A 1 Multi-sensor Upper Tropospheric Ozone Product (MUTOP) based on TES Ozone and GOES Water Vapor: I. Derivation S. R. Felker, J.L. Moody, A. J. Wimmers, G. Osterman, and K. Bowman Abstract The Tropospheric Emission Spectrometer, TES, produces vertical ozone profiles from the lower troposphere through the lower stratosphere, but provides limited polar-orbital coverage. This study combined TES ozone measurements with two synoptic dynamical tracers of stratospheric influence (specific humidity derived from geostationary GOES Imager water vapor, and potential vorticity (PV) from an operational forecast model) in order to create a product of layer-average UT ozone across the GOES-West satellite domain. Blending the advantages of two remotely sensed quantities (GOES water vapor and TES ozone) is at the core of the Multi-sensor UT Ozone Product (MUTOP), producing the spatial coverage of a geostationary image, while retaining TESs ability to vertically resolve UT ozone. Results show 70-80% of TES-observed UT ozone variability can be explained by correlation with these two dynamical tracers. MUTOP can reproduce TES retrievals across the GOES-West domain with a root mean square error (RMSE) of 19.2 ppbv. There are several major advantages to this multi-sensor product: (1) the product provides a high spatial resolution distribution of dynamically driven ozone in the UT; (2) created at 6 hour intervals, MUTOP traces the rapid movement of dynamically-driven ozone in the UT; and (3) calculated from operational fields (GOES Water Vapor and GFS PV), once validated, MUTOP can potentially be created and used in near real-time. This paper presents the scientific basis and methodology behind the creation of this unique ozone product, as well as statistical comparisons to independent TES observations.
2
1. Introduction 1.1 Ozone in the upper troposphere and the Tropospheric Emission Spectrometer
Ozone in the upper troposphere (UT) continues to receive great attention in the scientific literature due to its role in clear-sky radiative forcing [Wang et al., 1995; Gauss et al., 2003; Worden et al., 2008], its potential downstream movement into populated boundary layers [Cooper et al., 2004; Hudman et al., 2004], and its rather unique position within the mixing zone between stratospheric and tropospheric reservoirs [Shepherd, 2002; Bowman et al., 2007; Fairlie et al., 2007]. Stratosphere to troposphere transport (STT) is an important source of ozone for the UT, and a variety of methods have been employed to identify dynamical features associated with ozone exchange. Stohl et al. [2003] provide an excellent review of stratosphere troposphere exchange (STE). It is generally accepted that tropopause folding occurs in association with significant latitudinal displacements of the tropopause along sloping isentropic surfaces, conditions that occur in the vicinity of the polar and subtropical jet stream, and in association with cutoff lows. The literature also contains many references to the value of potential vorticity and upper tropospheric specific humidity as tracers of stratospheric ozone enhancements in the upper troposphere. Air in the extratropical troposphere is stirred quasi-adiabatically by large-scale transitory cyclones and anticyclones, and stratospheric signatures are reasonably well preserved by these quasi-conservative tracers. In this paper, we seek to characterize the variations in ozone associated with these time-varying synoptic dynamic conditions in the upper troposphere.
3 Although there are various methods for measuring ozone mixing ratios in the UT (e.g., ozonesondes, differential absorption LiDAR, and in-situ aircraft measurements), all of these provide very limited spatial and temporal coverage. With the advent of hyperspectral satellite remote sensing instruments such as the Tropospheric Emission Spectrometer (TES), it is now possible to retrieve a more extensive perspective of UT ozone. The TES instrument is a high-resolution infrared Fourier transform spectrometer with nadir view footprint of 5.3 x 8.5 km [Beer, Glavich, and Rider, 2001; Hilsenrath et al., 2006]. TES measures ozone in the nadir view from ~705 km above the earth onboard the NASA Aura satellite which is in a sun-synchronous polar orbit. One of the main proposed goals of the TES mission is to provide a global view of the vertically resolved ozone distribution within the troposphere [Beer, 2006]. The information provided by TES had only previously been available in the form of chemical transport model fields. Therefore, several studies have been undertaken to validate the observations of the TES instrument. There have been three papers focused on TES tropospheric ozone validation since the instrument became operational [Worden et al., 2007; Nassar et al., 2008; Richards et al., 2008]. Of these studies, two have focused on direct comparison of TES retrieval profiles to the closest and most time-coincident ozonesonde launches [Worden et al., 2007b; Nassar et al., 2008]. In order to compare to the TES profiles, the authors applied what is known as the TES observation operator (averaging kernel) to the ozonesonde retrievals. This is done to account for differences in vertical resolution between the sondes and the TES data. Applying the TES observation operator provides a way of viewing the ozonesonde data in terms of how its measurements would look with TES vertical
4 resolution and sensitivity. For the UT and lower troposphere (LT), Worden et al., using the original TES nadir retrieval version (V001), determined that TES has an UT and LT ozone mixing ratio bias of 16.8 ppbv and -2.6 ppbv, respectively, in the mid-latitudes. Nassar et al., using a newer version of the TES data retrievals (V002), determined ozone biases ranging from 2.9-10.6 ppbv for the UT and 3.7-9.2 ppbv for the LT. The comparisons with the sondes provide an estimate of the TES bias relative to sondes, but it is important to note that lack of temporal and spatial coincidence may have a significant impact on these types of analyses. The other validation paper [Richards et al., 2008] focuses on comparison of TES retrievals (V002) to Differential Absorption LiDAR (DIAL) column ozone measurements. These measurements were made from a NASA DC-8 which flew a series of flight legs over the North Pacific during the NASA Intercontinental Chemical Transport Experiment Phase B (INTEX-B) campaign (see below). Since DIAL curtains provide much more spatial data than ozonesondes, the authors were able to use tighter coincidences in their comparison. All DIAL ozone profiles fell within TES footprints and were within three hours of the TES overpass in time. Results of the study showed that the TES instrument has a positive ozone mixing ratio bias in the lower (surface to 500 hPa) and upper troposphere (500 to 300 hPa) of 4.71 and 9.05 ppbv, respectively.
1.2 The NASA INTEX-B Program
The NASA INTEX-B program involved an extensive field campaign over the North Pacific during the spring of 2006. A main focus of the campaign was analysis of ozone
5 and ozone precursor transport from Eastern Asia to the west coast of North America [Singh et al., 2006]. Independent ozone measurements were made over the North American portion of the study region with ozonesondes (the INTEX Ozonesonde Network Study 2006 (IONS06) [Thompson et al., 2007a and 2007b]), while in-situ FASTOZ aircraft measurements and DIAL curtains were made along flight tracks out of Hawaii and Anchorage (NASA DC-8) and in-situ ozone was measured onboard the NCAR C-130 flying out of Seattle. During the field campaign, TES special observation retrievals were planned to coincide with sonde launches and aircraft spirals.
1.3 Synoptic Dynamics and Upper Tropospheric ozone
The major goal of this study is to illustrate that TES observations in the upper troposphere effectively capture synoptic dynamic variations in ozone. Where past validation studies have examined correlations between large data sets of TES-observed ozone profiles and either ozonesonde, DIAL, or aircraft spiral ozone profiles without any significant consideration of large-scale weather patterns, this study is specifically focused on if, and how well, TES-observed upper tropospheric ozone responds to changes in the synoptic-dynamic conditions along each overpass. In order to assess the synopticdynamic state of the upper troposphere, the study makes use of two separate tracers for stratospheric influence: (1) lack of water vapor presence, or atmospheric aridity and (2) enhanced potential vorticity. Research has shown that the amount of water vapor present in the upper troposphere can be used to determine regions of stratospheric influence [Gray, Bithell,
6 and Cox, 1994]. Since water vapor mixing ratios are very low in the stratosphere and very high within the planetary boundary layer, portions of the upper troposphere which exhibit relatively low water vapor mixing ratios are most often the result of downward vertical motion and exchange from the stratosphere and portions which exhibit relatively high water vapor mixing ratios are most often the result of upward vertical motion and exchange from the boundary layer or lower free troposphere. Furthermore, Moody and Wimmers found that the location of the tropopause break and associated stratospheretroposphere exchange (STE) correspond with high-magnitude specific humidity gradients (GOES Layer Average Specific Humidity, GLASH) in the upper troposphere derived from Geostationary water vapor imagery [see Wimmers and Moody, 2001; Wimmers et al., 2003; Wimmers and Moody, 2004a; Wimmers and Moody, 2004b]. Potential vorticity (PV) has also been used as a tracer for air of stratospheric origin in the upper troposphere in many studies in the literature [Ancellet et al., 1994; Beuermann et al., 2002; Stohl et al., 2003; Pan et al., 2007]. PV is both a dynamic and a thermodynamic quantity; it is the product of atmospheric static stability which is a measure of the vertical temperature gradient, and the fluid circulation or absolute vorticity which is a measure of horizontal changes in wind velocity (speed or directional shear). PV increases rapidly in the lower stratosphere due almost entirely to the static stability component. Because PV is conserved under adiabatic conditions, as air moves out of the stratosphere into regions with lower static stability, the absolute vorticity of the air parcel must increase. This occurs in conjunction with speed shear in the vicinity of the jet stream, or with directional shear as air moves into troughs or meridional streamers.
7 1.4 Research Motivations and Study Goals
This paper has two objectives. The first is to demonstrate the ability of TES to observe synoptic-dynamic variation in upper tropospheric ozone. This is a largely qualitative assessment. The second is to derive a quantitative correlation between TESobserved UT ozone and synoptic-dynamical tracers in order to produce statistical retrievals of layer-averaged UT ozone across the INTEX-B study region. This multisensor upper troposphere ozone product (MUTOP) has the spatial and temporal resolution of the geostationary GLASH imagery, allowing for far greater spatial coverage than individual TES overpasses while gaining TESs abilities to characterize UT ozone. The motivations for this work are as follows: (1) the need for a unique TES tropospheric ozone validation method in which issues from a lack of temporal and spatial coincidence do not overly limit the quality of results, (2) the importance of knowing natural background ozone levels in transport regions in order to accurately characterize anthropogenic pollution transport (as in INTEX-B), (3) the need for knowledge of the spatial distribution of UT ozone as a function of time in order to assess clear-sky radiative forcing impacts, (4) the need for a synoptic-dynamic perspective when examining atmospheric chemistry, and (5) the need for integrative studies which interpret polar-orbiting satellite remote sensing measurements in the context of geostationary satellite retrievals and numerical model fields.
2. Data Sets 2.1 GOES Layer Average Specific Humidity Fields
8
GOES Layer Average Specific Humidity (GLASH) is a derived-product image of upper-tropospheric specific humidity based on the GOES (Geostationary Operational Environmental Satellites) Imager 6.7 micron water vapor channel. By demonstrating that the 6.7 micron channel brightness temperatures respond to specific humidity, atmospheric temperature, and satellite viewing angle, Wimmers and Moody [2001] were able to retrieve an image that shows just the contribution of layer average specific humidity by correcting for the way the 6.7 micron channel responds to variations in both uppertropospheric temperature and satellite zenith angle. The sensitivity of the GLASH product with respect to height in the atmosphere is defined by the vertical contribution weighting function (see Figure 2.1), which exhibits maximum sensitivity at 400 hPa and considerable weighting between 500 hPa and 300 hPa. In this study, GLASH brightness value fields derived from GOES-West 6.7 micron retrievals were used. High GLASH brightness values correspond directly to high water vapor mixing ratios in the mid-toupper troposphere, while low brightness values likewise correspond to low water vapor mixing ratios in the same layer. The GOES-West domain was chosen to correspond to the study area of the INTEX-B program. GLASH fields are used from April 16th, 2006 to May 16th, 2006, covering the majority of the INTEX-B study time period. GLASH fields were available throughout this period at intervals of 6 hours (00, 06, 12, and 18 UTC) in entirety.
2.2 GFS Model Potential Vorticity Fields
9 The Global Forecast System (GFS) model is a global numerical weather prediction model. This study makes use of GFS model analysis (0 hour forecast) grids and GFS calculation of potential vorticity (PV) from model temperature and wind fields. The version of the GFS model used in this study has a horizontal resolution of 1 degree by 1 degree and 26 vertical levels. Within the troposphere, these vertical levels are separated by 50 hPa each. GFS model PV was archived throughout the INTEX-B time period (from April 16th, 2006 to May 16th, 2006) at 6 hour intervals which correspond in time with the GLASH products (00, 06, 12, and 18 UTC). To achieve spatial correspondence with GLASH as well, only GFS PV data from within the GOES-West domain were used in this study.
2.3 TES Special Observation Step and Stare Ozone Profiles
TES special observation products known as nadir step and stare (SS) observations were performed throughout the INTEX-B time period. Every SS atmospheric curtain consists of 125 nadir column ozone measurements along the Aura satellite track, with each retrieved profile separated from its neighbor by ~45 km [Osterman et al., 2007]. The latitude range of the TES SS curtains corresponds well with the GOES-West latitude domain, with measurements from 15 degrees north to 65 degrees north. TES ozone retrievals are calculated using an optimal estimation approach based on contemporaneous temperature, water vapor, and ozone retrievals from TES [for details see Rodgers, 2000; Worden et al., 2004; Bowman et al., 2002; Bowman et al., 2006] and a priori values
10 based on a climatology from the Model of Ozone and Related Chemical Tracers (MOZART).
3. Methodology 3.1 Averaging of Equivalent Atmospheric Layers between Data Sets
In order to assess the correlation between TES ozone and the dynamical tracers of GLASH brightness values and GFS modeled PV, it was first necessary to layer average the three data sets such that all three are representative of the same general layer within the atmosphere. Since the GLASH products weighting function peaks at 400 hPa and has considerable weight between 500 and 300 hPa (see Figure 2.1), and since the purpose of this study is to examine the spatial and temporal evolution of ozone distributions in the upper troposphere, a simple step-function layer-averaging approach was used (as below):
PV = ( PV 300 + PV 400 2 + PV 500) 4 TES O 3 = ((OZ 287 + OZ 316) 2 + (OZ 383 + OZ 422) + OZ 511) 4
(2.1) (2.2)
The simplified layer averaging was chosen to represent the majority of the peak of the GLASH contribution weighting function while reducing computational time and eliminating excess weighting in the stratosphere. TES ozone layer-averaging involves a slightly more complicated equation in order to establish discrete measurement values closest to 300, 400, and 500 hPa. Since TES profiles almost always display a monotonic increase in ozone mixing ratio with height in the atmosphere (especially above 500 hPa),
11 it is unlikely that any bias has been introduced through this methodology. The abovedescribed layer-averaging approach results in GFS PV and TES ozone weightings which closely correspond to the atmospheric layer of highest weighting for the GLASH product retrievals. Therefore any further references to the upper troposphere in this paper will be specifically referring to the atmospheric layer from 500 to 300 hPa (see Figure 2.2 for example of TES ozone and GFS PV averaged layer). This is the same layer which Richards et al. [2008] used to define the upper troposphere in their DIAL-based TES validation paper published earlier this year.
3.2 Regression Analysis
To determine the statistical correlation between layer-average TES ozone measurements and layer-average GFS model PV estimates and layer-average specific humidity (i.e., GLASH brightness value retrievals), regression analysis was used. To match the data for regression analysis, each TES ozone measurement was matched to its spatially-closest GLASH brightness value and GFS PV counterparts. Since the GFS model resolution is significantly lower than that of TES nadir retrievals, in many cases more than one layer-average TES ozone value was matched with the same PV value along an overpass. Because this study is especially concerned with the impact of synoptic dynamics on upper tropospheric ozone, it was important to make sure that excessive error from a lack of temporal coincidence did not make its way into the data set. Since GLASH products and GFS model fields were available at 6 hour intervals, the greatest possible
12 time separation for TES overpasses from one of these fields was 3 hours. To reduce error from time separation and the potential for sampling entirely different air masses, a temporal coincidence criterion of 1.5 hours was established for the regression data sets. Although this likely still causes some error in comparison, it is a far tighter coincidence requirement than has been used in any other TES validation studies (see section 1.1). Based on these criteria, the final data set consists of 30 TES step and stare curtains over the North Pacific and within the GOES-West domain. These 30 TES profiles contain 3134 profiles, and thus the same number of individual layer-average TES ozone measurements. In order to set aside data for error analysis use later, the data set was split approximately in half (each profile in every other overpass in order of date was placed in one data set, and the others were placed into a separate data set). The final result was one data set of 1547 corresponding TES ozone, GFS PV, and GLASH BV values and another set of 1587 values in the opposing data set. Three different regressions were calculated based on each data set: (1) a quadratic regression with TES layer-average ozone as the dependent variable and GLASH BV as the independent variable, (2) a quadratic regression with TES layer-average ozone as the dependent variable and GFS layer-average PV as the independent variable, and (3) a multiple linear regression with TES layer-average ozone as the dependent variable and GLASH BV and GFS layer-average PV as the independent variables. The basic form of these regression equations is provided below:
(1) O3 = a GLASH BV + b GLASH BV + c (2) O3 = a PV + b PV + c
2
2
(2.3) (2.4)
13 (3) O3 = a GLASH BV + b PV + c (2.5)
By using the regression equations, which provide a numerical representation of the relationship between TES ozone and the synoptic dynamical tracers of GLASH and PV, GOES-West domain GLASH and GFS PV fields were converted to fields of layeraverage upper tropospheric ozone mixing ratios. This application of the regression equations thus allowed for the creation of three different statistically-derived layeraverage upper tropospheric ozone products available at 6 hour intervals. In order to understand how well the derived regression equations represent TES measurements of layer-average upper tropospheric ozone mixing ratios, the regression equations from data sets 1 and 2 were applied to the GLASH BV and GFS PV values from the opposing data set. This allowed for 2 independent measures of regression error based on the 2 data sets (a sort of self-validation) as well as for evaluation of the robustness of the regression between data sets. Residuals were also examined to see if there was any systematic bias to the regression error. Residuals were plotted over the statistically-derived upper troposphere layer-average ozone products in order to visually evaluate residual correlations with air mass boundaries and cloudy regions.
4. Results 4.1 Case Study Evaluations and Visual Correlations
14 Based on visual analysis of GLASH BV and GFS PV fields with TES overpass overlays (see Figure 2.3), it is clear that TES upper troposphere layer average ozone retrievals are responding to dynamical variations in the UT. As can be seen in these images, TES-observed ozone is elevated in and around the edges of upper-level troughs, dry-air streamers, and cut-off lows (which correspond with low GLASH BVs and high GFS PV values). This same correspondence was observed for all of the TES overpasses which occurred during the INTEX-B period and fell within the GOES-West domain. Figure 2.3 also shows the great amount of variation in the TES ozone retrieval between overpasses. In the first overpass (May 13th at 12 UTC), TES measurements follow the trough axis and mostly display two distinct UT ozone regimes: the southern portion with free tropospheric air and an absence of UT ozone and the northern portion with considerable stratospheric influence and elevated UT ozone. This situation is not unexpected, and fits well with the climatological mean within the Northern Hemisphere mid-latitudes during spring. In the second overpass (May 14th, 00 UTC), the TES instrument observed the very western edge of the same trough feature that it passed through 12 hours earlier. In this case, TES measured slightly heightened UT ozone mixing ratios along the western edge of the trough, then transitioned back into a warm, moist, ozone poor UT within the upper-level ridge. In the last example (May 15th at 12 UTC), TES observed an even greater climatological anomaly and still captured the dynamical variation in UT ozone quite well. By this time, the upper-level trough observed during the last two overpasses has evolved and a unique set of cut-off low features have formed at its base. Moving generally from south to north, TES first observed moist, free tropospheric air with little ozone present in the UT, then
15 encountered the western side of the cut-off low feature. Here TES observed heightened UT ozone mixing ratios, corresponding with signatures of atmospheric aridity and high PV in the two products. Ozone mixing ratios decreased on the northern side of the cut-off low, before increasing again as TES passed over the remnants of the upper-level trough.
4.2 Regression Results
Based on the regressions between layer-average TES ozone and the two synoptic dynamical tracers, three different numerical relationships were determined for each data set half (see Figures 2.4, 2.5, and 2.6 for regression plots). For the first data set half, the following regression equations were determined for (1) quadratic TES ozone as a function of GLASH BV, (2) quadratic TES ozone as a function of GFS PV, and (3) multiple linear TES ozone as a function of GLASH BV and GFS PV:
(1) O3 = 12.085 GLASH BV + 0.02808 GLASH BV + 1341.4 (ppbv) (2) O3 = 44.8 PV 4.47 PV + 47.4 (ppbv)
(3) O3 = 1.0899 GLASH BV + 20.187 PV + 257.3 (ppbv)
2
2
(2.6)
(2.7) (2.8)
For the second data set half:
(1) O3 = 18.154 GLASH BV + 0.04482 GLASH BV + 1891.3 (ppbv) (2) O3 = 48.8 PV 4.60 PV + 47.9 (ppbv)
2
2
(2.9) (2.10)
16 (3) O3 = 1.2016 GLASH BV + 19.333 PV + 281.1 (ppbv) (2.11)
The average R2 value between the two data sets was 0.65 for the quadratic GLASH regression, and 0.72 for the quadratic PV regression, and 0.76 for the multiple-linear regression (for specific R2 values, see Tables 2.1 and 2.2). The average mean absolute error (MAE) and root mean square error between the two data sets were 14.4 ppbv and 19.3 ppbv for the multiple-linear regression, 16.8 ppbv and 23.3 ppbv for the quadratic GLASH regression, and 15.9 ppbv and 21.1 ppbv for the quadratic PV regression, respectively (for specific MAE and RMSE values, see Tables 2.1 and 2.2). Clearly, the multiple-linear regression produces the best product, explaining more of the variance with less error. All images presented in this paper are on based the regression equations calculated from data set half 2 due to the slightly more robust fit to the TES observations.
4.3 Statistical Retrieval Ozone Products
Images produced from the regression equations above represent the spatial distribution of upper tropospheric ozone mixing ratios at 6 hour time intervals (see Figure 2.7 and the supplemental animation link). There are several major advantages to this derived product: (1) it is calculated from 2 operational fields (GLASH BV and GFS PV). This means that the multi-sensor derived UT ozone product can be created and used in near real-time; (2) the product provides high spatial resolution information on the distribution of dynamically driven ozone in the UT; and (3) the product, created at 6 hour
17 intervals, allows one to track the (sometimes very rapid) movement of this dynamicallydriven ozone in the UT. An analysis of residual magnitude revealed no systematic heightened regression error for the combined GLASH and PV multi-sensor product. Evaluation was not carried out for the GLASH-only and PV-only products, as the combined product is clearly superior in terms of TES retrieval correlation and error. While no systematic heightened errors are noted, as can be seen in Figure 2.8, there are certain cases in which residual magnitude exhibits a positive correlation with UT ozone gradients (i.e. air mass boundaries) and/or regions of cloud presence. A reason for the lack of systematic elevated error under these conditions may be due to the limited TES data set, in which there are too few TES/MUTOP comparisons near air mass boundaries and within clouded regions to identify a statistically significant relationship.
5. Discussion 5.1 Product Performance and Utility
The concept behind the multi-sensor UT ozone product (MUTOP) derived in this study is similar in its methodology to the EUMETSAT Multi-sensor Precipitation Estimate (MPE) product [Heinemann, Lattanzio, and Roveda, 2002]. This MPE product is a blending of geostationary IR retrievals with microwave sensors on polar-orbiting satellites through statistical correlation (just as MUTOP is a blending of GLASH and TES). The authors tout the utility of this product based on its high temporal and spatial resolution (geostationary component) and improved accuracy (polar-orbiting component).
18 This logic of blending the advantages of two remote sensing platforms is at the core of the MUTOP product, which allows for using the greater temporal and spatial coverage of geostationary specific humidity while gaining TESs ability to characterize UT ozone. Regression analysis suggests that approximately 70-80% of variation in upper tropospheric ozone (as observed by TES) can be explained by the two synoptic dynamic tracers of upper tropospheric specific humidity (GLASH) and potential vorticity (GFS PV) during spring 2006 within the INTEX-B GOES-West domain. This finding suggests that ozone in the UT is dominantly controlled by dynamical processes. This leads to the conclusion that to correctly characterize anthropogenic influences, it is first necessary to improve estimates of dynamically-driven variation. Since such a large portion of UT ozone is driven by synoptic dynamical processes, statistical GOES domain retrievals of ozone verified by comparisons to TES data are able to provide reasonably accurate estimates of average ozone mixing ratios in the UT. The significant scatter around each of the three regression lines suggests that other factors, such as in-situ chemical production/destruction of ozone and long-range transport of ozone from polluted boundary layers, are certainly important influences on UT ozone as well. As part of the 2003 Intercontinental Transport and Chemical Transformation (ITCT 2K2) study, for example, Cooper et al. [2004] addressed the capacity of warm conveyor belt systems as transporters of polluted boundary layer air from the Asian continent into the North Pacific mid-and-upper troposphere. They found that 44% of warm conveyor belt air masses passed through the lower troposphere, leading to heightened ozone mixing ratios in warm conveyor belts over the North Pacific. These findings may help to explain limited MUTOP performance under moist, low PV
19 conditions, as warm conveyor belt air masses and subtropical free tropospheric air masses have similar moisture and potential vorticity characteristics, but could display very divergent ozone chemistries. Overall, the MUTOP product is not as effective at characterizing ozone mixing ratio variability within moist air masses, as synoptic dynamics have little effect on UT ozone mixing ratios in these regions. Although MUTOP has certain limitations as noted above, it provides a capacity for upper tropospheric ozone estimation within the GOES domain that is a large step beyond what is currently available in the form of spatially-derived daily ozone plots based on TES retrievals alone (see Figure 2.9). This figure shows what was previously available in the form of a TES-derived daily ozone product within the GOES-West domain (bottom right of panel) in comparison to 5 corresponding images produced from the methodology described in this paper. The previously available image is a TES Level 3 Daily Ozone plot for April 24th, 2006, created by averaging the 464 hPa and 316 hPa plots. The Level 3 product is created by combining TES global survey overpass measurements from a 26 hour period (in this case from ~12 UTC on the 24th to ~12 UTC on the 25th) and spatially interpolating between overpass measurements. The TES Level 3 Daily Ozone products are quite useful as browse products for identification of regions of interest for further research and investigation, but are limited in several ways with regard to accurate synoptic- and meso-scale dynamical representation of ozone in the upper troposphere. First, the Aura satellite and the TES instrument are in sun-synchronous polar orbit, meaning that the satellite moves from east to west. This is opposite the movement of large-scale weather in the Northern Hemisphere mid-latitudes, and thus L3 ozone products may be distorting synoptic
20 features in the UT. Second, the L3 product is presented as a daily plot, but the measurements used in the interpolations are not all from the same times. Even measurements very near one another spatially can be from entirely different times of day (ascending and descending nodes). Therefore the products do not represent a daily mean, but instead a stitched-together representation of UT ozone from different times during a 26 hour period. Lastly, simple spatial interpolation is used to produce these daily plots, therefore UT ozone enhancements associated with small-scale features that fall between overpasses (e.g., cut-off lows) may be completely missed and any that features that are not continuous between overpasses may be incorrectly characterized in the spatial averaging. A comparison of five MUTOP images corresponding to the 26 hour interval over which the TES Level 3 Daily Ozone map is stitched together is provided in Figure 2.9. Here one can see that MUTOP provides a greater context for UT ozone observations than has been available previously in these L3 products. MUTOP is available at higher temporal resolution (every 6 hours) and is able to demonstrate synoptic and mesoscale features in the UT which are either missed or poorly resolved in the L3 products. For example, from the April 24th-25th case in Figure 2.9, one can see that the L3 product fails to capture certain features in the UT and has considerable problems with respect to meridional streamers due to the nature of polar-orbiting remote sensing, such as in the feature just off the Pacific Northwest and California coastlines. Also, the disadvantages of spatial interpolation become evident, as can be seen in the L3 products poor characterization of two distinct synoptic ozone enhancements over Eastern Canada in Figure 2.9. The TES L3 Daily Ozone plot suggests that these two features are connected,
21 while MUTOP sees the clear distinction between them in terms of dynamics. Once the separate panels of the multi-sensor product are viewed, it is much easier to understand the features characterized in the TES L3 product. Overall MUTOP allows for better interpretation of individual TES overpasses and associated Level 3 Daily Ozone plots and clearly demonstrates an improvement in UT ozone characterization over the GOES-West domain. Viewing of these two product types together may be most useful, as doing so allows for identification of TES-observed ozone enhancements in the UT which are not dynamically-controlled (i.e. are not present in MUTOP imagery).
5.2 Product Type Advantages and Disadvantages
Between the statistically-derived GOES retrievals of upper tropospheric ozone presented in this paper, it is clear that the product based on the multiple-linear regression (GLASH and PV) is most representative of the actual TES retrievals over the INTEX-B study region. This multiple regression was able to explain the greatest percentage of the variance in TES-observed UT ozone mixing ratios (highest R2 value) and produced the lowest error (both MAE and RMSE) of the three products. However, each product has its own advantages and disadvantages with respect to estimating layer-averaged upper tropospheric ozone. The product derived from only the GLASH relationship has the advantage of very high spatial resolution, making it capable of distinguishing smaller features in the upper troposphere, such as elongated troughs and dry air streamers. The GLASH-derived
22 product also has the advantage of being an actual remotely-sensed observation of the state of the atmosphere. A significant disadvantage of the GLASH-derived product, however, is that GLASH retrievals can not detect dry layers (e.g. tropopause folds) that lie beneath moist layers in the upper troposphere due to sensor saturation of retrievals under these circumstances. The PV-derived product, on the other hand, depends on a model of the atmospheric state. Although temperature fields in the upper troposphere are modeled fairly well, modeled wind fields are much more uncertain. If the model is incorrect in its placement of synoptic features, and mesoscale details, this could lead to substantial ozone estimation error in some cases. The PV-derived product is also limited by the horizontal spatial resolution of the GFS model (1 degree by 1 degree). The lack of fine-scale resolution can result in averaging out of mesoscale or smaller-scale synoptic features in the upper troposphere, resulting in both over and under-estimation of PV in and surrounding these features, respectively. However, the PV-derived product has an advantage over the GLASH-derived product in that there is no saturation issue, and GFS modeled PV does capture the presence of heightened potential vorticity associated with stratospheric intrusions (the existence is well modeled, even if the physical position and vertical extent is not always correctly captured). This difference results in smoother ozone gradients along the edges of synoptic-scale dynamical features in the PV-derived product. The combined product which takes into account both GLASH BV and GFS PV combines the advantages and disadvantages of both products, but results in a better fit to the actual TES observations. This improvement is most likely due to the fact that GLASH
23 is able to provide better estimates of ozone in regions of small-scale features and GFS PV is able to capture the smoother gradients associated with STE.
5.3 Conclusions and Recommendations for Future Work
In order to establish the legitimacy of the product derivation described in this paper, independent validation of MUTOP needs to be carried out. Validation will also provide an indirect assessment of TESs performance in UT ozone retrieval, since MUTOP estimates are derived from TES-tracer relationships. The companion to this paper provides a validation of MUTOP using ozonesonde data from the INTEX-B time period and within the GOES-West domain and compares these validation results to those of past TES validation studies [Felker et al., 2008]. Based on the results of this derivation paper, in the form of image analysis and TES correlations, MUTOP provides visualization of UT ozone that surpasses what is currently available in the form of TES Level 3 Daily Ozone maps (see section 2.5.1) and chemical transport models. Although direct comparison to chemical transport models was not possible in the confines of this study, the companion validation paper results [see Felker et al. ,2008] suggest that MUTOP displays a closer correlation to UT ozonesonde measurements [Tarasick et al., 2007]. MUTOP provides a dynamical background context of transient ozone enhancements and overall highly variable ozone mixing ratios in the UT that is well-resolved temporally (every six hours; see supplementary directory of all available imagery), allowing for the viewing of synoptic and mesoscale (dry air
24 streamers and cut-off lows) dynamical features and their evolution through image looping. While the derived products of layer-average UT ozone created as part of this study represent the TES retrievals during the INTEX-B campaign well, there are some limitations which should be addressed through further study. First, the regression relationships need to be tested in different seasons (other than spring) in order to assess robustness. Since STE is believed to be at a maximum in the northern mid-latitudes during spring, the current regression relationship may lead to overestimates in other seasons [Appenzeller, Holton, and Rosenlof, 1996]. It is quite plausible that different regression relationships would need to be derived for each season in order to accurately characterize UT ozone. Second, the regression relationships also need to be tested over the same period in another year in order to test the robustness with respect to inter-annual variation in UT ozone. Third, it would be beneficial to assess the correlation between high-magnitude regression residuals and a tracer for anthropogenic pollution such as CO. Doing this would help in understanding the causes for the large degree of scatter around the regression curves.
25
Table 2.1 Regression and Error Analysis results
Data Set Half #1 GLASH + PV 0.74 14.4 ppbv 19.4 ppbv GLASH 0.61 16.8 ppbv 23.1 ppbv PV 0.69 15.6 ppbv 20.8 ppbv
R2 MAE RMSE Data Set Half #2
R2 MAE RMSE
GLASH + PV 0.77 14.4 ppbv 19.2 ppbv
GLASH 0.68 16.7 ppbv 23.5 ppbv
PV 0.75 16.1 ppbv 21.4 ppbv
26
Figure 2.1 Contribution weighting function for GLASH product retrievals (solid line), where amplitude provides the relative weight contribution from each pressure level. The simplified layer-average, which is applied to GFS PV and TES Step and Stare profiles in this study, is plotted (dashed line) to show that it represents the peak of GLASH weighting, without over-emphasis of the stratosphere.
27
Figure 2.2 Plot of TES Step and Stare ozone profile for May 10th, 2006, in which the ozone volume mixing ratio (VMR) is plotted in color shading (ppbv). The location of the jet stream (black contours...