Passive tracers are employed in a relatively high spatial and temporal resolution three-dimensional transport model to analyze a stratosphere-troposphere exchange (STE) event over the eastern United States and western North Atlantic Ocean. The model is validated against measurements taken on board the National Oceanic and Atmospheric Administration WP-3D Orion aircraft during the North Atlantic Regional Experiment study in the spring of 1996. Overall, the model reproduces the measurements well during the early part of the flight where there is indication of a small stratospheric intrusion. However, the very strong signatures of STE and mixing contained in the measurements later in the flight are not captured. Use of a finer horizontal resolution (20 km as opposed to 60 km) brings the model results closer to the aircraft measurements and yields higher values (50% at 7–8 km altitude) of ozone, O3, with a deeper penetration into the troposphere (20% at 80–120 ppbv levels).
 Many studies have been undertaken to investigate the origin of ozone (O3) over the North Atlantic (NA) [e.g., Cooper et al., 1998; Parrish et al., 1993; Prados et al., 1999]. In particular, it has been argued that both photochemical and stratospheric sources are associated with the enhanced postfrontal O3 at Bermuda [Oltmans and Levy, 1992; Moody et al., 1995]. The importance of O3 formation from anthropogenic pollution has been highlighted through correlations of measurements of O3 and carbon monoxide (CO) from surface sites in the NA region [Parrish et al., 1998]. However, the interpretation of O3-CO correlations in the free troposphere can be complicated by the mixing of stratospheric air, characterized by high O3 and low CO, with boundary layer air, characterized by high CO and low O3 [Parrish et al., 2000].
 On 20 March 1996 the National Oceanic and Atmospheric Administration (NOAA) WP-3D Orion aircraft flew through a cutoff low located over a prominent anthropogenic pollutant source region in the western NA area during the 1996 North Atlantic Regional Experiment (NARE) [Fehsenfeld et al., 1996]. The measurements obtained indicate the strongest downward transport of stratospheric O3 and upward transport of anthropogenic CO of the entire NARE period, as well as mixing of the two air masses.
 Past research has demonstrated that chemical transport models (CTMs) are a useful means of examining the variation of pollutant transport, in both a spatial and temporal sense, that measurements alone cannot provide [e.g., Lin et al., 1998]. However, CTM modeling of strong, subsynoptic-scale stratospheric intrusions with nearly coincident anthropogenic pollutant uplifting [e.g., Fischer et al., 2002; Esler et al., 2003; Cooper et al., 2004] has not been so widely documented. In recent transport modeling studies, Kowol-Santen et al.  considered only upward transport of CO from the planetary boundary layer (PBL) which did not include any particularly high peaks. Beuermann et al.  modeled O3 measurements for a strong, subsynoptic-scale intrusion only qualitatively through reconstruction of potential vorticity (PV) fields. Stratospheric intrusions from Kentarchos et al. [2000a, 2000b] were much less strong than those encountered by the WP-3D aircraft, and only surface observations were available for model comparisons. The limitations encountered in simulating maximum O3 peaks with the coupled chemistry global circulation model (European Center/Hamburg 4) used by Kentarchos et al. [2000a, 2000b] were ascribed to insufficiently fine vertical resolution near the tropopause. In the current modeling study a finer vertical resolution of ∼1 km is employed at the tropopause level.
 The purpose of this paper is to examine the ability of the NOAA Aeronomy Laboratory three-dimensional transport model to simulate the strong signature of stratosphere-troposphere exchange (STE) in the aircraft measurements of 20 March 1996. Simulations of the O3 and CO transport on 20 March 1996 using the three-dimensional model are presented for two different horizontal grid resolutions (60 and 20 km), and the results are compared with the corresponding aircraft measurements. The relatively high spatial and temporal resolutions of the models are a prerequisite to simulating the tracer features observed over length scales <120 km often observed during STE events [Holton et al., 1995; Olsen et al., 2002].
2. Synoptic Situation
 Upper level pressure anomalies and intense surface fronts dominated the synoptic situation over the extratropical western NA region during the second half of March 1996 [Kentarchos et al., 2000a]. The development of pressure anomalies over the United States between 17 and 20 March are shown sequentially in the 500 hPa height fields within Figure 1. A broad trough encompasses the entire United States on 17 March as a secondary upper level trough moves southward from northern Manitoba. This trough continues southward and expands over the midwestern United States until 0000 UTC, 20 March, when the 5400 m contour becomes closed over the Ohio River Basin. After this time the newly formed cutoff low moves to the northeast along the U.S. coastline, as illustrated by the 500 hPa height contour overlays on the 2115 UTC, 20 March water vapor image (Figure 2). In Figure 2, warmer colors represent moist air in the middle and upper troposphere, and cooler colors represent dry air, thereby indicating the spiraling of dry polar air with moist subtropical air in the middle to upper troposphere surrounding the cutoff low on 20 March. The low deepens slightly during 20 March until it leaves the continental United States at 0000 UTC on 21 March.
 On 20 March 1996 the WP-3D aircraft flew from Tampa, Florida, at 1900 UTC to Providence, Rhode Island, at 2400 UTC (Figure 2), with an average speed of ∼100 m/s. While many species were measured during the flight, only O3 and CO are focused on in this study. The method for measuring O3 was nitric oxide (NO) chemiluminescence with an accuracy of 3% [Ryerson et al., 1998], while that for CO was vacuum ultraviolet fluorescence with an accuracy of 5% [Holloway et al., 2000], the temporal resolution of both being 1 s.
4. Model Description
 The three-dimensional model employed in the current study consists of two independently run components, namely, meteorology [Grell et al., 1994] and then photochemical transport [McKeen et al., 1991]. For the meteorology component the Pennsylvania State University/National Center for Atmospheric Research (NCAR) mesoscale modeling system (MM5) is used. MM5 is a three-dimensional, compressible, primitive-equation model which uses the terrain-following σ coordinate, where σ = (p − pt)/(ps − pt) for pressure p, model-top pressure pt, and surface pressure ps. In this case the model is run in nonhydrostatic mode, with Grell et al.  cumulus parameterization, high-resolution Blackadar  PBL, and Dudhia  simple ice schemes.
 Four-dimensional data assimilation is also employed within MM5, where nudging is used in conjunction with observations to ensure that the model results stay close to the true conditions [Stauffer et al., 1991]. The observations included are the National Centers for Environmental Prediction (NCEP) global analysis ds083.0 data set (on mandatory pressure levels up to 10 hPa and every 12 hours) and NCEP global upper air observations (archived at NCAR) which increase the vertical resolution. The generated three-dimensional fields of horizontal winds, temperature, and water vapor mixing ratios are used for initialization, assimilation, and boundary conditions.
 Winds and state variables from MM5 stored at half-hourly intervals are used to drive the photochemical transport component. The photochemical transport component has the capacity to simulate the photochemical and dynamical changes of 34 atmospheric species. However, for this case study, simulations that include full photochemistry are found to be almost indistinguishable from those in which O3 and CO are treated as passive tracers, owing to suppressed photochemistry in spring and latitudes being higher than 45°. A parameterized vertical transport scheme for convective clouds is used [Lin et al., 1994], but for the current study, simulations with and without this scheme give virtually identical results. Boundary and initial conditions for O3 and CO are based on Pacific Exploratory Mission (PEM)-West B observations, with background values of 30 and 140 ppbv, respectively. The stratospheric influence on O3 is parameterized by enforcing the condition of proportionality to PV at the top boundary of the model domain and above 5 km on the lateral boundaries [Ebel et al., 1991], with the proportionality constant of 50 ppbv per PV unit [Browell et al., 1987]. It should be noted that higher values of 60–70 ppbv per PV unit have been reported elsewhere in the literature for the PV proportionality constant in spring [Stohl et al., 2000]. For CO the stratospheric influence is parameterized by fitting a curve through observed ratios of CO to O3 [Murphy et al., 1993]. Further, CO emissions are calculated from a 1996 emissions inventory [U.S. Environmental Protection Agency, 1998].
 Simulations were performed between 15 and 23 March 1996 for a horizontal resolution of 60 km (80 × 80 grid cells) and then between 17 and 20 March 1996 for a horizontal resolution of 20 km (150 × 150 grid cells) using one-way nesting of the dynamics from the larger domain at the boundaries. The 20 km nested domain is shown in Figure 3a. Top and side boundary conditions for CO and O3 for the 20 km resolution case are taken from the half-hourly output of the 60 km resolution case. Model results for the 20 km resolution case with boundary conditions determined by PV, as in the 60 km resolution case, are found to be very similar to the results reported here. Horizontally, the model domain is centered at 40°N and 85°W (Figure 3). The vertical extent is from the surface to 50 hPa, divided into 23 unevenly spaced σ levels, with a concentration of layers in the PBL and lower troposphere. Nominal altitudes of the vertical grid interfaces above 6.5 km are 7.00, 7.86, 8.82, 9.91, 11.18, 12.96, 14.63, and 17.38 km.
5. Results and Discussion
5.1. Model Simulations and Differences due to Horizontal Resolution
 The model-simulated horizontal distributions of O3 and CO at 7.4 km on 20 March 1996 at 2230 UTC are shown in Figure 3 for the 60 km resolution case. At this height level the distributions of O3 and CO are anticorrelated, with maximum O3 values of ∼180 ppbv occurring over the eastern tip of West Virginia and corresponding CO values between 68 and 77 ppbv. These distributions reflect the spiral air mass depicted in Figure 2, and the elevated O3 with depressed CO levels within this spiral also indicates its stratospheric origin [Danielsen et al., 1987]. Height versus latitude cross sections (not shown) through the easterly regions of the spiral, in the vicinity of the most northerly third of the WP-3D flight path, additionally indicate downward transport of stratospheric O3, as well as an uplifting of anthropogenic CO.
 Similar horizontal distributions and height versus latitude cross sections were plotted for the 20 km resolution case (also not shown). It was found that the plots are qualitatively similar to the 60 km resolution case but display clear evidence of more O3 and less CO. For the horizontal distributions at 7.4 km with 20 km resolution the maximum value of O3 is 50% greater, and the minimum value of CO is 25% smaller, than for the 60 km resolution case. That is, the “downward moving” O3 effectively reaches lower altitudes with the 20 km resolution case (by a factor of 20% for the contour range 80–120 ppbv). The “upward moving” CO values represented by the 113–150 ppbv contours also reach 25% higher with the 20 km resolution case. Both the higher O3 values and their deeper penetration into the troposphere with the finer resolution are consistent with the findings of Kentarchos et al. [2000b]. Also as found by Kentarchos et al. [2000b], the finer resolution brings an increase in vertical wind shear, from which the deeper O3 penetration follows.
Figures 4a and 4b illustrate the differences in vertical wind shear between the 20 and 60 km resolution cases for 1200 UTC on 20 March 1996. Averaged over the interior part of the 20 km domain, the mean vertical velocities for both resolutions are very small (<10% of the average upward or downward vertical velocities), and vertical transport is dominated by broad but compensating areas of strong upward and downward motions at all altitudes. Figure 4a shows that the average winds in both the upward and downward directions increase with the 20 km resolution case between 11 and 14 km altitude and also below 3 km. The averaged standard deviations in the vertical winds are displayed in Figure 4b. Since the square of this quantity is equivalent to the turbulent kinetic energy in the vertical direction, it is clear that the finer-resolution case is contributing to a factor of 2 or more increase in this energy term for altitudes above 11 and below 7 km. Since the shortest wavelength scale resolved horizontally is 40 km for the 20 km resolution case, Figure 4b implies that the MM5 dynamics predict a significant contribution to the vertical turbulent energy associated with this intrusion event from horizontal length scales in the 40–120 km range.
 The resulting effect of model resolution on the large-scale vertical transport of O3 is illustrated in Figure 4c, where layer averages of vertical O3 flux have been integrated over 24 hours (20 March 1996) and most of the 20 km resolution domain. Since the 60 km resolution O3 concentrations are used as boundary conditions for the 20 km resolution case, the differences in O3 flux are due entirely to differences in the integrated transport patterns shown in Figures 4a and 4b, along with subsequent O3 redistribution. The maximum downward average O3 flux is nearly the same for both resolutions but occurs at 10 km for 20 km resolution and 4 km lower for 60 km resolution. This large increase in downward vertical flux for the 20 km resolution case at the 10 km level can be attributed to the increase in vertical turbulent kinetic energy and vertical wind shear in the 11–14 km region above, while the vertical profile below 7 km is significantly modified by the increased turbulent energy in the lower part of the model domain.
Figure 4d shows the resulting average O3 concentrations at 0000 UTC on 21 March 1996. The influence of vertical transport differences on layer average O3 is damped to a large degree by the effects of boundary inflow/outflow from the limited domain. Nonetheless, the model layers between 7 and 11 km show a 5–17% increase of layer average O3 with the 20 km resolution case. Figures 4a–4d illustrate the point that vertical transport and tropospheric O3 budget studies based on global three-dimensional model analysis [e.g., Roelofs et al., 1997; Hauglestine et al., 1998; Li et al., 2002] should be sensitive to model resolution, with finer horizontal resolution resulting in higher stratospheric O3 input to the troposphere.
5.2. Point-by-Point Comparisons Between the Model Simulations and the Aircraft Measurements
 Comparisons of the model-simulated values of O3 and CO, for both the 60 and 20 km resolution cases, with the aircraft measurements, are illustrated in Figure 5. Figure 5 demonstrates that the model captures the measurements reasonably well for both O3 and CO and for both resolutions. For further discussion of Figure 5 it is useful to divide the plot into sections, as shown, and to consider each in turn, namely, (1) the first ascent up to around 7 km, (2) the following descent down to around 0.5 km, (3) the level portion at around 0.5 km, (4) the second ascent up to around 7 km, and (5) the final descent to the surface, with section 4 being divided further to distinguish the peak of CO (section 4B) from the double peak of O3 (section 4A).
 For the first section the model at both resolutions does a fair job of reproducing the aircraft measurements, and in particular, the model at 20 km resolution captures well the two peaks of O3 and the peak of CO contained in this region. From back trajectory calculations of the MM5 forecast winds the transport pathway for CO during this portion of the flight path involved strong cyclonic advection and upward transport of CO from the PBL. The back trajectory's low point (3.7 km) occurs over eastern Virginia 16 hours before sampling and then moves northward over New York City at 4.5 km altitude, 12 hours prior to sampling. Six hours prior to sampling, the back trajectory traverses southwest over southern Ohio, before finally reaching the southeast and east United States, where the air mass was sampled over central South Carolina at 6 km. This transport pathway is consistent with the warm conveyor belt classification of air masses occurring to the east of midlatitude cyclones [Cooper et al., 2001].
 A key feature of both the second and last lower altitude sections of Figure 5 is the large peaks in CO (257 and 266 ppbv, respectively). It can be seen that at neither resolution does the model capture either of these peaks. However, examination of the horizontal distributions of CO at the corresponding altitudes, around 4 and 2 km, reveals 200 ppbv CO maximums for the 60 km resolution case and 235–260 ppbv CO maximums for the 20 km resolution case at locations horizontally displaced from the flight path. For the 2 km observed CO peak the horizontal displacements are 635 and 240 km for the 60 and 20 km resolution cases, respectively. For the 4 km observed CO peak the horizontal displacements are 155 and 240 km for the 60 and 20 km resolution cases, respectively. Upward transport of CO from the PBL and transport in cyclonic spirals are again revealed by MM5 back trajectories as being the corresponding transport pathways. The third section of Figure 5, being at such low altitude, will not be discussed further.
 The most interesting section of Figure 5 is the fourth section, which contains the strong signature of STE. Here the measurements show a double O3 maximum, with values of 189 and 265 ppbv, and a CO peak of 179 ppbv contained in between, all occurring in the portion of the flight path over the western North Atlantic Ocean. The flight time between the two O3 peaks in this feature is 0.23 hours, which corresponds to a horizontal distance of ∼91 km. Figure 5 shows that the model at neither resolution accurately simulates the measured maximum values of O3 nor captures the peak in CO, but the 20 km resolution model does at least begin to show the structure of the high O3 measurements. Further, similarly to sections 2 and 5, the horizontal distribution of CO at the altitude under consideration, i.e., 7.4 km, indicates CO values up to within 30 ppbv of the measured CO peak at locations horizontally displaced (380 km to the northwest) from the flight path. This is illustrated for the 60 km resolution case in Figure 3, where CO reaches 149 ppbv just to the north of flight path end (Rhode Island).
 There are several possible factors contributing to the decreased model performance in section 4 of Figure 5 compared to section 1. First, Figure 2 shows that the southern portion of the flight passed through a region of small gradients in the 500 hPa pressure height contours, while the northern portion was in the region of large gradients and thus higher dynamic activity. This is corroborated by the very high wind speeds and equivalent potential temperatures observed during section 4B of the flight. The model at both resolutions therefore appears to be simulating the large-scale broader features of the STE event fairly well but fails to adequately reproduce features near and along the frontal boundaries. Second, model displacements of specific regions with large updrafts and downdrafts are probably to be expected, since boundary conditions that drive the MM5 meteorology are based on NCEP analysis from upper air observations with effective temporal and spatial resolutions of 12 hours and a couple hundred kilometers, respectively. Third, the movement of the low-pressure anomaly leading up to the STE event of 20 March 1996 indicates that the event probably originated outside of the model domain. Fourth, in contrast to these large-scale features the spatial scale of the O3-CO feature in section 4B is only ∼90 km. Dynamical forcing associated with this feature is not expected to be resolved by the 60 km resolution case and perhaps may be only marginally resolved by the 20 km resolution case. Fifth, the effects of numerical diffusion in the tracer transport algorithm and wave dispersion within the MM5 numerical algorithms cannot be ruled out as contributing sources of error under the extreme conditions of STE events. Last, the convection and PBL parameterizations in the MM5 model were developed and tuned for mesoscale models with coarser horizontal resolution (60–120 km), and there is no guarantee that they should directly scale to model grids with finer horizontal resolution.
 Correlations of O3 versus CO for the aircraft measurements and the 60 and 20 km resolution simulations within the five sections of Figure 5 are shown in Figure 6a. Figure 6a illustrates the deficiency of the model in capturing the measured maximum values of O3 at either resolution. However, the measurements and model results for both resolutions generally exhibit a negative correlation between O3 and CO: a characteristic of polluted air masses within which no significant photochemical production of O3 has arisen, i.e., common during winter/spring [Parrish et al., 2000]. In order to further analyze the stratospheric air masses and the mixing of these with anthropogenic pollution the data of Figure 6a are repeated in Figures 6b, 6c, and 6d for the measurements, the 60 km model simulations, and the 20 km model simulations, respectively, with the data from the higher-altitude sections 1, 4A, and 4B being recolored to purple, yellow, and orange, respectively.
Figures 6b, 6c, and 6d reinforce the relatively good performance of the model with 20 km resolution in capturing the high O3 and CO values in section 1. They also demonstrate that while the model does not capture the measured high values of CO in section 4B at either resolution, in section 4A the model at both resolutions reproduces the higher CO measurements quite well. Further, considering background levels of O3 and CO to be around 30 and 140 ppbv, respectively, Figures 6b and 6d indicate partial mixing of anthropogenically and stratospherically influenced air masses [Parrish et al., 2000]. In particular, partial mixing within section 4 alone is apparent in Figure 6b. This partial mixing within section 4 is not present in the model data of Figures 6c or 6d, owing to the underestimation of O3 and CO as discussed above with Figure 5.
 This paper describes a tracer study of the O3 and CO transport associated with the cutoff low that occurred on 20 March 1996, using two different horizontal resolutions. The WP-3D aircraft encountered two distinct regions with O3 signatures from stratospheric intrusions on this date: a relatively weak intrusion over the southern United States at the beginning of the flight and a relatively strong intrusion off the northeast U.S. coastline that exhibited sharp anticorrelations of O3 with convectively uplifted CO from the surface. Both model resolution cases predict the O3-CO relationship for the weaker intrusion reasonably well but underestimate the degree of stratospheric O3 forcing and associated convective transport of CO for the stronger intrusion. However, the model case with 20 km resolution begins to reproduce some smaller-scale O3 features and CO values similar to the measurements that the 60 km resolution case is unable to resolve.
 From horizontal distributions at the altitude that the highest O3 and CO values were measured, the maximum simulated values of O3 are 50% greater, and there is a 20% deeper penetration of stratospheric O3, with the finer resolution than with the coarser. Analysis of the 20 March 1996 vertical winds from MM5 shows that the 20 km resolution case has twice as much turbulent kinetic energy associated with vertical velocity variance compared with the 60 km resolution case in the 11–14 km region and for altitudes <6 km. The MM5 model results therefore attribute significant dynamical forcing from horizontal wavelength scales between 40 and 120 km. This is consistent with the prominent 90 km wavelength scale feature within the O3 and CO measurements in the stronger intrusion event.
 The inability of the model, with coincident point-by-point comparisons, to fully capture the ranges of O3 and CO for the stronger intrusion event is probably due to a number of factors. One underlying reason could be the inherent limitation of using a finite domain model, since the dynamic and chemical fields associated with such intrusions are dependent on the accuracy, smoothness, and effective resolution of the boundary conditions. Point-by-point comparisons of model results with observations close to frontal activity would be expected to be in disagreement if the location of the fronts and associated dynamics were not adequately simulated. Also, numerical artifacts related to tracer transport and the MM5 dynamics cannot be ruled out as contributors to the decreased model performance in the strong intrusion event without more detailed model testing. Despite these model caveats both model results and aircraft measurements suggest that adequate simulations of such STE events with Eulerian-based models would ideally have ∼10 km horizontal resolution and encompass much of the Northern Hemisphere.
 This work provides a useful basis for validating the model in full photochemical transport model for the current STE case study and others. There is also scope for further investigation of model performance within the framework of this case study, including a more detailed examination of the effect of grid resolution on the O3-PV relationship, use of a hemispheric domain, and consideration of the implications for the use of such a model in calculating budgets of tropospheric O3.
 The authors thank the staff of the NOAA Aircraft Operations Center and the crew of N43RF for providing the aircraft support for the measurements.