Simulation of Summertime Ozone over North America

The concentrations of 03 and its precursors over North America are simulated for three summer months with a three-dimensional, continental-scale photochemical model using meteorological input from the Goddard Institute for Space Studies (GISS) general circulation model (GCM). The model has 4øx5 ø grid resolution and represents non linear chemistry in urban and industrial plumes with a subgrid nested scheme. Simulated median afternoon 03 concentrations at rural U.S. sites are within 5 ppb of observations in most cases, except in the south central United States where concentrations are overpredicted by 15-20 ppb. The model captures successfully the development of regional high-O3 episodes over the northeastem United States on the back side of weak, warm, stagnant anticyclones. Simulated concentrations of CO and nonmethane hydrocarbons are generally in good agreement with observations, concentrations of NOx are underpredicted by 10-30%, and concen- trations of peroxyacylnitrates (PANs) are overpredicted by a factor of 2 to 3. The overprediction of PANs is attributed to flaws in the photochemical mechanism, including excessive production from oxidation of isoprene, and may also reflect an underestimate of PANs deposition. Subgrid nonlinear chemistry as captured by the nested plumes scheme decreases the net 03 production computed in the United States boundary layer by 8% on average.

Several three-dimensional regional models for 03 over the United States have been developed in the past decade [Liu et al., 1984;Chang et al., 1987;Carmichael et al., 1991;McKeen et al., 1991;Roselle et al., 1991]. These models all use fine grid resolution (20-100 km). They are designed to simulate high-O3 episodes occurring over spatial scales of a few hundred km and temporal scales of a few days. We needed a model that could be applied conveniently over continental and seasonal scales. The (Table 1). The archive includes 4-hour averages of dynamical variables (winds, mixing depths, convective mass fluxes) and 5day averages of other meteorological variables (temperature, humidity, cloud reflectivities). Table 2 Prather, 19901. c The frequencies of dry, shallow wet, and deep wet convective events are archived separately. The dry convection frequencies do not include mixed layer convection, which is represented separately as the mixing depth. Shallow wet convection is defined as extending up to at most layer 3. The vertical distribution of convective events between pairs of layers within a column is computed at each 4-hour model time step by scaling the 5-day mean vertical distribution to the 4-hour totals of convective events in the column. The convective mass fluxes are computed from the frequencies of convective events as described by Praiher et al. [ 1987].

t, Vertical extent of dry convective instability initiated by surface heating; archived as a fractional number of GCM layers [Jacob and
a Cloud reflectivities at 800, 500, and 200 mbar, reconstructed to approximate the GCM vertical distribution of cloud optical depth [Spivakovsky et al., 1990a]. c A sinusoidal diel cycle of temperature with amplitude 5 K and maximum at 15 LT (local time) is applied to the lowest model layer over land. Isoprene emission and surface resistances to deposition are computed using surface air temperatures, which are derived from the temperatures in the lowest model layer by assuming a dry adiabat.
The reactivity of HC1 is defined as that of isoprene in the continental boundary layer in daytime and as that of a 70/30 propene/isoprene mix under other conditions. The reactivity of HC2 is defined as that of a mix 13% propane, 32% C•,5 alkanes, 20% C6-8 alkanes, 2% C4,5 alkylnitrates, 6% C6-8 alkylnitrates, 8% benzene, 12% toluene, and 7% ethylene [Jacob et al., 1989]. The partitioning of HC1 and HC2 is in units of atoms C. polynomials of 11 independent variables defining the chemical environment (Table 3). The parameterization procedure is described by Spivakovsky et al. [1990b], and details are given in Appendix A. The polynomials in Table 3 include 4-hour average (P-L) for Ox, NOx, and PANs and 4-hour average concentrations of OH and NO3 from which (P-L) for CO, HC1, and HC2 are computed. Yields of CO from the oxidation of HC1, HC2, and methane (1.7 ppm) are taken as 60%, 30%, and 90%, respectively [Lurmann et al., 1986]. The radiation field in the chemical parameterizations is defined by two photolysis rate constants, JNO 2 and JodD), which are computed in turn as polynomials of seven independent variables parameterizing results from the radiative transfer code (Table 3). All independent variables in Table 3 are either computed in the model or specified by the GCM archive, except for the 03 columns which are climatological means dependent on latitude and month [Spivakovsky et al., 1990a].

Emissions
The model includes anthropogenic emissions of NOx, CO, and non-methane hydrocarbons (NMHCs), and biogenic emission of isoprene, all released in the lowest model layer (Figure 3)    temperatures. Isoprene is the principal contributor to HC1 emission everywhere except in cities.

Deposition
Deposition velocities of 03, NO,,, and PANs are computed at the midpoint of the lowest layer (= 250 m above ground) using the resistance-in-series model of Wesely and Hicks [1977] with surface resistances and roughness heights from Wesely [1989], and surface-type information from Matthews [1983]. Friction velocities and Monin-Obukhov lengths are estimated with a Pasquill-Gifford parameterization dependent on wind speed, solar radiation flux, cloud cover, and roughness height [Seinfeld, 1986]. Under nighttime conditions over land the Monin-Obukhov length is generally less than 250 m, i.e., the atmosphere near the surface is strongly stratified, and we assume that deposition is negligible. Over water the atmosphere is assumed neutrally buoyant at all times, and the friction velocity is calculated from the local wind speed and roughness height.  Figure 4).     Table 4a).

Forest ([03] < 40 ppb) is associated with either rain or polar air. Rain suppresses mixing of the surface layer, an effect that is not resolved by the model, and intrusions of clean polar air are not well resolved either because the boundary conditions at 56øN represent mean observations from subpolar ozonesonde stations.
A more relevant test of the model variance is the ability to simulate regional high-O3 episodes over the eastern United States.

Logan [1989] presented 'seasonal episode statistics derived from O3 concentration data at the nine SURE sites and Whiteface
Mountain (sites 1-9 and A). She defined an episode as the occurrence of mean 10-16 LT O3 concentrations in excess of 70 ppb at four or more of these 10 sites for two or more consecutive days. She found five episodes in June to August of each of 1978 and 1979, with each episode lasting from 2 to 5 days, for a total of 15 episode days each summer. The same analysis in the model summer indicate four episodes, each lasting from 3 to 9 days, for a total of 21 episode days. The frequency of regional 03 episodes in the model is consistent with observations, although the model episodes tend to be longer than observed.
It is well known that regional 03 episodes over the eastern United States are associated with weak, stagnant, warm anticyclones [Decker et al., 1976;Altshuller, 1978;Wolff and Lioy, 1980]. This association is well reproduced by the model, as illustrated in Figure 11  Not all anticyclones passing over the northeastern U.S. generate regional 03 episodes, either in the model or in the observations. Important variables determining the occurrence of a regional 03 episode are the persistence, pressure, and temperature of the anticyclone [Logan, 1989]; pressure is important because weak anticyclones are associated with regional stagnation. We derived quantitative criteria for predicting regional 03 episodes in the observations by using 1978-1979 data for anticyclone tracks over the eastern U.S. [National Oceanic and Atmospheric Administration (NOAA), 1978, 1979] and concurrent statistics of 03 episodes from Logan [1989]. Results in Figure 12 show that regional 03 episodes are most likely to occur when anticyclones stagnate over the eastern U.S. or immediately offshore for N > 3 days, with anticyclone MSL pressure P < 1025 mbar and afternoon surface air temperature T > 298 K. The same criteria diagnose successfully the development of regional 03 episodes in the model (Figure 12), implying that the model generates episodes  (Table 4a).

Model results (thin lines) are compared to observations (thick lines). The abscissa is such that a normal distribution would plot as a straight line (probability scale).
with either P or T. We argue in the companion paper that stagnation rather than temperature is the primary driving force for the development of regional O3 episodes. [1986] mechanism. First is excessive production of higher peroxyacyhnitrates; calculations using the mechanism indicate typically a 30-40% contribution of higher peroxyacylnitrates to rural PANs [Sillman, 1988], but the observations suggest a lower contribution. It is assumed in the mechanism that all peroxyacylnitrates have the same formation and destruction rates as PAN, however the lifetimes of the higher peroxyacylnitrates could be short due to rapid thermal decomposition or oxidation by OH . Model results for HC1 (mainly isoprene) are not comparable to surface measurements because of the steep gradient between the surface layer and the mixed layer   Fig. 12. Criteria for the occurrence of regional 03 episodes in association with anticyclone passages over the eastem U.S. in June to August. All anticyclones persisting for N > 2 days over the eastern United States or immediately offshore (between 32øN and 44øN and between 90øW and 65øW) are diagnosed for the occurrence of a regional 03 episode based on the criteria of Logan [1989]. The diagnosis is then related to three anticyclone variables: (1) the number N of days of persistence of the anticyclone over the eastem U.S., (2) the mean anticyclone pressure P during the N days, and (3) the mean 14 LT surface temperature T during the N days at the four SURE sites experiencing the highest 03 concentrations. Model results for the GCM summer are compared to two summers of observations (1978)(1979). Anticyclone data for 1978-1979 are from NOAA [1978,1979], and corresponding data for regional 03

A likely reason for the underprediction of isoprene at the ROSE site is that local emission is underestimated.
Guenther et al.
[1993] report a.light-saturated isoprene emission of 6.7x10 •2 molecules cm -2 s -• at 303 K for the site, i.e., twice the value in the model, reflecting the high abundance of oak and sweetgum. Considering that O3 production is primarily NO,, limited, a factor of 2 uncertainty in isoprene emission has little effect on model results. We conducted a sensitivity simulation with isoprene emissions doubled uniformly from the standard model and found less than a 4 ppb increase in mean O3 concentrations anywhere. An additional explanation for the underprediction of isoprene at the ROSE site may be that OH concentrations Here F is the areal fraction of leaf biomass that emits isoprene (assumed independent of altitude), L is the total leaf area index of the canopy, and co is a light correction factor: L t0= Z g(hdL' (B5) where I is attenuated by the leaf area index overhead L'. We view the canopy as a grey absorber with light extinction coefficient k = 0.5 normalized to the leaf area index [Verstraete, 1987]'

I =/øe-SZ'/•ø (B6)
where P is the solar radiation flux at canopy top (specified by the GCM) and 0 is the solar zenith angle. Figure B1