Impact of U.S. Oil and Natural Gas Emission Increases on Surface Ozone Is Most Pronounced in the Central United States

Observations of volatile organic compounds (VOCs) from a surface sampling network and simulation results from the EMAC (ECHAM5/MESSy for Atmospheric Chemistry) model were analyzed to assess the impact of increased emissions of VOCs and nitrogen oxides from U.S. oil and natural gas (O&NG) sources on air quality. In the first step, the VOC observations were used to optimize the magnitude and distribution of atmospheric ethane and higher-alkane VOC emissions in the model inventory for the base year 2009. Observation-based increases of the emissions of VOCs and NOx stemming from U.S. oil and natural gas (O&NG) sources during 2009–2014 were then added to the model, and a set of sensitivity runs was conducted for assessing the influence of the increased emissions on summer surface ozone levels. For the year 2014, the added O&NG emissions are predicted to affect surface ozone across a large geographical scale in the United States. These emissions are responsible for an increased number of days when the averaged 8-h ozone values exceed 70 ppb, with the highest sensitivity being in the central and midwestern United States, where most of the O&NG growth has occurred. These findings demonstrate that O&NG emissions significantly affect the air quality across most of the United States, can regionally offset reductions of ozone precursor emissions made in other sectors, and can have a determining influence on a region’s ability to meet National Ambient Air Quality Standard (NAAQS) obligations for ozone.


■ INTRODUCTION
During 2010−2015, the combined U.S. oil and natural gas (O&NG) production grew at an unprecedented rate (Supplemental Information (SI) Figure 1), making the United States the largest O&NG producing nation in the world. This growth was stimulated by new technologies, in particular hydraulic fracturing, which enabled the exploration of previously uneconomical shale plays. Atmospheric emissions from heavy drilling equipment, power generation at drill sites, trucking, and controlled and fugitive emissions from well sites, have received attention because of their lasting and reoccurring impacts on air quality, atmospheric chemistry, and climate, from local to global scales. 1−4 Accidental releases, such as during the 2015−2016 Aliso Canyon natural gas blowout, can cause large episodic injections of pollutants into the atmosphere with severe short-term impacts on air quality and human health in the immediate surroundings. 5,6 Among the pollutants of concern are primary fossil-fuel hydrocarbons, i.e., methane and nonmethane volatile organic compounds (VOCs), as well as nitrogen oxides (NO x ), and black carbon emissions from diesel combustion and flaring. 1,3,7 Measurements within and downwind of O&NG basins have shown at times highly elevated atmospheric levels of VOCs. 8−12 Ground-based monitoring and remote sensing observations have also revealed increased NO x . 13,14 While elevated VOC levels have been attributed mostly to fugitive emissions from oil and gas drilling, increases in NO x can be linked to flaring, on-site power generation, and heavy trucking associated to drilling and establishing of the production infrastructure. O&NG emissions have been predicted to contribute to summertime photochemical ozone production based on their chemical reactivity, 15−17 as well as by regional photochemical modeling. 18−22 Two modeling studies have pointed out regional variability of the ozone effects, with potential ozone benefits from the transitioning of coal to O&NG power production. 23,24 O&NG emissions not only foster ozone chemistry during the summer ozone season but also can drive ozone production during snow-covered conditions in winter when ozone precursors are trapped in a shallow surface layer. The high albedo of the snow enhances solar irradiance and increases radical production from carbonyl photolysis, which in turn promotes ozone production. 25−28 As a consequence, the sparsely populated but heavily drilled Uintah Basin reported the highest number of exceedances of the U.S. National Ambient Air Quality Standard (NAAQS) for ozone within the whole United States in 2013, 26,28 thereby challenging the common view of ozone being exclusively a summertime pollution problem. Comparison studies between model outputs and observations of surface ozone levels have pointed to higher top-down flux estimates in O&NG drilling regions than inventory emission estimates, 27,29,30 indicating that modeling based on these inventories has likely tended to underestimate the influence of O&NG emissions on local and regional ozone.
The recent increases seen in the ozone precursor emissions suggest a possible increasing contribution of O&NG emissions to local and regional ozone production, bearing the potential to offset ozone pollution mitigation gains made in other emission sectors. 8 This has brought increasing attention to the role of O&NG emission and their possible compromising role in a region's ability to comply with NAAQS regulations for ozone. 22,31,32 Premature deaths from ozone and PM2.5 resulting from a medium O&NG emission scenario were predicted to increase by 200−460 annually for the Marcellus and Utica shales alone. 22 On the U.S. national scale, premature deaths due to exposure to ozone and PM 2.5 resulting from O&NG emissions have been estimated to reach 1100−2700 yr −1 by 2025. 33 In a previous preliminary modeling assessment, 10 we applied estimates for increases of C 2 −C 5 VOC emission from U.S. O&NG in photochemical modeling. These simulations showed the potential for O&NG emissions to result in up to 0.5 ppb mean summer ozone enhancements, with the largest effects seen over California and the central United States. 10 Here, we expanded on this work, using not only an improved total, spatial, and chemical species distribution of VOCs but also the contribution of NO x emissions from the growth of U.S. O&NG exploration. This improved representation was then applied to derive an assessment of the regional sensitivity of O&NG emission changes, the contribution of O&NG emissions to regional exceedances of ozone NAAQS, and resulting ozone enhancements across the contiguous United States. where pairs of whole air samples are collected weekly and shipped to a central laboratory in Boulder, CO, for C 2 −C 7 VOC analyses. 34−36 For this study, we considered the VOC species ethane, propane, iso-and n-butane, and iso-and npentane, which together constitute the bulk of O&NG nonmethane VOC emissions. 16,17 A list of sites considered in this analysis is provided in SI Table 1.
Ozone Observations. Surface ozone data were from the Tropospheric Ozone Assessment Report (TOAR) data portal. 37 As the focus of this study is on the continental U.S., the data used originated primarily from U.S. sources. For the evaluation of the model runs, only sites with a "rural" characteristic 37 were considered for calculating grid averages.
Monthly mean values were calculated at each site, and the data were then aggregated in 2 × 2°grid squares. 38 Estimation of Growth Rate of U.S. O&NG Emissions. We explored several methods for estimating the rate of increase in U.S. O&NG VOC emissions: (1) During 2009−2014, U.S. oil and natural gas production increased by 64 and 21%, respectively (SI Figure 1). A simple approach would be to use these growth rates as a surrogate for the emission increase, based on the assumption that emissions scale linearly with the production volume. This assumption is supported by data from a site downwind of the Marcellus Shale, where median ethane measured over four years correlated with the reported shale gas production with an r 2 of 0. 59. 9 For the combined oil and natural gas emission increase, one would need to weight these emissions by the relative fraction that oil versus natural gas emissions contributes to total O&NG emissions. An inherent uncertainty is the assumption that the development and production of a new well, mostly due to hydraulic fracturing extraction technologies, are subject to the same fugitive emission loss rate as conventional extraction technologies. This has been questioned by some studies that have pointed out higher emission rates from hydraulic fracturing than from conventional drilling operations. 39 (3) A number of recent publications have developed estimates of the tropospheric ethane increase based on column Fourier transform infrared (FTIR) spectroscopy measurements. 10,43−45 These observations are primarily from remote and/or high elevation sites and are therefore most representative of the ethane growth in the background atmosphere. A review of four publications with a total of ten data sets from seven Northern Hemisphere (NH) sites yields ethane mean/ median rates of change of 4.3/4.6% yr −1 (SI Table 2). Based on the geographical pattern of increases seen in the shorter-lived propane, the authors argued that this NH troposphere ethane increase largely stems from O&NG emission increases in the United States.
The modeling work presented here was based on the emissions growth that is outlined under point (3) above, which is a lower/conservative emission growth estimate, in comparison to the data and assumptions under (1) and (2). This emission growth was further evaluated by comparison between model and observations as detailed below.  52.0, in the T63L31 resolution, i.e., with a spherical truncation of T63 (corresponding to a quadratic Gaussian grid of approx. 1.9 × 1.9°in latitude and longitude), with 31 vertical hybrid pressure levels up to 10 hPa. Treatment of aerosols has been described previously. 47−49 The model was used in its Chemistry-Transport Model configuration, 50 i.e., without feedback between chemistry and transport. The model was nudged 51 toward the European Centre for Medium-Range Weather Forecast (ECMWF) reanalysis data (ERA-interim, 52 ); simulations are covering the period 2009−2014. Further descriptions and evaluations of the ECHAM5 chemistry and model are presented in the literature. 49,53−57 The chemistry mechanism for consideration of the O&NG VOCs was the same as detailed previously, 10 but augmented to include oxidation chemistry of simple C 4 −C 5 hydrocarbons (n-and iso-butane and n-and iso-pentane), as described in Pozzer et al. 49 Further, a simple mechanism for toluene chemistry was adopted, similar to the approach of Lelieveld et al. 58 The model simulations adopted emissions from the RCP85 database (Representative Concentration Pathway 8.5), 59 scaled as described in Pozzer et al. 60 To improve the agreement between model results and the observations from VOC monitoring, ethane emissions were further increased by 50% for all emission sources at latitudes north of 20°N. The resulting ethane emissions are as follows: in the NH, 12.1 Tg yr −1 of ethane is emitted by anthropogenic sources, 0.2 Tg yr −1 by biogenic sources, and 0.9 Tg yr −1 by biomass burning, totaling 13.2 Tg yr −1 for 2009. With the adjusted emissions, model outputs for the year 2009 agreed with GGGRN observations generally within 10% to the observations (average bias for all the stations worldwide) (SI Figure 2). The adjusted ethane emissions were therefore considered as our default emissions in this work. The propane, butane, and pentane emissions agree with the GGGRN observational data set as shown in Pozzer et al. 49 Modeled O&NG Emissions. In addition to the standard sector present in the RCP85 database, updated emissions from the O&NG sector were included. The emission map was based on shale O&NG well distribution, available at https:// fracfocus.org, which is a different approach than our previous work. 61 This database was considered the most complete, with approximately 2.5 million total well sites registered. This well inventory includes both active and inactive wells. This differentiation will likely only have a minor influence on the results, as the location and density of the wells are deemed to be reasonably proportional to the location of extraction activity, and the relatively large grid scale in the model reduces the uncertainties associated with the relative distribution of retired wells. We assumed that all wells emit the same amounts of VOCs and the same VOC signatures, neglecting differences in well production and leakage rates. Finally, the distributed well map was aggregated in a 0.5 × 0.5°regular map, and emissions were scaled based on the well number density in each grid cell. The resulting emissions map ( Figure 1) identifies regions that have experienced recent growth of O&NG development, with regions of large emission increases mostly in the central and northeastern United States.
As enumerated above, we applied an O&NG emission growth rate of ethane of 0.42 Tg/yr −2 , consistent with our previous work 10 and equivalent to a ∼17% increase of O&NG emissions over the five-year window (2009−2014). Based on the VOC ratios presented in, 10 we estimated 0.42, 0.30, 0.11, 0.08, 0.05, 0.06, and 0.15 Tg yr −2 increases for ethane, propane, n-butane, iso-butane, n-pentane, iso-pentane, and toluene emissions every year for five years, respectively. This  63 Although most of the global VOC emission increase identified by Helmig et al. 10 is probably from emissions in the United States, other global regions may have contributed to the flux increase. To reflect and compensate for this uncertainty, VOC emissions >C 5 were excluded as they could not unambiguously be identified as emitted by O&NG. Their relative fraction (of total VOCs) can vary over a wide range; on average, they constitute on the order of 10% of the total O&NG VOC emissions. 16,17,64 These longer chain and aromatic VOC generally have higher reactivity and ozone production potential than the lighter nonmethane hydrocarbons (NMHCs). 65 The methane data were nudged 66 on the surface mimicking the observed values for the simulated period. Although the background methane oxidation is considered in the model, we did not consider increases in regional O&NG methane Environmental Science & Technology pubs.acs.org/est Article emissions, which can have a further impact on regional ozone production. 67,68 The disregard of emissions of the O&NG longer chain (>C 5 ), aromatic VOCs, and methane emissions, make it more likely that the results for the estimated O&NG ozone production increase are a lower limit. In addition to VOCs, O&NG operations also emit nitrogen oxides (NO x ). 13,14 The NO x emission rates from oil and gas hydraulic fracturing well site development stem from a variety of diverse sources that depend on operator practices and State regulations. The relative ratio of NO x /VOC emissions from O&NG operations can be quite variable, depending on the abundance and technology of flaring, power generation, and other industry practices, 7,13,69 as well as emissions from the traffic at well sites. 70 Resulting emission rates are likely highly variable, and at this time poorly defined and uncertain. Here, we chose a representation of NO x emissions in our simulations based on Ahmadov et al. 27 and applied their estimated bottomup NO x /VOCs mass emission ratio of 0.023, which we considered the best available estimate for the time represented by the modeling window. We applied this ratio to a total VOC O&NG emission increase estimate of 1.2 Tg yr −2 . 10 resulting in a 0.026 Tg yr −2 increase of O&NG NO x emissions, emitted as NO in the model. Potential NO x emission reductions from the conversion of coal to O&NG power production and potentially associated ozone benefits 23,24 are considered in RCP85.
Model Sensitivity Runs. ECHAM5-MESSy tends to overestimate surface ozone in comparison to observations, 71 similar to what has been noted for other chemistry−climate models. 72,73 Because of the recognition of this potential bias, we intentionally avoided building interpretations on modeled absolute ozone results. Instead, we determined the sensitivity to emission changes from relative differences between model run scenarios. The absolute bias cancels out in these calculations, making results more robust since they are relatively immune to the bias in modeled absolute ozone. In this work, three numerical simulations were performed with the model.
Model_Run_Constant. All emissions were kept constant throughout the years 2009−2014 (i.e., every year has the same emissions as the year 2009). This simulation served as a baseline to determine the effect of meteorology alone on the ozone changes over the investigation period.
Model_Run_RCP85. All emissions were following the RCP85 database as described above, without any new O&NG emissions.
Model_Run_O&NG_Trend. The same as the Model_-Run_RCP85 simulation, with C 2 −C 5 NMHC, toluene, and NO x emissions from shale gas emissions added, increasing from May 2009 to May 2014 as discussed above. (Please note that changes in a reactive and highly variable gas such as ozone over a five-year window may not necessarily imply a "trend." While we avoided the use of the word "trend" in the text, this term was designated as a label for the model runs that used the increasing O&NG emissions.) NAAQS Exceedances and Ozone Increase Rates. The NAAQS exceedances were calculated as the number of days in a year when the daily maxima of 8-h running mean for ozone exceeded 70 × 10 −9 mol/mol (ppb), following the U.S. Environmental Protection Agency (EPA) definition (www.epa. gov/outdoor-air-quality-data/air-data-ozone-exceedances). The 2009−2014 rates of change in ozone were calculated following the NOAA trend analysis tool, 74 as described ■ RESULTS AND DISCUSSION Model Evaluation. The ethane model results of the simulation that gave the closest agreement with the data (Model_Run_O&NG_Trend) reproduced the observed ethane mixing ratios for most comparisons within 10% (SI Figure 2). Further, the ethane rates of concentration increase that were calculated from the data and the modeled rates of increase were compared for each location for the years 2009−2014 (SI Figure 3). In total, 91% of the results from both methods agree in the sign of the rate of change (increase) and 48% within a factor of two (modeled rate of increase relative to the observed value between 0.5 and 2). It needs to be emphasized that due to the variability in the data, and the relatively short (5 years) time window that was considered, for a significant number of cases, the calculations for the rate of changes in the measurement data had a relatively high uncertainty error (see Supplementary Figure 6) demonstrated the cause for this at first seemingly contradictory finding: 2009 and 2010 were relatively low ozone years, followed by 2 years with above-average ozone. These ozone differences are probably linked to meteorological differences in these years that affected ozone production. Over the north-central United States, average temperatures during 2012 were approximately 1.6 K higher than those during 2009. 78 It is therefore likely that these dynamic/meteorological differences and interannual variability in meteorological conditions, in particular the perseverance of high-pressure systems, caused above-average summer temperatures and above-average surface ozone. Previous research has estimated increases of 2−6 ppb of ozone for each K increase in the temperature. 79,80 This sensitivity and these interannual weather differences most likely caused a marked upward/increasing ozone change during the 2009−2014 time period for the north-central United States that defines the long-term ozone trends. These ozone changes are therefore likely largely due to the year-to-year variability and overall relatively warmer weather during the latter part of the chosen five-year time interval and are not a reflection of long-term changes in ozone precursor emissions, including from the O&NG sector.
Effect of O&NG Emissions on Summer Mean Ozone. Using a differential atmospheric modeling approach that considered actual meteorological conditions provided estimates for added ozone production from the growth of O&NG emissions. The sensitivity study included our current best estimate of O&NG emission increases of methane, VOCs, and NO x over the 2009−2014 growth period. To remove the influence of meteorology, the effect of the O&NG emissions on surface ozone was assessed by subtracting the 2014 model output with the scenario that did not include the O&NG emission increases, i.e., the Model_Run_RCP85, from the Model_Run_O&NG_Trend simulation. Their difference, shown in Figure 3, is insensitive to meteorological and other emission changes (as they cancel out), and therefore only show the ozone changes due to increasing O&NG emissions. There is high spatial variability in this ozone contribution across the United States. Mean summer ozone increases of ≥1 ppb are observed for more than half of the continental United States, These summer ozone increases are ∼20% higher than our previous estimates, 10 which is primarily due to the combined inclusion of O&NG emissions of higher VOCs and NO x . This relatively small difference illustrates that the contribution of the O&NG NO x emission is relatively small (<20%) compared to the O&NG VOC emissions. Furthermore, the increases in ozone are now more pronounced in the central United States, reflecting the updated distribution of emissions increase. Notable ozone increases are also seen over selected areas in southern California. This finding is a bit surprising, as emission increases in California are relatively small (Figure 1). This high sensitivity is primarily due to high NO x conditions in this region, which causes relatively modest increases in VOC emissions to have a relatively large influence on ozone production.
Two other previous studies have provided estimates of the recent O&NG increase in U.S. surface ozone. Kort et al. 21 show a sensitivity study that considered fugitive alkane (C 2 and higher) emissions from the Bakken Shale only. For one particular day in August 2014, they calculated ozone increases at and downwind of the primary emission locations in North and South Dakota, and eastern Montana, reaching 2 ppb. Tzompa et al., 81 applying updated O&NG emission fields to the GEOS-Chem CTM, predicted daytime summer ozone increases of up to 4 ppb, with the strongest effect seen over the central United States, which is reasonably consistent with our findings.
Effect of O&NG Emissions on Regulatory Summer 8-h Mean Ozone Metrics. The U.S. ozone compliance with NAAQS is defined by the "Design Value," which is calculated as the mean of three consecutive years of the fourth highest annual value for the daily maximum average 8-h ozone value (MDA8). 82 As we simulated a limited amount of years, we decided instead to estimate the number of days for individual years when the MDA8 exceeded 70 ppb, which is a metric also largely used by the EPA. 83 For instance, for the year 2014, the model predicts that the ozone production from the added O&NG emissions caused 0−10 additional days when the MDA8 exceeded 70 ppb over regions in the contiguous United States, with 90% of the United States affected by at least one additional exceedance day (Figure 4).
Additional exceedance days from the growth of the O&NG emissions are observed in most of the continental United States, with the exception of Alaska. Approximately 90% of the contiguous United States experienced at last one additional day with 8 h surface ozone >70 ppb. These results for ozone standard exceedance days display a higher regional sensitivity than the previously reported results for average ozone increases. 33,81 It must be stressed that these results are rather sensitive to the particular regional meteorological conditions in the chosen model year. The chosen year 2014 was, for instance, a relatively low ozone year in the central United States ( 32, , 84 SI Figure 6), which would tend to produce a relatively lower number of ozone exceedances for that region and particular year. It must, however, also be stressed that the model in general tends to overpredict the absolute ozone concentrations on the surface (see the discussion above and ref 71 ), which has the effect of overestimating the absolute number of MDA8 exceedances compared to the EPA observations. This is particularly pronounced in the few comparison locations in the central-southern United States, whereas a relatively better agreement was found for the western regions (see also section "Policy Implications" for a discussion on uncertainties). For estimating this bias, we performed an alternative calculation, where the year 2014modeled ozone from the added O&NG emission was subtracted from the observational data. This resulted in a sensitivity of -1−12 exceedance days (compared to 0−28 days) and an overall lower fraction of the stations with an increase of at least one exceedance day, i.e., ∼5 vs 67%. There is a notable overlap of regions with a high number of added high ozone days and areas that are in nonattainment with the ozone NAAQS (SI Figure 8). The impact is most pronounced on a wider geographical scale in the central United States. Hence, although meteorology is the main driver of ozone increase in the 2009−2014 period, the O&NG emissions have the effect of further promoting possible ozone NAAQS exceedances in this region.
A preliminary assessment of the health effects projected on the order of 320 (298−344 at a confidence level of 95%) additional premature deaths from the added ozone per year (SI Text 1; please note that this mortality estimate does not include possible reductions in mortality from the decreased emissions that result from the transition of coal-to natural-gaspowered electricity-generating plants). According to the assessment in Fann et al., 33 health effects from fine particulate matter produced by O&NG emission are of a similar magnitude; therefore, the total mortality rate from O&NG- Environmental Science & Technology pubs.acs.org/est Article produced ozone and PM2.5 is possibly on the order of two times the ozone estimate. Policy Implications. A notable finding is that ozone increases are predicted across large geographical scales across most of the United States. This also includes numerous regions and States that do not have seen significant increases of O&NG industries within their own borders. Results reflecting a smaller, i.e., regional geographical scale bear a higher uncertainty due to the relatively coarse (1.8 × 1.8°, i.e., ∼200 km × ∼200 km) model resolution. Additional sources of uncertainties include the neglect of the contribution of ozone production from regional methane emissions, the omission of higher O&NG VOC species, and a conservative emission growth estimate (see above) for primary O&NG development regions. Diurnal ozone production occurs on time scales that are shorter than the transport scales and mixing regimes within the model grids; consequently, ozone production and exceedances can be higher or lower at selected locations within a grid. It is therefore likely that our O&NG ozone production estimates are over predictions in some regions and under predictions in other regions. It is well possible that a higher number of added exceedances occur on smaller geographical scales and in closer proximity to emission sources. Further, our investigations exclusively focused on summer ozone, as winter ozone chemistry under conditions of surface snow cover, low boundary layer heights, and the particular chemical conditions, resulting in high ozone production rates, is currently not well presented in the model. Elevated ozone and NAAQS exceedances during winter have been observed in the Uintah Basin, Utah, and Upper Green River Valley, Wyoming. These conditions and events were not included in this evaluation. Consequently, our estimates of the increased O&NG emissions on ozone and air quality regulation compliance, excluding these winter ozone episodes must be taken with caution, and are likely a low estimate of the year-round impacts of emissions on ozone air quality in these O&NG basins.
This research had the objective to investigate the regional to large-scale effects of increased O&NG emission by a differential modeling approach. An inherent weakness stems from the lack of regional/basin specific emission speciation and emission rates. Regulations on O&NG operation and emission controls vary by states, and the rate of controlled and fugitive emissions varies widely depending on many conditions, including the chemical composition of the petroleum reservoir, operator equipment, practices, and on the stage of well development. 85 Our modeling attributed emissions equally among all reported wells, neglecting any of these dependencies, which contribute to the uncertainty in the regional sensitivity of our findings. There is also the potential of a reduction in ozone precursor emissions in the proximity and downwind of electrical power plants that were converted from coal to gas as a consequence of the lower cost and readily available gas for power generation. 24 As stated by the authors of this publication, 24 the combined effects will likely vary regionally depending on the location and overlap of O&NG basins and processing facility in relation to the location of power generation plants, and need to be evaluated on a regional case-by-case basis.
Our studies demonstrate that the assessment of air quality impacts of the rapidly growing U.S. O&NG industry would largely benefit from a more accurate characterization of the chemical emissions and their regional variations. Further, policy implementation for emission reductions from O&NG industries bear the potential to reduce ozone production and improve air quality within, and in adjacent and downwind areas, including neighboring states. These reductions would be particularly advantageous for several ozone non-attainment areas to reach compliance with the ozone NAAQS 18,21,86−88 and would furthermore be beneficial for downwind regions across wide areas of the United States.
List of sites, sorted by latitude, within the GGGRN (Table 1); growth rate of atmospheric ethane (Table 2); oil and natural gas production trends in the United States ( Figure 1)

■ ACKNOWLEDGMENTS
The chemical modeling work by MGS was funded through the PoF III program of the Helmholtz Association, Germany. The VOC analyses at INSTAAR were conducted in samples collected for NOAA's Global Greenhouse Gas Reference Network, which is operated by the NOAA Global Monitoring Laboratory. NOAA also provided support for the VOC analysis. We thank all personnel who are involved in the management and execution of this program. We also thank all colleagues and agencies who generate and publicly share meteorological and chemical data that were used as inputs in the atmospheric modeling. The authors acknowledge the use of the Ferret program for analyses and graphics in this paper. Ferret is a product of NOAA's Pacific Marine Environmental Laboratory (information is available at http://ferret.pmel.noaa. gov/Ferret/). (