next up previous contents
Next: 5.4 Case Study IV: Up: 5. CASE STUDIES FOR Previous: 5.2 Case Study II:

Subsections

  
5.3 Case Study III: A Three-Dimensional Urban/Regional Scale Photochemical Air Quality Model

Uncertainty Analysis of the Urban Airshed Model (UAM-IV) for the New Jersey/Philadelphia Domain

Photochemical modeling of ozone (O$_3$) levels in the atmosphere is complicated by the fact that ozone is a secondary pollutant; it is formed through complex chemical reactions involving oxides of nitrogen (NO$_2$ and NO), and volatile organic compounds (VOCs). The modeling process is further complicated by the presence of natural, model, and data uncertainties, as shown in Figure 5.3. Uncertainties in the input data and parameters (such as emission estimates and reaction rates) become especially significant when photochemical models are used to evaluate control strategies for the reduction of ozone levels [43].

Biogenic emissions have a significant impact on atmospheric ozone concentrations, depending on the availability of NO$_x$. Earlier studies on the effects of biogenic emissions on O$_3$ concentrations indicate that typically O$_3$ concentrations increase in response to increases in biogenic hydrocarbons. However, modeling studies suggest that in some extremely NO$_x$-limited areas, increases in biogenic emissions decrease O$_3$ concentrations [173]. Sensitivity tests in a case study show that local biogenic emissions are an important contributor to local O$_3$ production, relative to anthropogenic hydrocarbons (AHCs) [142]. Isoprene forms a main component of biogenic emissions, and affects the chemistry of the troposphere because its oxidation products are precursors for the photochemical production of ozone [11]. It is one of the most abundant phytogenic chemical species found in the ambient air [73], and accounts for a substantial portion of the atmospheric hydrocarbon load [185,88].

However, biogenic emission estimates are laden with large uncertainties [173]. In addition, there are significant uncertainties associated with the chemical mechanisms employed in photochemical modeling, such as uncertainty in the chemical reaction rates and uncertainties arising from the lumping of chemical species [7]. The focus of the present work is on the comparative evaluation of the effects of uncertainty in the biogenic emission estimates and uncertainties in the reaction rate coefficients on the estimated ozone levels, with respect to isoprene (2-methyl-1,3-butadiene), since it is one of the major components of biogenic emissions. More specifically, the aim is to compare the effects of uncertainties associated with the emission rates of isoprene with the effects of uncertainties in the rates of reactions involving isoprene. Such uncertainty analysis could provide information that could be useful in identifying the area to focus resources: studying the reaction mechanisms in more detail versus improving the inventorying methods.


 
Figure 5.15: Origins and types of uncertainty present in Photochemical Air Quality Simulation Models
\begin{figure} \centerline{\epsffile{unc-types-biog.eps} } \end{figure}

 

5.3.1 Uncertainties Associated with Biogenic Emission Estimates

There are significant uncertainties associated with biogenic emission estimates, and specifically with isoprene emission estimates. These uncertainties occur mainly due to the following factors:
Currently employed models and associated inventories for estimating biogenic emissions typically cannot treat the above variability in adequate detail. Hence significant uncertainties occur in the emission estimates.

The total U.S. emission estimates of biogenics are reported to range from 22 to 50 Tg/yr depending upon the formulation of different emission rate factors [135]. Some studies report the uncertainty in the biogenic emission estimates of an inventory for the United States to be of the order of 300% [136,173]. In another study, the uncertainties in the emission estimates of isoprene in Europe are reported to be of the order of 500% [188].

In addition, there are significant uncertainties associated with the description of isoprene chemistry. Recently reported studies show that different photochemical mechanisms with different treatments of isoprene chemistry give significantly different estimates of ozone levels [225]. The main sources of uncertainties in existing chemical mechanisms are: (a) lumping of various biogenic compounds, and (b) insufficient data to estimate reaction rate parameters associated with biogenic compounds. It is reported in the literature that some reaction rate parameters associated with isoprene chemistry are uncertain up to a range of 150% [7].

5.3.2 Uncertainty Analysis

Considering the importance of biogenic emissions, and the uncertainties involved in their estimates and reaction rates, it is important to study how these uncertainties propagate through the model and affect the predicted O$_3$ concentrations. For simple models, uncertainty propagation can be accomplished through conventional methods such as Monte Carlo methods. However, grid-based photochemical models are typically very demanding computationally, thus making conventional Monte Carlo methods prohibitively expensive.

Due to the lack of computationally efficient uncertainty analysis methods, a number of studies reported in the literature with respect to uncertainty in photochemical models involved running the model for selected sets of parameter or input values, and thus obtaining some estimates on the ``bounds'' of uncertainty in model results. The main advantage of this approach is that it is simple and requires few model runs. However, this approach (``sensitivity/uncertainty'' testing) does not provide insight into the distribution of the model results. Here, the regression based SRSM is applied to estimate the probability distributions of the estimated ozone concentrations.

As mentioned earlier, one goal of this approach is to characterize individual contributions of uncertainties in chemistry and emissions on the ambient ozone levels, and to identify important contributors to uncertainty, so that resources could be focused to reduce uncertainties in those factors. Another goal is to obtain an approximate estimate of the range of uncertainty in the predicted ozone concentration, resulting from the above mentioned input uncertainties; such an estimate could potentially assist in answering some regulatory questions.

5.3.3 Estimation of Uncertainties

The present work involves the study of the effects of uncertainty in the emission and reaction rates of isoprene, on the ozone levels in the atmosphere. This is pursued as follows:

More specifically, a photochemical box model is used to perform preliminary screening based on the effects of uncertainty in isoprene reaction rate constants and initial concentrations of isoprene on the time-dependent concentrations of ozone. The results from this exercise are utilized in identifying key factors contributing to uncertainty in the application of a grid-based PAQSM, specifically the UAM-IV.

A box-model consists of a single well mixed reactor cell, and the evolution of chemical species depends only on time dependent photochemical reaction rates. In the present work, a single cell option of the Reactive Plume Model (RPM-IV) is used for the screening runs. A single cell trajectory model, is equivalent to a box model, since there is no transfer of material occurring from outside the cell. The RPM-IV was used here since the same chemical mechanism, the Carbon-Bond IV (CB-IV) mechanism [216], is employed to describe the chemistry in both RPM-IV and UAM-IV. The CB-IV mechanism (see Appendix C) consists of a set of 95 chemical reactions among 35 surrogate chemical species corresponding to organic bonds and functional groups.


  
Figure 5.16: The Philadelphia/New Jersey modeling domain used for uncertainty analysis
\begin{figure} \centerline{\epsfig{figure=nj-dom2.eps,width=\textwidth}} \par\end{figure}

The Urban Airshed Model, version IV, [154] is a photochemical grid model that has been applied to several urban areas in the United States, Europe and the Far East [78,63]. This model solves the atmospheric diffusion equation (ADE) [181] on a three dimensional grid covering the airshed of interest. Atmospheric chemistry is described by the CB-IV mechanism. This model requires meteorological inputs and the emission information for both anthropogenic and biogenic emissions. Further details of the UAM-IV model are presented in Appendix C.

This case study considers the modeling of the Philadelphia-New Jersey region for July 19 and 20, 1991, when a severe ozone episode occurred in this region. The domain and the modeling grid (at a horizontal resolution of 5 km$\times$5 km) for the case study are shown in Figure 5.16.

The meteorological inputs were developed using the Diagnostic Wind Model (DWM) [57], the anthropogenic emissions were obtained from State-provided inventories, processed using the Emission Processing System (EPS) [31]. The biogenic emissions were obtained from the EPA's Biogenic Emissions Inventory System (BEIS) model [31]. These inputs were obtained from the input data from Georgopoulos et al. [78]

The procedure used for uncertainty analysis consisted of the following steps:


  
Figure 5.17: Probability distribution of the daily maximum ozone concentration
\begin{figure} \centerline{\epsfig{figure=uam.max.eps,width=\textwidth}} \par\end{figure}

5.3.4 Results of the Case Study

In the first phase (screening phase), nine parameters were considered, and the objective was to identify parameters that have a significant effect on the O$_3$ concentration predictions. The box model was used to to calculate the O$_3$ concentration. The nine parameters included initial concentrations of isoprene and monoterpenes, and the rates of seven reactions of isoprene with the O$_3$, NO$_3$, O and OH radicals. Of these, only three factors, namely the initial concentration of isoprene, and the reaction rates of the reactions of isoprene with NO$_3$ and with the OH radical were found to have an appreciable effect on the calculated O$_3$ concentrations.

In the second phase, representative estimates of uncertainties in these parameters were used to study their effects on urban scale O$_3$ levels. Truncated lognormal probability distributions were used to represent uncertainties in these parameters. The value of a given parameter, $x$, could lie between $x_m/f$ and $x_mf$, where $x_m$ is the reported value and $f$ is an uncertainty factor that is estimated a priori. This means that $\log(x)$ is normally distributed with a mean $\log(x_m)$, and a variance $\log(f)/2.5$. Here, 2.5 is used as a truncation factor to specify the probability distribution uniquely, and to limit the range of values $x$could have.


  
Figure 5.18: Probability distribution of the daily average ozone concentration
\begin{figure} \centerline{\epsfig{figure=uam.ave.eps,width=\textwidth}} \par\end{figure}


  
Figure 5.19: Probability distribution of the daily maximum 8 hr running average ozone concentration
\begin{figure} \centerline{\epsfig{figure=uam.8hr.eps,width=\textwidth}} \par\end{figure}


  
Figure 5.20: Probability distribution of the pervasiveness of the ozone episode
\begin{figure} \centerline{\epsfig{figure=uam.per.eps,width=\textwidth}} \par\end{figure}

More specifically, the uncertainties in isoprene emission estimates, and the rate constants for the reactions of isoprene with NO$_3$ and with the OH radical, were all assumed to be represented by a truncated lognormal distributions with uncertainty factors of 1.5; this is a rough estimate of the uncertainties based on the articles by Roselle [173] and by Atkinson [7].

The application of SRSMs, for this case study involving 3 uncertain parameters, required 10 simulations for a second order approximation method, and 20 simulations for a third order approximation. The approximation converged for the third order - there was an almost negligible difference between the estimates of the second and third order approximations. The third order approximations were used to estimate the uncertainties in the output metrics considered.

The effect of uncertainty in these parameters, on four representative metrics of photochemical air quality O$_3$ levels through the study domain was studied. The four metrics were: (a) the daily maximum of hourly average O$_3$ concentration, (b) the daily average O$_3$ concentration, (c) the daily maximum eight hour running average (an important metric in relation to the new ozone regulatory standard), and (d) the ``pervasiveness'', an alternative metric, defined as the number of grid cell hours that have ozone concentrations in excess of the hourly maximum O$_3$ standard of 120ppb [81,78].

Figure 5.17 shows the probability distributions describing the uncertainty in the daily maximum of hourly average concentration for the episode days of July 19 and 20, 1991, for the Philadelphia-New Jersey domain. Figures 5.18 and 5.19 describe the uncertainty in the daily average concentration, and the daily maximum eight hour running average concentration, for the same case, whereas Figure 5.20 shows the uncertainty in the pervasiveness.

In summary, the following conclusions were reached:


  
Table 5.2: Contribution of uncertainties in the individual parameters to the overall uncertainty in the UAM-IV, as estimated by the SRSM
  Jul 19 Jul 20
  Mean OH NO$_3$ ISOP Mean OH NO$_3$ ISOP
Max (ppm) 0.143 0.0003 0.0000 0.0009 0.178 0.0003 -0.0001 0.0003
Ave (ppm) 0.062 0.0000 0.0000 0.0002 0.072 -0.0001 0.0000 0.0001
8hr-Ave (ppm) 0.121 0.0000 -0.0001 0.0005 0.149 -0.0003 0.0000 0.0001
Pervasiveness 341.71 6.32 -1.22 47.53 4075.0 31.30 -9.17 157.14

Table 5.2 summarizes the uncertainty in the uncertainty estimates resulting from this analysis. The coefficients in the table indicate the uncertainty contributed to a metric, due to the uncertainty in the individual parameter. The coefficients indicate that the uncertainty contribution is very small relative to the mean value of the metric. Further, they indicate that the uncertainty in the isoprene emission rate contributes significantly to the overall uncertainty, and that the uncertainty in the reaction rate of the NO$_3$-isoprene reaction has an almost negligible effect. However, the uncertainty in the OH-isoprene reaction constant contributes to an extent comparable to the isoprene emission uncertainty. So, the above analysis indicates that the isoprene emission rates and the reaction rate constant of the isoprene-OH reaction should be studied in more detail.

This case study also demonstrates how the SRSM, while requiring a small number of simulations, can be used to estimate the uncertainties in the outputs of a complex model such as UAM-IV. The execution time, about five days (20 model runs $\times$ 6 hours per model run) is much less when compared with a Monte Carlo analysis involving, for example, 500 runs, requiring 125 days of computer time.

next up previous contents
Next: 5.4 Case Study IV: Up: 5. CASE STUDIES FOR Previous: 5.2 Case Study II:
Sastry S. Isukapalli
1999-01-19