Prepared for Marcel Haberstadt

American Automobile Manufacturers Association

7430 Second Avenue, Suite 300

Detroit, Michiogan 48202

Prepared by Ralph E. Morris, Keane Lee, Greg Yarwood

ENVIRON International Corporation

101 Rowland Way, Suite 220

Novato, CA 94945

31 January 1997


A preliminary analysis was conducted using readily available ambient air quality data to compare observed isoprene concentrations with concentrations predicted by the Ozone Transport Assessment Group (OTAG) UAM-V photochemical model simulations for the July 10-18, 1995 episode.


Opening Comments

A preliminary analysis was conducted using readily available ambient air quality data to compare observed isoprene concentrations with concentrations predicted by the Ozone Transport Assessment Group (OTAG) UAM-V photochemical model simulations for the July 10-18, 1995 episode. These UAM-V predictions are based on biogenic emissions inventory estimates generated by the BEIS2 model. It has been suggested by some observers that OTAG BEIS2 isoprene emission estimates, under which isoprene constitutes 70 - 80 percent of total VOC emissions in the OTAG domain, may be over stated. OTAG UAM-V model results show almost no sensitivity to VOC controls except in some of the very largest urban areas (Northeast urban corridor, Chicago). This result appears to be inconsistent with trends reports such as those reviewed by Morris (1996) which indicate that declines in ozone concentrations over the past ten years or so in the eastern U.S. coincided with substantial reductions in VOC emissions but little or no change in NOx emissions. Furthermore, inventory evaluation studies suggest that anthropogenic VOC emissions may be under-represented in the OTAG inventory, yet ozone is significantly overpredicted under this inventory in isoprene-rich areas such as the South. Previous evaluations of BEIS2 based on direct comparisons of isoprene fluxes from vegetation are based on a limited range of vegetation types and ambient conditions and therefore may not be representative of the full range of conditions represented by the OTAG model application. Reconciliation of the BEIS2 isoprene inventory with ambient measurements is complicated by the high reactivity of isoprene which must somehow be accounted for. UAM-V isoprene concentration predictions account for atmospheric dispersion, reactivity, and deposition processes. Therefore, comparison of UAM-V predictions with isoprene measurements is the best methodology currently available for performing real-world evaluation of the BEIS2 emission estimates used in the OTAG modeling.

Ambient measurements of isoprene, ozone, total non-methane hydrocarbons (TNMOC), and NOx were compared with UAM-V predictions for the surface layer grid cell in which each ambient monitoring site is located. The UAM-V OTAG surface-layer grid cells are approximately 12 km x 12 km wide by 50 m high. Comparisons were made for July 10-18, 1995 at 15 monitoring sites: 11 PAMS sites in the Northeast, two PAMS sites in Baton Rouge, LA, one Southern Oxidants Study site in Nashville, TN, and the Harvard Forest, MA research site. Time series of observed and predicted concentrations were compared for each species. Additional statistical comparisons were conducted using data from the mid-morning to early evening (hours 11 - 19) during which isoprene emissions peak and vertical mixing is usually most vigorous, thus minimizing differences between surface-based measurements and the volume-average predictions. Statistical distributions of observed and predicted concentrations (unpaired in time) were compared, and summary statistics (mean observed and predicted, bias, bias as percent of mean observed, gross error, gross error as percent of mean observed, and normalized bias and normalized gross error above preselected threshold concentrations) were calculated for values paired in time. All analyses were done on a site-by-site basis.

For isoprene, biases as a percent of mean observed were less than plus or minus 50 percent at eight out of the 15 monitoring sites. Most (six) of the remaining sites exhibited large over predictions with biases of between 150 to 250 percent at four of these sites. One site (Ware, MA) exhibited a bias of less than -50 %. These results are in qualitative but not quantitative agreement with those from a parallel study being conducted by Dr. Eric Edgerton. Reasons for the quantitative discrepancies between the two studies have not been fully explored in this preliminary analysis. Although preliminary and not yet conclusive, the results to date suggest there may be a bias toward overestimating isoprene in the OTAG Modeling System.

For TNMOC, biases were within plus or minus 50 % at over half of the sites examined; observed TNMOC values were significantly less than predicted values at the two Baton Rouge sites; significant overpredictions were noted at sites in Delaware and Maine. Paradoxically, at many of the sites where isoprene was overpredicted, TNMOC was underpredicted. Thus, the isoprene weight percents at these sites were severely overpredicted. Bias in the Isoprene/TNMOC ratio was within plus or minus 50 % at 7 of the 13 sites with available TNMOC measurements; the ratio was underpredicted by 59 percent at one site (Ware, MA where isoprene was underpredicted by just over 50% ) and overpredicted at four sites.

A comparison of predicted and observed TNMOC/NOx ratios revealed significant differences at some sites, whereas at others the TNMOC/NOx ratios were comparable. When there were significant differences, the OTAG Modeling System tended to overstate the observed TNMOC/NOx ratios.

Despite the relatively large discrepancies in observed and predicted precursor concentrations and ratios noted above, UAM-V ozone predictions were biased by more than plus or minus a relatively modest 15 percent at only 3 of the 13 sites with ozone measurements. However, it is somewhat disturbing that at these 13 sites there is a tendency toward systematic overestimation of the average afternoon observed ozone concentrations.

Further analysis of the OTAG Modeling System isoprene concentrations is ongoing. Data from additional sites are being included. In addition, comparisons aloft with the NARSTO-NE aircraft measurements are also planned.



Many States in the eastern U.S. performed preliminary photochemical modeling for their 1994 State Implementation Plans (SIPs) which suggested transported ozone and ozone precursors from upwind regions were contributing significantly to ozone exceedances in their nonattainment areas. The presence of significant amounts of transported ozone from upwind made it difficult for many of the nonattainment regions to demonstrate attainment with any reasonable level of local emissions controls. In recognition of the transport problem, the U.S. Environmental Protection Agency (USEPA) established a two­phase program for states to develop approvable ozone SIPs. In a policy memorandum dated March 2, 1995 ("Ozone Attainment Demonstrations"), USEPA outlined the major elements of this program. Phase I requires states to complete pre­November 1994 SIP requirements, submit regulations sufficient to meet the initial Rate of Progress requirements, and submit modeling analyses. Phase II calls for a two­year (1995­1996) consultative process to assess national and regional strategies to deal with ozone transport in the eastern U.S., and subsequent revisions of local control plans, as necessary, based on any new national or regional strategies. To accomplish the Phase II consultative process, the Environmental Council of States (ECOS), in conjunction with USEPA, established the Ozone Transport Assessment Group (OTAG).

OTAG is performing regional meteorological (RAMS and SAIMM), emissions (EMS95), and photochemical (UAM-V) modeling of the eastern U.S. to identify regional emissions control strategies to reduce ozone transport into ozone nonattainment regions during periods of ozone exceedances. Four episodes were selected for regional photochemical modeling:

The EPA, Midwest, Southeast, and Northeast Modeling Centers are taking the lead for modeling, respectively, the 1988, 1991, 1993, and 1995 ozone episodes.

Early on in the OTAG process, a decision was made to use BEIS2 biogenic emission estimates over BEIS1. BEIS2 uses more recent biogenic emission measurements and is generally believed by the scientific community to provide more accurate estimates of biogenic emissions. However, preliminary regional model simulations performed using BEIS2 exhibited serious overprediction of the observed ozone concentrations (e.g., EPA ROM and MOCA UAM-V sensitivity simulations). OTAG formed an Ad Hoc Biogenic Emissions Committee to investigate the use of BEIS1 versus BEIS2. It was recommended that BEIS2 should be used and that the CBM-IV isoprene chemistry be updated using more recent smog chamber experiment data which would lower the ozone formation potential of isoprene.

The latest (D2) UAM-V base case simulations for the four OTAG episodes using BEIS2 and the new isoprene chemistry exhibited "reasonable" performance in the Northeast and Midwest, but still exhibited an overprediction tendency in the South, especially over Atlanta. Biogenic emissions (mainly isoprene) constitute 70-80 percent of the total VOC emissions in the OTAG modeling domain. The OTAG VOC/NOx control sensitivity tests exhibit almost no ozone sensitivity to VOC controls, except for some of the very largest urban areas such as Chicago and New York City. Clearly, the BEIS2 isoprene emissions are driving the conclusions of the OTAG emissions sensitivity simulations. There is circumstantial evidence that BEIS2 may be overestimating biogenic isoprene emissions:

Several researchers (e.g., S.T. Rao of the NYSDEC, George Wolff of General Motors, B. Cox and S.-H. Chu of EPA, and K. Jones of Zephyr Consulting, see OTAG Review Report, Morris, 1996) have noted that over the last decade or so there is a definite downward trend in observed meteorology-adjusted ozone concentrations in the Northeast. Since NOx emissions in the region have been fairly stable but VOC emissions have been reduced substantially, this suggests that the reduction in ozone might be due to the reductions in VOC emissions. Yet the OTAG UAM-V ozone estimates exhibit very little sensitivity to reductions in VOC emissions.
Anthropogenic VOC emissions may be under-represented in the OTAG emissions inventory (e.g., absence of off-cycle VOC emissions from mobile sources), with Canadian emissions being particularly suspect, yet ozone is significantly overpredicted in the South. One possible hypotheses is that the biogenic emissions may be overstated compensating for low anthropogenic emissions in the Northeast and Midwest but, because biogenic emissions are much greater in the South, overcompensating for low anthropogenic VOC emissions in the South resulting in ozone overprediction tendency.

Given the importance of BEIS2 emissions estimates in driving the OTAG VOC versus NOx emission control decision making, a real-world evaluation of the BEIS2 biogenic emission estimates as derived in the OTAG modeling is needed. There have been several studies shich compred BEIS2 emission fluxes with measured fluxed off of a forest canopy (e.g., Guenther et al., 1996). However, comparison of the BEIS2 emissions estimates with special study measurements of emissions from vegetation is complicated by the limited range of vegetation types and ambient conditions for which the experiments were performed. Further, these measurements were collected under ideal conditions whereby microscale meteorological and landuse characteristics were quantified which is not possible when generating regional inventories as for the OTAG domain. There has been some success in the past in performing emissions/ambient measurement reconciliation, however implementation of such an approach for BEIS2 would be problematic due to the high reactivity of isoprene; any comparison of the BEIS2 emission estimates with ambient data must account for the reactivity of isoprene. Since the UAM-V takes isoprene reactivity into account and the OTAG UAM-V version contains an update of the isoprene chemistry, the comparison of the UAM-V isoprene estimates with ambient data provides the best methodology currently available for performing a real-world evaluation of the isoprene emissions estimates being used in the OTAG modeling.

Overview of Approach

The objective of this study is to evaluate the OTAG UAM-V/BEIS2 isoprene estimates against ambient measurements to determine whether BEIS2 as applied in the OTAG modeling study accurately represents observed isoprene (biogenic) emissions. Historically, speciated VOC ambient measurements have only been collected during specialized field studies (e.g., LMOS and SOS) or at specialized research chemical measurement sites. However, for the 1995 OTAG modeling period, there are several sources of isoprene measurements with which to compare with the UAM-V/BEIS2 model estimates:

By 1995, many Photochemical Assessment Monitoring Sites (PAMS) were in operation providing a fairly large database of measured speciated VOC (and other species) concentrations, at short (hourly and 3-hourly) time resolution.
The NARSTO-Northeast 1995 field study included 9 air quality research stations in the eastern U.S. which collected speciated VOC data during Intensive Operating Periods (IOPs) (including July 12-16 of the 1995 OTAG episode). NARSTO-Northeast also operated aircraft which collected VOC samples aloft during the IOPs.
The Southern Oxidant Study (SOS) operated an extensive field study, with surface and aloft air quality sampling, in the Nashville area during the summer of 1995. Although data from the SOS study are slow in becoming available, data may be available upon request from some of the individual researchers (e.g., TVA).
There are other special study data associated with universities, laboratories, or States that may be available (e.g. Harvard Forest data).

This study was planned in Two Phases. In Phase I we would provide a preliminary evaluation of the OTAG UAM-V/BEIS2 isoprene predictions using currently available observed data that were available by mid-December and did not involve extensive processing. If the results from the preliminary evaluation are enlightening, then the comparisons of the UAM-V/BEIS2 model predictions with a more "complete" (as available) observational data set will be performed and the result documented in a report. We will also look at other comparisons of ozone and ozone precursors concurrent with the isoprene measurements in order to obtain a complete and comprehensive understanding of the photochemical system.

Speciated ambient VOC measurements in the eastern U.S. for the July 1995 OTAG episode that were available by Mid-December 1995 were used in the preliminary (Phase I) analysis discussed in this document. Unfortunately, the speciated surface and aircraft VOC data from the NARSTO-95 field study were still undergoing quality assurance at the time this study was performed and so could not be included in the preliminary analysis. In addition, data from many of the PAMS sites were not yet available or contained incomplete or possibly suspect data and so could also not be included. Also, except for one site, data from the SOS 1995 Nashville study have not yet been released.

Complimentary Studies

There have been several independent evaluations of BEIS2 that bear mention before discussing this study's results. Some of these studies have involved the collection of detailed VOC emissions factors from different vegetation types which were used to develop and evaluate the BEIS2 emission factors, whereas others collected ambient speciated VOC and other species and microscale meteorological measurements which were used along with detailed site specific biomass to provide independent evaluation of the BEIS2 isoprene and VOC emission fluxes from a canopy (e.g., Guenther et al., 1993; 1994; 1996; Monson et al., 1995; Geron et al., 1995; Lawrimore et al., 1995; Geron et al., 1994). There have also been some indirect evaluations of the BEIS2 emissions estimates using photochemical models whereby modeled ozone and precursor concentrations were compared against observations using BEIS1 and BEIS2 biogenic emission estimates (e.g., Sillman et al., 1995). However, none of the aforementioned studies evaluated the biogenic emissions estimates being used in the OTAG process. Currently, there are several groups performing such analysis including each of the OTAG Modeling Centers, the OTAG model performance evaluation contractor (STI), work by the one of the NARSTO-NE contractors (STI), and work being performed by Eric Edgerton formerly of Environmental Science Engineering (ESE) for Southern Company Services (SCS). In this Chapter we breifly discuss some of these studies to set the stage and aid in the interpretation of our analysis presented in Chapter 3.

Comparison of Measured Versus BEIS2 Isoprenes Fluxes from a Forest Canopy

As noted above, there have been several studies comparing measured versus model estimated isoprene fluxes from a forest canopy. Probably the most detailed and extensive analysis was performed at a forested site near Oak Ridge Tennessee during July and August 1992 using 8 different measurement techniques (Guenther et al., 1996). Individual leaf emissions and midday forest isoprene fluxes were measured along with meteorological eddy measurements on a tower in the canopy. In addition, detailed information on vegetation biomass was collected as a function of direction from the measurement tower. When using the detailed information on meteorology within the canopy (e.g., temperature and UV) and accounting for the biomass distribution along the sector of forest that was upwind of the tower, they found that the isoprene emission fluxes derived using the BEIS2 emissions factors were within 25 percent of the measured fluxes.

OTAG Model Performance Evaluation

As part of the model performance evaluation of the OTAG UAM-V D2 base case simulations, comparisons were made between predicted and observed ozone precursors (RHC, NOx, and NOy) at five NARTO-Northeast monitoring sites: Holbrook PA; Arendtsville PA; Kunkeltown PA; Brookhaven NY; and Truro MA (OTAG, 1996). In general, the model overestimated the NOy concentrations and underestimated the RHC concentrations. However, the comparisons at rural sites, where the modeled RHC was biased low on average, were different than at the urban PAMS sites, where an average overestimation bias for RHC existed.

OTAG Isoprene Performance Evaluation by ESE/SCS

Dr. Eric Edgerton, formerly of ESE Environmental Inc, (ESE), is performing a parallel study to this one for Southern Company Services (SCS) comparing modeled and measured isoprene concentrations during the July 1995 OTAG episode. Because ESE was the monitoring contractor for the NARSTO-Northeast and SOS-Nashville studies, he had advanced copies of the data before it underwent the final QA and release to the public. At the December 1996 OTAG meeting, Dr. Edgerton presented model/measured isoprene comparisons at 14 sites in the OTAG domain as shown in Figure 2-1. Note that in this study we initially looked at 23 potential sites with isoprene concentrations of which only four sites overlapped with the Edgerton database. Because the OTAG UAM-V isoprene predictions are 12 km x 12 km grid cells averaged over a 50 m lowest layer, the focus of the model/measured comparison was during the afternoon when the atmosphere is "well-mixed" so that the predicted 50 m layer-average isoprene concentrations are more representative of the surface measured isoprene concentrations. Dr. Edgerton also performed a height adjustment of the measured surface isoprene concentrations to adjust to the average height of layer. However, in the afternoon during vigorous vertical mixing, this height adjustment did not significantly alter the surface concentration (just a few percent).

Figure 2-2 summarizes the average percent bias in the afternoon OTAG UAM-V isoprene concentrations at the 14 sites analyzed. Because of the uncertainties in the modeled grid cell average versus point measurement and uncertainties in the emissions and meteorology, Dr. Edgerton divided his sites into those that predicted the observed afternoon isoprene on average within ± 50%, those with an overprediction bias (> 50%), and those with an underprediction bias (< -50%). He noted that 5 of the 14 sites (36%) had average isoprene agreement within ± 50%. Of the remaining sites, 7 (50%) showed modeled isoprene concentrations that were substantially higher than observed (53-311% overestimation on average), and 1 showed modeled concentrations substantially lower than observed. As shown in Figure 2-3, Dr. Edgerton did not see a regional bias in the OTAG/UAM-V isoprene over- or under-prediction bias. Dr. Edgerton concluded that additional work is needed to understand the magnitude and causes of the differences between model output and field observations.

[Figures not currently available.]

Figure 2-1. Locations of the 14 monitors where model/observed isoprene concentrations were made in the Edgerton SCS Study (Source: Edgerton, 1996).

Figure 2-2. Percent bias of afternoon predicted isoprene concentrations (100 BIAS/AVG OBS) from the Edgerton SCS Study (Source: Edgerton, 1996).

Figure 2-3. Spatial distribution of modeled isoprene bias from the Edgerton SCS Study (Source: Edgerton, 1996).

Preliminary Results

Measrement Database

There were three main sources of measured isoprene data for the preliminary comparison with model estimates from the OTAG July 1995 UAM-V base case simulation:

As noted previously, it was hoped that data from the 1995 NARSTO-Northeast monitoring study could be integrated into the analysis and more of the data from the 1995 SOS-Nashville Study could be included. However, by mid-December such data had not yet been released. In addition, the number of PAMS sites where data were available was less than anticipated. We processed data for 23 sites across the OTAG region for which isoprene measurements were collected.

Because the speciated VOC sites where isoprene measurements were collected tended to be enhanced chemistry sites (e.g., PAMS), we also processed the measured data for total nonmethane hydrocarbons (TNMOC), ozone, and oxides of nitrogen (NOx) for comparison with the model estimates to aid in understanding the complete photochemical oxidant system. For each of the 23 sites for which data were available, we first plotted the time series of predicted and observed isoprene, TNMOC, ozone, and NOx concentrations to determine the reasonableness and data completeness for each site. Of the 23 sites processed, 8 were not used in the statistical comparisons either for having insufficient data capture to calculate meaningful statistics, or, in the case of measured data obtained on the UAM-V base case tape from the OTAG Clearing House, data were of unknown origin and appeared questionable.

Table 3-1 identifies each of the 23 sites originally processed for inclusion in the OTAG UAM-V/BEIS2 isoprene evaluation study and the reason for the elimination of the 8 sites from the statistical comparisons. Most of the data that were available by mid-December 1996 were PAMS sites. There are four classifications of PAMS sites: Type 1 = upwind sites; Type 2 = urban sites; Type 3 = downwind ozone maximum sites; and Type 4 = downwind outer edge.

Figure 3-1 displays the locations of the resultant 15 sites where statistical comparisons were made between the OTAG UAM-V/BEIS2 modeling system predicted and the measured values.

Table 3-1. Preliminary set of monitoring sites where speciated VOC measurements were collected that were processed and included in the OTAG UAM-V/BEIS2 isoprene model performance evaluation database.

Site Identifier

Site Location

Site Type

Reason for Exclusion


E. Hartford, CT

Type 2 PAMS

No isoprene observations in database


Stafford, CT

Type 3 PAMS


Lums Pond, DL

Type 1,4 PAMS


Braidwood, IL


Low capture, unknown source


Camp Logan, IL


Low capture, unknown source


Capitol, LA

Type 2 PAMS


Pride, LA

Type 1 PAMS


Agawam, MA

Type 1 PAMS

Low isoprene data capture


Borderland, MA

Type 1,3 PAMS

No data on record


Chicopee, MA

Type 2 PAMS


Harvard Forest, MA

Research Site


Lynn, MA

Type 2 PAMS


Newbury, MA

Type 3 PAMS

Low isoprene data capture


Ware, MA

Type 3 PAMS


Lake Clifton, MD

Type 2 PAMS


Cape Elizabeth, ME

Type 3 PAMS


Rider College, NJ

Type 3 PAMS


The Bronx, NY

Type 2 PAMS


Arendtsville, PA

Research Site

No data currently (12/96) available


Philadelphia, PA

Type 2 PAMS


East Providence, RI

Type 2 PAMS


Youth Inc., TN

SOS Site


Corbin, VA


Insufficient data capture

Data Model Intercomparisons

Using the preliminary database described above, we performed model-data comparisons between isoprene and related species. Figures showing the complete comparison for all of the measures are provided in the Appendices to this report. A few descriptive and summary figures are presented in this section. The following comparisons were made for this study using the preliminary (data available by mid-December cut-off time) database:

Time Series Analysis
Appendix A contains time series of predicted and observed isoprene, TNMOC, ozone, and NOx concentrations at each of the 23 sites listed in Table 3-1.

East Hartford, Connecticut (CT HAR): At the time the data from the CT HAR site were downloaded there was no total or speciated VOC measurements available for the OTAG July 1995 modeling period. The agreement between the predicted and observed ozone concentrations is reasonable, except the model fails to capture the ozone spikes on the high observed ozone days (e.g., July 13, 14, and 18). As this is an urban PAMS site (Type 2), this difficulty of the model to replicate the observed peaks may be due to the low model resolution (12 km ) which may be insufficient to characterize urban plumes.

Stafford, Connecticut (CT STR): Stafford CT is one of four sites, along with Chicopee MA (MA CHI), Harvard Forest MA (MA HAV), and Ware MA (MA WAR), that are located fairly close to each other in northern CT/middle MA (see Figure 3-1). There is a fairly consistent trend in the model/observation comparison among the four sites in this region. Although the model tends to track the diurnal variations in the observed isoprene concentrations fairly well at Stafford and Chicopee, it fails to capture the very highest observed afternoon isoprene spikes (e.g., on July 14). The observed isoprene peak at the MA CHI site that occurred around midnight on July 18 is suspect as there should be minimal isoprene emissions at this time. At the MA HAV and MA WAR sites, the model is underestimating the observed daytime isoprene concentrations. In general, it appears that isoprene concentrations are underestimated at these four sites. The model does a poorer job in reproducing the observed TNMOC concentrations at these four sites than for isoprene. The modeled and observed diurnal TNMOC profiles appear to be out of phase with each other. The comparisons of the predicted and observed ozone concentrations range from fairly reasonable (e.g., MA WAR and CT STR) to suspect (MA CHI). The predicted and observed comparisons for NOx are more suspect due to uncertainties in the measurement techniques (e.g., the measured values include more than just NO and NO2) and there is also the potential for subgrid-scale impacts of local plumes. Not surprisingly, the model fails to capture localized spikes and, in general, the observed NOx is higher than predicted, which may be partly due to additional observed nitrogen species that are included in the observed "NOx".

Lums Pond, Delaware (DL LUM): At DL LUM the model estimated isoprene concentrations have a curious double peak every day of the episode, the expected one in the afternoon when isoprene emissions are at the greatest and an additional one in the early morning. On some days the observed diurnal profile also contains such a double peak (e.g., July 18), but not on every day of the episode as estimated by the model. It is speculated that this early morning modeled isoprene spike may be due to inconsistencies between the RAMS generated atmospheric mixing profiles and the BEIS2 derived isoprene emissions. The RAMS meteorology was not used to develop the BEIS2 emission estimates, thus, there may be inconsistencies in the meteorology used (e.g., temperature and UV) to define biogenic emissions and vertical mixing. For example, a time zone shift or, different light algorithms, or presence/lack of clouds may result in sufficient temperature and light for BEIS2 to estimate isoprene emission fluxes which are not present in the RAMS meteorology so there is very slow vertical mixing. The model fails to capture the magnitude and diurnal variations in the observed TNMOC. The model/observed ozone comparisons are reasonable on some days but poor on others (e.g., July 15). The model estimates an earlier onset of ozone production than observed, which may possibly be related to the higher than observed predicted morning isoprene concentrations.

Camp Logan and Braidwood, Illinois (IL CAM and IL BRA): Isoprene observations for the two Illinois sites were included on the data tape of the UAM-V base case model predictions from the OTAG Clearing House. Only isoprene measurements were included and these data were very sparse. Thus, data from these two sites were not included in the statistical model/data comparisons.

Capitol and Pride, Louisiana (LA CAP and LA PRD): The model exhibits very poor agreement with the magnitude and diurnal variations in the observed isoprene, TNMOC, ozone, and NOx concentrations. Care should be taken is using the model to evaluate alternative control plans in this region. These sites are located in the outer 30 km resolution model grid.

Agawam and Borderland, Massachusetts (MA AGA and MA BOR): Both of these sites suffer from poor data capture, so were not used in the statistical model/observed comparison. For the one day in which isoprene observations did exist at the MA AGA site (July 14), the model agreed quite well with the observed values.

Lynn, Massachusetts (MA LYN): In general, the predicted and observed hourly isoprene concentrations agree reasonably well, with the exception of modeled very high afternoon isoprene concentrations on some days (e.g., > 50 ppbC on July 14) that are not reflected in the observed values. The model and observed TNMOC hourly concentrations also agree reasonably well, however again it appears that the model diurnal profile is out of phase with the observations. The observed ozone diurnal profile on the two higher ozone days (July 13-14) are simulated quite well by the model.

Newbury, Massachusetts (MA NEW): For the few days in which observed isoprene concentrations were available at this site, the model tended toward overestimation. Data capture for isoprene, TNMOC, and ozone were low at this site so it was not included in the statistical performance evaluation.

Lake Clifton, Maryland (MD LKC): At the Maryland site near Baltimore the model is significantly overestimating the observed isoprene concentrations -- the two-peaked diurnal profile is present in both the measurements and model estimated time series. TNMOC appears to be also slightly overestimated, again the model and observed diurnal profiles are quite similar. The modeled ozone performance ranges from extremely good on some days (July 14) to questionable on others.

Cape Elizabeth, Maine (ME CAP): During periods of very low isoprene concentrations, both the model and observations agree quite well (e.g., July 10-12, 16-19). However, during higher isoprene concentration conditions it appears that the model overestimates the observed afternoon peaks. Although the model estimated TNMOC appears to be greater than observed over most hours, the observed TNMOC spikes are not replicated by the model. Until the end of the episode when the predicted and observed ozone concentrations are both near background concentrations, the model does a poor job in reproducing the observed hourly ozone concentrations.

Rider College, New Jersey (NJ RID): The diurnal variation in the observed hourly isoprene concentrations are reproduced by the model reasonably well at this site, with the exception of underestimating the observed spikes on July 14 and July 15.

The Bronx, New York (NY BNX): As seen for some of the other sites, the model estimated diurnal bimodal isoprene peaks do not appear to be reflected in the observations. The modeled isoprene concentrations vary from underestimation (e.g., July 12 and 17) to overestimation (July 15). The diurnal variations in the observed ozone concentrations have large amounts of hour-to-hour variability reflecting the influences of local NOx sources at this urban site. The modeled diurnal ozone profile is much smoother due to the inability of the model to simulate the subgrid-scale impacts of local sources.

Arendtsville, Philadelphia (PA ARE): When data were acquired and processed for this site, there was no data available during the July 1995 OTAG episode.

Philadelphia, Pennsylvania (PA PHL): The model appears to systematically overestimate the observed isoprene concentrations in Philadelphia. Except for high (> 200 ppbC) TNMOC peaks on July 10 and July 12, the model appears to reproduce the magnitude of the observed TNMOC reasonably well. For most days the observed hourly ozone concentrations are not simulated well by the model.

East Providence, Rhode Island (RI PRO): The model significantly overestimates the observed isoprene concentrations at this site on July 13-16. However, TNMOC is generally underestimated. The general day-to-day variation in the observed ozone concentrations are reproduced reasonably well, although the model fails to reproduce the observed peak on July 14.

Youth Inc., Tennessee (TN YTH): Only isoprene concentrations were acquired for the Tennessee site. Except for the underestimation of the high observed isoprene concentrations on July 11 and 15, the model reproduces the observations reasonably well.

Corbin Virginia (VA COB): Only one day of isoprene data was available at this site (July 14) during which it appears that the model exhibits significant overprediction. Due to the limited data availability, this site was not included in the statistical data/model comparison.

Statistical Comparisons of Predictions and Observations
We performed two types of statistical comparisons of the predicted and observed concentrations and their ratios at the 15 selected sites:

Figure 3-2 displays the statistical model/data comparison for isoprene at the four sites located in reasonably close proximity in northern CT and mid-MA (see Appendix B for all sites). At these four sites it appears that the model has a tendency toward underestimating the observed isoprene concentrations. An examiniation of the Box Plots reveals that at these sites the observed isoprene concentrations have much greater variability than predicted. The variability across the sites is also much greater in the observed values than predicted; the average observed isoprene conentrations vary by almost a factor of three at the four sites that are in reasonably close proximty (11.5 ppbC at MA CHI to 30.8 ppbC at MA WAR). Whereas the average observed values only varies by about 50% (8.1 ppbC at MA CHI to 12.3 at MA WAR). Similar Box Plots and model performance statistics for the remainder of the sites are provided in Appendix B. More analysis is needed to fully interpret and understand the results of these displays.

Isoprene Summary: Figure 3-3 summarizes the percent differences between the predicted and observed afternoon (1100-1900) isoprene concentrations at the 15 monitoring sites under study in the Phase I preliminary analysis. Unlike the Edgerton analysis presented in Figure 2-2, we see that over half of the sites (8 out of 15) have average isoprene concentrations that agree within ± 50%. However, like the Edgerton analysis discussed in Chapter 2, there is more of a tendency toward significant (bias > 50%) overprediction (6 out of 15 sites) than significant underprediction (< - 50%) (1 out of 15 sites). Furthermore, at four of the sites, the isoprene overprediction bias is quite large (150% to 250%). At the four sites that are in common between the Edgerton and this analysis (DE LUM, MA LYN, ME CAP, and TN YTH), there is qualitative agreement, but quantitative differences: (1) at DE LUM both studies estimate an overprediction bias, but Edgerton's analysis suggest that the overprediction bias (60%) is more significant than this analysis (30%); (2) at MA LYN the reverse is true with Edgerton estimating an approximate 30% bias whereas our analysis is closer to 60%; (3) the two studies agree that the OTAG UAM-V/BEIS2 modeling systemn significantly overestimates isoprene concentrations at the ME CAP sites by 200-230%; and (4) the Edgerton study calculates a larger overprediction bias at the TN YTH site (55%) than this study (10%). More details on the procedures used in the Edgerton analysis are needed to resolve these differences.

TNMOC Summary: A summary of the percent bias for the comparison of predictd and observed TNMOC concentrations are provided in Figure 3-4, the complete comparison is provided in Appendix C. As seen for isoprene, over half of the sites predict the average observed TNMOC concentrations to within ± 50%. The model appears to be understating the observed TNMOC concentrations at the two Baton Rouge sites, even though isoprene was overestimated at one of them (LA CAP). It is interesting to note that at many of the sites where there was significant overestimation of isoprene (e.g., LA CAP, MA LYN, MD LKC, PA PHL, RI PRO), the model underestimates TNMOC.

Isoprene-to-TNMOC Ratios: Appendix D compares the predicted and observed isoprene-to-TNMOC ratios, with a summary of the percent bias provided in Figure 3-5. The predicted isoprene fraction of the TNMOC agrees to within ± 50% at over half of the sites (7 out of 13). At the remaining six sites, there is one with a slightly significant underestimation (-59% at MA WAR), one with a slightly significant overestimation (68% at ME CAP), and four with very signifiant overestimation (> 200%) of the observed isoprene-to-TNMOC ratio.

TNMOC-to-NOx Ratios: Because of the uncertaintis in the NOx measurements and potential for subgrid-scale impacts at monitors, care should be taken in the interpretation of the TNMOC-to-NOx ratio comparisons. Further, just examining the summary figure of percent bias in Figure 3-6 provides an incomplete picture of the comparison; the magnitude of the ratios is much more important as they suggest the chemical regime of the photochemical system and whether VOC or NOx control will be more effective at reducing ozone concentrations. For example, if both the predicted and observed TNMOC-to-NOx ratios are above approximately 20, then the predicted and observed photochemical systems are both in the NOx-sensitivity regime even if there is signifiant bias. Similarly, a TNMOC-to-NOx ratio of less than 10 suggests a VOC-sensitive regime. An examination of the predicted and observed TNMOC-to-NOx ratios in Appendix E suggests there are serious shortcomings of the model at the DL LUM site where the observed ratio (3) is in the VOC-sensitive regime whereas the predicted ratio (25) is in the NOx-sensitive regime. At the two Baton Rouge sites, both the model and observations have ratios greater than 20 which suggest NOx-sensivity regime. The average observed TNMOC-to-NOx ratios at the three Massachusett sites are all < 10 which suggest more VOC-sensitivity, but except for one site (MA WAR) the predicted values (18, 10, and 30) are tilted more toward NOx-sensitivity conditions. The remaining four sites (MD LKC, NJ RID, NY BNX, and RI PRO) exhibit reasonable agreement between the predicted and observed TNMOC-to-NOx ratios.

Ozone Concentrations: The final model/data comparison is for ozone, which is shown in Appendix F with the average bias summarized in Figure 3-7. A more complete evaluation of the OTAG UAM-V/BEIS2 modeling system ozone model performance evaluatuion is provided in the OTAG model evaluation report (OTAG, 1996). The comparison of the predicted and observed ozone concentrations are much closer than seen for the ozone precursors. At only three sites does the bias exceed 15%, the two Baton Rouge sites and PA PHL site. Although the 13 sites analyzed represent just a small fraction of available ozone measurements, it is somewhat disturbing that at these 13 sites there is a tendency toward systematic overestimation of the average afternoon observed ozone concentrations.

24-Hour Comparisons: Figures comparing the predicted and observed isoprene, TNMOC, isoprene-to-TNMOC ratio, TNMOC-to-NOx ratio, and ozone for the full 24-hour diurnal period are provided in Appendix G. Given the difficulty in interpreting modeled 50 m concentrations with surface values under nocturnal stable comnditions and the limited resources available for Phase I of this study, at this time we have not interpeted these results.

Figure 3-1. Locations of the 15 monitoring sites where statistical comparisons of predicted and observed isoprene, TNMOC, isoprene-to-TNMOC ratios, TNMOC-to-NOx ratios, and ozone where made in the Phase I preliminary analysis.

Figure 3-1

Figure 3-2a. Statistical comparison of predicted and observed isoprene concentrations at the Stafford, Connecticut (CT STR) monitoring site. Box Plots display the 5th , 25th, median, mean (symbol), 75th, and 95th percentile of the distribution.

Figure 3-2a

Figure 3-2b. Statistial comparison of predicted and observed isoprene concentrations at the Chicopee, Massaschusetts (MA CHI) monitoring site. Box Plots display the 5th , 25th, median, mean (symbol), 75th, and 95the pecentile of the distribution.

Figure 3-2b

Figure 3-2c. Statistial comparison of predicted and observed isoprene concentrations at the Harvard Forest, Massaschusetts (MA HAV) monitoring site. Box Plots display the 5th , 25th, median, mean (symbol), 75th, and 95the pecentile of the distribution.

Figure 3-2c

Figure 3-2d. Statistial comparison of predicted and observed isoprene concentrations at the Ware, Massaschusetts (MA WAR) monitoring site. Box Plots display the 5th , 25th, median, mean (symbol), 75th, and 95the pecentile of the distribution.

Figure 3-2d

Figure 3-3. Summary comparison of the percent average bias between the OTAG UAM-V/BEIS2 predicted and observed afternoon isoprene concentrations at the 15 monitoring sites (more details are found in Appendix B).

Figure 3-3

Figure 3-4. Summary comparison of the percent average bias between the OTAG UAM-V/BEIS2 predicted and observed afternoon total nonmethane organic compund (TNMOC) concentrations at 13 monitoring sites (more details are found in Appendix C).

Figure 3-4

Figure 3-5. Summary comparison of the mean OTAG UAM-V/BEIS2 predicted and observed isoprene-to-TNMOC ratios at 13 monitoring sites (more details are found in Appendix D).

Figure 3-5

Figure 3-6. Summary comparison of the mean OTAG UAM-V/BEIS2 predicted and observed TNMOC-to-NOx ratios at the 10 monitoring sites (more details are found in Appendix E).

Figure 3-6

Figure 3-7. Summary comparison of the percent average bias between the OTAG UAM-V/BEIS2 predicted and observed ozone concentrations at the 13 monitoring sites (more details are found in Appendix F).

Figure 3-7


The preliminary model evaluation of the OTAG UAM-V/BEIS2 biogenic (isoprene) emission estimates for the July 1995 D2 base case simulation has raised several issues that need further investigation and analysis. In addition, within the limited resources of Phase I of this study, the results generated could not be analyzed in as detailed fashion as desired. Additional data sources (e.g., NARSTO-Northeast 1995 data, more PAMS sites, SOS data) have also become available since the Phase I work was initiated and need to be integrated into the analysis. The following are areas that we see require further investigation under Phase II of this work effort:

Detailed Investigation into Procedures Used to Generate the OTAG Biogenic Emissions: We need to clearly understand the procedures used and assumptions made in running BEIS2 to generate the OTAG biogenic emission estimates. In our analysis we noticed that frequently there was an unusual modeled diurnal isoprene pattern that was not supported by the observation which may have been due to inconsistencies between the meteorology used for biogenic emissions (observations) versus the meteorology used to define mixing (RAMS/SAIMM). In order to interpret any discrepancies between the theoretical basis of BEIS2 and its evaluation under ideal circumstances versus the way it was applied in the OTAG study to generate a regional biogenic emission inventory, the procedures used to generate the OTAG biogenic emissions need to be clearly documented so that their implications on the OTAG modeling can be determined.
Comparison with Detailed Flux Measurements: OTAG biogenic emission estimates and estimated isoprene concentrations for the grid cells containing the locations where detailed vegetation emission fluxes were measured in the special studies (e.g., the site near Oak Ridge, TN as reported by Guenther and co-workers, 1996) should be extracted and compared with the historical detailed measured fluxes under similar meteorological conditions. This may help establish a link between the BEIS2 experimental derived emission factors and evaluation under ideal well-characterized conditions and the real-world application of the model for generating regional biogenic emission estimates as done for OTAG.
Isoprene Bias: The analysis suggest that there may be a net overprediction bias for isoprene concentrations in the OTAG UAM-V/BEIS2 modeling system. The more rural NARSTO sites used by Dr. Edgerton appear to confirm this hypothesis. More data sites need to be included in the analysis in order to have a more comprehensive and statistically significant comparison.
Reconciliation with Edgerton Analysis: Although this study and the Edgerton analysis produced qualitatively similar results for the four sites in common, there were some quantitative differences. Discussions are needed with Dr. Edgerton to obtain details on the procedures he used to make the isoprene comparisons so they can be resolved with this study's results.
Isoprene versus Other TNMOC: One of the most provocative findings of our Phase I work effort is that at many of the sites where isoprene was overestimated by the OTAG modeling system, TNMOC was underestimated. Although of lower mass concentrations than TNMOC, isoprene is much more reactive (approximately five times more reactive than automobile exhaust). The results to date suggest that isoprene (biogenic VOC) may be overstated in the OTAG modeling system, but may be compensated for by an understatement of anthropogenic VOC emissions. More investigations into this issue is needed (e.g., urban-rural stratification) to understand the validity of this hypothesis and its implications.
Spatial Representativeness/Subregional Analysis: Related to the urban-rural analysis is the need to perform spatial analysis of the model/data comparisons to identify any spatial trends. Additional subregional analysis may also be beneficial, for example the results at the two Baton Rouge sites raise serious questions concerning the adequacy of the model in that subregion.
TNMOC/NOx Ratios: Some of the model/data comparisons of the TNMOC-to-NOx ratios were quite disturbing. In several cases, the model estimated ratios that would characterize the photochemical system as NOx-sensitive, whereas the observations suggested VOC-sensitivity. A closer look at the TNMOC and, especially the NOx measurements is needed before determining the implications of model/data differences in the TNMOC-to-NOx ratios.
Site Characterization: Comparison of the individual site characteristics (e.g., local biomass, exposure, vegetation type) with the grid cell average may provide some insight into the model/data differences.
Comparisons Aloft: Comparison of the isoprene and other measured concentrations from the NARSTO aircraft observations with the measured values aloft will provide additional insight into any potential bias in the OTAG BEIS2 emissions estimates.

Each of the above items should be examined in more detail under Phase II of this study. In addition, Phase II would generate a report and presentation quality graphics.


Andronache, C., W.L. Chameides, M.O. Rodgers, J. Martinez, P. Zimmerman, and J. Greenberg. 1994. Vertical distribution of isoprene in the lower boundary layer of the rural and urban southern United States. J. Geophys. Res. Vol. 99, p. 16,989.

Edgerton. 1996. "Status Report on Model vs. Observed Isoprene", presented to OTAG Air Quality Analyses Workgroup. December 1996.

Geron, C.D., T.E. Pierce, and A.B. Guenther. 1995, Reassessment of biogenic volatile organic compound emissions in the Atlanta area. Atmos. Envt. Vol. 29, p. 1569.

Geron, C.D., A.B. Guenther, and T.E. Pierce. 1994. An improved model for estimating emissions of volatile organic compounds from forests in the eastern United States. J. Geophys. Res. Vol. 99, p. 12,773.

Goldan, P.D., M. Trainer, W.C. Kuster, D.D. Parrish, J. Carpenter, J.M. Roberts, J.E. Yee, and F.C. Fehsenfeld. 1995. Measurements of hydrocarbons, oxygenated hydrocarbons, carbon monoxide, and nitrogen oxides in an urban basin in Colorado: implications for emission inventories. J. Geophys. Res. Vol. 100, p. 22,771.

Guenther, A., P. Zimmerman, L. Klinger, J. Greenberg, C. Ennis, K. Davis, W. Pollack, H. Westberg, G. Allwine, and C. Geron. 1996. Estimates of regional natural organic compound fluxes from enclosure and ambient measurements. J. Geophys. Res. Vol. 101, p. 1345.

Guenther, A., P. Zimmerman, and M. Wildermuth. 1994. Natural volatile organic compound emission rate estimates for U.S. Woodland Landscapes. Atmos. Envt. Vol. 28, p. 1197.

Guenther, A.B., P.R. Zimmerman, P.C. Harley, R.K. Monson, and R. Fall. 1993. Isoprene and nonterpene emission rate variability: model evaluations and sensitivity analyses. J. Geophys. Res. Vol. 98, p. 12,609.

Lawrimore, J.H., M. Das, and V.P. Aneja. 1995. Vertical sampling and analysis of nonmethane hydrocarbons for ozone control in urban North Carolina. J. Geophys. Res. Vol. 100, p. 22,785.

Monson,R.J., M.T. Lerdan, T.D. Sharkey, D.S. Schimel, and R. Fall. 1995. Biological aspects of constructing volatile organic compound emissions inventories. Atmos. Envt. Vol 29, p. 2989.

Morris R.E. 1996. "Review of Recent Ozone Measurement and modeling Studies in the Eastern United States". Draft Final Report. Prepared for the Ozone Transport Assessment Group (OTAG). February.

OTAG. 1996 Evaluation of the UAM-V Model Performance in OTAG Simulations: Summary of Performance Against Surface Observations. Draft. Prepared for Ozone Transport Assessment Group. October 25, 1996.

Sillman, S., K.I. Al-Wali, F.J. Marsik, P. Nowacki, P.J. Samson, M.O. Rodgers, L.J. Garland, J.E. Martinez, C. Stoneking, R. Imhoff, J.H. Lee, L. Newman, J. Weinstein-Lloyd, and V.P. Aneja. 1995. Photochemistry of ozone formation in Atlanta, GA: models and measurements. Atmos. Envt. Vol 29, p. 3055.


Appendix A:

Appendix B:

Appendix C:

Appendix D:

Appendix E:

Appendix F:

Appendix G:



Submit your comments, feedback, questions, and ideas pertaining this page. Your input will be automatically added to the existing annotations. In order to add a new comment, you must be registered with the OTAG/AQA Peoples Page.