Estimates of regional natural volatile organic compound fluxes from enclosure and ambient measurements

. Natural volatile organic compound (VOC) emissions were investigated at two forested sites in the southeastern United States. A variety of VOC compounds including methanol, 2-methyl-3-buten-2-ol, 6-methyl-5-hepten-2-one, isoprene and 15 monoterpenes were emitted from vegetation at these sites. Diurnal variations in VOC emissions were observed and related to light and temperature. Variations in isoprene emission from individual branches are well correlated with light intensity and leaf temperature while variations in monoterpene emissions can be explained by variations in leaf temperature alone. Isoprene emission rates for individual leaves tend to be about 75% higher than branch average emission rates due to shading on the lower leaves of a branch. Average daytime mixing ratios of 13.8 and 6.6 ppbv C isoprene and 5.0 and 4.5 ppbv C monoterpenes were observed at heights between 40 m and 1 km above ground level the two sites. Isoprene and monoterpenes account for 30% to 40% of the total carbon in the ambient non-methane VOC quantified in the mixed layer at these sites and over 90% of the VOC reactivity with OH. Ambient mixing ratios were used to estimate isoprene and monoterpene fluxes by applying box model and mixed-layer gradient techniques. Although the two techniques estimate fluxes averaged over different spatial scales, the average fluxes calculated by the two techniques agree within a factor of two. The ambient mixing ratios were used to evaluate a biogenic VOC emission model that uses field measurements of plant species composition, remotely sensed vegetation distributions, leaf level emission potentials determined from vegetation enclosures, and light and temperature dependent emission activity factors. Emissions estimated for a temperature of 30øC and above canopy photosynthetically active radiation flux of 1000 gmol m '2 s -i are around 4 mg C m -2 h -1 of isoprene and 0.7 mg C m '2 h -1 of monoterpenes at the ROSE site in western Alabama and 3 mg C m -2 h -1 of isoprene and 0.5 mg C m -2 h -1 of monoterpenes at the SOS-M site in eastern Georgia. Isoprene and monoterpene emissions based on land characteristics data and emission enclosure measurements are within a factor of two of estimates based on ambient measurements in most cases. This represents reasonable agreement due to the large uncertainties associated with these models and because the observed differences are at least partially due to differences in the size and location of the source region (flux footprint) associated with each flux estimate.


Introduction
Ambient mixing ratios of volatile organic compounds (VOC) are important variables in the chemistry and transport models used to investigate regional tropospheric oxidant Oak-pine savanna occurs in upland areas and is dominated by several species of oaks including turkey oak (Quercus laevis), sandpost oak (Q. margaretta), and bluejack oak (Q. incana). Longleaf pine (Pinus palustris) and loblolly pine (P. taeda) typically are found in these areas as well. The pine-oak forests also occur in uplands and are characterized by a codominance of pine and oak in terms of leaf biomass, though pines comprise most of the woody biomass. These forests include loblolly pine (P. taeda), slash pine (P. elliottii), bluejack oak (Q. incana), and water oak (Q. nigra), along with persimmon (Diospyros virginiana) and black cherry (Prunus serotina). Age structure analyses comparing the oaks and the pines in these forests indicate that the oak are, on average, older that the surrounding pines. This suggests that the oakpine savanna is a relatively early successional community arising after a disturbance, probably fire, and that the pine-oak forest is a later successional community in the development of upland forests.
The bottomland hardwood forests are highly productive, tall-canopied communities occurring in moist and wet lowlying areas. These are characterized by a wide variety of species which vary in abundance from site to site. The two common trees in these forests are blackgum (Nyssa spp.) and red maple (Acer rubrum), with cypress (Taxodium spp.) in permanently inundated areas. There is a well-developed subcanopy tree layer in these forests that includes red titi (Cyrilla racemifiora) and sweetbay (Magnolia virginiana).
A land characteristics database was generated using the four visible and near-infrared bands of the Landsat multispectral scanner (MSS)sensor and is referred to in this paper as the LCC-MSS database. The MSS sensor uses broadbands that cover a relatively wide range of wavelengths dnd has a nominal spatial resolution of 80 m [Jensen, 1986]. supervised classification, in which spectral signatures are generated for each land use category using "training areas" and then used as a basis for classifying the rest of the image, and (2) unsupervised classification, in. which spectrally similar pixels are classed together and later identified as a particular land use category by the analyst. Both methods rely on ground truth information gathered from other sources, which here consisted of a color infrared aerial photo obtained on February 16, 1988, and site observations made during the 1991 SOS field study. Image input for each of the classifications consisted of the normalized difference vegetation index (NDVI) for each of the three scenes [Jensen, 1986]. This vegetation index is computed as the ratio of the difference between the near-infrared (NIR) and visible red band (RED) reflectances, divided by the sum of those two reflectances: where Co is the concentration in the outlet airstream, Ci is the concentration in the inlet airstream, f is the flow rate into the enclosure, and B is the dry weight foliar mass within the enclosure. Air was passed over an enclosed emission source at a constant rate. Ambient air flowing into an enclosure was pumped through Teflon tubing from locations selected to minimize background concentrations. Uncertainties in Co, f, and B result in a total uncertainty of about _+10%.

NDVI = (NIR-RED)/(NIR + RED)
Uncertainties in Ci result in approximately _+0.03 gg C g-1 h-I errors in flux estimates which are typically less than 5% of the emission rate for major VOC species. A sufficient flow rate, 1 to 10 L min-1 depending on enclosure size, was maintained so that the temperature, humidity, and CO 2 mixing ratios within the chamber were similar to ambient conditions. Individual leaves were sampled with a photosynthesis measurement system (LI-6200, LICOR, Lincoln, NE) with a 1.5 -L cuvette. This commercial system was modified to operate in an open-flow configuration and to allow sampling of inlet and outlet airstreams. Tree branches, shrubs, and ground cover were enclosed with bag enclosures that ranged in volume from 15 to 30 L. The bag enclosures consisted of a rigid aluminum frame covered by a flexible Teflon bag. Bag enclosures were supported by a tripod to minimize contact with the enclosed vegetation. Leaf temperature, enclosure temperature, relative humidity, photosynthetically active radiation (PAR), and general sampling conditions were recorded for each enclosure measurement. Photosynthesis, transpiration, and stomatal conductance of representative leaves were measured with a LI-6200 system. When an emission rate measurement experiment was completed, the foliage was cut and leaves were dried in an oven and weighed to obtain dry-weight biomass. The ratio of dry-weight leaf biomass to leaf area (measured prior to drying) was determined for each species to provide a conversion factor. Some enclosure air samples were analyzed for a variety of VOC and used to develop VOC emission potentials. These samples were analyzed by gas chromatography (GC) systems with cryogenic preconcentration using a flame ionization detector (FID) (HP5890, Hewlett Packard, Palo Alto, California) to quantify concentrations and a mass spectrometer detector (MSD)(HP5971, also Hewlett Packard) to identify compounds. The lower detection limit is approximately 10 parts carbon per trillion (pptv C) by volume for a 1-L sample using the GC-FID instrument [see Greenberg and Zimmerman, 1984]. This corresponds to a lower bound flux of less than 0.001 gg C g-1 h-1 for individual VOC. Most samples were analyzed at the field site within 48 hours. Representative samples were stored in stainless steel canisters and transported to a permanent laboratory to provide quality assurance and positive identification of all VOC compounds.
Other enclosure air samples were analyzed only for isoprene, ct-pinene, •-pinene, and limonene and used to investigate diurnal emission rate variations. These samples were collected in 20-mL glass syringes and analyzed at the field site using an isothermal gas chromatographic system with a reduction gas detector (Trace Analytical RGD-2).

Ambient Measurements
Ambient air samples were collected in Teflon bags using a whole air sampling unit similar to the system described by Zimmerman et al. [1988]. The samplers were attached to the tether line of a helium-filled tethered balloon or to a pulleymounted line on a tower. Automatic timers on the sampling pumps allowed air samples to be collected simultaneously at two to four heights between 10 m and 1 km above ground level. Sample periods were typically 15 min for the ROSE study and 30 min for the SOS-M study. All ambient air samples were analyzed by the GC-FID and GC-MSD systems described above.
The planetary boundary layer at both sites was characterized using an Advanced Data Assimilation System, tethersondes, and airsondes manufactured by AIR (Boulder, Colorado). The tethersonde provided vertical profiles of wind direction, wind speed, temperature and humidity up to 1 km AGL. Temperature and humidity profiles up to 5 km AGL were measured with airsondes.
Mixing layer heights at ROSE were estimated from Doppler radar measurements [White and Fairall, 1991]. Temperature and latent heat fluxes and momentum flux were measured above the forest canopy with instruments deployed on a tower at the ROSE site using the eddy correlation technique (R.T. McMillen, private communication, 1991).

Enclosure Measurements and Emission
Modeling Zimmerman [1979] estimated natural VOC emissions using a simple inventory approach, where an emission rate was multiplied by a leaf biomass factor and a temperature correction factor. Subsequent efforts have used increasingly more accurate and highly resolved input variables and have employed algorithms that provide a more realistic simulation of variations in input variables and the response of emissions to these variables [see Lamb et al., 1993;Guenther et al., 1995;Geron et al., 1994]. An area flux F is estimated from the product of the following three components: foliar mass estimates, emission potentials representative of a specific temperature and PAR, and an emission activity level that accounts for the actual temperature and PAR conditions. Each of the three model components was investigated in this field research program and is described in this section.

Foliar Mass Estimates
Estimates of vegetation distributions around the two field sites were obtained from the following five land cover databases: geoecology, U. S. Department of Agriculture  Table 1 show that forest cover estimates range between 66 and 77% around the ROSE site and 58 to 61% at the SOS-M site. These estimates agree quite well, given the different categorization schemes, e.g., some databases grouped areas as mixed forest and cropland, and that the databases represent different years, e.g., the LCC-MSS ROSE database is based on 1988 data while the geoecology and USDA sources represent data compiled between 1970 and 1982. As discussed in the following section, VOC emissions from the foliage of different forest species vary considerably. The emission potentials for different forest types can vary b y more than a factor of 5. We have grouped the forests at the two sites into three categories that roughly correspond to the three forest types used in early emission inventory procedures (e.g., Table 1 do not agree well at this level of landscape characterization. This is partly due to differences in categorization schemes. The geoecology database greatly underpredicts the amount of coniferous forest because it does not account for the conversion of native mixed forests into pine plantations.

Zimmerman 1979). The four methods shown in
Guenther et al. [1994] have shown that three forest categories are not sufficient for natural VOC emission modeling and that the contribution of each plant genus to the total leaf biomass should be estimated when possible. Table 2 contains estimates of foliar mass of the dominant plants at each field site. Over 90% of the total foliar mass at either site can be accounted for by fewer than 10 genera of plants. The    Genus  Examvie  EWDB  AVHRR  MSS  GEO  EWDB  AVHRR  MSS  GEO  I  T   Acer  maple  22  0  23  0  11  0 Table 2 indicate that the estimates for less common trees often differ by more than a factor of 10.

SOS-M Foliar Mass g m -2 ROSE Foliar Mass g m -2 EP
Estimates of some of the dominant trees including pines (Pinus spp.), sweetgum (Liquidambar styracifiua), and gum (Nyssa spp.) also differ considerably. Estimates of oak (Quercus spp.) foliar mass, the major source of isoprene emission in these forests, are fairly consistent among the four databases.

Emission Potentials
VOC emission potentials (the emission rate at a specified temperature and light intensity) of 30 plant species representing 20 genera were characterized by the measurements described in section 2.2. These plants include all of the species that contribute a significant portion (>0.5%) of the total foliar density at the two field sites. Average isoprene and monoterpene emission rates for each plant genus are shown in Table 2. The emission algorithms described by Guenther et al.
Most foliar emission rate measurement surveys [e.g., Zimmerman, 1979] have used whole branch enclosure techniques. This method works well for compounds which are not light dependent but complicates the measurement of isoprene emission rates which are strongly dependent on light conditions [Guenther et al., 1991[Guenther et al., , 1993. The shaded leaves on the lower portion of a branch have a considerably lower emission rate than leaves that are in direct sunlight. As a result, the average emission rate for a branch will be lower than the emission rate of a leaf. Early efforts to model regional VOC emissions [e.g., Lamb   In addition to the hemiterpene isoprene, 15 monoterpene (ct-pinene, [3-pinene, limonene, sabinene, myrcene, terpinene, tricyclene, ct-thujene, camphene, t-ocimene, ctphellandrene, A3-carene, ct-terpinolene, ct-terpinene, and rcymene) compounds were emitted from one or more plant species. Three monoterpenes ( (

Humidity (g/kg), wind (m/s), and VOCs
where C should be interpreted as a mixed-layer average. We now discuss the limitations of these simplifying assumptions. We can evaluate the errors in our flux estimate which stem from neglecting the entrainment flux (WC)zi, time rate of change zi[dC/dt), and advection ziU[dC/dx) terms in (4). We estimate entrainment using a simple jump model [Lilly, 1968]. Since the chemical lifetime of biogenic VOCs is fairly short, we assume that their mixing ratio is zero above the boundary layer. The jump in VOC mixing ratio across the planetary boundary layer top is then roughly the mean boundary layer mixing ratio. The entrainment flux is given b y the product of the jump in mixing ratio and the mixed-layer growth rate (typically, about 0.05 m s -1 during the day). Since entrainment dilutes the mixed layer, neglecting entrainment in the box model causes a systematic underestimate of the surface flux. This underestimate is, at most, about 1 mg C m -2 h -1 for isoprene, 0.2 mg C m-2 h-• for (t-pinene, and 0.1 mg C m -: h-• for [•-pinene. Nonzero mixing ratios of biogenic VOC above the boundary layer will minimize this underestimate. We cannot distinguish the mixing ratio time rate of change zi[dCIdt), from advection ziU[dC/dx) using these observations, but we can estimate the magnitude of the sum of these terms by observing the evolution in the mean mixing ratio profile over time for days with more than one balloon profile. We expect the mixing ratio to increase over the course of the day, neglecting advection. This would mean that the box model, which assumes steady state, again underestimates the surface fluxes. The observations, however, show significant but random trends in the mixed-layer average mixing ratio over the course of the day. This indicates that the steady state approximation is, on the average, reasonable and that advection, random in sign, is the dominant term. set OH concentrations at 4 x 106 molecules cm -3 for an analysis that includes our ROSE data, and Jacob et ai. [1993] use an OH concentration of 6 x 106 molecules cm -3 in a comparison of the hydrocarbon mixing ratios we measured at the ROSE site. We use a maximum OH concentration of 4 x The predicted concentrations are based on fluxes estimated using the LCC-MSS landscape data and emission potentials from Guenther et al. [1994]. All samples were collected between 0800 and 1700 LST at heights above 40 m above ground level. Concentrations are in parts per billion of carbon. 10 6 molecules cm -3 and the OH diurnal variation described b y Lu and Khalil [1991]. Jacob et al. [1993] note that the few direct measurements of OH concentrations in rural air tend to be lower than those computed from photochemical models. This is probably because the models underestimate OH sinks such as oxygenated VOC. This 50% uncertainty in OH concentration translates directly into a 50% uncertainty in the box model flux estimates. Any error is likely to be systematic. Our estimates of the mean mixing ratio C added some uncertainty. For any one biogenic VOC profile the typical standard deviation of the mean mixing ratio was approximately 25% of the mean. We fit a reasonable vertical profile (Chameides et al. [1992] below 150 m and Moeng and Wyngaard [1989] for the rest of the mixed layer) to each observed mixing ratio profile and computed the vertical average of the fit to obtain C. Since this vertical variability is accounted for, the 25% standard deviation of the mean is an overestimate of the uncertainty in C. The uncertainty in C is likely to be random in sign from one profile to the next.
Another source of uncertainty in (5), are the estimates of zi from boundary layer radar during ROSE [White and Fairall, 1991] and airsonde and tethersonde profiles during SOS-M. The zi estimates typically have an uncertainty of about 15%, and errors are likely to be random.
We conclude that the flux estimated by equation (5) is subject to a total random uncertainty of about 25% due to the roughly 20% and 15% random, independent uncertainty in our estimates of C and zi, respectively. OH introduces an approximately 50% uncertainty in the flux and is likely to be a systematic error whose sign we do not know. Direct OH measurements will resolve this issue.
Mixed-layer gradient estimates. The isoprene and monoterpene mixing ratio profiles were also used to estimate surface fluxes using the mixed-layer gradient MLG technique. This technique assumes that boundary layer mixing is dominated by convective turbulence and that boundary layer conditions evolve slowly compared to the convective turnover time zi/w. of about 10 min ( Table 6 shows estimates of the lifetimes of isoprene and monoterpenes for these experiments. The roughly 1-hour lifetimes are fairly long compared with the 10-rain convective turnover time, but some overestimates of the fluxes may result from not accounting for this reactivity. Using the very simple argument of Davis et al. [1994], we estimate that for a mean mixed-layer isoprene mixing ratio of 7.5 ppbv C, a surface flux of 4 mg C  There is some uncertainty in accounting for the displacement height of the forest canopy [Stull 1988]. We have chosen to neglect displacement height for these estimates. This may cause, at the extreme, about a 30% overestimate in the fluxes calculated at these sites. While this discussion is somewhat helpful, the observations show that the actual variability in mixing ratio differences is probably larger than can be accounted for by sampling error. Violations of the mixed-layer assumptions could be causing significant variability in the measured vertical gradients and the resulting flux estimates. The most likely violations of the mixed-layer assumptions are vertically varying horizontal advection caused by heterogeneity in surface fluxes on the scale of one to a few kilometers and irregular mixing caused by extensive convective cloud activity. We have attempted to eliminate periods where cloud activity may perturb boundary layer mixing. We cannot evaluate an expected magnitude for errors in the surface flux measurements due to surface heterogeneity with the current set of observations. Flux estimates from ROSE are presented by Davis et al. [1994]. The profiles which remained after meteorological screening for convective mixing showed some evidence of variability which could be indicative of heterogeneous emissions. There was more profile variability, for example, than existed in a similar data set collected in an Amazon forest preserve [see Davis et al., 1994]. The SOS-M results, presented here, show considerable evidence of poorly mixed profiles, even after screening the data for conditions without strong convective mixing. Figure 1 shows the landscapes around the two sites. It seems reasonable from the varying landuse seen in these aerial photos that the flux estimates at the SOS-M site are influenced from more spatial heterogeneity than the ROSE site, that variability exists at both sites on the scale of one to a few kilometers needed to disturb the mixed-layer assumptions, and that the pattern of heterogeneity is fairly random with respect to wind direction at both sites. We proceed therefore assuming that any variability in the vertical profiles caused by heterogeneous emissions is random and that our results averaged over all profiles are meaningful. Comparison of box model and mixed-layer gradient methods.

In practice, the MID technique depends on accurate measurements of small vertical gradients in mixing ratio
Fluxes estimated using the BM and MLG methods are compared in Table 6 isoprene (25 to 43%), higher for a-pinene (36 to 72%), and slightly higher for [3-pinene (9 to 20%)relative to the BM estimates at the two sites. The similarity at both sites in the comparison of the techniques is very encouraging. Although we cannot rule out coincidence, especially given the large variability in individual flux estimates present in both techniques, the mean fluxes obtained appear to be meaningful and reproducible. It is interesting to note that, as shown in Table 6, the most reactive (with respect to OH) of these three compounds, isoprene, has consistently lower flux estimates with the • technique (relative to the BM estimates), while for the least reactive compound, a-pinene, the MLG technique predicts the highest relative fluxes. The systematic errors which we have identified due to neglecting entrainment in the BM and neglecting scalar reactivity and displacement height in the MI• technique could only help resolve the discrepancy in terpene fluxes since they result in increasing the BM flux estimates and decreasing the MI• flux estimates. If OH is overestimated in the box model, this would help resolve the discrepancy in isoprene fluxes but would exacerbate the comparison of terpene fluxes. Another possible explanation is the mismatch in flux footprints. The range of the MLG footprint is a few kilometers, while the BM footprint is over 10 km. The density of monoterpene emitters decreases beyond a few kilometers fetch due to the high density of pine trees near the center of each site. Since the BM is sensitive to a larger footprint, this nearby concentration of pine trees would result in higher monoterpene and lower isoprene fluxes estimated by the MLG technique. Comparison with emission model estimates. Best estimates of isoprene and monoterpene emission rates were calculated using the emission potentials of Guenther et al. The predicted values are based on fluxes estimated using the land cover characteristics-multi spectral scanner landscape data and emission potentials from Guenther et al. [1994].
Uncertainties in emission potentials contribute a large part of the overall uncertainty in emission model estimates. Isoprene emission rates for individual leaves tend to be about 75% higher than emission rates averaged over an entire branch due to shading on the lower leaves of a branch. Emission potentials of VOC compounds other than isoprene and monoterpenes are especially uncertain but may contribute significantly to the total flux. Average daytime mixing ratios of 13.8 and 6.6 ppbv C isoprene and 5.0 and 4.5 ppbv C monoterpenes were observed at the two sites. Together, these biogenic compounds contain about 35% of the total carbon in nonmethane VOC and over 90% of the VOC reactivity with OH. VOC fluxes estimated from ambient mixing ratios using a box model technique and a mixed-layer gradient technique agree within a factor of 2. Fluxes estimated by extrapolating enclosure measurements (emission model) and based on ambient mixing ratios (box and gradient models) were within a factor of 2 in most cases and within a factor of 3 in over 90% of all cases. Emission model estimates for isoprene were within 5% of those based on ambient mixing ratios at one site and 42% lower at the other site. Monoterpene emissions estimated by the emission model were about 60% lower than observed at both sites. A qualitative assessment suggests that higher monoterpene flux estimates should be expected from the ambient mixing ratio data which represent fluxes averaged over a smaller area surrounding the site. These results show that emission models can provide reasonable estimates of ambient isoprene and monoterpene mixing ratios. This comparison can be made with more certainty by obtaining (1) accurate estimates of the source region (flux footprint) associated with each ambient mixing ratio profile, (2) a better understanding of when the mixed-layer assumptions are valid, and (3)more accurate estimates of the variables used to estimate fluxes from ambient measurements (e.g., CI-t concentration).