Influence of increased isoprene emissions on regional ozone modeling

The role of biogenic hydrocarbons on ozone modeling has been a controversial is- sue since the 1970s. In recent years, changes in biogenic emission algorithms have resulted in large increases in estimated isoprene emissions. This paper describes a recent algorithm, the second generation of the Biogenic Emissions Inventory System (BEIS2). A sensitivity analysis is performed with the Regional Acid Deposition Model (RADM) to examine how increased isoprene emissions generated with BEIS2 can influence the modeling of elevated ozone concentrations and the response of ozone to changes to volatile organic compound (VOC) and nitrogen oxide (NOx) emissions across much of eastern North America. In- creased isoprene emissions are found to produce a predicted shift in elevated ozone concen- trations from VOC sensitivity to NOx sensitivity over many areas of eastern North America. Isoprene concentrations measured near Scotia, Pennsylvania, during the summer of 1988 are compared with RADM estimates of isoprene and provide support for the veracity of the higher isoprene emissions in BEIS2, which are about a factor of 5 higher than BEIS 1 during warm, sunny conditions.


Introduction
The role of biogenic hydrocarbons on photochemical modeling has been the subject of considerable debate since the 1970s [Dimitriades and Altshuller, 1977;Lurmann et al., 1984;Trainer et al., 1987;Chameides et al., 1988;Chock et al., 1995;Simpson, 1995;dackson et al., 1996]. The National Academy of Sciences [National Research Council (NRC), 1991] emphasized the importance of including biogenic volatile organic compounds (VOCs) in ozone modeling, but recent modeling studies performed for the eastern United States [Morris et al., 1997] have raised concerns about the accuracy of current biogenic emission estimates by noting gross overpredictions of isoprene concentrations when comparing model estimates with near-surface measurements of isoprene. Concern about the biogenic VOC emission inventories is understandable, because isoprene emissions from the second version of the Biogenic Emissions Inventory System (BEIS2) are about 5 times higher than those previously estimated with BEIS 1 [Geron et al., 1994]. Furthermore, this radical change in the magnitude of isoprene emissions may cause a shift in the simulated effectiveness of different ozone-precursor emissions abatement strategies from ones that are more effective with VOC reductions to ones that are more effective with NOx reductions [Sillman et al., 1997].  Rules are followed sequentially until land use area in a county totals 100%.

•Air Resources Laboratory, National Oceanic and Atmospheric
lists the rules that were used to extract key features from various data sets for the BELD.
The spatial resolution of the BELD is at the county level in the United States and at the subprovince level in Canada. For air quality model applications, land use data are allocated to model grid cells using the ARC/INFO TM Geographical Information System (GIS). Table 2 shows the abundance of the most prevalent land use types estimated for the RADM domain.

Emission Factors
The emission flux factors assumed in BEIS2 are listed in Table 3 and represent leaf-on (full biomass) conditions. These fluxes have been normalized to a leaf or soil temperature of 30øC and, for isoprene, a photosynthetically active radiation (PAR) of 1000 lamol m -2 s -1.
In Table 3, factors for tree genera have been adapted from Table 3 of Geron et al. [1994]. The emission factors in BEIS2, expressed as lag m -2 m 'l, were obtained from the Geton et al. factors (expressed as lag C (g foliar dry mass) '• h 'l) by multiplying by the foliar density (given by Geron et al. as g m '2) and by the compound/carbon ratio (i.e., isoprene is 68 g/60 g C). Tree genera in Geron et al. for which emission factors were not given, because of a lack of field measurements, have been arbitrarily assigned the lowest emission rate for each VOC category. Geron et al. note that these genera represent only 3.5% of forest leaf biomass in the eastern United States. An additional category, open forest (OFOR), has been added to Table 3 to account for the area in a county that is designated as forest by the U.S. Forest Service inventory statistics, but for which crown area calculations did not yield 100% coverage. The open forest category is assumed to be underbrush and is assigned a grassland emission rate for VOC and a forest emission rate for NO. Soil NO emission factors for forests were taken from Williams et al. [ 1992].
VOC agricultural emission factors with a few exceptions are reported by Lamb et al. [1993] and are derived primarily    , and the emission rates for crops are sorted by nitrogen fertilizer application rates. Because rice is grown in a wet environment, the wetlands emission rate is assumed for NO. The USGS land use data are used mostly in Canada and in portions of the western United States, where genus-level forest cover and agricultural crop data were unavailable. The VOC emission fluxes proposed by Guenther et al. [1994] were assigned to our synthesis of the USGS database to arrive at 18 land use/emission factor classes. The most prevalent USGS classes in the RADM domain are northern mixed forest (NMXF, 5%), conifer forest (CONF, 5%), woodland/cropland (WDCP, 2%), and hardwood forest (HARF, 1%). The VOC emission factors for NMXF are an area-weighted average of the three northern mixed forest classes from Table 4  Special treatment has been given to urbanized areas. Atlases of potential vegetation were consulted to determine whether the potential natural vegetation for an urban area is either forest or grassland. Then based on work of Nowak et al. [1996], urban areas are assumed to consist of either 32% tree cover (forested regions), 22% tree cover (grassland regions), or 11% tree cover (desert regions, not applicable in this modeling domain). In the eastern United States, where forest inventory data were available, tree genus distributions are taken from nearby county inventory statistics. Urban forest areas outside of the eastern United States are assigned to the urban tree (UTRE) category, which is simply assumed to consist of 50% hardwood forests (HARF) and 50% coniferous forests (CONF). The remainder of the urban area (UOTH) is assumed to consist of 20% grass and 80% barren. where NO is the adjusted soil NO emissions flux, NOo is the emissions flux standardized to a soil temperature Ts of 30øC, and T is the soil temperature. Soil temperature is based on minor adjustments of the ambient air temperature using empirical relationships given by Williams et al. [1992].

Differences in BEIS1 and BEIS2
In this paper, we assume that isoprene emissions from BEIS2 are a factor of 5 higher than BEIS1. The increase in isoprene arising from the BEIS2 algorithm is caused mostly by changes in the base emission factors assumed for trees and may be attributed to several interrelated causes. The BEIS1 isoprene factors were based mostly on branch enclosure measurements collected near Tampa  where PAR: is computed as a function of height z, PARo is estimated for the top of the canopy, and LAI: is the leaf area index summed from the top of the canopy (z=0) down to height z. Total LAI is assumed to be 3 (m 2 m -2) for pines, 5 for deciduous trees, and 7 for other coniferous genus types.
In BEIS2, leaf temperature is assumed simply to equal ambient temperature, because of the difficulties in establishing a reliable leaf energy balance model [Fuentes et al., 1995;Lamb et al., 1996]. Sensitivity studies of Lamb et al. [1993] showed that ignoring the difference in leaf temperature relative to ambient temperature can change biogenic VOC emission fluxes by -33 to +50%. Future versions of BEIS will include a more robust canopy model, when the leaf energy balance algorithm has been improved.
Leaf biomass is distributed uniformly with height, except for deciduous trees where the specific leaf weight is assumed to vary as a function of height, so that the distribution of leaf biomass varies from top to bottom (27, 21, 18, 17, and 17%) according to Geron et al. [1994]. This adjustment increases isoprene emissions from deciduous genus types by as much as 10% because light levels are higher in the upper portions of the canopy.
Emissions for other biogenic VOCs, E, are assumed to vary only as a function of leaf temperature as follows:  Geron et al. [1995] for the Atlanta, Georgia, area indicate that daily isoprene emissions with BEIS2 were about a factor of 5 higher than BEIS 1. Therefore, in the modeling performed for this paper, we assume simply a factor of 5 difference in isoprene emissions to allow for a systematic comparison of BEIS 1-equivalent emissions and BEIS2.

Ozone Model Sensitivity
The sensitivity of RADM to changes in biogenic VOC emissions is examined for the period July 28 to August 6, 1988. This period had two pulses of relatively high regional ozone levels in the eastern United States. Further details on the meteorology and air quality patterns during this period are given by Dennis et al. [ 1990]. Only those forest and agricultural categories comprising at least 2% of the grid are listed.

Modeling Assumptions
Simulations from version 2 of the Regional Acid Deposition Model (RADM) [Chang et aL, 1987] are used to examine ozone sensitivity and in section 4 to compare against measured isoprene concentrations. For this study, we have employed a 21-layer version of RADM, consisting of 35 x 38 grid cells each having a horizontal resolution of 80 km. The modeling domain extends from (23.9 ø N, 98.4 ø W) to (49.2 ø  N, 62.2 ø W). Because of computational restraints, horizontal grid resolution in the model has been sacrificed to allow more detailed simulation of vertical processes, which are believed to be important for isoprene distributions. Near the surface, modeled layer depths are-40 m. The relatively coarse 80-km grid resolution is deemed adequate for examining isoprene at Scotia, Pennsylvania, because the land use around this site is similar to that for the corresponding model grid cell as listed in Table 4. An 80-km grid is also probably adequate for examining regional ozone, as a sensitivity analysis performed using RADM with a 20-km and an 80-km grid resolution for the northeastern United States showed very similar percentage changes in regional ozone due to emission changes. We are, however, less confident about using the coarse grid cell reso- lution in RADM to examine ozone sensitivity for urban areas, and the sensitivity of modeled ozone reported here should be viewed from a regional, nonurban perspective. RADM handles numerous physical and chemical processes, including advection, vertical diffusion, cloud effects on vertical transport and actinic flux, dry and wet deposition, and gas and aqueous chemistry. Vertical eddy diffusivity is treated using first-order similarity theory equations, as described by Byun Table 5. For all metrics, it is readily apparent that changes in biogenic emissions influence elevated levels of regional ozone, with higher biogenic emissions generally resulting in higher ozone levels. While the ozone metrics show sensitivity to biogenic emissions for the base case emission scenarios, perhaps of greater interest is the relative sensitivity of elevated ozone levels to biogenic emissions as a function of VOC versus NOx emissions reductions. To examine this, we define a relative VOC/NO,, effectiveness index (REI) as follows: where the metrics are taken from Table 5 for the 50% NOx, 50% VOC, and base emissions scenarios. REIs for each metric and biogenic emissions scenario are listed in Table 6. As an example, using the number of grid cell hours >80 ppb metric, the REI for "no BEIS" scenario is calculated to be +0.1%. This value is calculated from data in Table 5 as (2380-2371)/6326, and then multiplied by 100. The value of +0.1% implies that the VOC emission reduction scenario is +0.1% more effective than the NOx emission reduction scenario in reducing the number of grid cell hours >80 ppb for the "no BEIS" emissions base case. These sensitivity results indicate why there is so much interest in biogenic emissions. First, the predicted ozone levels  Modeled results from the RADM for July 28 to August 6, 1988, excluding results from 22 "urban" grid cells. Grid cell (1,1) is positioned at the southwest comer of the domain. and the regional extent of elevated levels are increased with a change from BEIS1 to BEIS2. Second, the relative effectiveness of VOC versus NOx control is shifted toward NOx control with a change from BEIS 1 to BEIS2.

Comparison of Measured and Modeled Isoprene
Isoprene concentrations from the RADM have been extracted from the grid cell and model layer(s) corresponding to the measured concentrations. As noted in Table 4, the grid cell overlaying Scotia is comprised of 73% forest and 33% oak crown cover, reflecting forest inventory data collected in this area [Hansen et al., 1992]. Emissions computed for this grid cell are indicative of the rural character of the Scotia site and as noted in Table 7 are dominated by biogenic VOC emissions and relatively small emissions of anthropogenic and biogenic NOx. Table 8 compares measured isoprene concentrations with two sets of RADM base case simulations. The comparison consists of three vertical groupings: (1) a near-surface group consisting of 150 5-m measurements and concentrations from the first model layer; (2) a 100-m profile group consisting of available tower and tethered balloon measurements and concentrations averaged from model layers 1-3; and (3) a 1600-m profile group consisting of available tower, balloon, and aircraft measurements and concentration profiles aver-aged from model layers 1-13. Concentrations from layer one of RADM/BEIS2 average 25% less than the measured 5-m concentrations (4.5 ppbv versus 6.0 ppbv), while concentrations from RADM/BEIS1 are an order of magnitude smaller (0.6 ppbv). For both the 100 and 1600 m average profiles,

RADM/BEIS2
concentrations are about 50% lower than that measured. RADM/BEIS1 concentrations are more than an order of magnitude lower than that measured. Profiles of isoprene obtained for the two periods when aircraft data were available are shown in Figure 6. It is evident that, although RADM/BEIS2 shows a tendency to underpredict, it clearly performs better than RADM/BEIS 1. Figure 7. Each binned point represents the mean (time and concentration) of 10 values. The binned points appear closer together during the daylight hours because of the higher frequency of measurements. For the daylight hours, RADM/BEIS2 is consistently about a factor of 2 lower than the measurements. Both modeled and observed concentrations peak around sunset, but observed values decrease much more rapidly than modeled values. Concentrations with RADM/BEIS1 are consistently lower than those measured except just before sunrise, when observed values reach a minimum.

The average diurnal plot of observed and modeled nearsurface concentrations is shown in
To probe the processes responsible for the modeled diurnal behavior, we have extracted information from the RADM/BEIS2 simulation for July 31 to August 1, 1988, a period with a high frequency of observations. Figure 8 shows the observed and modeled isoprene concentrations, which behave similar to the mean diurnal plot in Figure 7. The processes from RADM model layer one contributing to isoprene production and destruction are plotted in Figure 9. The dominant processes in layer one are emissions (source) and vertical diffusion (sink). Although chemical reactions in layer one are important, they are nearly an order of magnitude less than vertical diffusion. The OH plus isoprene reaction contribution peaks near midday, while isoprene destruction from the O3 and NO3 reactions exceed that from the OH reaction beginning around sunset. The net production curve shows that competing processes nearly offset each other during much of the day, but the net production rate increases rapidly to 10 ppbv/h just before sunset (2300 UT). During this hour, vertical diffusion is only about a half (-17 ppbv/h) as large as emissions (+31 ppbv/h). Chemical destruction only removes about 4 ppbv/h. After the large peak in isoprene production, the only significant processes reducing isoprene at night are the O3 and NO3 reactions, which remove 1-2 ppbv/h. This analysis suggests that modeled isoprene concentrations in layer one are very sensitive to the delicate balance of isoprene emissions (which are driven by solar radiation) and vertical diffusion (which is driven by sensible heat flux during the daytime). At night, the modeled isoprene concentrations appear to be affected mainly by a combination of vertical model resolution and the modeling of the stable boundary layer because the concentration time series is relatively insensitive to known uncertainties in the nighttime chemistry.

Discussion
Concern about the significant increase in isoprene emission estimates is understandable when viewed from a historical perspective. After a period of speculation during the 1970s, biogenic hydrocarbon emissions were dismissed as being unimportant to urban ozone formation [Altshuller, 1983;Lurmann et al., 1984]. The issue was reexamined in the 1980s by Trainer et al. [1987] and Chameides et al. [1988] who pointed out that biogenic VOC emissions were probably important for ozone formation in rural areas and for affecting VOC/NOx sensitivity near some urban areas. Meanwhile, a national biogenic emission inventory emerged for ozone modeling yielding VOC emissions on the same order of magnitude as anthropogenic emissions [Lamb et al., 1993]. This inventory was quickly superceded by a methodology yielding even higher isoprene estimates [Geron et al., 1994], which formed the primary basis of BEIS2. As the biogenic emission inventories were rapidly changing, ozone modelers in the air quality modeling community were grappling with the new challenge of predicting regional as well as urban ozone. The complexities associated with the move to regional models were increased by the large uncertainties with the biogenic emission estimates. We have described the algorithm for BEIS2, which at the time of this study represented a contemporary approach for generating a biogenic emissions inventory. Although not presented in this paper, the underlying basis of BEIS2 for estimating isoprene fluxes from deciduous forests has been evaluated with several field studies [Fuentes et al., 1995;Guenther et al., 1996;Pier and McDuffie, 1997]. It can be inferred from these micrometeorological and whole-tree field studies that isoprene flux estimates are accurate to within about 50% and that estimates of emissions with BEIS2 are superior to BEIS 1.
We have examined how the evolution in biogenic emission estimates can affect ozone predictions and have shown that these changes have a major influence on ozone predicted with the RADM model (Table 5). Changes in modeled elevated ozone levels are quite sensitive to VOC versus NOx emission reductions. In addition, across much of the RADM modeling domain, elevated levels of ozone went from being VOC sensitive to NOx sensitive with the increased levels of biogenic VOC emissions resulting from BEIS2 (Table 6). These results are similar to those of Roselle [1994] who used the Regional Oxidant Model (ROM), but because of the relatively coarse resolutions in both modeling studies, further examination of urban areas using models with finer grid resolutions is needed.
While field studies support the assertion that isoprene fluxes from BEIS2 are superior to BEIS1, only a scant amount of information has been published on modeled versus observed isoprene concentrations. Comparing near-surface observations of isoprene versus calculations from the Urban Airshed Model, Morris et al. [1997] suggest that estimates from BEIS2 are much too high. These suggestions contrast with the results reported here, which indicate that using BEIS2 in RADM resulted in mean near-surface isoprene predictions that were slightly lower (25%) than observed. Although it can be instructive to examine near-surface isoprene concentrations, it is difficult to separate model shortcomings from measurement variability. In fact, Andronache et al. [1994] recommend that isoprene be measured well above the surface (starting about 40 m) because of the difficulties associated with analyzing isoprene measured near the surface, difficulties that can be attributed to rapid and spatially heterogeneous processes of emissions, vertical diffusion, and chemistry. Our analysis of daytime data collected above 40 m (Figures 6a-6b) indicated that BEIS2 resulted in isoprene concentrations that are about 50% lower than observed values, with results from BEIS 1 an order of magnitude too low. It is our judgement that the comparison of surface and upper air isoprene concentrations from the Scotia data set provides additional corroboration to the published field flux studies on the veracity of the BEIS2 isoprene estimates relative to BEIS 1.
As shown in the diurnal plots (Figure 7), however, further improvement on the treatment of isoprene in regional ozone models such as the RADM is needed. In particular, the proc- Another source of uncertainty in the modeling of isoprene is the possibility that deposition may provide a small but important 'sink of isoprene. Cleveland and Yavitt [1997] note that soils may remove as much as 5% of globally emitted isoprene on an annual basis. Currently, most air quality models assume negligible amounts of isoprene deposition.
While uncertainties exist with the photochemistry of isoprene and its reaction products, the primary destruction reactions with OH, NO3, and 03 are reliably known [Carter, 1996]. We further examined isoprene decay rates in sensitivity simulations with changes in NO3 and N2Os production and loss rates, and, although not shown in this paper, we found only minor impacts on the nighttime decay of isoprene. Thus the discrepancy in modeled and observed nighttime decay of isoprene probably is not caused by uncertainty in the isoprene chemistry, but it may result either from the air quality model's representation of horizontal advection, deposition, or vertical mixing or from the spatial variability of measured concentrations within the stable boundary layer.
Although uncertainty in the isoprene chemistry did not seem to cause the nighttime discrepancy we observed here, there is some uncertainty in the isoprene photochemistry. of ozone production. It is important that future modeling studies include more complete descriptions of isoprene chemistry, like that proposed by Carter [1996]. Indeed, future evaluation studies would be well advised to examine longer-lived species associated with isoprene chemistry, such as methacrolein, methy vinyl ketone, and their PAN analogues. Unfortunately, most compressed photochemical mechanisms currently used in air quality models, including RADM, do not represent these important decay products. Because of the complexity involved with evaluating nearsurface concentrations of isoprene, future studies would be advised to take an aggressive three-dimensional view, by examining extensive measurements of isoprene and its oxidation products throughout the first few hundred meters of the atmosphere and by examining horizontal variability. As we gain further understanding and make improvements in the chemical and physical processes in the models, work will undoubtedly continue on improving the accuracy of isoprene emission algorithms with upgrades in land use distributions, base emission factors, and environmental corrections.