudhia, 2001] and then defines the surfaces to be wet when CMC is >0, which is more sophisticated compared with that in WDDM.

[12] The WDDM was extracted from the WRF-Chem model V3.1.1 and executed in a 1-D mode. The Noah LSM V3.1 (with GEM) was executed in the same fashion. The prescribed surface parameters were modified according to the site conditions (z0 = 1.7 m, LAI = 5.6 m2 m−2). The 30 min interval measurements of air temperature (T), relative humidity (RH), wind speed (WS), wind direction (WD), atmospheric pressure (Pa), precipitation rate (Precip), and downward shortwave radiation (Rg_in), gap filled by the data from the Duke Forest Loblolly Pine AmeriFlux site, were used to drive Noah-GEM. The forcing input of downward longwave radiation (Rlong_in) was from the one-eighth degree hourly North American Land Data Assimilation System (NLDAS) outputs [Mitchell et al., 2004; see also http://www.emc.ncep.noaa.gov/mmb/nldas/]. Noah simulated friction velocity (u*) and Obukhov length (L) via a Monin-Obukhov similarity based iterative process, using air temperature, humidity, pressure, and wind speed [Chen et al., 1997; Chen and Zhang, 2009]. WDDM requires inputs of T, Rg_in, RH, Precip, u* and L (L = −ρCpu*3θ/kgH) obtained from measurements.

3. Results and Discussion

3.1. Initial Model Performance

[13] Figure 1 compares simulated Vd(PAN) by WDDM and Noah-GEM against the observations, with different boxes noting different surface conditions (dry or wet, determined by Noah-GEM).Table 1 presents the statistical results of the comparison.

Figure 1.

(a) Comparison of time series for observed and modeled Vd(PAN). The shaded (white) boxes correspond to the wet (dry) surface conditions as determined from the water content on the canopy surface computed in the Noah land surface model coupled with a photosynthesis-based Gas Exchange Model (Noah-GEM). (b) Time series of observed relative humidity (RH) and precipitation (Precip).

Table 1. Statistical Results of Observed and Modeled Vd(PAN) (cm s−1)a
 All (N = 349)Day, Dry (N = 121)Day, Wet (N = 30)Night, Dry (N = 41)Night, Wet (N = 108)
  • a

    Ave and Med are average and median values; N is the number of data samples; R is the correlation coefficient between observation and model simulation; Day is 08:00–18:00 (LST); Night is 20:00–05:00 (LST); “Dry” or “Wet” indicates the surface conditions as shown in Figure 1; A1 indicates that Ri was adjusted from the initial value of 130 s m−1 to 40 s m−1; A2 indicates that Ri was adjusted from the initial value of 130 s m−1 to 40 s m−1 and Rlu of 650 s m−1 and 125 s m−1 were assigned for dry and wet surfaces, respectively; S1 indicates the measured u* and Lare used instead of that calculated by Noah-GEM; S2 indicated the measuredu* and L are used and the reactivity scaling factor β was adjusted from the initial value of 0.6 to 2.

WDDM A10.240.050.310.520.570.480.310.330.
WDDM A20.500.550.570.610.680.480.680.710.460.270.180.580.400.420.49
Noah-GEM S10.320.260.400.550.590.380.400.410.510.110.080.610.150.120.58
Noah-GEM S20.530.510.520.710.730.310.840.860.520.250.180.610.350.290.58

[14] The observed Vd(PAN) displayed large variations, but exhibited a diurnal pattern with peak values (on average) of ∼1.0 cm s−1 at midday. Nocturnal Vd(PAN) was small, but significantly different from zero with typical values ranging from 0.2 to 0.6 cm s−1. There were several detectable precipitation events during the measurement period and the Duke Forest site was also quite humid (Figure 1b). Careful examination of the time series in Figure 1 reveals a clear tendency toward larger Vd(PAN) under wet conditions because of rain or dew [see also Turnipseed et al., 2006]. Also statistical results in Table 1 shows that the median Vd(PAN) values at night(day) were 0.14 (0.53) and 0.43 (0.69) cm s−1 for dry and wet conditions, respectively, exhibiting significantly enhanced PAN uptake by the wet canopy surface. The increase of Vd(PAN) because of wetness enhancement during the daytime was smaller than that during nighttime, which should be due to the offset effect of stomatal blocking under daytime wet conditions, similar to what was found for O3 [Zhang et al., 2002].

[15] WDDM predicted much smaller Vd(PAN) compared to observations under all conditions (i.e., day or night, dry or wet) with a correlation coefficient of 0.32. Vd(PAN) modeled by WDDM peaked at ∼0.2 cm s−1 during noon and remained at a very low value of 0.04 cm s−1, showing almost no differences between dry and wet conditions. Noah-GEM predicted largerVd(PAN) than WDDM, but still did not agree well with the observations, with a correlation coefficient of 0.24. During nighttime, Vd(PAN) modeled by Noah-GEM was larger and exhibited more variations than WDDM, more closely approaching the observations in terms of both magnitude and correlation. During the daytime, Noah-GEM underestimatedVd(PAN) around noontime and overestimated Vd(PAN) in the afternoon, resulting in relatively low correlations (R = 0.17–0.32). Wu et al. [2011]also found Noah-GEM tended to overestimateVd(O3) in the afternoon, which was explained by the lack of considering the accumulation effect of stomatal fluxes on plant leaf uptake in the model. Other possible reasons include that the stress function currently being used for soil moisture interaction or the coupling coefficient being used are insufficient [Chen and Zhang, 2009; Charusombat et al., 2012] and will need to be modified in future work.

[16] As previously discussed, WDDM and Noah-GEM employ differentRs and Rns schemes. Figure 2shows that there were large differences between WDDM and Noah-GEM in estimatingRs and Rns. WDDM simulated a nearly constant Rns of ∼2400 s m−1, while Rnssimulated by Noah-GEM exhibited a significant diurnal trend with smaller values during the daytime (∼500 s m−1) and larger values at night (∼1500 s m−1). Diurnal profiles in Rs were similar as the Rsschemes in both models assumed complete stomatal closure at dark. However, Noah-GEM predictedRs of ∼250 s m−1 at noon, only half of that modeled by WDDM.

Figure 2.

Average diurnal cycles of stomatal resistance (Rs) and nonstomatal resistances (Rns) for peroxyacetyl nitrate (PAN) modeled by WRF-Chem dry deposition module (WDDM) and Noah-GEM.

[17] At night, stomatal uptake and thermal decomposition are all suppressed because of stomatal closure, larger NO2/NO ratios, and lower temperature. The nonstomatal uptake to other surfaces, including leaf cuticles, bark, soil, or ground litter should be considered as the only pathway for PAN decay. So Rns equals Rc at night. As in the work of Zhang et al. [2002], we determined the observed Rc according to

display math

where Ra and Rb were calculated in WDDM using measured meteorological variables (i.e., u* and L), and the observed Vd were used in equation (8). We derived the observed Rnsat night by only using the nighttime data. We then divided the whole data into data sets under dry and wet conditions which was determined from the water content on the canopy surface computed in Noah-GEM (seesection 2.2 and Figure 1). The median Rns was calculated to be 650 s m−1 under dry conditions and decreased to 125 s m−1 under wet conditions, much smaller than the estimates by WDDM (∼2400 s m−1) and Noah-GEM (∼1500 s m−1). Mean Rns values are strongly influenced by a small number of extreme large data points. Turnipseed et al. [2006] adopted a similar approach and suggested 244 s m−1 when dry and 125 s m−1 when wet. The main difference in calculating the Rns between this study and the work by Turnipseed et al. [2006] is the definition of wetness. In the work of Turnipseed et al. [2006], an oversimplified definition of wetness similar to WDDM was used where the surface is defined to be wet either during and immediately following precipitation events or when ambient RH was >96%. The data subset under dry conditions obtained using that wetness definition should include some time points with residual precipitation on the canopy surface which was divided into the wet period in this study, and thus had a larger average Vd and lower Rns.

[18] Because direct measurements of Rs were not available at the Duke Forest site, examining the observed surface heat fluxes against the model outputs provided an independent assessment of Rs(surface heat fluxes are the primary outcomes for LSMs such as Noah-GEM, but are not provided by WDDM).Figure 3shows that Noah-GEM–modeled heat fluxes are in good agreement with measurements, thereby demonstrating some confidence in the simulation ofRsby Noah-GEM, which in turn illustrates that WDDM overestimatedRs, by almost a factor of 2.

Figure 3.

Comparison of observed and Noah-GEM–modeled average diurnal variations of latent heat flux (LE) and sensible heat flux (H).

3.2. Chemical Influence on Vd(PAN)

[19] The measured flux of PAN to the canopy can be contributed by surface uptake (deposition) as well as air space chemical reactions below the sensor height [Doskey et al., 2004; Turnipseed et al., 2006; Wolfe et al., 2009].

[20] The PA radical generated from the thermal decomposition of PAN (reaction (R2r)) can also deposit on the surface, potentially contributing to an observed downward flux of PAN. As the PA radical is one of the most powerful oxidants among the peroxy radicals [Schuchmann and von Sonntag, 1988] and the measurements [Villalta et al., 1996; Roberts et al., 1996] demonstrated that the heterogeneous hydrolysis could be a very efficient loss process for PA on an aqueous surface, we would expect a fast surface uptake of PA under both dry and wet conditions. Wolfe et al. [2009] estimated the contribution of PA deposition to PAN flux in the Blodgett Forest case and found the fraction was <5% even when considering the surface as a perfect sink for PA, i.e., the maximum Vd of PA (Vmax = (Ra + Rb)−1) was given. The small fraction was mainly attributed to the much lower PA concentrations compared with PAN. Turnipseed et al. [2006] also suggested a negligible influence of PA surface deposition on PAN flux in the Duke Forest case from the viewpoint of time scales since the calculated rate of PAN thermal decomposition is much slower than the rate of turbulent diffusion.

[21] Both Doskey et al. [2004] and Wolfe et al. [2009] showed evidence that within the atmospheric surface layer, thermal decomposition of PAN (reaction (R2r)) and subsequent loss via reaction (R3) can contribute to the measured above canopy flux when the thermal decomposition rate varies significantly with altitude below the flux measurement height. Although this also depends on the gradient of the chemical partitioning between NO and NO2, Wolfe et al. [2009] showed that the change in the decomposition rate was the major contributing factor. If a large temperature gradient due to strong daytime surface heating is present, the equilibrium of reaction (R2) shifts toward PA and NO2 in the region closer to the surface, which can lead to a gradient in PAN concentrations and, thus, a downward flux of PAN (as was the case in the study of Wolfe et al. [2009]).

[22] Reliable absolute temperature gradient measurements (as well as chemical gradients of [NO]/[NO2]) were not available during the Duke Forest campaign; however, temperatures derived from a vertical profile of sonic anemometers (at z = 3, 8, 16, and 26 m) showed the expected pattern of a warmer daytime canopy and colder temperatures within the canopy at night (Figure 4). Although the uncertainty of these gradients is likely substantial, this data suggested that midday temperature gradients were typically <1°C (often only around 0.5°C). Further evidence to support small temperature gradients comes from the observations that latent heat fluxes were typically twice as large as sensible heat fluxes (Figure 3), indicating that energy partitioning in this ecosystem strongly favored water evaporation over sensible heating of the air. Wolfe et al. [2009] observed that higher values of surface conductance (gc(PAN)) were associated with increased temperatures. Reevaluating the data set of Turnipseed et al. [2006], we do not see such a dependence (compare our Figure 5a with Figure 8 of Wolfe et al. [2009]). Figure 5b shows the variation of the residual conductance (gr(PAN) = gc(PAN)-gs(PAN)) with increasing temperature. There was no increase of gr(PAN) with increasing temperature, suggesting that thermal decomposition was relatively unimportant for the observed PAN fluxes during this study. However, as shown by Wolfe et al. [2009] and Doskey et al. [2004], the thermal decomposition effect was an important contributing factor during periods of strong sensible heat flux and must be considered when comparing with deposition models.

Figure 4.

Air temperatures measured by the sonic anemometer at the height of 3, 8, 16, and 26 m above the surface (3, 8, and 16 m were at the bottom, middle, and top of the canopy, respectively; 26 m was the flux measurement height). Data has been removed for the hours during and immediately following precipitation events. For median midday mixing ratios of [NO] = 198 pptv and [NO2] = 1666 pptv, the thermal lifetime for PAN is estimated to be 1.1 h at 32°C; for median midnight mixing ratios of [NO] = 91 pptv and [NO2] = 1662 pptv, the thermal lifetime for PAN is estimated to be 12.3 h at 20°C.

Figure 5.

(a) Scatterplot of canopy stomatal conductance for PAN (gs(PAN)) versus surface conductance for PAN (gc(PAN)), color-mapped by air temperature. Only points during 08:00–18:00 (LST) are shown, and data has been removed when precipitation was measured. The line drawn is the 1:1 line. (b) Whisker box plot of the residual conductance (gr(PAN) = gc(PAN) − gs(PAN)) for 2°C intervals of the air temperature. On each box, the central mark is the median, the edges of the box are the 25th and 75th percentiles, and the whiskers extend to the extreme data points. The gswas calculated using the inversion of the Penman-Monteith equation;gc was determined as (Vd(PAN)−1RaRb)−1. Further details are provided by Turnipseed et al. [2006].

3.3. Sensitivity Tests and Model Improvements

3.3.1. WDDM Stomatal Pathway (Rs)

[23] WDDM estimates Rs based on the prescribed minimum canopy stomatal resistance (Ri) (equation (3)). Many studies [e.g., Cooter and Schwede, 2000; Wu et al., 2011] have shown that uncertainty in the prescribed Ri can dominate the errors in estimating Rs and thus Vd of the gases controlled by stomatal pathways. Here, we conducted sensitivity tests, differing only by the value of Ri in WDDM and compared the Rsoutput with that calculated from Noah-GEM as it was evaluated by comparing against observations of water vapor exchange. We found thatRi of 40 s m−1 instead of the initial value of 130 s m−1 [see Wesely, 1989] generated values of Rsclose to Noah-GEM. The modeled daytimeVd(PAN) by WDDM increased by more than a factor of two, more closely approaching the observations (Table 1 and Figure 6a). The correlations did not improve significantly, however, suggesting that some key factors could not be captured by the model if only modifying Ri.

Figure 6.

(a) Comparison of observed and WDDM-modeled average diurnal cycles ofVd(PAN). (b) Scatterplots of observed and WDDM-modeledVd(PAN). (c) Same as Figure 6a but for Noah-GEM. (d) Same as Figure 6b but for Noah-GEM. The shaded (white) cycles correspond to the wet (dry) surface conditions; A1, A2, S1, and S2 are shown as inTable 1; the lines drawn are the 2:1, 1:1, and 1:2 lines, respectively.

[24] Although reducing Ri (from 130 to 40 s m−1) improved the model performance (on averaged Vd, but not on the correlation between model and measurement) for this site, the reduced value might not be suitable for different forest sites. The real reason for the poor performance of WDDM is that the simple formula for Rs (equation (3)) does not consider factors dominating stomatal resistance, such as the canopy biological information (e.g., LAI) as well as some environmental stress functions (e.g., vapor pressure deficit, and soil moisture) which are essential in order to obtain adaptable Rs prediction over different vegetated surfaces and regions. Many stomatal resistance submodules following the more sophisticated Jarvis style empirical equation are available [e.g., Baldocchi et al., 1987; Hicks et al., 1987; Meyers et al., 1998; Zhang et al., 2001]. Photosynthesis-based semiempirical equations have also been developed for estimating stomatal uptake [e.g.,Niyogi et al., 2009; Wu et al., 2003; Hirabayashi et al., 2011]. A comparison between these two types of stomatal equations by Niyogi and Raman [1997]found that the photosynthesis-based equation is more responsive to the external environmental conditions and thus consistently performs better in simulatingRs than the Jarvis style equation. Nonstomatal Pathway (Rns)

[25] WDDM simply parameterizes the Rns components based on specified constant values, which minimizes the complexity and computational cost. Here we replaced the constant values with those derived from observations (see section 3.1). Turnipseed et al. [2006] suggested that because of the low within canopy transport in the stable conditions of night (the transfer resistance within the canopy were greater than 2000 s m−1 in the models), the primary loss pathway at night was most likely via cuticular deposition over the upper canopy. Therefore we assigned 650 s m−1 when wet and 125 s m−1 when dry to cuticular resistance (Rlu) in WDDM and tested its influence on Vd(PAN) estimates (WDDM A2). One limitation for this sensitivity test is that we also applied those values derived from nighttime measurements to daytime simulations. The Rlu during daytime may be different from that during nighttime, but the difference is expected to be relatively small [Zhang et al., 2002].

[26] Table 1 shows that WDDM A2 modeled a very close Vd(PAN) to the observation at night and the correlation also increased significantly. The adjustment of Rlu also improved the daytime simulation as the bias was reduced and the correlation increased. Figure 7 shows that WDDM A2 was in much better agreement with observations than the initial model results capturing the shape and magnitude of the observations. The simulated Vd(PAN) by WDDM A2 was usually below 1.0 cm s−1, however, around noon on most days, very high Vd values (i.e., >1.0 cm s−1) were observed. Therefore the averaged diurnal curves (Figure 6a) show that WDDM A2 still underestimated the noon peak of Vd(PAN), but when looking at the median values, the model was much closer to the observations (Table 1). Approximately 66% of the WDDM A2 estimates are within a factor of 2 of the observed values (Figure 6b).

Figure 7.

Comparison of time series of observed and modeled Vd(PAN). The shaded (white) boxes correspond to the wet (dry) surface conditions; A2 and S2 are shown as in Table 1.

3.3.2. Noah-GEM Atmospheric Resistances (Ra and Rb)

[27] The standard Noah-GEM model requires the input of routinely available meteorological variables (seesection 2) but calculates some micrometeorological variables (e.g., u* and L) itself via a Monin-Obukhov similarity based iterative process, using air temperature, humidity, pressure, and wind speed [Chen et al., 1997; Chen and Zhang, 2009]. Figure 8shows that Noah-GEM generally reproduced the magnitude and shape of observedu*, but a large discrepancy can be observed especially during nocturnal stable conditions. In Noah-GEM,Ra, Rb and Rns are dependent on u* or L. Figure 6cshows that Noah-GEM modeled a slightly largerVd(PAN) if the measured u* or Lwere used to calculate the resistances instead of that modeled by Noah-GEM. Statistical results (Table 1) reveal that the correlations improved significantly in every condition (day or night, dry or wet). The correlation coefficient for all the data increased from 0.24 to 0.40, implying that Noah-GEM captured a much better variation of the observedVd(PAN) because of improved inputs of u* and L. The sensitivity of Noah-GEMVd estimates to errors in meteorological inputs as well as model parameters will be evaluated systematically in future work.

Figure 8.

Comparison of time series of observed and Noah-GEM–modeled friction velocity (u*).

[28] WDDM and Noah-GEM S1 were driven by the same set ofu*, L, and other standard meteorological variables without any tuning of the model parameters. Table 1shows that Noah-GEM S1 modeled nocturnalVd(PAN) closer to the observations than WDDM, and the correlation for Noah-GEM S1 (R = 0.58–0.61) during nighttime was significantly higher than that for WDDM (R = 0.19–0.35).Zhang et al. [2003] has the explicit dependence of Rns on the meteorological (u* and RH), biological (LAI), and surface (canopy wetness) conditions and the results in this study suggest it is thus superior to the approaches simply using constant values. Future work to improve the model will involve assimilating more environmental factors (e.g., pH of soil and canopy surface water) and leaf scale processes (e.g., lipid dissolving into cuticle membrane, enzymatic conversion within leaf interior [Karl et al., 2010]) that essentially affect the canopy uptake of various pollutants. Nonstomatal Pathway (Rns)

[29] As PAN has high oxidizing capacity and relatively limited solubility, the dry deposition of PAN was assumed to be similar to that of O3 and the solubility scaling factor α and the reactivity scaling factor β (equation (7)) were specified to be 0 and 0.6, respectively, in Noah-GEM [Zhang et al., 2003]. A few studies that simultaneously measured the uptakes of PAN and O3 by vegetation during daytime or nighttime are available [Hill, 1971; Garland and Penkett, 1976; Shepson et al., 1992; Wolfe et al., 2009], but since the daytime fluxes can be influenced by chemical reactions as well as stomatal uptake, the nighttime results are more appropriate for comparing Rns between PAN and O3. There have previously been two studies that simultaneously measured the deposition of PAN and O3 over vegetative area at night. Shepson et al. [1992] obtained average Vd(PAN)/Vd(O3) of 2.4 over three vegetative rural sites by measuring the nighttime concentration losses of PAN and O3 and relating them to deposition, namely the boundary layer budget technique. EC fluxes of PAN and O3 measured at Blodgett Forest by Wolfe et al. [2009] indicated Vd(PAN)/Vd(O3) of 0.50 at night. Thus these two studies indicate that the ratio Vd(PAN)/Vd(O3) ranges from 0.50 to 2.4 at night. Model estimates of this ratio by Noah-GEM S1 was from 0.60 to 1.0 with a median of 0.68, close to the results ofWolfe et al. [2009].

[30] We conducted the sensitivity tests by tuning the reactivity scaling factor βin Noah-GEM. The results show that increasingβcan reduce the bias and enhance the correlation between observations and simulations by Noah-GEM. A value of 2 forβis found a better fit at this specific site. Also note that in Noah-GEM a decrease in cuticular resistance of PAN when the canopy was wet was assumed, which in fact resulted from the assumption of enhanced O3 cuticle uptake by wet canopy in the model [Zhang et al., 2002, 2003] and is consistent with field measurements [Turnipseed et al., 2006]. As shown in Figure 7, with the adjustment of the βparameter from the initial value of 0.6 to 2, Noah-GEM was in much better agreement with the observations than the initial model results. The shape and magnitude of the observations was reproduced well except for a few very high observedVd values >1.3 cm s−1. However, the average diurnal cycles (Figure 6c) show that the model did not capture the midmorning decrease in the observed Vd(PAN) and overestimated Vd(PAN) in the afternoon. Approximately 68% of the Noah-GEM S2 estimates are within a factor of 2 of the observed values (Figure 6d). The ratio Vd(PAN)/Vd(O3) during nighttime modeled by Noah-GEM S2 ranged from 1.0 to 1.8 with a median of 1.6 which was in the middle of the range determined in field (0.50–2.4). Thus increasingβ to the value of 2 still keeps the model outcomes consistent with the field observations. Recently the Community Multiscale Air Quality model (CMAQ) also increased its surface reactivity scaling parameter for PAN to obtain a larger Vd(PAN) prediction based on the findings from recent measurements [Pleim and Ran, 2011]. However, generalization of this assumption is cautioned before measurement data are available over various canopies. It is very likely that the relative magnitude of Vd(PAN)/Vd(O3) and thus the β value varies under different conditions because of the chemical processes involved in the dry deposition processes.

4. Conclusions and Recommendations

[31] Modeled dry deposition velocities for PAN from two community dry deposition models, WDDM and Noah-GEM, were evaluated using eddy covariance flux measurements over a coniferous forest canopy. Both models underestimatedVd(PAN) by a factor of 2–5 and the correlations between the model results and measurements were also very low, e.g., R = 0.32 and 0.24 for WDDM and Noah-GEM, respectively. The underprediction of the modeledVd was largely attributed to the modeled stomatal and nonstomatal uptake since the influence of PAN thermal decomposition on measured fluxes was small.

[32] Through the adjustments of the model parameter or variable, WDDM and Noah-GEM can finally reach a satisfactory performance. However, a large challenge still exists for modelingVd(PAN) as our knowledge of the dry deposition process of PAN is still limited and the current adjustments would be considered more of an empirical fit to observations.

[33] Noah-GEM estimated more realistic stomatal uptake than WDDM (in terms of the averageVd values, but not the correlation) as was shown by a comparison with measured water vapor fluxes. This is because the former included the physiological process including the CO2 assimilation rate while the latter used a prescribed minimum stomatal resistance without key biological factors (e.g., LAI). The newly adopted nonstomatal resistance parameterization in Noah-GEM also had a better capacity to capture the variations inVd(PAN), because of its dependence on biological (LAI), meteorological (u*, RH), and surface (canopy wetness) conditions, than the simple Rns parameterization in WDDM. This was demonstrated by the better modeled average Vdand better correlations from Noah-GEM than from WDDM using nighttime data only (when nonstomatal uptake dominates). The correlation between Noah-GEM–modeledVd and measurements was also improved if using observed u* and L instead of model calculated values, implying the importance of meteorological variables on the dry deposition processes.

[34] On the basis of sensitivity tests conducted in the present study and our current understanding of dry deposition processes, the following recommendations are provided for dry deposition modeling approaches suitable for use in chemical transport models. The stomatal uptake of PAN and many other gaseous pollutant species should be modeled using stomatal resistance submodules that at least consider key biological and meteorological variables that are available as input parameters to host models. As an example, the sunlit/shade stomatal resistance modules provide results as good as multilayer stomatal resistance modules but require less computer resources and fewer input parameters and are much more accurate than single-leaf modules [e.g.,De Pury and Farquhar, 1997; Zhang et al., 2001]. Further study is needed for the comparison of more sophisticated Jarvis type stomatal resistance modules with photosynthesis approach–based modules to help the community choose proper stomatal uptake modules.

[35] Nonstomatal uptake contributes ∼50% of total dry deposition fluxes for PAN, O3,and many other species and need to be modeled more accurately. Most nonstomatal resistance parameterizations currently used in chemical transport models are very simple and require further evaluation and improvement. Many recent studies confirmed the strong dependence of the nonstomatal uptake on meteorological conditions, especially the friction velocity and canopy wetness. Thus, it is recommended that nonstomatal uptake include key meteorological and biological factors. High-quality dry deposition flux measurement data for PAN and other lesser understood species over various canopies are also needed in order to improve the nonstomatal resistance modules.

[36] Leaf area index is a key parameter for both stomatal and nonstomatal uptake; high temporal and spatial resolution data for LAI from remote sensing should be used as input for dry deposition modules within chemical transport models. Consistence of input parameters and/or meteorological variables (e.g., roughness length, soil moisture, canopy wetness, and many boundary layer micrometeorological variables) between the dry deposition module and its host model is preferred.

[37] The implementation of Noah-GEM calculatedVdin the WRF-Chem model is underway. This work is a first step and preliminary study to evaluate the dry deposition of nitrogen-bearing compounds which are critical for estimating the surface and atmospheric nitrogen budget in atmospheric chemistry models. Further evaluations using measurements over various sites and covering longer time periods are needed to better understand the dry deposition of PAN and narrow the uncertainties in model parameters.


[38] This work was supported in part by the following grants: the Natural Science Foundation of China (grants 40875076 and U0833001), the National Program on Key Basic Research Project of China (973) (grant 2010CB428504), the Fundamental Research Funds for the Central Universities, the National Center for Atmospheric Research (NCAR) Advanced Study Program (ASP), the NCAR BEACHON (Bio-hydro-atmosphere interactions of Energy, Aerosols, Carbon, H2O, and Organics and Nitrogen) Program, the NCAR Water System Program, and the U.S. NSF DRInet (OCI and Hydrology). NCAR is operated by the University Corporation for Atmospheric Research (UCAR) under the sponsorship of the National Science Foundation (NSF). The authors thank the Duke FACE site and its staff for allowing us to work at their research site as well as the invaluable assistance they provided during this experiment. We would also like to thank Gabriel Katul and Paul Stoy for their permission to use data from the Duke Forest AmeriFlux data archives.