University of Wisconsin-Madison, Madison, Wisconsin, USA
Corresponding author: D. Tobin, Cooperative Institute for Meteorological Satellite Studies, Space Science and Engineering Center, University of Wisconsin-Madison, 1225 W. Dayton St., Madison, WI 53706, USA. (email@example.com)
 The Cross-track Infrared Sounder (CrIS) is the high spectral resolution spectroradiometer on the Suomi National Polar-Orbiting Partnership (NPP) satellite, providing operational observations of top-of-atmosphere thermal infrared radiance spectra for weather and climate applications. This paper describes the CrIS radiometric calibration uncertainty based on prelaunch and on-orbit efforts to estimate calibration parameter uncertainties, and provides example results of recent postlaunch validation efforts to assess the predicted uncertainty. Prelaunch radiometric uncertainty (RU) estimates computed for the laboratory test environment are less than ~0.2 K 3 sigma for blackbody scene temperatures above 250 K, with primary uncertainty contributions from the calibration blackbody temperature, calibration blackbody reflected radiance terms, and detector nonlinearity. Variability of the prelaunch RU among the longwave band detectors and midwave band detectors is due to different levels of detector nonlinearity. A methodology for on-orbit adjustment of nonlinearity correction parameters to reduce the overall contribution to RU and to reduce field of view (FOV)-to-FOV variability is described. The resulting on-orbit RU estimates for Earth view spectra are less than 0.2 K 3 sigma in the midwave and shortwave bands, and less than 0.3 K 3 sigma in the longwave band. Postlaunch validation efforts to assess the radiometric calibration of CrIS are underway; validation results to date indicate that the on-orbit RU estimates are representative. CrIS radiance products are expected to reach “Validated” status in early 2014.
 The path from research to operations for high spectral resolution infrared sounding has made a major step forward with the flight of the Cross-track Infrared Sounder (CrIS) on Suomi NPP [Glumb and Predina, 2002; Y. Han et al., CrIS measurements, sensor data record algorithm, calibration and validation activity overview, and record data quality, submitted to Journal of Geophysical Research: Atmospheres, 2013]. The CrIS is a high spectral resolution Fourier Transform Spectrometer that is the operational counterpart to the Atmospheric Infrared Sounder (AIRS) on the NASA EOS Aqua Platform [Pagano et al., 2003; Aumann et al., 2003]. It has spectral resolution/coverage and spatial sampling properties similar to AIRS and the same wide range of potential applications. While developed primarily as a temperature and water vapor profiling instrument for weather forecasting, its high accuracy and extensive information about trace gases, clouds, dust, and surface properties make it a powerful tool for climate applications as well.
 For applying CrIS data to numerical weather prediction and climate process studies similar to those explored with AIRS and the Infrared Atmospheric Sounding Interferometer (IASI) [e.g., Chahine et al., 2006; Smith et al., 2009; Hilton et al., 2012], it is important that its radiometric accuracy be rigorously understood. This type of characterization is also required for its proper use in satellite intercalibration efforts [Goldberg et al., 2011; GSICS Traceability Statement for IASI and AIRS, 2011; Chander et al., 2013]. Furthermore, accurate characterization is also crucial to the new application of this type of data for benchmarking the current climate regime of the Earth using approaches defined for the NASA Climate and Absolute Radiance and Refractivity Observatory (CLARREO) Decadal Survey mission [Wielicki et al., 2013]. While the CLARREO observations will have more complete coverage of the infrared emission spectrum, including the far-infrared (16–50 µm), and will provide an accuracy of better than 0.1 K 3 sigma, proven on-orbit using new on-orbit verification and test system technologies [e.g., Taylor et al., 2012; Best et al., 2012], CrIS data is expected to augment the temporal and spatial coverage of CLARREO observations. These types of demanding applications provide motivation for the overview of CrIS radiometric accuracy provided in this paper. Also, by understanding the physical basis of CrIS uncertainties, in the future it should be possible to transfer the even higher accuracy of CLARREO-type observations to CrIS. (3 sigma is traditional terminology for a “not to exceed” uncertainty estimate. In this paper, 1-sigma and 3 sigma are used to express the 68 and greater than 99% confidence intervals in the estimated uncertainties, and are used interchangeably with the metrological coverage factors, k, of 1 and 3).
 The primary goal of this paper is to describe the uncertainty in the radiometric calibration of CrIS based on prelaunch and on-orbit efforts to estimate calibration parameter uncertainties, as well as to provide a few examples of postlaunch validation efforts to assess the predicted uncertainty. As described in (Y. Han et al., submitted manuscript, 2013) the Interface Data Processing Segment (IDPS) CrIS radiance products are expected to reach Validated status in early 2014. Postlaunch validation studies are underway, and as such, there is the potential for further refinements to calibration parameters and their associated uncertainties. Results to date, however, show that the CrIS radiometric accuracy is very good and substantially better than program requirements that were established primarily for weather applications. It is clear that the advantages of high spectral resolution IR for weather forecasting and climate demonstrated on-orbit by AIRS and IASI [Siméoni et al., 2004; Blumstein et al., 2007] predecessor observations are being matched, and in some regard even exceeded, by CrIS. Many of the successes are due to the knowledge of the spectral response functions inherently provided by high spectral resolution observations [Goody and Haskins, 1998; L. L. Strow et al., Spectral calibration and validation of the CrIS satellite sounder, submitted to Journal of Geophysical Research: Atmospheres, 2013] as well as excellent noise performance [V. Zavyalov et al., Noise Performance of the CrIS Instrument, submitted to Journal of Geophysical Research: Atmospheres, 2013].
 We begin by describing the CrIS radiometric uncertainty (or calibration accuracy) expected for observing blackbody spectra based on preflight thermal/vacuum testing (section 2), present techniques and results for on-orbit calibration refinement analyses (section 3), followed with an assessment of radiometric uncertainty for Earth spectra (section 4), and finally present example results of on-orbit radiance validation by comparison with calculated spectra and other sensors (section 5). Results are summarized in section 6.
2 Preflight Radiometric Uncertainty for Blackbody Spectra
 The Radiometric Uncertainty (RU) of CrIS characterizes the accuracy of the observed radiance spectra. RU represents an upper limit of the bias with respect to the true radiance for a large ensemble of observed spectra; it does not include effects such as detector noise which vary randomly from one spectrum to another. Additionally, these RU estimates do not include data transfer and quality control parameter related artifacts [Y. Han et al., submitted manuscript, 2013, section 5.3], which are unrelated to the inherent accuracy of the CrIS observations and which can be removed with future reprocessing efforts. This section compares the CrIS RU sensor specification to an estimate of the CrIS RU relevant to prelaunch thermal vacuum test conditions. In terms of sensor specification, the CrIS RU is stated as a percent of 287 K blackbody radiance with 1-sigma values of 0.45%, 0.58%, and 0.77% for the longwave, midwave, and shortwave bands, respectively. Converting to brightness temperature using the Planck function for blackbody radiation, 3 sigma RU specifications are shown in Figure 1. The 3 sigma specifications are greater than 0.5 K in the shortwave band and greater than 1 K at the longwave end of the longwave band. As shown here and in section 4, both the prelaunch and on-orbit estimates of the actual RU are approximately a factor of three times better than specification.
 Estimates of the prelaunch RU are computed using a parametric perturbation of the radiometric calibration algorithm, using parameter uncertainties determined from prelaunch characterization tests. RU estimates are complimentary to, but should be distinguished from, the majority of postlaunch calibration/validation efforts that aim to estimate product uncertainties through comparison with correlative observations or calculations. For the purpose of this paper, the CrIS radiometric calibration algorithm includes two primary components: a radiometric nonlinearity correction applied to the complex (uncalibrated) spectra followed by radiometric calibration of the resulting linear complex spectra using a responsivity and radiometric offset determined from two calibration targets. During prelaunch thermal vacuum testing, the CrIS Internal Calibration Target (ICT) and an external Space Target (ST) were used to calibrate views of an External Calibration Target (ECT). Both the ECT and ST are large, well-characterized high emissivity blackbodies designed to support CrIS thermal vacuum testing. Differences between predicted and calibrated ECT view spectra are used to assess the overall radiometric calibration of CrIS, and ECT view data collected over a range of ECT temperatures are also used to determine nonlinearity. Following Revercomb et al. , the equation for the calibrated ECT view radiance spectra, RECT, is
where RICT and RST are predicted radiance spectra for the ICT and ST calibration views, and C′ECT, C′ICT, and C′ST are nonlinearity corrected versions of the ECT, ICT, and ST view complex spectra, respectively. The radiometric nonlinearity occurs in the interferogram domain but for the CrIS optical bandpasses and quadratic nature of the nonlinearity, the nonlinearity correction simplifies to:
where C is the measured (nonlinear) complex spectrum, a2 is the detector/Field-Of-View (FOV) dependent quadratic nonlinearity coefficient, and VDC is the photon-induced DC level voltage at the first stage of the detector preamplifier which varies for each interferogram [Knuteson et al., 2013]. It should be noted that equations (1) and (2) are simplified versions of the calibration dealing only with the radiometric contributions to the CrIS RU; spectral contributions are addressed in (L. L. Strow et al., submitted manuscript, 2013).
 To verify the radiometric calibration of CrIS, ECT view data were calibrated and assessed for a range of ECT set point temperatures: 200 K, 233 K, 260 K, 287 K, 299 K, and 310 K. The RU is also estimated for each of the calibrated ECT view spectra. Table 1 lists the various calibration parameters and their uncertainties used to compute RU. These include the temperatures and emissivities of the ICT and ST used to compute RICT and RST and the a2 values used to perform the nonlinearity corrections. Each of the terms is discussed further below. Other potential contributions, such as stray light and polarization effects are less significant due to the sensor design [Stumpf and Overbeck, 2002], and are not included here.
Table 1. Prelaunch Calibration Parameters and Uncertainty Estimates
3 sigma Uncertainty
TICT, Refl, measured
TICT, Refl, modeled
a2, Longwave FOVs
0.01 to 0.02 V−1
a2, Midwave FOVs
0.00 to 0.05 V−1
 CrIS utilizes an ambient temperature sensor design with the optical bench, interferometer, ICT, and the majority of optical components at similar ambient temperatures. In the CrIS calibration algorithm, the predicted radiance when viewing the ICT is computed using the ICT “environmental model” [JPSS Configuration Management Office, 2012], which includes an emissive term as well as several reflected terms due to the nonunit emissivity of the ICT. Furthermore, there are optical components of the reflected terms that have representative temperature sensors and those which require thermal modeling (the baffle of the Scene Selection Module, SSM), leading to a simplified yet representative expression for the ICT predicted radiance:
where B is the Planck function, ɛICT is the effective cavity emissivity of the ICT, TICT is the effective temperature of the ICT, and TICT, Refl, Measured and TICT, Refl, Modeled are the effective temperatures of the reflected optical components which have temperature sensors and those whose temperatures require a thermal model. The 3 sigma uncertainty of TICT is 112.5 mK. This value was determined from engineering estimates and characterization tests of the effective temperature of the ICT taking into account the inherent uncertainties in the temperature sensors, and thermal and aging effects. The ICT emissivity, shown in Figure 2, was measured using a special thermal vacuum test that provided increased sensitivity to the ICT reflected components and from an independent test of the emissivity for one wavelength region in the shortwave band. The ICT emissivity has values ranging from as low as 0.974 in the shortwave band and at the shortwave end of the midwave band, to as high as 0.996 in the longwave window region. The 3 sigma uncertainty in the knowledge of ɛICT is 3%, or ~0.03 in effective emissivity. Due to the ambient design of the CrIS sensor (e.g., that TICT, Refl is similar to TICT), this relatively high uncertainty in the ICT emissivity does not result in large uncertainties in the calibrated spectra. The optical components that have active temperature monitors included in the ICT reflected term include the frame, Optical Mechanical Assembly (OMA), interferometer beam-splitter, and ICT baffle. The 3 sigma uncertainty in TICT,Refl,Measured is conservatively estimated as 1.5 K. The SSM baffle accounts for roughly half of the solid angle of the ICT. Its temperature is not measured; a thermal model is used to predict the SSM baffle temperature based on its measured mount temperature, and this has an orbital variation with range of approximately 6 K. As a very conservative estimate, the prelaunch 3 sigma uncertainty in TICT,Relf,Modeled is taken to be this full range, 6 K.
 The predicted radiance for the ST involves the ST effective temperature, emissivity, and reflected temperatures:
where ɛST is the ST effective emissivity, TST is the effective ST temperature, and TST,Refl is the temperature of optical and structural components in view of the ST. The ST is a large 5-bounce blackbody with effective emissivity of 0.9995, with a 3 sigma uncertainty of 0.0009. The ST is cooled to ~105 K to serve as the cold calibration target during prelaunch testing, and TST has a conservative 3 sigma uncertainty estimate of 6 K. Because the CrIS sensor is at ambient temperature and views the ST, the reflected term contributes the largest uncertainty to RST. The nominal value of TST,Refl is therefore taken to be the nominal, ambient temperature of CrIS (e.g., 280 K) with a conservative 3 sigma uncertainty of 9 K.
 The detector nonlinearity corrections (equation (2)) require knowledge of the characterization parameters a2 and VDC for each detector. VDC is known well from telemetry observation, and the primary uncertainty is in the determination of the a2 values. Using prelaunch data, the a2 values are determined using two independent methods, and the differences between the resulting a2 values are used to estimate systematic uncertainties in a2. The first method involves CrIS viewing the ECT over a range of ECT temperatures, with the a2 value for each detector determined empirically to create optimal agreement of calibrated radiance spectra among the nine CrIS FOVs and with respect to the predicted ECT view spectra. These a2 values, derived from ground TVAC characterization using a stepped temperature ECT, are referred to as a2TVAC-ECT.
 The effect of a quadratic nonlinearity in the interferogram domain is to produce low-resolution artifacts that peak outside the optical pass band that are proportional to the nonlinearity magnitude [e.g., Knuteson et al., 2004]. The second method therefore involves the use of CrIS diagnostic mode (DM) interferogram data collections (bypassing the normal onboard numerical filtering and decimation process which limits the band pass of the resulting spectra) and analysis of the out-of-band harmonics to characterize the nature of the nonlinearity and estimate a2 for each detector. These values are referred to as a2TVAC-DM. Figure 3 shows both sets of prelaunch a2 values, with the a2TVAC-DM values derived from DM views of the ST from the TVAC3 Mission Nominal data set. There is relatively good agreement between the two independent methods of estimating a2. Taking the mean over FOVs, the structural uncertainty derived from these independent methods leads to 3 sigma uncertainties of 29% for longwave FOV values and 47% for midwave FOV values. Figure 3 also shows that all longwave FOVs have appreciable yet similar values of nonlinearity correction (a2), while the midwave band FOVs have a larger range of correction values ranging from the highly linear FOV 6 and FOV 9 to maximum nonlinearity for FOV 7. CrIS utilizes photovoltaic HgCdTe detectors in all three spectral bands [Masterjohn et al., 2003]. Subsequent investigation has identified that the set of nine midwave detectors were cut from two different wafers. The same nonlinearity analysis performed for the shortwave band detectors shows that they are all highly linear. Therefore, the shortwave band a2 values are set to zero and the corresponding nonlinearity uncertainty is zero in this analysis.
 Using equations (1) through (4) and the uncertainties listed in Table 1, the prelaunch CrIS RU is shown as 3 sigma brightness temperature uncertainties in Figures 4 and 5. For each calibration parameter, the calibration is perturbed by the 3 sigma parameter uncertainty. The uncertainties are largely independent of one another, and the total RU is computed as the root sum square (RSS) of the individual terms. Figure 4 shows the individual contributions to the RU budget and the total RU for FOV 7 for calibrated views of the ECT at 287 K. The leading uncertainty terms are the TICT, ɛICT, TICT,Refl,Modeled, and a2 terms. For blackbody view spectra, the uncertainties are generally smoothly varying with wave number, while the TICT,Refl terms follow the signature of the ICT emissivity versus wave number. Figure 5 shows the RU for all nine CrIS FOVs as a function of ECT brightness temperature for one representative spectral channel in each of the three CrIS spectral bands. Two versions of the RU are shown. The top row in Figure 5 shows RU estimates using all terms in Table 1, where the ST is used in thermal vacuum test conditions along with the ICT to perform radiometric calibration. The bottom row does not include the ST uncertainty terms and represent the on-orbit case where Deep Space (DS) views would be available for calibration. These results show that the ST terms introduce significant additional uncertainty for the prelaunch test conditions, particularly for cold scene temperatures and shorter wavelengths. However, for warmer scene temperatures, the estimated prelaunch RU is considerably less than the CrIS specification values shown in Figure 1. The spread in RU among the various FOVs is due to FOV-dependent nonlinearity magnitudes, with midwave FOV 7 having the largest magnitude and uncertainty along with FOV 9 in the longwave band. The shortwave band detectors are linear, and the resulting RU estimates are independent of FOV for that band.
 Also shown in the top row of Figure 5 are 3 sigma uncertainties in the ECT view predicted brightness temperatures as well as the absolute value of the predicted minus calibrated brightness temperatures for each ECT temperature set point and each FOV (“ECT residuals”). These ECT residuals are used to verify the overall radiometric calibration of CrIS, and the ECT residuals should therefore be bounded by the uncertainty of the ECT view predicted radiances and the CrIS RU estimates. The ECT used in TVAC is a large five-bounce specular blackbody. The ECT radiance uncertainty is computed using an ECT temperature uncertainty of 89.1 mK 3 sigma, cavity emissivity of 0.9995, emissivity uncertainty of 0.0009 3 sigma, reflected temperature of 290 K, and reflected temperature uncertainty of 5 K 1 sigma. Due to the contribution of the ECT reflected radiance from a warm external environment, the uncertainty in the predicted ECT view radiance becomes larger for the cold ECT temperature set points and for shorter wavelengths. For all wavelengths and FOVs, Figure 5 shows that the ECT residuals are smaller than both the uncertainty of the predicted ECT views and the uncertainty in the CrIS calibrated spectra. In other words, the ECT residuals are consistent with the ECT and calibrated CrIS uncertainty estimates.
3 Methodology for On-Orbit Nonlinearity Confirmation and Refinement
 As discussed in the previous section, Figure 5 represents an estimate of the radiometric uncertainty based on the prelaunch estimates of uncertainties in the calibration parameters and verification testing performed in a thermal vacuum chamber. If the assumption is made that the relevant calibration parameter uncertainties have not changed postlaunch, then the bottom row of Figure 5 (computed assuming a real view of Deep Space) represents the CrIS on-orbit radiometric uncertainty for viewing a Planck spectrum. This assumption is valid for all of the parameter uncertainties except for the nonlinearity coefficients. Repeated CrIS prelaunch testing cycles demonstrated that the linearity of some detectors can change upon detector warm-up prelaunch and subsequent detector cool-down postlaunch. The exact cause of this effect is still under investigation, but the implication was that the nonlinearity of the longwave and midwave detectors would have to be reevaluated in orbit.
 This section describes the methodology used to determine the on-orbit a2 parameters and estimate their uncertainties. A remarkable outcome of this approach is the ability to reduce the prelaunch RU of the most nonlinear detectors by tying these to reference detectors with smaller nonlinearity, particularly for the midwave band FOVs. At the same time, because the largest source of radiometric differences among FOVs is due to the varying degrees of nonlinearity, the nonlinearity coefficients have been adjusted to produce optimal radiometric agreement among FOVs. Ensuring radiometric uniformity of the nine CrIS FOVs is important for users of the data, e.g., data assimilation.
 The methodology used to refine the on-orbit nonlinearity parameters involves two steps. The first step provides an estimate of the potential change in nonlinearity from prelaunch to on-orbit, and involves the collection of on-orbit DM data for DS views and subsequent analysis of the out-of-band harmonics. This is the same process used prelaunch to provide the a2 estimates using DM ST data discussed in section 2. The fractional change in a2 from prelaunch to on-orbit is therefore accurately estimated from the ratio of these DM derived on-orbit and prelaunch a2 values. The results suggest that some detectors, e.g., longwave FOV 9 and midwave FOV 7, changed significantly while other detectors stayed close to their prelaunch estimates. The longwave FOV 5 a2 value is estimated to have changed approximately 6%, and these change estimates have an uncertainty of approximately 5% 1 sigma. The on-orbit DM data are very useful because they have provided the ability to verify the quadratic nature of the nonlinearity while on-orbit, and also to verify that the shortwave band detectors are linear. Also, importantly, they have been used to verify that two of the midwave band FOVs (6 and 9) have negligible nonlinearity on-orbit.
 The second step for refining the on-orbit nonlinearity correction involves the selection of a “reference FOV” for each spectral band, followed by an Earth view FOV-to-FOV analysis in which adjustments to the a2 values of the remaining eight FOVs are made to create optimal agreement with the reference FOV brightness temperature observations. For the midwave band, FOV 9 is highly linear and is the reference FOV. In other words, it serves as a linear reference for the other nonlinear midwave FOVs. Unlike the midwave band, all longwave FOVs display similar levels of nonlinearity. FOV 5 has the lowest nonlinearity and is chosen as the longwave reference FOV. The Earth view FOV-to-FOV analysis approach makes use of the fact that, in a statistical sense, all nine FOVs observe the same Earth scene distributions (ignoring the small angular spread among the FOVs within the 3 × 3 array, 1.1 to 1.5°). Brightness temperature differences relative to the selected reference FOV are computed for each FOV for a large set of Earth observations for spectral regions which have high sensitivity to nonlinearity and also low spatial variability (typically high altitude peaking channels). The data set is restricted to include only the center four Fields-of-Regard (FORs) nearest the nadir view to avoid artifacts due to atmospheric opacity differences for higher view angles. The data set is also filtered to only include FORs with low spatial variability (standard deviation of nine FOVs within the FOR less than 1 K). The a2 values are then adjusted to minimize the observed FOV-to-FOV brightness temperature differences. Resulting on-orbit a2 values are shown in Figure 3. A sample time series of brightness temperature differences is shown in Figure 6 for spectral means of the 672–682 cm−1 longwave region and the 1585–1600 cm−1 midwave region, using CrIS radiance products generated at IDPS. The abrupt transition in April 2012 is due to the upload of Engineering Packet version 33 (EP 33) to replace the prelaunch nonlinearity parameters (EP 32) with those determined postlaunch in the early checkout period using Earth view FOV-to-FOV analysis of a relatively small data sample. After upload of EP 33 the FOV-to-FOV agreement is greatly improved, yet further refinement of these parameters utilizing a larger data set is expected before CrIS reaches validated status. The time variations of the longwave band FOV-to-FOV differences (amplitude of ~0.015 K) are not yet understood and under investigation. Note, however, that they are substantially smaller than the overall uncertainty estimated for Earth view spectra at this wavelength region (see section 4). Furthermore, an adjustment to the longwave reference FOV 5 a2 value is expected prior to reaching validated status.
 The FOV-to-FOV brightness temperature differences are shown to be Gaussian in nature, leading to remarkably small estimated uncertainties of 3 mK (3 sigma) in the longwave band and 6 mK (3 sigma) in the midwave band when using a month of FOV-to-FOV differences. These brightness temperature uncertainties are then converted to uncertainties in a2 using a2 Jacobians, dR/da2. The uncertainties in the final on-orbit a2 values therefore have contributions from the prelaunch determination of a2 for the reference FOV (in Table 1), the change from prelaunch to on-orbit for the reference FOV estimated from Diagnostic Mode data (15% 3 sigma), and the FOV-to-FOV adjustments based on Earth view analysis. Since the FOV-to-FOV analysis creates optimal agreement with the reference FOV, the a2 uncertainty contributions are combined in units of a2 rather than percent. The on-orbit 3 sigma uncertainties of the quadratic nonlinearity parameter are listed in Table 2 for each detector in the longwave and midwave focal planes. For both spectral bands, the uncertainty in the Earth view FOV-to-FOV adjustments are relatively small and the total uncertainties are largely dependent on the uncertainty of the reference FOV. For the midwave band, this results in small uncertainties in units of a2 for all FOVs due to the excellent inherent linearity of reference FOV9, yet large percentage uncertainties for FOVs when the a2 value is small. For the longwave band, all FOVs have similar a2 values and the final uncertainties are approximately equal to the FOV 5 uncertainty in terms of both percent and in units of a2.
Table 2. 3 Sigma Uncertainties in On-Orbit a2 Values, Given in Units of a2 and Percent
 To illustrate the impact of these reduced a2 uncertainties, RU estimates computed for ECT views using these uncertainties are shown in Figure 7. These are computed assuming that a DS view is available, and can be compared directly to the RU estimates shown in the bottom row of Figure 5. The primary impact of the on-orbit a2 adjustments is to remove the majority of the FOV dependence of the estimated RU. Additionally, the midwave RU is reduced to be similar to that predicted for the linear shortwave band due to the high linearity of midwave reference FOV 9, while the longwave band RU for all FOVs remains slightly elevated with respect to the midwave and shortwave bands.
4 On-Orbit Radiometric Uncertainty for Earth Spectra
 On-orbit RU estimates for Earth view spectra are computed using the same approach as described in section 2 for prelaunch conditions. However, as opposed to prelaunch blackbody spectra and prelaunch parameter uncertainties, the on-orbit RU estimates are shown here for representative Earth view spectra using on-orbit parameter uncertainties. The on-orbit calibration uses a DS view with unit emissivity rather than a less perfect ST in TVAC. Thus, the uncertainty of the on-orbit nonlinearity parameters differs from the prelaunch uncertainties, with on-orbit uncertainties described in section 3 and given in Table 2. Additionally, postlaunch validation efforts have verified that the ICT environmental model does not introduce significant artifacts (either in terms of the shape of the ICT reflectivity or in orbital variations), and the uncertainty of TICT,Refl,Modeled is reduced to 3 K (3 sigma) for these on-orbit RU estimates. This allows CrIS to produce very low RU from its ICT even though its emissivity is less than perfect.
 Given the on-orbit uncertainties contained in Tables 1 and 2, the RU of an Earth scene can be computed. Figures 8 and 9 show 3 sigma RU estimates for representative warm (nominally clear sky) and cold (high thick cloud) Earth view FOV 9 spectra collected on 24 February 2012. Overall, the RU is less than 0.2 K in the midwave and shortwave bands, and less than 0.3 K in the longwave band. In both cases, and as with the prelaunch estimates, the uncertainty in the ICT temperature is a leading contributor. For Earth scenes, the nonlinearity corrections and resulting RU contributions are dependent on the spectrally integrated signal of the complex spectra, as well as the spectral shape of the Earth view spectra, and thus can vary significantly from one spectrum to another. For typical warm scene spectra, the longwave band nonlinearity corrections and associated uncertainties are small in the longwave window region and larger in the more opaque CO2 absorption region. For the example cold scene RU estimate, the Earth view spectrum is more similar to a blackbody and the resulting RU contribution is relatively flat versus wave number with ~0.1 K contribution throughout the opaque and window regions of the longwave band. The on-orbit nonlinearity contributions to the RU in the midwave band are very small. Additionally, due to efforts discussed in section 3, there is little FOV-to-FOV variability in the Earth scene RU estimates. Finally, to capture a larger range of Earth scene radiances, RU for an orbit of data on 24 February 2012 is shown in Figure 10. The distributions include RU for all spectral channels and FOVs. The largest RU values observed in the longwave band correspond to 14 µm channels for nominally clear sky scenes such as that shown in Figure 8 but with warmer surface temperatures.
5 Example Postlaunch Radiance Validation Results
 RU estimates presented in section 4 based on calibration parameter uncertainty estimates suggest an overall on-orbit RU of less than 0.2 to 0.3 K 3 sigma. Various postlaunch efforts are underway to independently assess the radiometric accuracy and stability of CrIS, as well as the spectral calibration, noise performance, and geolocation accuracy. The radiometric assessments include a variety of approaches including, for example, comparisons with other satellite sensors and with clear sky calculated spectra [e.g., Tremblay et al., 2012; Y. Han et al., submitted manuscript, 2013] and underflights by high altitude aircraft [e.g., Larar et al., 2011; Taylor et al., 2013]. While these validation efforts are ongoing, preliminary results are generally consistent with the on-orbit RU estimates, with observed biases on the order of a few tenths of K, stable with time, and small FOV-to-FOV differences. Future publications will present the postlaunch validation findings in greater detail, including differences presented as a function of scan angle, scene brightness temperature, and orbital phase, for example. Some example results are presented here.
 Following the methodology of Strow et al. , comparisons of clear sky observed and calculated spectra are shown in Figure 11. The ensemble includes tropical (−30 to +30 degree latitude) ocean nighttime spectra collected between April 2012 and April 2013. The determination of clear scenes is also described in Strow et al. [2006, section 3.2], with only a small percent (1%) of the nighttime ocean scenes accepted. The calculations are performed using SARTA [Strow et al., 2003] developed for CrIS with ECMWF 3 h analysis/forecast fields used for ocean surface temperature and atmospheric state profiles. The observed biases are similar to biases computed for IASI and AIRS [e.g., Strow et al., 2006, Figure 9]. This includes the largest observed residual at 2380 cm−1; these channels have peak sensitivity to atmospheric temperature around 25 mbar, but with an unusual bimodel shape with a secondary peak near 4 mbar which gives it unique sensitivity to the ECMWF profile shape. Given the uncertainties in the calculated spectra, the observed minus calculated residuals are reasonable and not inconsistent with the CrIS RU estimates that are under 0.3 K everywhere.
 Utilizing simultaneous nadir overpasses (SNOs) and intercomparison techniques described in Tobin et al. , example comparisons of CrIS and METOP-A IASI and CrIS and AIRS are shown in Figures 12 and 13. To assess the FOV dependence of the CrIS calibration, IASI minus CrIS spectra for northern hemisphere SNOs are shown for each CrIS FOV in Figure 12, with the mean IASI-CrIS difference removed. Differences are shown for IDPS CrIS products using the prelaunch EP 32 and on-orbit EP 33 nonlinearity parameters. The results show significant improvement using the on-orbit adjusted nonlinearity parameters described in section 3 and are consistent with reduced FOV dependence of the predicted on-orbit RU presented in section 4. EOS Aqua and Suomi-NPP are in similar orbits, resulting in frequent SNOs covering a wide range of observed spectra. Figure 13 shows time series of daily mean AIRS minus CrIS brightness temperature differences for spectral regions with sensitivity to various CrIS calibration parameters. Due to the imprecise methodology for normalizing the spectral response functions of CrIS and AIRS L1B spectra, the comparisons are shown for ~10 wave number averages; this averaging produces a more meaningful assessment of the radiometric differences between CrIS and AIRS. The 677, 1395, and 1592 cm−1 regions have sensitivity to CrIS nonlinearity and the discontinuities in April 2012 are due to the operational processing change from EP 32 to EP 33 as discussed in section 3. Otherwise, the differences are well within the estimated RU of CrIS and stable with time.
 The current on-orbit RU estimates do not include several other effects currently under study. These include spectral ringing artifacts for unapodized spectra, potential calibration artifacts in opaque regions of the shortwave band, and differences with other on-orbit sensors observed for cold scene temperatures in some spectral regions [Y. Han et al., submitted manuscript, 2013, section 5.4], as well as a potential adjustment to the longwave reference FOV 5 a2 value based on reanalysis of the thermal vacuum test data. Pending further diagnoses of the validation analyses and determination of the root cause of the differences, the on-orbit RU estimates will be assessed and refined as needed.
 This paper has described the uncertainty in the radiometric calibration of CrIS based on prelaunch and on-orbit efforts to estimate calibration parameter uncertainties, and provided example results of postlaunch validation efforts to assess the predicted uncertainty. RU characterization is important for weather, climate, and intercalibration applications of the data. Prelaunch RU estimates computed for the laboratory test environment utilizing ST and ICT views for calibration are less than ~0.2 K 3 sigma for blackbody scene temperatures above 250 K, with primary uncertainty contributions from the ICT temperature, ICT reflected radiance terms, and detector nonlinearity correction parameters. Significant FOV-to-FOV variability in the longwave and midwave band prelaunch RU is due to prelaunch uncertainty in nonlinearity contributions. A methodology for on-orbit adjustment of detector nonlinearity correction parameters, utilizing Diagnostic Mode data of Deep Space and Earth view data, results in reducing the overall nonlinearity contribution to the RU. In addition, this methodology reduces FOV-to-FOV calibration differences better than can be achieved with ground test methods. The resulting CrIS on-orbit RU estimates, shown in section 4 for representative warm and cold scene Earth view spectra, are less than 0.2 K 3 sigma in the midwave and shortwave bands, and less than 0.3 K 3 sigma in the longwave band. Postlaunch validation efforts to assess the radiometric calibration of CrIS are underway; validation results to date indicate that the on-orbit RU estimates presented here are representative. IDPS CrIS radiance products are expected to reach Validated status in early 2014. Pending additional findings from ongoing validation analyses in this time frame, refinements to the CrIS radiometric calibration parameters and associated RU estimates will be investigated. However, it is already clear that the high accuracy of CrIS makes it an exceptional asset for weather applications, and with final refinements of calibration coefficients and associated data reprocessing efforts, also for climate applications.
 This research was supported by the NOAA Joint Polar Satellite System Office under grant NA10NES4400013, by the former Integrated Program Office, and by NASA Suomi-NPP Science Team grant NNX11AK21G. The authors would like to extend their thanks to the NASA Atmospheres Product Evaluation and Analysis Tools Elements and Community Satellite Processing Package teams at the University of Wisconsin-Madison for various data access and data processing efforts that contributed to results presented here. IASI L1C data and IDPS generated CrIS data were obtained from NOAA's Comprehensive Large Array-data Stewardship System, and AIRS L1B data were obtained from the Goddard Earth Sciences Data and Information Services Center.