Satellite observations of carbon monoxide (CO) from the Measurements of Pollution in the Troposphere (MOPITT) instrument are combined with measurements from the Transport and Chemical Evolution Over the Pacific (TRACE-P) aircraft mission over the northwest Pacific and with a global three-dimensional chemical transport model (GEOS-CHEM) to quantify Asian pollution outflow and its trans-Pacific transport during spring 2001. Global CO column distributions in MOPITT and GEOS-CHEM are highly correlated (R2 = 0.87), with no significant model bias. The largest regional bias is over Southeast Asia, where the model is 18% too high. A 60% decrease of regional biomass burning emissions in the model (to 39 Tg yr−1) would correct the discrepancy; this result is consistent with TRACE-P observations. MOPITT and TRACE-P also give consistent constraints on the Chinese source of CO from fuel combustion (181 Tg CO yr−1). Four major events of trans-Pacific transport of Asian pollution in spring 2001 were seen by MOPITT, in situ platforms, and GEOS-CHEM. One of them was sampled by TRACE-P (26–27 February) as a succession of pollution layers over the northeast Pacific. These layers all originated from one single event of Asian outflow that split into northern and southern plumes over the central Pacific. The northern plume (sampled at 6–8 km off California) had no ozone enhancement. The southern subsiding plume (sampled at 2–4 km west of Hawaii) contained a 8–17 ppbv ozone enhancement, driven by decomposition of peroxyacetylnitrate (PAN) to nitrogen oxides (NOx). This result suggests that PAN decomposition in trans-Pacific pollution plumes subsiding over the United States could lead to significant enhancements of surface ozone.
 Quantifying the continental outflow and intercontinental transport of air pollutants is one of the major challenges of atmospheric chemistry today. Integration of satellite and aircraft observations with three-dimensional (3-D) models is a promising new approach for addressing this challenge. We present here such an integration through a combined model interpretation of aircraft observations of Asian pollution outflow and trans-Pacific transport from the Transport and Chemical Evolution Over the Pacific (TRACE-P) aircraft mission, together with satellite observations of carbon monoxide (CO) from the Measurement of Pollution in the Troposphere (MOPITT) instrument. In the process we demonstrate the value of the MOPITT observations for mapping intercontinental transport of pollution, and we present the first observationally based analysis of the production of ozone during trans-Pacific transport of Asian pollution.
 The NASA TRACE-P mission focused on the study of Asian chemical outflow over the northwest Pacific in February–April 2001 [Jacob et al., 2003]. Two aircraft intensively sampled the region from bases in Hong Kong and Japan. Transit flights from the United States sampled a major trans-Pacific pollution transport event. The MOPITT instrument [Drummond and Mand, 1996; Edwards et al., 1999] provided global coverage of CO concentrations from a polar orbit. Validation profiles for MOPITT conducted by the TRACE-P aircraft [Jacob et al., 2003] set the stage for a combined analysis of the TRACE-P and MOPITT data sets. A global chemical transport model is used here to provide the necessary transfer function between the MOPITT and aircraft observations, and to place these observations in the context of our current understanding of sources and transport from Asia.
 Carbon monoxide, a product of incomplete combustion, is present at concentrations of 50–500 ppbv in the troposphere. It is the principal sink of OH, the main tropospheric oxidant, and as such plays a key role in determining the oxidizing power of the atmosphere. It has a mean atmospheric lifetime of 2 months against oxidation by OH, its main atmospheric sink, and is therefore also a sensitive tracer for continental outflow and intercontinental transport of pollution [Staudt et al., 2001; Liu et al., 2003].
 The TRACE-P mission involved two NASA aircraft (DC-8 and P3-B) deployed from bases in Hong Kong and Japan to sample outflow from Asia. Transit flights from the United States sampled a major event of trans-Pacific transport of Asian pollution on 26 and 27 February, described in section 4. The aircraft payload included instrumentation to measure O3, CO, reactive nitrogen oxide (NOy) species, and a suite of hydrocarbon and halocarbon tracers [Jacob et al., 2003]. Observations of CO were made continuously with 1 Hz frequency using the Differential Absorption of CO Measurement (DACOM) instrument [Sachse et al., 1987]. The measurement accuracy is approximately 2%.
Fuelberg et al.  give an overview of the meteorological conditions during TRACE-P. Asian pollution was primarily exported to the Pacific in WCBs ahead of cold fronts [Liu et al., 2003]. This pollution included both anthropogenic (defined herein as fossil fuel and biofuel sources only) and biomass burning influences, the latter due to seasonal biomass burning in Southeast Asia [Heald et al., 2003]. The two influences were often mixed in the outflow [Carmichael et al., 2003; Ma et al., 2003]. Following standard practice we distinguish “anthropogenic” from “biomass burning” sources, even though biomass burning here is largely due to human activity. The distinction is useful because biomass burning has a separate geographical and seasonal signature, and because the inventories are constructed from very different types of bottom-up information.
 Detailed inventories for anthropogenic emissions [Streets et al., 2003] and daily biomass burning emissions [Heald et al., 2003] were constructed for the TRACE-P period from bottom-up economic and fire information, respectively. These inventories were used by Palmer et al.  in an inverse model analysis to determine the constraints from the TRACE-P CO observations toward better quantifying CO sources from east Asia. They concluded that anthropogenic emissions from China in the Streets et al.  inventory were underestimated by 35%, and that biomass burning emissions from Southeast Asia in the Heald et al.  inventory were a factor of three too high.
 Other TRACE-P studies examined the Asian outflow of ozone and of NOy. Large NOy enhancements were observed downwind from Asia and consisted mainly of HNO3 and PAN [Miyazaki et al., 2003; Talbot et al., 2003]. PAN was the dominant form of NOy in outflow associated with WCBs [Miyazaki et al., 2003]. Examination of NOy-CO correlations by Koike et al.  indicated an NOy export efficiency to the free troposphere from northeastern China of 15%, with only 0.5% of the emitted NOx remaining as NOx in the free tropospheric outflow. Significant ozone production in Asian outflow during TRACE-P appeared to be limited to biomass burning plumes originating from low latitudes [Tang et al., 2003]. Ozone production in anthropogenic plumes was generally insignificant, reflecting the weak photochemical activity at that time of year [Pierce et al., 2003].
2.2. MOPITT Observations
 The MOPITT instrument is a nadir IR correlation radiometer launched aboard the NASA Earth Observing System Terra spacecraft in 1999 [Drummond, 1992; Edwards et al., 1999]. It has solar backscatter channels to measure total CO columns, and IR emission channels to measure CO vertical profiles. Correlation spectroscopy provides two signals for each channel. So far only the IR emission channels have been retrieved. The satellite is deployed in a polar Sun-synchronous orbit with a 10:30 local equator crossover time. Horizontal resolution is 22 km × 22 km, and cross-track scanning achieves approximate global coverage in 3 days. CO concentrations are reported on 7 vertical levels (surface, 850, 700, 500, 350, 250, and 150 hPa), and as a total column, for all cloud-free scenes.
 The MOPITT observations reported here are version 3 data (http://www.eos.ucar.edu/mopitt), which have a 10% precision on both column and mixing ratio [Pan et al., 1998; Edwards et al., 2003]. Current CO retrievals are based on only 4 signals which limits the vertical resolution, as described below [Deeter et al., 2003]. The retrieval is ill conditioned without the inclusion of external information in the form of an a priori profile xa and its error covariance matrix Sa. The vertical profile retrieved by the instrument is related to the true vertical profile x by the averaging kernel matrix A, as described by Rodgers :
where ε is the measurement error and G is the gain matrix. The MOPITT retrieval uses a single profile of xa (and Sa) with mixing ratio decreasing from 120 ppbv at the surface to 75 ppbv at 250 hPa and less than 60 ppbv at 150 hPa [Deeter et al., 2003]. The averaging kernel matrix is given by
where I is the identity matrix and is the retrieval error covariance matrix, different for each retrieval and supplied with the MOPITT retrieval product. The rows of A (averaging kernels) describe the vertical resolution of the retrieval. A sample set of averaging kernels for the North Pacific is shown in Figure 1. The column averaging kernel (right panel of Figure 1) specifies the total column sensitivity to a change in the true mixing ratio at each level and is calculated as follows:
where t is the transfer operator which includes both the conversion from volume mixing ratio to number density and the integration over pressure. Figure 1 shows roughly two independent regions of information in the MOPITT retrieved profile: one in the middle troposphere (300–800 hPa) and one in the upper troposphere (pressure < 500 hPa) [Deeter et al., 2003]. For details on how the averaging kernels vary with land types and time of day, see Deeter et al. . We find that the global average number of degrees of freedom for signal in the MOPITT vertical profiles, which is the total of the eigenvalues of the averaging kernel and describes the number of independent pieces of information inferred from the retrieval [Rodgers, 2000], is 1.21 ± 0.45 for 1 sample day (15 March 2001). One can think of this value as the number of modes of vertical structure retrieved by MOPITT [Rodgers, 1990]. The first eigenmode for MOPITT is representative of a broad column-like signal with an eigenvalue close to one. The remaining contribution to ds, which can exceed 0.5 in the tropics but is generally less than 0.2 in the extratropics, can be thought of as the contribution from a broad vertical gradient. We choose to focus on the column data in our analysis.
 Seven MOPITT validation profiles for a range of conditions were performed during TRACE-P underneath coincident MOPITT overpasses. Four were conducted under sufficiently clear-sky conditions to enable MOPITT retrievals. Results presented by Jacob et al.  show that MOPITT captures the variability in the observed aircraft columns, with a high bias of 6 ± 2%, but provides no apparent information on vertical structure for Asian plumes over the Pacific [Jacob et al., 2003]. One of the validation profiles (DC-8 flight 5) sampled a trans-Pacific pollution transport layer event with CO in excess of 250 ppbv, as discussed in section 3. The 6% MOPITT bias observed during TRACE-P is consistent with MOPITT validation results for a suite of aircraft campaigns and profiles (L. Emmons et al., Validation of MOPITT CO retrievals with aircraft in situ profiles, submitted to Journal of Geophysical Research, 2003).
 We conducted a simulation for February–April 2001 with full O3-NOx-hydrocarbon chemistry. We also conducted CO-only simulations with tagged sources and monthly mean OH concentration fields archived from the full chemistry simulation. Initial conditions were taken from a 1-year spin-up simulation and model output was sampled every 3 hours. Anthropogenic sources of CO and its precursor hydrocarbons are as described by B. N. Duncan et al. (Model study of the variability and trends of carbon monoxide (1988–1997): 1. Model formulation, evaluation, and sensitivity, submitted to Journal of Geophysical Research, 2003, hereinafter referred to as Duncan et al., submitted manuscript, 2003). Our anthropogenic source of CO from China (181 Tg CO yr−1) is 66% higher than in the inventory of Streets et al.  and 8% higher than the a posteriori estimate from the inverse model by Palmer et al.  constrained with TRACE-P observations. Biomass burning emissions are from a bottom-up inventory constrained with daily satellite fire counts for the TRACE-P period [Heald et al., 2003]. Most of the global biomass burning for this period takes place in Southeast Asia and the corresponding regional source for February–April 2001 is 69 Tg CO. The inverse model analysis of Palmer et al.  suggests that this regional biomass burning source is too high by a factor of 3; as we will see, simulation of the MOPITT data leads to a similar conclusion.
 Comparisons between full-year model results for 2001 and ground station CO measurements by the National Ocean and Atmospheric Administration (NOAA) Climate Monitoring and Diagnostics Laboratory (CMDL) Carbon Cycle Group for representative Northern Hemispheric sites are shown in Figure 2. Several recent changes to the GEOS-CHEM chemical mechanism, including a faster N2 + O(1D) rate constant [Ravishankara et al., 2002] have decreased the global mean OH concentration by 10–15%. The most recent global model evaluation of CO by Duncan et al. (submitted manuscript, 2003) which does not include this change in rate constant, shows a model underestimate in the northern tropics and midlatitudes by 10–40 ppbv, with better agreement achieved in the arctic latitudes. The model simulation reported here is generally unbiased in its simulation of the CMDL Northern Hemispheric sites; however high-latitude sites are overestimated by 10–30 ppbv in summer-fall. Our 10–30 ppbv spring underestimate at Midway (MID) does not appear to be driven by an underestimate in Asian sources since neither the TRACE-P observations nor the MOPITT observations (discussed below) reveal such a bias.
3. Interpretation of MOPITT Observations for the TRACE-P Period
3.1. Global Distributions
Figure 3 shows the mean CO column observed by MOPITT during TRACE-P (20 February to 10 April 2001) and the corresponding GEOS-CHEM model column with MOPITT averaging kernels applied (according to equation (1)). Here the MOPITT data for each day have been averaged over the 2° × 2.5° GEOS-CHEM model grid and the resulting gridded product is then averaged over the 49-day TRACE-P period. MOPITT retrievals are not performed for cloudy scenes and are limited by the orbital coverage, but the effective coverage is increased here because of averaging over the 2° × 2.5° grid. Temporal coverage of the gridded MOPITT product over the North Pacific is 20–70% (Figure 4). The GEOS-CHEM model results in Figure 3 were constructed by binning the model on the MOPITT pressure levels for each retrieval scene, weighting by the MOPITT averaging kernels (such as shown in Figure 1), and integrating the column.
 Both model and satellite observations show high CO levels over continental source and outflow regions, as well as a general latitudinal gradient. Columns are low over high terrain (e.g., the Himalayas, the Rocky Mountains). A difference plot (Figure 3, bottom panel) shows no mean global bias between the model and measured CO. Mean MOPITT and GEOS-CHEM columns for the TRACE-P period are highly correlated (R2 = 0.97, slope = 0.97); and the correlation also holds for data points on individual days (R2 = 0.87, slope = 0.95; Figure 5). The slope indicates that the model underestimates MOPITT observations by 5%. As pointed out in section 2.2, the MOPITT observations appear to be biased high by 6% relative to the TRACE-P aircraft data.
3.2. Constraints on Asian Emissions
 The pattern of model bias relative to MOPITT matches the distribution of biomass burning, notably over Southeast Asia [Heald et al., 2003]. MOPITT column observations over Southeast Asia (10°–25°N, 80°–110°E), where biomass burning dominates anthropogenic emissions by a factor of two in the model, are on average 18% lower than the model column. The bias extends to southern China, where biomass burning effluents from Southeast Asia are lifted to the middle troposphere during transport in WCBs [Liu et al., 2003]. The discrepancy suggests that biomass burning emissions in the model for Southeast Asia are too high, consistent with the inverse model analysis of the TRACE−P CO data by Palmer et al. . In order to estimate how the overestimate of MOPITT observations translates to an overestimate of biomass burning emissions, we reduced the mean simulated CO profile over Southeast Asia by a fraction of the mean biomass burning contribution (determined from a sensitivity simulation). We then passed the resulting profile through the averaging kernels for 1 sample day (16 March 2001). We thus estimate in the model that reducing the biomass burning source over Southeast Asia by 60% (to 39 Tg yr−1 CO over the 70°–140°E, 0°–50°N region) would correct the MOPITT overestimate, consistent with the previous inverse analysis of TRACE-P data by Palmer et al. . An important caveat in the interpretation of model vs. MOPITT discrepancies in terms of errors on CO sources is the possibility of model bias in the vertical structure of CO. This could arise, in particular, as a result of excessive convection transporting CO to the upper troposphere where MOPITT is most sensitive. Comparison of GEOS-CHEM and MOPITT vertical profiles over Southeast Asia shows no significant difference in shape (Figure 6), but this reflects the lack of MOPITT information on vertical structure as previously discussed. The model vertical profiles over Southeast Asia show an enhancement in the upper troposphere due to deep convection. If we were to change the shape of the vertical profile in the model while conserving the total CO column, such that we reduce upper tropospheric CO to tropical background concentrations (55–70 ppbv) and correspondingly increase lower-tropospheric CO, we would match the MOPITT observations (Figure 6). Cloud data show that frequent deep convection did in fact occur over Southeast Asia during TRACE-P [Fuelberg et al., 2003], and an intercomparison of model simulations of CO for the TRACE-P period reveals no systematic bias in the GEOS-CHEM vertical distribution [Kiley et al., 2003], but in the absence of measured vertical profiles over Southeast Asia some ambiguity remains.
 The MOPITT observations reveal no evident model bias in the anthropogenic source of CO from China. This is consistent with the TRACE-P aircraft observations, since the source used here is only 8% higher than the a posteriori estimate derived by Palmer et al.  by inverse modeling of the TRACE-P data. Previous inverse model studies by Kasibhatla et al.  and Pétron et al.  required high anthropogenic emissions from eastern Asia (350–380 Tg yr−1 and >500 Tg yr−1, respectively, compared to our 323 Tg yr−1) to achieve agreement with global CMDL ground station observations in 1994 and 1990–1996, respectively. The constraints from TRACE-P and MOPITT observations imply a weaker Asian source.
3.3. Asian Outflow and Trans-Pacific Transport
 We now turn to the variability of Asian outflow to the Pacific and subsequent trans-Pacific transport, as observed by MOPITT and simulated by GEOS-CHEM. Figure 7 shows MOPITT and GEOS-CHEM time series for the TRACE-P period over the northwest, central and northeast Pacific. As before, the GEOS-CHEM model results are sampled for each MOPITT scene and are processed through the MOPITT averaging kernels. The MOPITT observations over the northwest Pacific show outflow events every 5–8 days associated with the passage of cold fronts across the Asian Pacific Rim, as also observed during TRACE-P [Fuelberg et al., 2003; Liu et al., 2003]. GEOS-CHEM and MOPITT are correlated (R2 = 0.61) in this region, reflecting the ability of the model to capture the WCBs driving the Asian outflow ahead of the frontal passages [Liu et al., 2003]. Farther downwind across the Pacific the agreement between model and observations degrades (R2 = 0.35 over the northeast Pacific), possibly due to accumulation of model transport errors or to vertical displacement of the Asian pollution layers (J. H. Crawford et al., Exploring the relationship between MOPITT and in situ observations of CO based on a large-scale feature sampled during TRACE-P, submitted to Journal of Geophysical Research, 2003). Column concentrations are most sensitive to CO enhancements in the midtroposphere (5–6 km) due to the application of the MOPITT averaging kernels (Figure 1). The MOPITT data over the northeast Pacific identify four major events of trans-Pacific transport of Asian pollution during the TRACE-P period, and these are indicated by arrows in Figure 7. These events are all captured by the GEOS-CHEM model but with a dampened magnitude. The four trans-Pacific transport events identified by MOPITT during the TRACE-P period were also seen by in situ observations. Event 1 was observed on the outbound TRACE-P transit flights, as discussed further in section 4. Events 2 and 3 were observed at the Cheeka Peak Observatory (CPO, 48.30°N 124.62°W) on the coast of Washington state on 11 and 20 March, respectively [Jaeglé et al., 2003]. There were no observations at CPO during events 1 and 4. Event 4 was observed from aircraft during the PHOBEA II experiment and was characterized as an air mass that had descended from high latitudes but originated from east Asian sources [Jaffe et al., 2003].
Crawford et al.  showed that high CO concentrations observed over the northwest Pacific during TRACE-P were correlated with cloud cover, reflecting the role of WCBs as the dominant pathway of outflow. They suggested that this might lead to bias in satellite observations of continental outflow. To investigate this issue we compared the average CO column from GEOS-CHEM over the northwest Pacific during the TRACE-P period (2.62 ± 0.11 × 1018 molecules cm−2) to the average of the subset corresponding to MOPITT observation points (2.57 ± 0.16 × 1018 molecules cm−2). There is no significant difference between these two numbers, implying that at least in the model world there is no cloud bias in the MOPITT sampling of the northwest Pacific; however, the model representation of clouds is not particularly robust.
4. Ozone Production During Trans-Pacific Transport: A Case Study
4.1. Trans-Pacific Transport Event on 22–27 February
 Layers with elevated CO from rapid trans-Pacific transport of Asian pollution were observed on the two outbound transit flights of the DC-8 during TRACE-P (flight tracks shown in Figure 8). In flight 4 from California to Hawaii (26 February 2001), vertical profiles indicated CO concentrations in excess of 200 ppbv in midtropospheric layers at 600–250 hPa (Figure 9, profiles A, B, and C; locations shown in Figure 8). The southern layers (B and C) were also enhanced in O3, whereas the northern layer (A) at 40°N off the coast of California was not. In flight 5 from Hawaii to Guam conducted the following day (27 February 2001), strong CO and O3 enhancement layers were observed in the lower free troposphere at 800–600 hPa (Figure 9, profiles D and E). CO concentrations were greater than 200 ppbv, and were accompanied by O3 in excess of 70 ppbv.
 We find that the pollution layers observed in flights 4 and 5 all originated from one single Asian outflow plume on 22 February 2001 associated with a WCB lifting Chinese pollution ahead of a cold front. Back trajectories (Figure 8) offer evidence for this common origin; further evidence from chemical tracer observations and the GEOS-CHEM model simulation will be presented below. The Asian plume was rapidly transported across the northwest Pacific and encountered a blocking high-pressure system near Hawaii on 25 February (Figure 10). It then split into northern and southern branches. The northern branch continued to North America and was sampled on flight 4 (26 February). The southern branch veered south to the tropical Pacific, subsided and was sampled on flight 5 (27 February). The back trajectories for profiles B and C in Figure 8 suggest some mixing of the Asian plume with tropical marine (B) and upper tropospheric (C) air masses. We limit further analysis to profiles A, D, and E, for which the clusters of back trajectories identify most clearly a common origin.
4.2. Source Characteristics of Trans-Pacific Asian Plume
 Both the northern and southern plumes sampled in TRACE-P exhibit the characteristics of recently emitted pollution, as indicated in Table 1. The high C2H2:CO concentration ratios (2.7–3.6 pptv/ppbv) are characteristic of fresh Asian outflow [Russo et al., 2003]. The C3H8:C2H6 ratio for the northern plume is also characteristic of fresh pollution [Russo et al., 2003], but the values for the southern plume are lower and indicate some photochemical aging.
Table 1. Average Species Concentrations and Ratios Observed in the Trans-Pacific Asian Plume of 26–27 February 2001
 Perchloroethene (C2Cl4), a chemical used in dry cleaning, is a tracer of urban pollution whereas methyl chloride (CH3Cl) is often viewed as a tracer of biomass burning [Blake et al., 2003]. Blake et al.  examined the chemical tracer signatures for the high-CO layers in flight 5. They find correlated enhancements of CH3Cl, C2Cl4, ethyne and halocarbons, implying a mixture of fossil fuel and biomass burning (or biofuel) sources. We further examined the CH3CN/HCN enhancement ratio in an attempt to separate biomass burning from biofuel influence, as this ratio in TRACE-P was much higher in biomass burning plumes (0.46) than in anthropogenic Chinese plumes including biofuel but no biomass burning influence (0.23) [Li et al., 2003; Singh et al., 2003]. We find CH3CN/HCN ratios > 0.6 in the pollution layers of flights 4 and 5 (Table 1), implying a biomass burning contribution to the Asian outflow. Blake et al.  suggest that lower levels of halocarbons and C2Cl4 in profile E vs. profile D are indicative of somewhat different origins. We find here that the CH3Cl/CO ratios were similar in profiles A, D and E, and the C2Cl4/CO ratio was consistent within a factor of 2 (Table 1). This together with the back trajectories supports a common origin, however there is clearly some heterogeneity present in the plume, which is not inconsistent with WCB lifting air masses from different sources over east Asia.
4.3. Model Simulation and MOPITT Observations
 The simulation of this trans-Pacific transport event by GEOS-CHEM is shown in Figure 11. The pollution transport in the model is consistent with the back trajectories shown in Figure 8. The model plume is transported across the Pacific in the midtroposphere (∼6 km) and encounters the blocking high-pressure system near the dateline where it splits into southern and northern branches. The southern branch subsides (to 4–5 km altitude) while the northern branch is forced upward (to 7–8 km altitude). The northern enhancement is displaced upward in the model relative to the observed aircraft profiles (Figure 9), which would reduce its sensitivity to observation by MOPITT as discussed in section 3. The tagged tracer simulations indicate that the model enhancement was 70–80% anthropogenic and 20–30% from biomass burning. Such mixing of anthropogenic and biomass burning effluents in the WCBs driving Asian outflow to the Pacific was frequently observed in TRACE-P [Carmichael et al., 2003; Liu et al., 2003; Ma et al., 2003].
 Although the model successfully describes the pathways for trans-Pacific transport of the 26–27 February plume, it greatly underestimates (by 50–100 ppbv) the CO maxima observed by the aircraft. Numerical diffusion in the model may be responsible. Such an underestimate was noted previously in the simulation of MOPITT observations of trans-Pacific transport events over the northeast Pacific (Figure 7) and is seen in particular in comparison to the MOPITT column observations for 23–27 February 2001. The daily MOPITT data are relatively sparse, because of incomplete orbit coverage and clouds over the Pacific, but comparison to the model results with averaging kernels applied (Figure 11, center panels) allows one to visualize the Asian outflow plume as it progresses across the Pacific. MOPITT observations indicate high levels of CO reaching North America on 27 February.
4.4. Photochemical Evolution and Ozone Production
 Although the pollution layers sampled by the TRACE-P aircraft on flights 4 and 5 originated from the same outflow event from Asia, as confirmed by the back trajectories and the chemical tracer signatures, they show evidence of very different photochemical evolution. We previously commented on the lower C3H8:C2H6 relationship observed in flight 5. The northern plume sampled in flight 4 exhibited no significant O3 enhancement (Figure 9, profile A). Conversely, the subsiding southern plume sampled on flight 5 contained large O3 enhancements (Figure 9, profiles D and E). Differential absorption lidar (DIAL) observations of O3 vertical cross sections along the tracks of flights 4 and 5 (Figure 12) show no O3 pollution enhancement at high latitudes (>35°N) consistent with the in situ observations in the northern Asian plume. The DIAL observations show a pronounced high-O3 layer between 2 and 4 km on flight 5, with levels exceeding 80 ppbv, consistent with the in situ observations of Figure 9 and demonstrating the broad horizontal extent (at least 2500 km) of this subsiding pollution layer. A positive correlation between O3 and aerosols was also found in these pollution plumes by Browell et al. , who attributed the enhancement to biomass burning sources in southeast Asia. The differences in O3 enhancements between the northern and southern plumes reflects not only the effect of solar radiation but also, and more importantly, the effect of subsidence driving PAN decomposition to NOx and hence O3 production. Figure 13 shows the concentrations of NOy species observed in the northern plume (profile A, flight 4) and in the southern plume (profiles D and E, flight 5). The total NOy concentration in both plumes is about the same (470–530 pptv), consistent with their common origin as NOy is conserved. Mean NOx levels are 18 pptv in the northern plume (4% of total NOy), and 163 pptv in the southern plume (30% of total NOy). Total NOy also includes 13% HNO3+nitrate aerosol and 74% PAN in the northern plume, vs. 51% HNO3+nitrate aerosol and 14% PAN in the southern plume (Figure 13). The dominant PAN contribution in the northern plume is consistent with the mean PAN/NOy molar ratio (0.75 ± 0.10) observed in the warm conveyor belt outflow in the free troposphere during TRACE-P [Miyazaki et al., 2003]. The difference in speciation between the northern and southern plumes can be explained by PAN decomposition to NOx during subsidence following the split of the original plume on 25 February. From the temperature measurements in the layers sampled in TRACE-P (Table 1), we find that the PAN lifetime drops from 105 days in profile A to 5–6 hours in profiles D and E.
 We can relate the observed PAN decomposition and O3 production in the southern plume. From Figure 8, O3 in the southern plume appears to be enhanced by 40 ppbv above the local tropical background. However, considering the midlatitude origin of this plume, a background O3 concentration of 57 ppbv (such as seen in profile A, Table 1) is more appropriate as a baseline for quantifying the O3 enhancement in the southern plume (65–74 ppbv, Table 1). We deduce a net ozone production of 8–17 ppbv in the southern plume in the 2 days following the splitting of the plume. Photochemical model calculations by A. J. Hamlin et al. (manuscript in preparation, 2003) initialized with the conditions observed in the northern plume yield an O3 production of only 2–5 ppbv in 2 days over the subsiding trajectory of the southern plume and do not reproduce the observed PAN loss. Increasing the rate of PAN decomposition by a factor of 2.1–2.3 in that calculation, within the range of the uncertainty in the rate constant [Sander et al., 2003] can match the observed PAN decomposition and produces an additional 2–3 ppbv of O3. This O3 increment matches the O3 enhancement observed in profile D but still underestimates that in profile E. Interpretation of this discrepancy is difficult as a result of uncertainty in the initial conditions. Heterogeneity within the plume, which was not captured in the model calculations, is a factor of uncertainty in the analysis.
 We have presented an integrated analysis of continental outflow and trans-Pacific transport of Asian pollution by using aircraft observations from the TRACE-P mission (February–April 2001), concurrent satellite observations of CO columns from the MOPITT instrument, and a global 3-D simulation with the GEOS-CHEM model. We used this analysis to determine the utility of the MOPITT observations to quantify Asian outflow and trans-Pacific transport, and to examine in more detail the sources and photochemical evolution (in particular O3 production) in a major trans-Pacific event observed during TRACE-P.
 The GEOS-CHEM model, previously evaluated with TRACE-P observations of Asian outflow of CO over the northwest Pacific, is found to be highly correlated and globally unbiased with respect to the MOPITT observations during TRACE-P (R2 = 0.87, slope = 0.95). The agreement between MOPITT observations and the GEOS-CHEM model simulation over China suggests no obvious bias in anthropogenic sources in this region. There are some regional biases, the largest of which is an 18% model overestimate of MOPITT CO columns over Southeast Asia. We estimate that a 60% decrease in seasonal biomass burning from that region would resolve this discrepancy, and such a correction would be consistent with an inverse model analysis of the TRACE-P CO observations using GEOS-CHEM [Palmer et al., 2003]. An important caveat is that model simulation of the MOPITT column CO is sensitive (through the instrument averaging kernel) to the shape of the CO vertical profile. As an extreme case, suppression of deep convection over Southeast Asia in the model would resolve the regional discrepancy with MOPITT without any change to the biomass burning source. This result highlights the importance of in situ measurements of vertical profiles from aircraft to assist the interpretation of satellite observations.
 The MOPITT observations for February–April 2001 identify four major events of trans-Pacific transport of Asian pollution to North America. The GEOS-CHEM model reproduces these events and shows high correlation with the corresponding MOPITT data, degrading however from the northwest to the northeast Pacific. Both numerical diffusion and vertical bias in the model may contribute to the underestimate over the northeast Pacific. All four events observed by MOPITT were also sampled in situ over the northeast Pacific by either the TRACE-P or the PHOBEA-II aircraft missions, or by the Cheeka Peak Observatory in Washington State.
 A series of major Asian pollution layers sampled in TRACE-P transit flights from California to Hawaii (26 February) and Hawaii to Guam (27 February) provided an opportunity for a more detailed investigation of the trans-Pacific transport and chemical evolution of these layers, including in particular O3 production. Concentrations of CO in the layers approached 300 ppbv. Elevated O3, approaching 80 ppbv, was observed in the low-latitude layers only. Back trajectories and chemical tracers reveal that these layers all originated from a single Asian outflow event on 22 February when a warm conveyor belt ahead of a cold front lifted Asian anthropogenic and biomass burning pollution to the free troposphere. The plume split into southern and northern branches upon encountering a blocking high-pressure system over the mid-Pacific. The MOPITT observations and the GEOS-CHEM model simulation allow a reconstruction of the evolution of this Asian outflow plume during trans-Pacific transport, although GEOS-CHEM does not reproduce the magnitude of the enhancements, perhaps because of numerical diffusion. Elevated O3 (ΔO3 = 8–17 ppbv) was observed in the southern but not in the northern plume. Reactive nitrogen in the northern plume was present mostly as PAN, but subsidence in the southern plume was found to drive PAN decomposition and from there O3 formation. The ΔO3 observed in the southern plume was high relative to values obtained from a photochemical model calculation initialized with the conditions of the northern plume. This represents the first field observation of PAN decomposition driving O3 production in polluted plumes transported to the remote troposphere.
 This work was supported by the Global Tropospheric Chemistry Program (GTCP), the NASA Atmospheric Chemistry Modeling and Analysis Program (ACMAP), and by NASA Headquarters under the Earth System Science Fellowship Grant NGT5-30466. CLH was also supported by the Natural Science and Engineering Research Council of Canada. We gratefully acknowledge Daniel Ziskin and Jarmei Chen for accelerating the data processing for TRACE-P and Bob Yantosca for HDF programming assistance. We would also like to acknowledge useful discussions with Paul Palmer and Jennifer Logan as well as the comments from two anonymous reviewers.