Abstract

UAM-V model output for the 1995 OTAG episode was compared with field observations from 16 rural and suburban sites operated by N-NE, SOS/Nashville and PAMS. Afternoon values were compared in order to eliminate, or at least minimize, vertical gradients and transitory excursions in isoprene concentration.

• INTRODUCTION
• APPROACH
• RESULTS AND DISCUSSION
• SUMMARY AND CONCLUSIONS
• POLICY IMPLICATIONS
• REFERENCES

1.0 INTRODUCTION

Photochemical ozone (O₃) formation is a by-product of the atmospheric oxidation of hydrocarbons and reactive nitrogen species (e.g., NO and NO₂). In a highly simplified view, hydrocarbon oxidation provides a source of oxygen in the form of oxy- and peroxy- radicals. Reactive nitrogen acts as a catalyst to release oxygen atoms from peroxy- radicals, which can then react with atmospheric oxygen (O₂) to produce O₃. The process is extremely complicated and is dependent on the mix and relative abundance of hydrocarbons and reactive nitrogen (Atkinson, 1990).

Isoprene is a highly reactive, naturally produced hydrocarbon. Important sources of isoprene include oak forests and a variety of woodland landscapes (Guenther et al., 1994). Recent estimates suggest that isoprene is the most abundant hydrocarbon (anthropogenic or biogenic) across much of the eastern U.S. (Geron et al. 1994). Analyses of isoprene reactivity, in the context of ozone formation, suggests that isoprene is the dominant hydrocarbon in the rural atmosphere and an important hydrocarbon in the urban atmosphere (Chameides et al., 1992; Cardelino and Chameides, 1990).

This paper presents a comparison of isoprene measurements with UAM-V model calculations for the July 1995 OTAG episode. The scope of this analysis was restricted to isoprene because: 1) isoprene is generally abundant and, hence, relatively easy to measure; 2) isoprene has a distinct emission inventory system within the UAM-V model; 3) isoprene is treated explicitly in the UAM-V chemical reaction scheme; and 4) the reactivity of isoprene is sufficiently high that a bias or uncertainty in concentration estimates could have a significant impact on model sensitivity to control strategies or other factors.

The July 1995 OTAG episode was selected for analysis because it provided the most robust set of field measurements available. The episode coincided with two major field experiments (NARSTO-Northeast and the SOS-Nashville Intensive) which provide unprecedented spatial coverage, especially in rural areas. In addition, the PAMS network had reached a stage of maturity where it could contribute high quality information for a variety of suburban locations. None of the other OTAG episodes enjoyed such a wealth of concurrent observational data.

As with any comparison of measurements and model calculations, perfect agreement between the two quantities is not possible and is not expected. The primary reason for this involves a mismatch between point measurements and volumetric model calculations. Measurements at field sites are influenced by local features that cannot be resolved with a regional scale model. By definition, surface measurements are ordinarily taken within 5-15 meters of ground level. Model calculations, in contrast, attempt to represent the atmosphere as a series of vertical layers, including a surface layer which is 50 meters deep. Horizontal and vertical gradients, to the extent they exist, have significant potential to influence comparisons. Although these fundamental differences between observations and model calculations cannot be eliminated, we have attempted to minimize them in the data selection process.

2.0 APPROACH

Isoprene data were obtained from NARSTO-NE (N-NE), SOS and PAMS data archives according to network specific data exchange protocols. Data were examined for completeness and the characteristics of monitoring sites were evaluated for representativeness. As a rule, only rural and suburban sites with at least five days of data during the 1995 OTAG episode were chosen for analysis. A total of 16 sites satisfied the above selection criteria.

The sites selected cover a broad geographic range and form a diagonal across the eastern U.S. from the Gulf of Mexico to the Gulf of Maine (see Figure 1). The density of sites is particularly high in the northeast and Tennessee-Kentucky area, due to field experiments conducted by N-NE and SOS. In contrast, sites are sparse, or non-existent, along the Atlantic Coast from North Carolina to Florida, in the deep south and throughout the midwest. The comparability of field measurements and model calculations thus cannot be determined for certain parts of the OTAG domain.

Figure 1

Site characteristics and sampling information are presented in Table 1 (below). N-NE and SOS each operated 5 sites, while PAMS operated 6 sites. The N-NE and SOS sites were almost entirely rural and the PAMS sites were, with one exception, suburban. Land-use was predominantly hardwood forest for the rural sites and predominantly transportation or mixed for the suburban sites.

Isoprene sampling and analysis protocols differed in several important respects from network to network. N-NE sites collected 3-hour (daytime) or 6-hour (nighttime) canister samples on selected intensive operating periods during the 1995 field season. These periods included the July12-July 16 portion of the 1995 OTAG episode. Canister samples were analyzed at a central laboratory facility following the 1995 field season.

SOS sites generally collected a single canister sample each day of the SOS/Nashville Intensive. Samples were collected over a 1-hour period typically from 1200 to 1300 local standard time. Sample dates covered the entire 1995 OTAG episode. Canister samples were analyzed within a few days of collection at a central laboratory facility in Nashville that was set up for the purposes of the study. The suburban site at Youth, Inc. collected and analyzed samples on roughly a 1-hour cycle using a manually operated, on-site gas chromatograph-mass spectrometer system. Manual samples were collected on most intensive days from approximately 0500 to 2300 local standard time.

PAMS sites generally collected and analyzed hourly samples using automated, on-site gas chromatographic systems equipped with flame ionization detectors. One site, Baton Rouge, LA, collected 3-hour canister samples that were subsequently analyzed at a central laboratory. PAMS samples were collected around the clock at all sites throughout the OTAG episode.

For comparison purposes, model output from the 1995 D2 basecase was obtained and data corresponding to sampling intervals at each site were extracted. Isoprene concentrations from the model were then converted from units of parts per billion-volume (ppb-V) to part per billion-carbon (ppb-C), in conformance with the reporting conventions of the analytical laboratories. Finally, data from the N-NE and PAMS sites were averaged over the hours 1100-1700 to provide a single daily value for comparison. These hours were selected to isolate periods of highest isoprene emission and relatively well-mixed atmospheric conditions (i.e., minimal vertical gradients).

Additional reasons for selection of the 1100-1700 time period are illustrated in Figure 2, which shows a 3-day time series of isoprene concentrations for the N-NE site at Kunkletown, PA. Modeled concentrations at this site show a number of interesting features. The most prominent features are the significant spikes that occur around sunrise and sunset each day. These appear to be the result of interaction between emissions and boundary layer structure, as portrayed in the model. The morning spike apparently results from the onset of emissions before break-up of the nocturnal boundary layer, while the afternoon spike results from the continuing emissions of isoprene as the nocturnal boundary layer becomes established. These periods were excluded from the comparison because they represent transient situations that the model may not be designed to handle.

Figure 2

The other prominent feature of the time series is the routine decrease of isoprene overnight to near zero concentration. This occurs at virtually all sites, both in the modeled values and in the observations. The nightly return to baseline reflects the degradation of isoprene by ozone (and possibly nitrate radical) in the absence of surface emissions. Although it is reassuring to see similar behavior in nighttime concentrations, nighttime hours were excluded from the comparison because isoprene emissions are negligible in the absence of sunlight.

3.0 RESULTS AND DISCUSSION

Means and 95-percent confidence intervals for observed isoprene, modeled isoprene and various comparison statistics are shown in Table 2 (below). In general, results show a broad range of concentrations across sites, as well as significant differences between observed and modeled isoprene. Mean observed isoprene ranged from 2.1 ppb-C at Cape Elizabeth, ME to 28.6 ppb-C at Land Between the Lakes, KY. Modeled isoprene concentrations ranged from 4.3 ppb-C at Holbrook, PA to 48.2 ppb at Truro, MA.

The overall average observed isoprene concentration was 10.2 ppb-C, while the overall average modeled isoprene was 15.7 ppb-C. Modeled and observed isoprene were significantly different (95-percent confidence interval) for 10 of the 16 sites, and mean differences were within 2 ppb-C for only 2 sites. Mean percent differences (PDs) were generally large and ranged from -71.4 percent at Brookhaven, NY to 244.8 percent at Cove Mountain, TN.

Absolute differences (ADs) and absolute percent differences (APDs) provide useful measures of typical comparability between two or more quantities, because positive and negative deviations are not allowed to cancel. Mean ADs between observed and modeled isoprene ranged from 1.1 ppb-C at Holbrook, PA to 34.9 ppb-C at Truro, MA and averaged 10.7 ppb-C over the 16 sites. Mean APDs ranged from 32.9 at Holbrook, PA to 245 percent at Cove Mountain, TN and averaged 119 for the study. In other words, modeled and observed isoprene were typically only within about a factor of two of each other.

Figure 3 shows mean modeled and mean observed isoprene for all sites. Also shown is the line of ideal agreement (1:1) and envelopes of agreement within +/- 25 percent and +/- 50 percent. Results show that overall agreement between the model and observations is relatively poor. Of the 16 sites, only 2 (Holbrook, PA and South DeKalb, GA) agreed within +/- 25 percent and only 6 agreed within +/- 50 percent. Of the 10 remaining sites, 9 fell above and 1 (Brookhaven, NY) fell below the +/- 50 percent envelope.

Figure 3

Mean PDs between modeled and observed isoprene are shown in Figure 4. Brookhaven, NY and Chicopee, MA exhibit significantly and moderately negative PDs, respectively, while all other sites exhibit positive PDs. Five of the 16 sites have mean PDs between -50 percent and + 50 percent and another 5 sites have PDs between 50 and 100 percent. PDs in excess of 100 occur at 5 sites, representing each of the networks and covering the entire geographic range of the study (i.e., from Louisiana to Maine). Model overpredictions thus are not linked to any particular sampling regime or restricted to a particular part of the model domain. By the same token, sites that exhibit better agreement between observations and model predictions (i.e., +/- 50%) represent all networks (sampling protocols) and a broad geographic range.

Figure 4

PDs between mean observed and mean modeled concentrations are depicted in Figure 5. The overall range of PDs is similar to Figure 4; however, the overall distribution is somewhat different. Six sites exhibit PDs of +/- 50 and these are evenly divided between positive and negative values. One site (Brookhaven, NY) is below -50 and 5 sites are above 100.

Figure 5

The scattergram of mean difference versus mean modeled isoprene concentration is shown in Figure 6. Also shown is the linear least squares regression line through the data and corresponding regression statistics. Results show a surprisingly linear relationship (R²=0.78) and one that suggests an increasing tendency to overpredict with increasing concentration. For modeled concentrations below 10 ppb-C, there is significant scatter around a mean difference of zero. Eight sites have mean modeled concentrations of less than 10-ppb-C and 3 of these exhibit an underprediction, on average. In contrast, eight sites have mean modeled concentrations of 10 ppb or greater and all but one of these show significant overpredictions. For modeled concentrations of 20 ppb-C or higher, 4 of 5 sites show overpredictions of at least 10 ppb-C.

Figure 6

The physical significance of the regression statistics shown in Figure 6 is not entirely clear. The slope of the regression line suggests that, in the limit, model overpredictions could amount to roughly 80 percent. A negative intercept, on the other hand, implies that some underprediction may occur at the low end of the concentration range. Given the limited number of data points in the regression, a literal interpretation of the regression statistics appears to be unwarranted. Nevertheless, the tendency for the model to overpredict seems clear. It is the precise nature of the overprediction that is uncertain. Additional sites, especially those with high predicted isoprene concentrations, are needed to confirm, or refine, the present observations.

The complexity of the UAM-V modeling system is such that a wide variety of factors could be responsible for isoprene overpredictions. These factors include, among others, vegetation specific emission rates, chemical destruction rates and meteorological advection and dilution rates. Site representativeness is another factor that must be considered in any model evaluation or comparison exercise. We have selected rural and suburban sites for this analysis based on the assumption that these sites are better characterized with respect to isoprene emissions than urban sites. We have also selected midday data to avoid large concentration excursions that may occur when the nocturnal boundary layer is breaking up or forming. Nevertheless, it must be recognized that we are using point measurements near the surface to represent a three-dimensional cell approximately 12 kilometers on a side and 50 meters deep.

Two potential sources of bias that we have been able to evaluate to a limited extent include vertical gradients and surface temperature error. The first of these addresses the ability to represent a finite layer with a measurement near the bottom of that layer. As noted earlier, isoprene measurements were taken at a height of about 5-10 meters above ground level (agl), whereas the model calculates an average concentration through a layer 50 meters deep. Andronache and Chameides (1993) have examined near-surface concentration gradients of isoprene. Based on available data and theoretical considerations, the authors concluded that isoprene typically exhibits an exponential decay with height that is related to rates of vertical mixing and chemical destruction. The resulting vertical gradient produces a 50 percent reduction in isoprene concentration between the surface and roughly 300 meters. However, the reduction between the surface and 50 meters (i.e., the top of the lowest UAM-V layer) is much less. A measurement at 10 meters is expected to be only 2-6 percent higher than the average through a 50 meter layer. Not only is the difference small relative to model overpredictions, it is of the wrong sign to moderate the overpredictions. In other words, a gradient of the form predicted by Andronache and Chameides (1993) would only increase differences between the model and observations.

The other area of investigation involved surface temperature measurements used to drive the isoprene emissions module. Temperature data for the isoprene module come from measurements made at roughly 1.5 meters agl. This height may be appropriate for short vegetation (e.g., grass and crops) but may not be appropriate for taller vegetation such as forest canopies. Examination of delta temperature information from the CASTNet network (a rural ozone and meteorology network of 40-50 sites) shows that there is typically a daytime gradient of about -0.5 to -1.0 degrees C (surface warmer) between 2 meters and 9 meters agl. Based on the temperature correction factors of Guenther et al. (1993), a temperature gradient of this magnitude implies a possible error (overestimate) in emissions on the order of 7-15 percent. This source of bias is not negligible, but it is generally much smaller than differences between observations and model output reported here.

4.0 SUMMARY AND CONCLUSIONS

UAM-V model output for the 1995 OTAG episode was compared with field observations from 16 rural and suburban sites operated by N-NE, SOS/Nashville and PAMS. Afternoon values were compared in order to eliminate, or at least minimize, vertical gradients and transitory excursions in isoprene concentration. The following conclusions obtain from the present study:

1. The overall accuracy of UAM-V isoprene predictions is relatively poor. Of the 16 study sites, only 2 showed agreement within +/- 25 percent and only six showed agreement within +/- 50 percent. Modeled isoprene was at least 50 percent higher than observation for 9 sites and at least 50 percent below observations for 1 site. Mean absolute percent differences averaged 119 over all sites and all samples, indicating that modeled values are typically only within a factor of two of observations.
2. Examination of model performance as a function of predicted concentration indicates that the model tends to overpredict at higher concentrations and that the bias increases with increasing concentration. At concentrations below approximately 10 ppb-C the model exhibits relatively small bias. At concentrations above 20 ppb-C the model exhibits a strong positive bias. In other words, model accuracy tends to decrease as predicted concentration increases. Since high isoprene concentrations generally imply NOx limitation of ozone production, this finding suggests that the model may be less reliable under conditions of NOx limitation than under other conditions.
3. Model overpredictions occur across all three operating networks and across the geographic range of sites. The bias does not appear to be related to a particular sample collection strategy or site setting (i.e., rural versus suburban). Data are not available to assess model performance in several areas of the OTAG domain; nevertheless, the areas that are represented in the analysis appear to cover much of the range of predicted concentrations (i.e., <5 ppb-C to > 40 ppb-C).
4. Estimated vertical gradients in isoprene and temperature do not appear to account for differences between modeled and observed isoprene. The vertical concentration gradient of isoprene is relatively weak between 10 meters and 50 meters and of the wrong sign to mitigate model bias. Anticipated vertical temperature gradients can account for only a small portion of the model bias (i.e., 7-15 percent).
   

5.0 POLICY IMPLICATIONS

Isoprene concentrations are seriously overpredicted by UAM-V for the 1995 modeling episode. In the absence of other information, similar overpredictions are expected for the 1988, 1991 and 1993 episodes as well. Isoprene overpredictions can predispose the model to a NO_x limited condition and can predispose the model to overstate the frequency and spatial extent of NO_x limitation. This, in turn, makes the model overly sensitive to NO_x control strategies and not sensitive enough to VOC control strategies. Control strategy results must be interpreted in light of this model predisposition.

6.0 REFERENCES

Atkinson, R. 1990.Gas-phase tropospheric chemistry of organic compounds: A review. Atmos. Environ. 24A:1-41.

Guenther, A., P. Zimmerman and P. Harley. 1994. Natural volatile organic compound emission rate estimates for U.S. woodland landscapes. Atmos. Environ. 28A:1197-1210.

Geron, C.A., A. Guenther and T. Pierce. 1994. An improved model for estimating emissions of volatile organic compounds from forests in the eastern U.S. J. Geophys. Res. 99:12773-12791.

Chameides, W. L., F. Fehsenfeld, M. O. Rodgers and C. Cardelino. 1992. Ozone precursor relationships in the ambient atmosphere. J. Geophys. Res. 97:6037-6055.

Cardelino, C. A. and W. L. Chameides. 1990. Natural hydrocarbons, urbanization and urban ozone. J. Geophys. Res. 95:13971-13979.

Andronache, C. and W. L. Chameides. 1994. Vertical distribution of isoprene in the lower boundary layer of the rural and urban southern United States. J. Geophys. Res. 99:16989-16999.

Guenther, A., P. Zimmerman and P. Harley. 1993. Isoprene and monoterpene emission rate variability: Model evaluations and sensitivity analyses. J. Geophys. Res. 98:12609-12617.