2.1. The MLS Measurement System
 MLS observes thermal microwave emission by the atmosphere in five spectral regions from 115 GHz to 2.5 THz. The temperature and GPH measurements described in this paper are taken from observations near the 118-GHz O2 spectral line and the 234-GHz O18O spectral line. MLS looks forward from the Aura spacecraft and scans the Earth's limb vertically from the ground to ∼90 km every 24.7 s. The vertical scan rate varies with altitude, with a slower scan providing greater integration time in the lower regions (∼0–25 km). The MLS vertical scans are synchronized to the Aura orbit such that vertical scans are made at essentially the same latitudes each orbit, with 240 scans performed per orbit (∼3500 scans per day).
 The geophysical products described in this paper are retrieved from calibrated MLS limb radiances using MLS version 2.2, level-2 software [Livesey et al., 2006]. Temperature and GPH are reported on 12 levels per decade from 1000–22 hPa, 6 levels per decade from 22–0.1 hPa and 3 levels per decade from 0.1–0.001 hPa. 240 profiles are retrieved per orbit, spaced at 1.5° great circle angle along the suborbital track at an essentially fixed set of latitudes. The MLS v2.2 data quality document [Livesey et al., 2007] gives more information on the format and contents of the MLS data files.
 Version 2.2 (v2.2) is the second public release of MLS data and has been used to process the incoming data stream since March 2007. Reprocessing of data collected since MLS became operational in August 2004 is also in progress using the v2.2 algorithms. For this validation effort, 93 days were processed with v2.2 algorithms with an emphasis on days of particular interest for validation of various MLS data products.
2.2. MLS Temperature and GPH Measurements
 The MLS measurement system [Livesey et al., 2006] uses optimal estimation theory [Rodgers, 2000] to retrieve an atmospheric state vector. The initial phase retrieves temperature/GPH/ptan using radiances from the vicinity of O2 spectral lines at 118 GHz and 234 GHz. The state vector contains temperature on 47 pressure levels and GPH on the 100-hPa reference surface. The temperature profile and the single reference GPH level determine the rest of the GPH profile through assumed hydrostatic balance. The state vector also includes the tangent pressures (ptan) of the limb observations, and ptan, temperature and refGPH (GPH at 100 hPa) are interrelated through the instrument scan model and assumed hydrostatic balance, and are simultaneously retrieved.
 The retrieval obtains temperature information from both “saturated” (optically thick) and “unsaturated” (optically thin) limb radiances. Saturated radiances are a weighted average of the black-body emission over some layer of the atmosphere, so each radiance provides temperature information about some layer. Unsaturated radiances near pressure-broadened O2 lines provide estimates of limb-tangent pressure from line shape. Pressure as a function of limb-pointing height (from the antenna/spacecraft scan model) determines temperature through assumed hydrostatic balance.
 The a priori temperature used in the MLS v2.2 retrieval between the surface and 1 hPa is the GEOS-5 analysis, discussed in section 3.1. At levels above 1 hPa, CIRA86 climatology [Fleming et al., 1990] is used. There is an approximately 5-km layer at 1 hPa over which the two sources of a priori transition smoothly. The assumed a priori temperature precision is piecewise linear in log pressure: 5 K at 1000 hPa, 10 K at 220 hPa and 20 K at 68 hPa and lower pressures. The a priori 100-hPa GPH is 16 km and its precision is 5 km. Assumed a priori precisions are chosen conservatively (loosely) so that a priori information is weighted less heavily than information from measured radiances.
2.3. Data Usage Guidelines
 In addition to describing file formats and contents, the data quality document [Livesey et al., 2007] also gives detailed instructions on the proper use of MLS data products. Each MLS Level 2 data point is reported with a corresponding precision. If the retrieval does not improve precision by at least a factor of two from its a priori value, then it has failed to extract sufficient additional information from radiances and retrieved values will be significantly influenced by a priori. Such precisions are set negative, and these data are not recommended for use in scientific studies. There is further discussion of temperature and GPH precision in section 2.5.
 Three additional data flags are provided for every vertical profile. “Status” is a bit field indicating operational abnormalities or problems with the retrievals. The meanings of its bits are given by Livesey et al. . Profiles for which Status is an odd number should not be used in any scientific study. Nonzero but even values of Status indicate that the profile has been marked as questionable by the data processing software, usually because of possible significant influence by thick clouds. At pressures of 147 hPa and lower (higher in the atmosphere), the cloud bits may generally be ignored. In the troposphere an attempt has been made to screen out radiances that have been influence by clouds, but some cloud-induced negative biases in retrieved temperature of up to 10 K are still evident, particularly in the tropics. The low-cloud bit of Status is meant to flag possible instances of such impacts, but validation work has shown this flag to be lagged 1–2 profiles from the impacted profile. Temperatures in the tropopause (316–178 hPa) should be rejected as possibly influenced by cloud if the low-cloud Status bit (Status AND 32) is set in either of the two profiles following the profile in question.
 The “Quality” field indicates the degree to which the measured MLS radiances have been fit by the Level 2 algorithms. Larger values of Quality generally indicate better radiance fits, whereas values closer to zero indicate poorer radiance fits, and thus less reliable data. The Quality reported with the v2.2 temperature is based only upon the χ2 of Band 8 radiances (radiances used in the temperature retrieval are discussed in section 2.4) and so is primarily an indication of the fit of the retrieval in the troposphere. However, low values of Quality are not consistently associated with profiles that are obviously outliers. Profiles having Quality values less than 0.6 are generally not recommended for scientific use. This threshold for Quality typically excludes ∼4% of temperature profiles.
 The “Convergence” field is a somewhat arbitrarily defined function of χ2 which can be used to flag profiles within “chunks” of ∼10 profiles in which the retrieval failed to converge well. Convergence values around 1.0 typically indicate that the retrieval has converged. Profiles with Convergence greater than 1.2 are not recommended for use in scientific studies. This screening typically rejects 2 percent of profiles, and 0.5 percent in addition to those flagged by Quality <0.6 while flagging most poorly converged profiles.
 Temperature and GPH at pressures lower than 0.001 hPa or higher than 316 hPa are not recommended for use in scientific studies. Unlike in v1.5, these levels are not, as a rule, marked with negative precision. An additional cost for departures of the retrieved-profile curvature from a priori curvature has been added to v2.2 to constrain smoothness, as described by Livesey et al. . Neither the degree to which this smoothness constraint leaks a priori information into the retrieval, nor the degree to which it extrapolates retrievals to levels where MLS lacks direct observations, have been properly accounted for in the setting of negative precision in the v2.2 algorithms.
 A large set of v2 temperature profiles with low Quality (between 0.4 and 0.55) and high (poor) convergence exists poleward of 70° latitude in autumn and early winter. Figure 1 shows such a cluster of points in data from four days in April and May of 2006, when more than half of profiles south of 70°S had Quality less than 0.6. Here, Quality and Convergence reflect the convergence failure in the final “phase” of the temperature retrieval, which combines radiances from the isotopic O18O line at 234 GHz with radiances from the vicinity of the 118-GHz O2 line. This nonconvergence may be related to poor treatment of O3 in the retrieval. In these cases, the retrieved state falls back to the generally well-behaved output of a previous “phase” that uses only 118-GHz radiances and has somewhat degraded vertical resolution in the troposphere. Biases between the temperatures of the two phases are generally less than 1 K in the stratosphere. For polar stratospheric studies, users may want to relax the Quality threshold to 0.4 to fill in missing data, but the added profiles should not be used at pressures greater than 215 hPa.
Figure 1. Temperature Quality flag for April–May 2006. In the polar autumn, south of 70°S, profiles for which the final phase of the temperature retrieval failed to converge have Quality less than 0.7.
Download figure to PowerPoint
 In summary, temperature and GPH should be used only when the associated precision is positive, Quality is greater than 0.6, Convergence is less than 1.2, Status is even and the Status low-cloud bit (Status AND 32) is not set in either of the two subsequent profiles. Status low-cloud bits may be ignored at retrieval levels with pressures less than 178 hPa.
2.4. Radiance Spectra and Radiance Residuals
 Figures 2, 3 (left), and 4 show typical MLS radiances in the vicinity of the 118-GHz O2 line and Figures 3 (right) and 5 show radiances in the vicinity of the O18O line at 234 GHz. Emission in these spectral regions is dominated by oxygen, and is used to infer limb tangent-point pressures of MLS observations and to derive MLS temperature and GPH. The O2 mixing ratio is assumed constant from the surface to 0.008 hPa [Schwartz et al., 2006].
Figure 2. (top) Sample radiances (in units of brightness temperature) from MLS Band 1, centered on the 118-GHz O2 line. Global average radiances from observations on 24 September 2004 are shown for eight scan positions with approximate tangent altitudes: ∼7.5 km (purple), ∼11 km (dark blue), ∼15 km (light blue), ∼18 km (dark green), ∼22 km (light green), ∼26 km (yellow green), ∼36 km (orange), and ∼46 km (red). This is a single-sideband radiometer, so all radiances are from below the 126-GHz local oscillator. These radiances are the primary source of MLS temperature in the stratosphere and lower mesosphere. The widths of the various MLS spectral channels are denoted by the horizontal bars. (bottom) The average fit achieved to these radiances by the MLS v2.2 retrieval algorithms. The v2.2 retrieval does not use channels 6–20 for limb-pointings with tangent pressures less than 50 hPa, so these channels do not have residuals for the four highest altitudes shown. Diamonds are radiance precisions used in the retrieval's χ2 calculations.
Download figure to PowerPoint
Figure 3. As in Figure 2, except for Band 32 (solid line) and Band 34 (dashed line) overlaid on left and Band 33 overlaid on the right. Radiances are shown for six scan positions with approximate tangent altitudes: 3.5 km (purple), 11 km (blue), 18.4 km (dark green), 22 km (light green), 50 km (yellow), and 87.9 km (red). Diamonds on bottom plots show theoretical precisions based upon radiometer noise and channel bandwidth, but these channel precisions are inflated to 1 K in χ2 calculations of the retrieval to account for systematic uncertainties.
Download figure to PowerPoint
Figure 4. As in Figure 2. Band 22 radiances cover the opaque line center of the 118.75-GHz O2 line, which is Zeeman-split by the Earth's magnetic field. Radiances are shown for five scan positions with approximate tangent altitudes: 20 km (purple), 60 km (blue), 71 km (green), 81 km (yellow-green), and 93 km (red).
Download figure to PowerPoint
Figure 5. As in Figure 2, except for Band 8. Radiances are global averages from 24 September 2004 for eight scan positions with approximate tangent altitudes: 7.3 km (purple), 11 km (dark blue), 14.7 km (light blue), 18.4 km (dark green), 22 km (green), 26 km (yellow green), 35 km (orange), and 45.5 km (red).
Download figure to PowerPoint
 MLS band 22 provides radiances within ±4 MHz of the 118-GHz line center with 100 kHz resolution. Band 1 provides radiances within ±575 MHz of the 118-GHz line center with channel bandwidths ranging from 6 to 96 MHz. Bands 32 and 34 (not shown) have 500-MHz-wide channels centered 1.75 GHz and 3.5 GHz from the 118-GHz line center.
 Band 8 has channels within ±575 MHz of the 234-GHz line center with channel bandwidths ranging from 6–96 MHz. Band 33 channel 3 has sidebands with 500-MHz passbands at 232.5 GHz and 246.9 GHz. Radiance precision for this channel is inflated to 1 K to account for systematic uncertainties. This channel is fit to within 0.1 K at the top and bottom of a typical scan, but in the UT/LS it has an average positive residual of 1.5–2 K.
 The bottom parts of Figures 2, 3, 4 and 5 show the average fits achieved to these measured radiances by the retrieval algorithms. The scatter about these averages (not shown) is generally consistent with the levels of noise seen in the radiances.
 The fits for the saturated channels of Band 1, shown in Figure 2, are generally within ∼0.3 K. The outer channels of Band 1, which are used in the midstratosphere and above, have residuals of magnitude 0–1 K, with more variability among channels than is seen in the saturated radiances. Residuals in the unsaturated outer channels have an asymmetry about the line center which is not understood.
 Bands 32 and 34 have 500-MHz wide channels centered 1.75 GHz and 3.25 GHz from the 118-GHz O2 line for horizontal and vertical linear polarizations, respectively, and provide temperature information in the lowermost stratosphere and tropical tropopause layer. Polarization is not significant in these channels, and signals should be approximately symmetric about the line center, so these channels should be essentially quadruply redundant, but the mean residuals in presumably redundant channels differ by as much as 1 K, large compared to the 0.25- to 0.4-K estimated measurement precisions. These residuals are apparently the result of instrumental issues rather than deficiencies in geophysical modeling of radiances. The precisions of these channels have been inflated to 1 K so that the retrieval's χ2 calculation does not force the retrieved state to closely fit these systematic errors.
 Band 22 radiances and residuals are shown in Figure 4. Band 22 radiances cover the line center of the 118 GHz O2 line, which is Zeeman-split by the Earth's magnetic field [Schwartz et al., 2006]. The relative orientation of the Earth's magnetic field to the MLS R1A radiometer's field-of-view and polarization results in a pair of Zeeman components being received by Band 22 for most parts of an orbit. These radiances contribute to the temperature retrieval primarily in the mesosphere and lower thermosphere (∼0.1–0.001 hPa). The blue lines show radiances and residuals for a pointing with an average tangent height of ∼60 km (∼0.22 hPa,) where the radiance is just starting to come out of saturation at the band edges. Residuals at this level are as large as 3–4 K in the band edges, and the inability of the retrieval to fit these radiances better may, in part, result from gain compression, which distorts spectral line shapes, as is discussed in section 2.6. Unlike in the case of Band 1 radiances, Band 22 radiances are used from where they are saturated to where they are optically thin. Residuals in the highest pointings shown (∼81 km: yellow, ∼93 km: red) display a line shift due to some combination of error in the Doppler-shift correction for spacecraft-Earth relative motion and unmodeled mesospheric along-track wind, and provide an opportunity for along-track wind retrieval [e.g., Limpasuvan et al. 2005].
2.5. Precision, Scatter, and Spatial Resolution
 Each point in an MLS retrieved profile is accompanied by a “precision” estimate taken from the diagonal of the solution covariance matrix [Livesey et al., 2006]. Precision quantifies random error, expected to beat down when repeated measurements are averaged, which results from the propagation of radiometric noise and of uncertainties in virtual measurements through the measurement system. Temperature precisions range from 0.6 K in the lower stratosphere to 2.5 K in the mesosphere and to 1 K at 316 hPa. In this discussion “precision” is distinguished from “accuracy,” which is the RMS of the difference from truth. Accuracy is estimated through analysis of sources of systematic error (discussed in section 2.6) and through comparisons with correlative measurements (section 3).
 Differences between repeated measurements of similar scenes provide an empirical upper bound on precision which can be compared to the precision returned by the retrieval system. Successive profiles generally see very similar scenes but have correlation due to shared calibration data, so their difference gives an unrealistically low value for precision. Profiles exactly one orbit apart are at the same latitude and local time, separated by 21 degrees of longitude, and may provide a useful upper bound on precision when the atmosphere is zonally symmetric. Precision estimates based upon such pairs in high-latitude summer are slightly larger than those returned by the measurement system in the troposphere and lower stratosphere and a factor of ∼1.4 larger from the middle stratosphere through the mesosphere.
 Poleward of 70°S on 7 February 2005, precision inferred from differences between successive orbits is 1.5 K at 316 hPa, 1 K or less from 100 hPa to 10 hPa, 1.4 K at 1 hPa, 2.3 K at 0.1 hPa, 3 K at 0.01 hPa and 3.5 K at 0.001 hPa. Tropical orbital crossings from the same day (within 50 km in distance but 12 hours different in local time) have RMS differences of 1 K at 316 hPa, providing an even lower limit on precision at this level.
 As the 100-hPa level is the only GPH element in the state vector, GPH precision is calculated at other levels by adding the contributions of this reference level precision and the temperature profile precisions, in quadrature. The resulting GPH precision profile always has a minimum value at the reference level, a nonphysical result of the neglect of off-diagonal elements in the solution's error covariance.
 The 100 hPa level is, in fact, close to the level where line width information provides the best pointing/pressure reference and the overestimate of GPH precision at other levels is expected to be of order 10 m or less. Calculated GPH precisions, given in column 7 of Table 1, are ∼35 m from 316 hPa to 100 hPa, 44 m at 1 hPa, 110 m at 0.001 hPa.
 The MLS retrieval algorithms operate in a two dimensional, tomographic manner [Livesey and Read, 2000; Livesey et al., 2006] that permits the direct modeling of line-of-sight gradients. Two dimensional averaging kernels [Rodgers, 2000] describe both vertical and horizontal resolution. Figure 6 shows the vertical temperature averaging kernels which result from the horizontal integration of the 2-D kernels. The vertical resolution of the MLS temperature measurement, taken to be the full width at half maximum of these averaging kernels, is 5.3 km at 316 hPa, 5 km at 100 hPa, 3.5 km at 32 hPa, 4 km at 10 hPa, 8 km at 1 hPa, 9 km at 0.1 hPa, 14 km at 0.01 hPa and 15 km at 0.001 hPa. In the along-track horizontal direction (not shown), the temperature data have single profile resolution (∼165 km) through most of the profile, degrading to 185 km at 0.01 hPa and to 220 km at 0.001 hPa. The cross-track horizontal resolution is defined by the horizontal width of the MLS field of view. For the 240-GHz radiometer, which provides information in the troposphere, this width is ∼6 km, and for the 118-GHz radiometer, which provides information from the tropopause upward, it is ∼12 km.
Figure 6. MLS v2.2 temperature vertical averaging kernels resulting from horizontal, along-track integration of 2-D averaging kernels for 35°N, September climatology. Individual colored lines show the contribution of atmospheric temperatures at each level to a given MLS retrieved temperature, with the retrieval level marked by a plus sign of the same color. The full width at half maximum (vertical resolution, in kilometers) is shown by the thick black dashed line. The solid black line shows the integrated area under the kernels as a function of MLS retrieval level. Where the integrated area is close to unity, the majority of the information comes from the atmosphere. Lower values are associated with increased contributions from a priori information.
Download figure to PowerPoint
2.6. Accuracy and Systematic Error Budgets
 Systematic uncertainties arise from instrumental issues (e.g., radiometric calibration, field of view characterization), spectroscopic uncertainty, and approximations in the retrieval formulation and implementation. This section summarizes the relevant results of a comprehensive quantification of these uncertainties that was performed for all MLS products. More information on this assessment is given in Appendix A of Read et al. .
 For each identified source of systematic uncertainty, the impact of a 2-σ perturbation of the relevant parameter upon MLS measured radiance and/or pointing has been estimated. For primary sources of systematic uncertainty, a full day of simulated, cloud-free radiances is generated for each perturbation, and these perturbed measurements are run through the routine MLS v2.2 processing algorithms. Comparison of these results with those of a retrieval on unperturbed radiances identifies both resulting systematic bias and scaling as well as additional scatter in the retrieval results. The extent to which scatter can be expected to average down is estimated to first order by these “full up studies” through their separate consideration of the bias and scatter each source of uncertainty introduces into the data.
 The difference between the retrieved product in the unperturbed run and the original “truth” model atmosphere is taken as a measure of uncertainties due to retrieval formulation and numerics. Another retrieval of the unperturbed radiances is performed with 3 K added to the temperature a priori to test sensitivity to its value.
 The impact of an additional set of minor systematic uncertainties has been estimated through calculations based on simplified models of the MLS measurement system Read et al. . Unlike the “full up studies”, these calculations only provide estimates of gain uncertainty (i.e., possible multiplicative error) introduced by the source in question.
 Figures 7, 8, 9 and 10 summarize the results of this quantification for temperature and GPH. These show the magnitudes of expected biases and additional scatter the various errors may introduce into the data, and should be interpreted as 2-σ estimates of their likely magnitude.
Figure 7. The estimated impact of various families of systematic uncertainties on the MLS temperature observations. (left) Magnitude of possible biases of unknown sign and (middle) additional scatter introduced by the various families of errors, with each family denoted by a different colored line. Cyan lines denote errors in MLS radiometric and spectral calibration. Magenta lines show errors associated with the MLS field of view and antenna transmission efficiency. Red lines depict errors associated with MLS pointing uncertainty. The impact of possible errors in spectroscopic databases and forward model approximations are denoted by the green line, and the combined effect of error in retrieval numerics and sensitivity to a priori is shown in grey. The blue lines show the impact of similar “knock on” errors in other species. Finally, the typical impact of cloud contamination is denoted by the purple line. (right) The root sum squares (RSS) of all the possible biases (thin solid line), all the additional scatters (thin dotted line), and the RSS sum of the two (thick solid line).
Download figure to PowerPoint
Figure 8. The modeled contribution of gain compression to systematic temperature error. Gain compression distorts spectral features, making temperature inferred from line-width-based pressure measurements and hydrostatic balance inconsistent with those made from saturated radiances. (left) The mean difference between profiles retrieved from simulated radiances with and without gain compression. (right) Closed circles are the RMS scatter in the difference between the two retrievals. Open circles are the precision of the single-profile difference.
Download figure to PowerPoint
Figure 9. The estimated impact of various families of systematic errors on the MLS GPH observations. The description of the lines is the same as in Figure 7 for temperature.
Download figure to PowerPoint
Figure 10. The modeled systematic error in GPH due to “gain compression.” Plots are as in Figure 8. Gain compression has the largest magnitude contribution to systematic error of any effect considered, and, unlike the others shown in Figure 9, the sign of this bias is significant. A retrieval run on simulated radiances with modeled gain compression matching best current estimates has a high bias in GPH at 100 hPa of ∼140 m compared to the control run, which is similar to the observed bias between MLS and GEOS-5 GPH.
Download figure to PowerPoint
 Recent laboratory work by the MLS instrument team in estimating the impact of amplifier nonlinearity finds that observed spectrally contrasting signals are “compressed” by ∼1.5 percent when viewed against a background scene of 300 K rather than against a scene close to 0 K. The magnitude of this distortion was not recognized until late in the development of version 2.2 algorithms, so there has been no attempt to correct for its impact on retrievals. Of the sources of systematic uncertainty considered in this work, gain compression makes the largest contributions to both temperature and GPH biases, and both the magnitude and sign of the resulting biases can be estimated. Other sources of bias have been modeled by propagating the uncertainty in some parameter (e.g., radiometer pointing offset) through the retrieval system, and the signs of the resulting errors are unknown. The impact of gain compression on retrieved temperature and on GPH is shown in Figures 8 and 10, respectively. Gain compression causes vertical oscillations in MLS v2.2 retrieved temperatures between 316 hPa and 10 hPa which are strikingly similar to those seen in comparisons with correlative data in section 3. However, modeled gain compression also causes a 1–3 K high bias at levels above 10 hPa, while comparisons with correlative data generally suggest that MLS has a low bias at these levels. Thus, initial estimates are that correction for gain compression in a future version of the MLS retrieval will improve agreement with correlative measurements at lower retrieval levels but make it worse at higher levels.
 The contribution of cloud effects to the systematic uncertainty, both from the presence of clouds not thick enough to be screened out by the cloud filtering and from the loss of information through omission of cloud-impacted radiances, has been quantified by adding scattering from a representative cloud field to the simulated radiances and comparing retrievals based on these radiances to the unperturbed results. The cloud-induced effects on temperature shown in Figure 7 are estimated by considering only the cloudy profiles (as defined by the known amount of cloud in the “truth” field). Cloud is estimated here to contribute 0.2 K or less to temperature bias at 100 hPa, increasing to ∼1.5 K with a ∼3.5 K standard deviation at 316 hPa.
 The largest contributions to systematic temperature uncertainty, apart from gain compression, are from “radiometric/spectroscopic” sources and “spectroscopy/forward model” sources, shown in cyan and green, respectively, on Figure 7. The forward model contributes a bias uncertainty of ∼3 K in the upper troposphere where temperature information is primarily supplied by unsaturated radiances from Band 8, as discussed in section 2.4. Uncertainty in the O2 line width parameter results in systematic bias uncertainty of 0.5 K or less in the lower stratosphere and of 1 K or less in the troposphere. The contributions of antenna transmission and field-of-view shape uncertainties (the magenta line on Figure 7) are a ∼0.5 K systematic uncertainty bias which is nearly uniform with retrieval level. Their contribution to scatter is less than 0.3 K.
 Over the range (316–0.001 hPa) of retrieval levels recommended for scientific use, this study indicates a bias uncertainty of up to 2–2.5 K between 100 hPa and 0.01 hPa, of up to 5 K at 316 hPa, and up to 3 K at 0.001 hPa. Additionally, gain compression contributes a generally positive bias to temperature between −2 K and 5 K with oscillatory vertical structure. The aggregate contribution to scatter is ∼1 K between 100 hPa and 0.01 hPa, increasing to ∼4 K between 100 hPa and 316 hPa and to ∼3 K between 0.01 hPa and 0.001 hPa.
 Systematic uncertainty of GPH can be broken into uncertainties that affect GPH on the 100-hPa reference level, and those that affect GPH profiles through retrieved temperature uncertainties. The contribution of gain compression, shown in Figure 8, results in a positive bias of ∼140 m in the 100-hPa reference GPH as well as an increasingly positive bias with height. Other sources of systematic uncertainty contribute on the order of 150 m of bias of unknown sign. The largest terms are due to uncertainty in the O2 line width (∼100 m) and uncertainty in the 118-GHz radiometer field-of-view pointing offset from the 240-GHz radiometer (∼100 m,) both of which are components of “pointing” on Figure 9. Retrieval numerics contribute up to 100 m of bias and up to 250 m of scatter in mesospheric GPH.
2.7. Comparison of v2.2 and v1.5 Temperature and GPH Data
 This paper describes temperature and GPH produced by v2.2 of the MLS data processing algorithms. The previous publicly released MLS data product, v1.5, has been produced for 95 percent of the days from August 15, 2004 to the end of February 2007. Both v1.5 and v2.2 use radiances from MLS Bands 22, 1, and 32/34, centered on the 118.75 GHz O2 line and described in section 2.4. V2.2 also uses radiances from MLS Band 8 and Band 33 channel 3 near the isotopic O18O line at 236 GHz to improve resolution in the troposphere. V2.2 temperature, GPH and water vapor are retrieved on a higher-resolution grid in the troposphere and lower stratosphere, with 12 levels per decade from 1000 to 22 hPa rather than the six levels per decade of v1.5. The MLS v2.2 temperature retrieval uses GEOS-5 temperature (discussed in section 3.1) as its a priori while v1.5 used GEOS-4 [Bloom et al., 2005].
 Figure 11 shows the mean difference between MLS v2.2 and MLS v1.5 temperature profiles from the first 93 days selected for processing with v2.2 algorithms. V2.2 has a ∼2.5-K cold bias relative to v1.5 throughout the stratosphere and mesosphere, and an additional ±2 K of persistent vertical oscillation. The exclusion of unsaturated radiances in the center of Band 1 in v2.2 retrievals changes the net effect of gain compression (discussed in section 2.6) on retrieved temperature, and is believed to be the main cause of bias between versions.
Figure 11. MLS v2.2 temperature minus v1.5 temperature, globally averaged for 93 days (287,000 profiles). (left) The mean difference between v2.2 and v1.5. V2.2 has a general ∼2.5-K cold bias relative to v1.5 throughout the stratosphere and mesosphere, with an additional ±2 K of vertical oscillation. Latitudinal variation (not shown) is small compared to persistent vertical structure shown. (right) Solid dots are the 1-σ scatter in individual pairs of profiles, and open dots are the average combined estimated individual profile precisions from the two retrievals. MLS v1.5 has been linearly interpolated to the higher-resolution pressure grid of v2.2 in the UTLS.
Download figure to PowerPoint
 Once reprocessing of the first 30 months of the mission is complete, v2.2 temperature will generally be preferred over v1.5 for scientific studies. Both retrieval versions have biases. V2.2 has slightly better vertical resolution in the upper troposphere, although somewhat poorer vertical resolution in the upper stratosphere. V2.2 has been more extensively validated and is consistent with the improved suite of v2.2 standard products. Routine processing of MLS with v1.5 algorithms was discontinued after February, 2007.
 Figure 12 shows the difference between v2.2 and v1.5 retrieved GPH, averaged for the same 93-day period. The difference in GPH retrieved by the two versions is smallest at 100 hPa (0 ± 40 m), but the cold bias in v2.2 relative to v1.5 temperature results in a GPH low bias in v2.2 relative to v1.5 that reaches −600 m at 0.01 hPa. As shown in Figure 9, absolute accuracy of 100 hPa is limited by MLS absolute pointing accuracy, and by the effects of gain compression, and is expected to be poorer than 200 m for both versions. The use of GEOS-5 100 hPa GPH rather than climatology as a priori, along with a tighter a priori precision, would improve this accuracy, but at the possible cost of some geophysical information.
Figure 12. MLS v2.2 GPH minus MLS v1.5 GPH, averaged for the first 93 days of v2.2 processing. The plots are as in Figure 11. The two versions agree at 100 hPa to within a 1-σ scatter of 40 m, which reflects their use of the same pointing information. The relative temperature bias integrates into a negative bias of v2.2 relative to v1.5 that reaches 600 m at 0.01 hPa.
Download figure to PowerPoint