Biogenic volatile organic compound emissions (BVOCs). II. Landscape flux potentials from three continental sites in the U.S.

Landscape flux potentials for biogenic volatile organic compounds (BVOCs) were derived for three ecosystems in the continental U.S. (Fernbank Forest, Atlanta, GA; Willow Creek, Rhinelander, WI; Temple Ridge, CO). Analytical data from branch enclosure measurements were combined with ecological survey data for plant species composition and biomass. Other quantitative flux measurements at the leaf and landscape level were incorporated to scale the results from the enclosure measurements to the landscape level. Flux estimates were derived by using a one week ambient temperature and light record (30 min time resolution) and adjusting all emission rates to these conditions with temperature and light correction algorithms.


2190
Northern red oak were the dominating species at the Wisconsin site with 68% of the total emissions. At the Colorado site, Gambel oak and service berry made up 86% of the BVOC emissions. Total daily average BVOC landscape fluxes determined by this method were 2.0, 4.5 and 2.4 mgC m2h ~ for the Atlanta, Wisconsin and Colorado sites, respectively. The contribution of isoprene to this overall BVOC flux was calculated to be 45, 38 and 54% at the three sites, respectively.
The branch enclosure technique combined with ecological surveying and landscape-scale isoprene flux measurements proved to be a valuable tool for seroerting a high number of plant species and for identifying the major . and most important emitters for a more thorough investigation. The obtained data, the suitability of this approach and its limitation to derive BVOC fluxes on the ecosystem level are critically evaluated and factors introducing experimental errors are identified.

INTRODUCTION
The role of emissions of biogenie volatile organic compounds (BVOCs) in atmospheric processes and the formation of tropospheric ozone have been extensively discussed (Fehsonfeld et al., 1992;Helmig et al., 1998) Selected results from these studies have been reported (Greenberg et al., 1994;Guenther et al., 1996a;Isebrands et al., 1998). Here, we use data from the branch enclosure experiments reported by Helmig et al. (1998) to derive landscape scale flux potentials at three sites using the results of extensive biomass surveys, ambient temperature and light data series, and scaling to measured landscape-level isopt~ae fluxes. This approach relies on the current knowledge of temperature and light dependence of isoprene and other BVOC emission rates and developed algorithms to scale the data from the experimental conditions to normalized standard values.

EXPERIMENTAL
At each of the three sites investigated transoets were used to characterize the composition, successional status, and environmental setting of vegetation communities. Biological data included tree species identification, tree diameter at breast height (DBH) (-1.5 m), canopy leaf area index (LAI), and understory cover. Specific leaf density was determined for all dominant tree species from leaf area and leaf dry weight measurements on representative leaves. Canopy foliar density was estimated from LAI and specific leaf density measurements. Environmental data included measurements of terrain, slope, aspect, location and elevation.
Site Descripn'on 2191 Fembartk Forest. The Fembank Forest site is within a 25 ha mixed hardwood and conifer woodland located in suburban East Atlanta, GA (latitude 33 ° 24'N, longitude 840 14'W) and is managed by the DeKalb County School System. Fernbank Forest has been preserved since the early 1800s and represents a mature remnant stand of natural Georgia Piedmont forest (Skeen, 1974). Mean canopy height is 30 m and mean overstory LAI is 4.8 m 2 m 2 (integrated leaf area/ground area). Dominant tree species include Quercus alba (white oak), Liquidambar styraciflua (sweet gum), Liriodendron tulipifera (tulip poplar) and Pinus taeda (loblolly pine). Major understory plants include Hedera helix (English ivy), Wisteria sp. (Wisteria), and several pteridophytes. Field measurements at this site were performed during June 1993. Ambient air and landscape flux measurements were also made during this experiment and are described by Guenther et al. (1996).  (Gambel oak) and Amelanchier alnifolia (serviceberry) shrubland. Ground cover includes Artemisia tridentata (big sagebrush) and Symphoricarpos occidentalis (snowberry). The open canopy has a mean height of about 2 m and a mean overstory LAI of 0.92. Landscapes within 30 km of the site are dominated by cropland and range land (~ 50% of the surface area) and scrub woodland (~ 15%). In addition to Gambel oak, aspen (Populus tremuloides) and spruce (Picea spp.) are common tree species in this region. This experiment was conducted during August 1993. Other flux measurements performed at the landscape level are described by Guenther et al. (1996).

Ecological Characterization
Representative vegetation communities at each site were identified, gridded, then sampled in 10 m x 10 m contiguous plots along 100-200 m transects. In each plot, all trees were identified, and the stem DBH and stem height was recorded. Trees were defined as living individuals ___ 4 cm DBH and _> 1.5 m tall. Smaller individuals (saplings) were identified and counted. Tree layer LAI, understory layer LAI, shrub and non-woody understory species cover, slope, and elevation were also recorded for each plot. Slope and tree height were measured using a clinometer, elevation was measured using an altimeter. LAI of green foliage was determined using two (above and below canopy) LAI-2000 meters (LICOR, Lincoln NE). Four LAI measurements were taken per plot, two at 1.5 m height to determine ~ree layer LAI and two on the ground to determine understory LAI. Conifer LAI measurements were adjusted according to LICOR manual procedures.
Branches of each tree species occurring in the transects were collected and measured in the laboratory for leaf area, using a leaf area meter (CID, Moscow, ID) and leaf dry weight, by drying leaves for 24 h at 50°C and weighing. Data were averaged to determine the conversion ratio (CR [cm2g-~]) of leaf area to leaf biomass for each species.
where LAI, is the measured tree layer LAI and BA~ is the sum of BA~ for that species in the plot. It is recognized that this method may underestimate (by ~ 25%) the LAI of broad-crowned trees such as oaks and overestimate (by ~ 35%) the LAI of narrow-crowned trees such as sweetgum and tulip poplar (Geron et al., 1994).
Green foliage dry biomass for a given species (FB~[g]) was calculated as where CP~ is the empirically derived leaf area to leafbiomass conversion ratio for that species.
For the Willow Springs site, Landsat Multi-Spectral Scanner (MSS) images acquired on May 11, 1989 andOctober 8, 1991 were used in concert with transect data for an overall areal estimation of species composition. The Landsat data was analyzed using the ENvironment for Visualizing Images (ENVI version 1.1; RSI, Boulder, CO) software package to classify ground cover types within a 14 km radius of the site. The two images were corrected prior to quantitative analysis to minimize atmospheric effects on the radiometric signal. In addition, the two images were co-registered with the use of ground control points and USGS topographic quadrangle maps to enable multi-2193 temporal pixel-to-pixel comparisons. This co-registration procedure provided eight data points at each pixel (2 samples with 4 spectral channels each) for spectral analysis and classification. The multi-temporal nature of these data aided classification by inherently incorporating changes in vegetation phenology (Wolter et al., 1995).
Seven ground training areas (open water, mixed conifers, black spruce bog forest, open bog, aspen-dominated hardwoods, mixed hardwoods, and grassy fields) were selected for image classification purposes using a USFS vegetation survey map as a digital overlay (USFS, Park Falls, WI). The average radiomelric signal from each training area was used to classify the remainder of the image pixels based on a spectral angle mapper (SAM) classification procedure. The SAM classification assigns classes based on the highest similarity between an individual pixel's spectrum and each training class average spectrum (Kruse et al., 1993).
At this site, the seven subjectively located 100 m 2 belt transects used for quantifying species composition and biomass of various forest types on the ground all fell within four of the seven cover classes. Three transects were classified as aspen-dominated hardwoods, two transects were classified as mixed hardwoods, and one transect each fell into the mixed conifer and black spruce bog classes, respectively. To estimate the overall biomass of each tree species within the sample area, transect data within each class were averaged and multiplied by the respective absolute class percentage cover. In total, these four classes comprised 78% of the ground cover types within the sample area, with open water, open bog, and grassy fields covering the remainder of the area. For comparison, a forest inventory database of the region that contains the study area estimates that these four forest classes comprise 85% of the ground cover (Hansen et ai., 1992). Within a 14 km radius of the study site, the aerial coverage of all seven ground classes were (in %) grassy fields (1), open bog (6), mixed conifers (7), mixed hardwoods (11), open water (12), black spruce bog (21) and aspen-dominated hardwoods (39). Table 1 lists all plant species sampled and their percentage leaf biomass at the three sites as determined from the transect and remote sensing analyses. A number of the plant species that were growing at the sites were not present in any of the transects. These species are listed with the biomass < 0.01% to indicate that they were present and sampled in the respective area. The majority of these species were exotics (i.e. they were planted, such as Gingko at Fembank). The reason these species were sampled was purely for inventory purposes. The total percentage of the overall tree/shrub leaf biomass in the transect area that was covered by the branch enclosure measurements was 92.8%, 92.3%, and 100% for Fernbank Forest, Willow Springs and Temple Ridge, respectively.

RESULTS
Cumulative relative landscape fluxes (CLF) were derived by multiplying the normalized (30°C/1000 ~tmol m-2s -~) BVOC/species emission rates (ER) from the branch enclosure experiments (Tables 2 -4 in Helmig et al. (1998)) by the biomass contribution (%BM) of each plant species (i) ( The respective data for isoprene were then normalized to the estimated and normalized (30°C/1000 ~mol m2s ~) landscape flux (LF) for isoprene: Different techniques were used at the three sites to measure isoprene landscape fluxes (Isebrands et al., 1998;Guenther et al., 1996). At Fembank Forest, tower gradient flux measurements of isoprene (Greenberg et al., 1994) gave a normalized isoprene flux of approximately 2.3 mg C m2h ~ (mean value for all wind directions). This method covers emissions fi~m the direct tower environment with a footprint area of approximately 300 m radius around the tower base. For the Willow Springs site, we used landscape scale isoprene flux measurements from sampling with a tethered balloon sampling platform. The balloon measurements cover a much larger footprint area than tower measurements. Footprint areas with a radius of approximately 10 km can be achieved. More detailed descriptions of these measurements are given by Guenther et al. (1996) and Isebrands et al. (1998). The normalized mean isoprene flux determined here was 4.4 mg C m~ "~. For the Temple Ridge site isoprene gradient measurements from a 10 m high radio tower were used, resulting in a mean and normalized isoprene flux of 3.3 mg C m2h ~.
The calculated landscape emission rates for all compounds at the three sites are tabulated in Table 2. The data listed in the first column for each site is normalized for the environmental conditions of 30°C and I000 ~tmol m-2s ~.
Actual BVOC emissions will vary because of changes in ambient temperature and light. While most BVOC emissions respond primarily to temperature, isoprene has been shown to respond to light; isoprene emissions subside during the night. Over a given diurnal scenario of light and temperature conditions, non-isoprene BVOC emissions will therefore have a significant higher conlribution to the total BVOC flux than at standard conditions of 30°C and 1000 ~tmol m2s t. In order to demonstrate the magnitude of this effect and to derive correction factors from standard conditions to actual ambient conditions, estimates of total and percentage BVOC fluxes under real-ambient conditions were derived by the approach detailed in the following.  ground level using a Vaisala sensor (HMP35C, Woburn, MA) for temperature and a quantum sensor (LI-COR LI-190SA, Lincoln, NE) for PAR. Light and temperature response factors were calculated from these meteorological data in the same manner as described for the correction of the branch enclosure conditions .
Isoprene was corrected for temperature and light effects. All other compounds were corrected for temperature only.
The normalized emission rates were then multiplied by these canopy-level factors at 30 min time intervals to give the emission rates for each shrub and tree species at the three sites are given in Table 3.
A surprising result is found for the Willow Springs site. Isoprene emissions are ranked second, exceeded by emissions of hexenol (incl. hexanal) derivative compounds. Hexenol derivative emissions were observed from 18 different plant species growing at this site. The major plant contributor to this landscape flux is quaking aspen. With a normalized total BVOC emission rate of 105 I, tg C h~gdw ~ and a biomass of 32 g m -z this tree species alone contributes to 57% of the overall BVOC flux, with the significant fraction in the form of hexenol compounds.
Unfortunately, the data were derived from just one individual measurement and the inclusion of literature reference data . Because of these reasons, there is a likely possibility that this sample may have been artificially contaminated fi'om disturbances to the plant. The degree of this possible error can not be assessed at this time. It is also striking that the overall landscape BVOC flux (time-averaged 4.5 mg C m-2h -~) is about a factor of Isoprene (890) Others (  gave fairly low biomass percentages for all oak trees in this region (4% total biomass) which are expected to be a major isoprene emitter. It is possible, however, that the biomass of isoprene emitting species within the footprint area of the tethered balloon sampling was higher than the result from the remote sensing studies on the survey area or that some major isoprene emitting species were not measured by the branch enclosure experiment. Also, the isoprene emission rate and percentage biomass of sphagnum moss are uncertain. Our sampling of sphagnum moss showed some isoprene emissions, however, it could not be determined if these emissions were from the moss or from Table 3 Total VOC flux from vegetation species at the three sites at standard conditions (30oC and 1000 umol/(s m2)) and under an ambient scenario of light and temperature conditions. Fluxes a~e given in microgram carbon per square meter and hour (ug C m-2h-l).  mosses in this area was not available.

Species
An accurate biomass estimation of BVOC fluxes at the Temple Ridge site are clearly dominated by isoprene emissions from Gambel oak ( Figure   3, Table 3). This site was the most homogenous of the sites studied with 56% of the biomass from Gambel oak. The total time-averaged contribution of isoprene to the overall BVOC flux is 54% with 68% of the total BVOC flux being from Gambel oak. The isoprene flux (1300 ~tgC m2h ~) is followed by cis-3-hexenyl acetate (440), cis-3-hexene-l-ol (190) and ct-pinene (39).

METHOD UNCERTAINTIES AND IMPROVEMENTS
Uncertainties and errors arising from the branch enclosure technique and the chemical analysis methods were discussed in detail by Helmig et al. (1998). The major sources of error identified were the temperature and lighl control during the branch enclosure, sample collection, water management during analysis, chromatography and mas,, spectrometric detection, and the sample representativeness and plant disturbance. However, it was also concludex that random errors partially cancel out and that scaling of the bag enclosure results to emission rates obtained frorr cuvette experiments improves the data quality. Additional factors that impact the accuracy of the data reported her~ are mainly from the biomass survey, from canopy effects and errors in the landscape isoprene flux measurements.
Biomass Survev Methods of species composition and biomass estimation varied between sites: at Willow Springs, ~ more regional sampling procedure was used, transects to sample representative species assemblages were distributec over a large area and the isoprene flux measurements were made using a tethered balloon technique. A rigorou,, accuracy assessment of the remote sensing classification has not been completed and thus the percentages of biomas,, determined from this analysis must be treated as preliminary in nature. In contrast, at the other two sites, transect,, radiated out from a central tower from which flux estimates were derived. This discrepancy in methodology ma t contribute to the variation between the Wisconsin site and the two other sites. The accuracy of the biomass survey it estimated to be on the order of_+ 50%.
~ Harley et al. ~1996) showed that emission rates from leaves can be significantly different dependin~ on where a leaf is growing within a canopy environment. Differences in the emission rates are related to the changing light and temperature conditions as a result of vertical attenuation of PAR through the canopy. For the calculations reported here, the use of reference isoprene fluxes determined on the landscape level partially corrects foi these effects. This approach assumes, however, that all vegetation species investigated experience the same degree ot shading, which is an inaccurate approximation. Likewise, this method assumes that all species considered are experiencing the same ambient temperatures, which, again, within a diverse forest stand, is only a rough approximation. We estimate that the fluxes reported here are closest to the actual flux for the Temple Ridge site, due to a lower biomass density that allows more of the total canopy leaf area to be exposed to the same levels of PAR and temperature. We estimate errors from canopy effects to be on the order of + 30%.

CONCLUSIONS
Despite the remaining large margins of error the developed technique and data presented allow to identify future research needs for improvements in the experimental setup and data analysis and to identify focus areas for BVOC measurements. The overall error for this method is difficult to assess. It is estimated that data for individual plant species are accurate to within a factor of-~ 2-3. Since all branch enclosure experiments were performed using the same experimental technique and under similar experimental conditions statistically systematic errors are to some extent corrected for by treating data for all species relative to each other and then sealing the cumulative data to the absolute landscape level results for isoprene. The BVOC emission potentials on the landscape level are therefore estimated to be within an overall error margin of a factor ~ 2.
The branch enclosure method in combination with a detailed biomass survey is a valid approach to identify the major emitting plant species in a given ecosystem. By thoroughly investigating a few selected plants, the bulk fraction of the total BVOC emissions can be assessed. Since reliable methods for the measurement of BVOC fluxes (other than for isoprene) at the ambient level are scarce, more attention should be given to refining the branch enclosure technique. This method currently appears to be the best experimental approach to screen a large number of vegetation species.
Within the discussed margins of error, the results of this study indicate that at the three sites investigated, nonisoprene BVOC emissions can be significant and contribute to as much as 46 to 63% of the overall BVOC flux.
Therefore, BVOC emission studies need to include the investigation of these compounds and not focus on isoprene fluxes only. The consideration of this portion of the BVOC emission is also important for atmospheric chemistry models.