Spectral calibration and validation of the Cross-track Infrared Sounder on the Suomi NPP satellite



[1] The Cross-track Infrared Sounder (CrIS) radiometric accuracy depends upon accurate frequency calibration. Here we present both the prelaunch calibration of the sensor and the minor modifications needed to this calibration post launch. Particular emphasis is given to ensuring that all nine detectors on each of the three CrIS focal planes are on a common frequency scale with accurate off-axis apodization corrections. Radiances from the current operational algorithm have a frequency calibration that is stable to 2 ppm, although this work suggests that the CrIS instrument is capable of a frequency calibration stability of better than 1 ppm.

1 Introduction

[2] The Cross-track Infrared Sounder (CrIS) [Han et al., 2013] flying on the Suomi NPP (SNPP) satellite represents NOAA's first high-spectral resolution operational sounder, replacing the High resolution Infrared Radiation Sounder instrument that has been flying on the NOAA polar operational satellites for decades. EUMETSAT has flown the Infrared Atmospheric Sounding Interferometer (IASI) high-spectral resolution sounder on the METOP platforms since 2006, and NASA continues to operate the Atmospheric Infrared Sounder (AIRS) that was launched on the EOS-AQUA platform in 2002. CrIS and IASI are Michelson interferometers, while AIRS is a grating spectrometer.

[3] These instruments were developed primarily to support numerical weather predictions (NWP) by providing improved temperature and water vapor profile information but are also used extensively by the scientific community for global measurements of trace gases, land surface properties, cloud properties, and soon, medium-term climate trending. These applications require an accurate and stable record of global infrared radiances.

[4] This paper addresses the accuracy of the frequency calibration of CrIS, which is substantially different than the approaches used on IASI and AIRS. The chain of frequency calibration is followed from thermal vacuum (TVAC) testing of CrIS to post-launch corrections, to long-term (1 year) validation. Section 2 summarizes the design features of CrIS relevant to frequency calibration and associated accuracy requirements. Section 3 presents the results of the CrIS TVAC testing during which extensive data was taken to characterize the spectral performance of CrIS. The in-orbit spectral calibration and longer-term validation of the CrIS frequencies is discussed in sections 4 and 5.

[5] The CrIS instrument started producing high-quality interferograms in late January 2012, but the NOAA Interface Data Processing Segment (IDPS) that processes these interferograms into calibrated CrIS radiances, called Sensor Data Records (SDRs), was not operational until several months later. Therefore, much of the CrIS spectral calibration work described here used our CrIS Calibration Algorithm Sensor Testbed (CCAST) [Motteler et al., 2013] software to produce CrIS SDRs. An off-line version of the IDPS SDR processing software, called ADL, was subsequently used to test the calibration coefficients derived using CCAST-derived SDRs. The ADL (and essentially identical operational IDPS) software contained some coding errors that affected spectral calibration. Section 4.3 describes our work around these relatively small coding errors, which will be fixed in the IDPS processing in late 2013.

2 CrIS Instrument Design and Spectral Sensitivity

[6] CrIS is a Michelson interferometer incorporating a nominal ±0.8 cm optical path difference (OPD) [Stumpf and Overbeck, 2002; NASA Goddard Space Flight Center, 2011]. The primary design feature of CrIS that is relevant to frequency calibration is the geometry of the focal planes. CrIS has three focal planes: band 1, longwave from 650 to 1095 cm−1, band 2, midwave from 1210 to 1750 cm−1, and band 3, shortwave from 2155 to 2550 cm−1. Each focal plane contains a 3×3 array of HgCdTe detectors as illustrated in Figure 1. We will refer to the nine different detectors on each focal plane as fields of view (FOVs) 1–9. The interferometer axis is nominally centered in the middle of FOV5. All nine detectors detect off-axis rays and thus require software corrections for the instrument self-apodization. We can divide the detectors into three groups; (1) the center detector, FOV5, which has very little instrument apodization, (2) the side detectors (FOVs 2, 4, 6, and 8), and (3) the corner detectors (FOVS 1, 3, 7, and 9) which are the most heavily apodized.

Figure 1.

Nominal geometry of the CrIS focal planes containing a 3×3 array of detectors. Tables that refer to each detector in this paper will be arranged as shown here.

[7] The calibrated CrIS radiances are reported on a fixed wave number grid with a sinc lineshape (boxcar apodization). CrIS requires rather large corrections to the measured radiances for the off-axis detectors to remove the instrument line shape (ILS) distortion that occurs for off-axis rays processed by the interferometer. These are well-defined and predictable consequences of Michelson interferometry that receive detailed explanation in the CrIS Algorithm Theoretical Basis Document [NASA Goddard Space Flight Center, 2011] and elsewhere [Genest and Tremblay, 1999; Desbiens et al., 2002, 2006].

[8] The apodization corrections, and frequency calibration of the CrIS wave number grid, depend on accurate knowledge of the CrIS focal plane geometry (detector position and shape relative to the interferometer boresight) and the wavelength of the metrology laser that drives the interferogram sampling; all of which are examined in detail in this paper.

[9] The CrIS longwave optical path difference of ±0.8 cm produces an unapodized spectral resolution of 0.625 cm−1. The CrIS midwave interferograms are truncated at ±0.4 cm (1.25 cm−1 resolution), while the shortwave interferograms are truncated at ±0.2 cm (2.5 cm−1 resolution). CrIS is capable of recording full 0.8 cm long interferograms for all three bands and has been commanded to this full-resolution mode twice in-orbit. The higher- resolution midwave (MW) and shortwave (SW) spectra are of high quality but will not be examined here since they only cover slightly more than 1 day of operation. Plans are underway at NOAA to permanently place CrIS into the full-resolution mode in the near future.

2.1 CrIS Neon Calibration System

[10] The CrIS SDR algorithm makes the reasonable assumption (shown to be true in section 4) that the alignment of the nine detectors on each of the three CrIS focal planes relative to the interferometer optical axis is essentially fixed once CrIS thermally stabilizes in-orbit. The metrology laser wavelength, however, is expected to change over time, primarily due to slight changes in operating temperature. Although the metrology laser is temperature controlled, the control range about the ambient temperature is small. If the instrument temperature changes significantly, the metrology laser temperature control set point must be changed, thus changing the metrology laser wavelength. So far, after 15 months of operation, the CrIS instrument temperature has remained very stable, with diurnal and seasonal changes that are too small to force a change in the metrology laser temperature set point.

[11] CrIS has a calibration subsystem that very accurately measures the metrology laser wavelength using a narrow emission line at 703.45 nm from a low-pressure Neon bulb co-aligned with the interferometer axis. The Neon lamp is activated for 10 s once every 109 min, and the number of interferometric neon fringes (including fractional fringes) is measured during the sweep of a single interferogram. Since this sweep contains a fixed number of metrology laser fringes, one can easily calibrate the wavelength of the metrology laser using the known Neon emission line wavelength. The transfer accuracy of this comparative process is better than 0.6 parts per million (ppm), as described in more detail in section 4. A repeat calibration interval of 109 min versus the 101 min orbit period was chosen in order to spread the Neon emission line measurements over the full range of orbital positions (latitudes) over the long term. This enables any orbital variations in the metrology laser calibration process to be characterized and tracked for all satellite views.

[12] An alignment offset between the Neon path in the interferometer and the path of the infrared signal will produce a frequency calibration error. In addition, engineering studies indicate that the Neon bulb spatial emission pattern could change during the mission and introduce small wavelength calibration errors. In practice, an effective Neon lamp wavelength is used, which is measured using either gas cell spectra during ground testing (see section 3) or using the upwelling radiance in-orbit (see section 4); both of which provide an absolute wave number scale. This effective wavelength will be slightly different than the true Neon lamp wavelength due to the Neon beam spread about the interferometer optical axis (computed to be about 18 ppm) and any offsets of the Neon ray path from the interferometer axis.

2.2 Spectral Calibration Accuracy Requirements

[13] The CrIS SDR algorithm is required to report the radiances using a sinc function ILS. The accuracy of the CrIS ILS can be compromised by two separate effects: (1) inaccurate characterization of the geometry of the focal plane detectors that determines the exact alignment of the detectors to the interferometer boresight axis and (2) inaccurate knowledge of the CrIS metrology laser wavelength that determines the interferogram OPD sampling positions.

[14] Effect (1) could produce systematic differences among the different FOV radiances. For a well-aligned instrument, such as CrIS, FOV 5 will be largely immune to focal plane position uncertainties since it is nominally centered on the interferometer axis. In addition, effect (1) is expected to be very static once CrIS is stabilized in-orbit so that any systematic differences due to optical alignment can be permanently removed through an in-orbit CAL/VAL process (see section 4.1).

[15] Effect (2) will influence all FOVs identically and is expected to change very slowly and smoothly by less than 3.5 ppm (max) over the mission life time due Neon lamp aging effects that impact optical alignment and hence the bulb's effective reference wavelength. This systematic degradation over time can also be removed through calibration updates of the in-orbit CAL/VAL process (see section 2.1).

[16] The metrology laser wavelength is updated via the Neon bulb calibration every 109 min thus producing accurate laser metrology wavelength knowledge on a near-continuous basis. Ground processing of the interferograms uses this knowledge to generate the proper wave number scale. Currently, a threshold of 2 ppm is used by the operational ground processing so that laser metrology wavelength drifts smaller than this value go uncorrected. Thus far, CrIS has experienced less than a ±1 ppm seasonal variation in the laser metrology wavelength as measured by the neon calibration subsystem.

[17] The formal specification for the accuracy of the CrIS wave number scale is 10 ppm through the life of the mission. This rather high uncertainty partially reflects past concerns for how difficult it might be to correct all nine CrIS FOVs to a single equivalent on-axis FOV measurement.

[18] CrIS does not have a requirement for relative differences in spectral calibration among the nine FOVs. However, a consistent spectral calibration among all FOVs is very important for a number of applications, especially for NWP data assimilation where sensor bias correction is performed to levels well below 0.1 K, and there is no system to provide different biases for different FOVs.

[19] The IASI interferometric sounder on METOP has four detectors in each of three focal planes, with each detector nominally placed the same distance off of the interferometer axis. As with CrIS, the IASI radiances must be corrected to equivalent on-axis values and any errors in the detector placement used in the correction software will introduce radiometric differences among the four detectors. Radiometric differences among detectors can also arise from inaccuracies in nonlinearity corrections. Whatever the cause, the initial assimilation of IASI radiances at ECMWF [Collard and McNally, 2009] was restricted to only a single detector because the radiometric calibration was found to vary among detectors by approximately 0.1 K, making it difficult to apply dynamic bias correction to the radiances.

[20] If the CrIS apodization corrections create differences among the nine detectors, it is likely that NWP data assimilation users will similarly need to select a single CrIS detector, ignoring 8/9 of the CrIS SDR radiances. Therefore, a careful prelaunch and post-launch calibration effort was designed to ensure that the CrIS apodization corrections and frequency scale were as identical as possible for all nine FOVs. We show here that this effort was successful, and FOV differences due to spectral calibration are well below 0.1 K. Experience with the AIRS sensor [Strow et al., 2006] suggested that in-orbit spectral calibration of upwelling radiances is possible to the ∼1 ppm level, although at present the AIRS L1b radiances are not corrected for orbital frequency shifts of about ±2 ppm and a slow ∼4 ppm shift that occurred during the first 7 years of operation.

[21] Figure 2 shows the CrIS brightness temperature errors due to a 1 ppm frequency calibration error, for both boxcar-apodized radiances and Hamming-apodized radiances. NWP assimilation centers to date only use IASI/CrIS radiances with an external apodization (close to Hamming shown here). This figure indicates that a 1 ppm frequency calibration error, for a Hamming-like apodization, should limit FOV-to-FOV frequency calibration differences to about 0.02 K, which should be small enough for successful NWP bias correction. Note that other radiometric differences between detectors (nonlinearity) may also cause FOV-to-FOV differences but are not the subject of this paper.

Figure 2.

Radiometric errors for the LW and MW bands that will result for a 1 ppm error in the CrIS metrology laser wavelength, for the SDR boxcar (sinc) ILS, and for a Hamming apodization.

3 Prelaunch Calibration

[22] The CrIS detector focal plane geometry and the Neon lamp effective wavelength were determined during thermal vacuum testing (TVAC) by recording transmittance spectra of CO2, CH4, and HBr, which have strong spectral features in bands 1, 2, and 3 respectively.

[23] Transmittance spectra are derived by measuring four types of radiance spectra for each gas:

  1. [24] FT1: cell full, hot blackbody.

  2. [25] ET1: cell empty, hot blackbody.

  3. [26] FT2: cell full, cold blackbody.

  4. [27] ET2: cell empty, cold blackbody.

[28] A very stable, but not necessarily high accuracy, blackbody was used for these measurements. The 12.54 cm long gas cell was placed in a vacuum chamber between the CrIS input aperture and the source blackbody, which was mounted inside the vacuum chamber. Nominal hot/cold blackbody temperatures were 340 K and 310 K. The radiative transfer equations for this setup can be solved for the gas cell transmittance:

display math(1)

Radiometric calibration drops out of the above equation for an instrument that is stable over the four data collects, so it was not performed. Instead, we used the original uncalibrated complex spectra (in counts) that are generated from the Fourier transform of the complex CrIS raw data records. This expression for τgas is similar to the standard interferometer complex calibration [Revercomb et al., 1988] and returns a mostly real result with a small complex residual-containing noise. Each transmittance was the mean of several minutes of data, resulting in very low noise.

[29] Figure 3 is a CO2 transmittance spectrum taken with CrIS showing a zoom of the Q-branch at 720 cm−1. These spectra have not been corrected for apodization effects, and one can clearly see the single FOV 5 spectrum centered at the highest wave number, with the four-sided FOV spectra centered at a slightly lower wave number, followed by the spectra of the four corner FOVs at even lower wave numbers. The apodization correction removes these shifts and distortions.

Figure 3.

Sample gas cell spectrum of CO2 taken with CrIS during TVAC. The zoomed inset shows the frequency separation of the middle (purple), side, and corner FOVs before correction.

[30] For this analysis the transmittances are not resampled to the CrIS SDR reporting wave number grid but kept at the “instrument grid” which is determined by the metrology laser wavelength for each test. Observed transmittances are fit to calculated transmittances at the instrument grid.

[31] After forming the transmittance according to equation (1), we apply a band-pass filter with a flat passband spanning the CrIS-reporting wave number grid to suppress data in the CrIS guard bands (wave number regions outside of the reporting grid). The CrIS apodization correction, based on the best estimate focal plane geometry and metrology laser wavelength, is then applied.

[32] Alternatively, one could use the CrIS SDR algorithm to radiometrically calibrate the gas cell spectral measurements and then form the transmittance spectra on the CrIS-reporting wave number grid.

[33] Computed transmittances are derived from monochromatic gas absorption coefficients using GENLN2 [Edwards, 1992], converted to transmittances, and inverse Fourier transformed to simulate an extended interferogram at the instrument-sampling grid. This extended interferogram is then truncated to the instrument OPD. The instrument grid and OPD are both functions of the current value of the metrology laser wavelength. The Fourier transform of this simulated interferogram is used as the reference truth, and the RMS difference between the observed and computed transmittances can be minimized by varying the nominal metrology laser wavelength.

[34] The differences between the fitted metrology laser wavelengths and the metrology laser wavelength derived from the Neon calibration subsystem are due to the combination of inaccurate detector focal plane positions and an offset error for all detectors if the Neon wavelength is incorrect or the Neon source is not aligned perfectly along the interferometer axis. This can be written in units of ppm frequency as

display math(2)

where i from 1 to 9 represents the nine FOVs, ri is the radial position of center of detector i from the interferometer axis and dν/dri is the sensitivity of the detector frequency to its radial distance from the axis. Δri is the offset to the nominal detector radial position that causes the shift of Δνi in the observed frequencies. Since FOV 5 is nearly centered on that axis, it is very insensitive to focal plane position errors, and dν/dr5 is nearly zero. Thus, for FOV 5, the difference between the fitted metrology wavelength and the metrology wavelength derived from the Neon lamp is solely due to Neon lamp wavelength calibration error.

[35] If the FOV 5 Neon calibration error is subtracted from the effective frequency calibration errors of the remaining eight detectors, these remaining differences are due to incorrect focal plane positions that were used in the apodization correction process, or

display math(3)

where i now includes all FOVs except FOV 5. Nominal values for the sensitivity of these eight FOVs to radial position errors, dν/dri in equation (2), are 0.027 ppm/μrad for corner FOVs (1, 3, 7, and 9) and 0.019 ppm/μrad for the side FOVs (2, 4, 6, and 8).

[36] A two-step approach was used to find the final focal plane configuration. The exact derivative of the ppm offset with respect to a radial error for each detector's position relative to the interferometer axis is easily calculated. A least square fit was then performed which only allowed a rigid x,y offset of the focal plane relative to the interferometer axis. While this step dramatically lowered the frequency calibration errors, further small changes to each detector's radial position (except FOV 5) were required to drive the residual errors to zero. The initial x,y offset of the whole focal plane also serves to move FOV 5 closer to the interferometer axis.

[37] Spectra for all three focal planes were fit using this approach, which yielded almost identical radial offsets of 0.03° for the TVAC test temperature closest to the instrument temperature on SNPP.

[38] The gas cell spectra and RMS over the nine FOVs of the final observed minus computed residuals are shown in Figures 4-6. These fits also included a per-spectrum scale factor and offset for each transmittance that was needed to minimize the residuals. The scale factor was generally around 0.99 and the offset near 0.01 in transmittance, and together they account for small shifts in signal levels during the data collection. The RMS is always well below 0.01 in transmittance in the regions fit and only grows to be near 0.01 for the lowest wave numbers in the CO2 spectra. Examination of the individual obs-calc transmittances indicates that this is systematic error in all FOVs and may be related to slight errors in the theoretical monochromatic calculations.

Figure 4.

RMS, over the nine FOVs, of observed minus computed CO2 transmittance after fitting for the CrIS longwave focal plane parameters during TVAC. This indicates very small differences among FOVs. The RMS grows near the longwave end of the band, possibly due to inaccuracies in the computed CO2 spectra used in these fits.

Figure 5.

Same as Figure 4 except for a CH4 spectrum in the midwave band.

Figure 6.

Same as Figure 4 except for an HBr spectrum in the shortwave band.

[39] The gas cell tests were performed at three different instrument temperatures, for a total of nine tests (3 instrument temperatures by 3 bands). The band 3 shortwave test using HBr was invalid due to electronics saturation for one of the three instrument temperatures. The mean value for the Neon bulb effective wavelength was 3.3±0.8 ppm higher than the National Institute of Standards and Technology (NIST) value (after taking the off-axis divergence of the Neon beam into account). This is very good agreement and indicates excellent co-alignment of the Neon beam with the infrared optical axis. The consistency among the eight valid tests of 0.8 ppm also validates our test methodology.

[40] Once both the rigid focal plane position and the Neon calibration are adjusted, the frequency calibration errors for all FOVs are generally less than 3 ppm, within specification. These remaining errors were driven to zero by adjusting the radial position of each FOV (except FOV 5) for each focal plane as discussed above. The resulting values for the focal plane positions and Neon wavelength calibration for the TVAC test closest to the SNPP mission temperature were inserted into the CrIS engineering packet, which the CrIS SDR software reads to generate the apodization corrections and set the metrology laser wavelength.

[41] The above fits were all performed with the as-built detector sizes, which are all identical with a diameter of 0.963°. Using the final fitted focal plane positions, we then varied the detector diameter and found that any changes to the as-built value increased the observed minus computed residuals.

4 Post-Launch Calibration

[42] CrIS first began producing valid interferograms in late January and early February 2012 once the detectors were cooled to their operating temperature. However, calibrated radiance SDRs were not produced by the NOAA IDPS for several months due to software issues. Consequently, we used our SDR test bed software, CCAST [Motteler et al., 2013], to produce SDRs in order to calibrate the focal plane positions and Neon lamp wavelength. CCAST has many similarities to the operational SDR software but is based on an independent implementation of the apodization corrections.

[43] Figure 7 is a sample of a clear tropical ocean CrIS spectrum. This particular spectrum is Fourier interpolated to a finer grid than the official CrIS SDR product and provides a better relative view of the structure of the brightness temperature spectrum compared to the Nyquist-sampled version. A zoom of this spectrum shown in Figure 8 shows the difference between the Fourier-interpolated spectrum and the Nyquist-sampled spectrum (which is the official SDR product) in the 725–745 cm−1 CO2 sounding region. This figure clearly illustrates why the two post-launch spectral calibration algorithms presented here must Fourier interpolate the observed and computed spectra in order to determine the correct frequency calibration of these spectra.

Figure 7.

Sample boxcar CrIS spectrum for a clear tropical scene. This spectrum has been Fourier interpolated to a very small spectral point spacing.

Figure 8.

Zoom of Figure 7 for a portion of LW CO2 region with crosses marking the locations of the official CrIS wave number grid used for the SDR output.

4.1 Relative Frequency Calibration

[44] Two very different methods were used for determining the detector positions relative to the interferometer axis and the Neon lamp calibration. The first method, which we call relative frequency calibration, compared the observed radiances for FOVs 1–4 and 6–9 to the observed radiance for FOV 5. Fields of regard (FORs, meaning a 3×3 scene recorded simultaneously) with very uniform window channel radiances at 901.25 cm−1 were selected. The spectra were Fourier interpolated, and RMS difference between the radiance for FOV 5 and the other FOVs was computed for a range of frequency offsets. The frequency offset with the lowest RMS difference was deemed the frequency error relative to FOV 5. Using equation (3), one can then determine the radial positions of the corner and side FOVs. This method works very well because it does not depend on radiances computed from a model (such as ECMWF). An additional advantage of this relative approach is that the inter-FOV calibration differences are largely independent of the wavelength region chosen for the minimization process. It provides the best information on the spectral shifts relative to FOV 5 used to determine the focal plane geometry and works well for all three bands (LW, MW, and SW) but provides no information regarding absolute calibration. The SDR radiances with boxcar apodization are used in this technique; no additional apodization is applied.

[45] The initial relative frequency errors for band 1, longwave, are shown in Table 1(LW: Relative to FOV 5). Note, these results were derived from radiances produced by the CCAST test bed algorithm. Results for the midwave and shortwave bands are similar. A least squares fit of these errors to a rigid focal plane model gave a focal plane radial shift of 0.008° (140 μradians) relative to the TVAC value. To completely minimize the ppm errors, the individual FOVs (other than FOV 5) had to be moved very slightly towards FOV 5, indicating a slightly tighter CrIS telescope focus in-orbit relative to TVAC. These were all very small changes, and in fact the initial in-orbit frequency errors were all within the 10 ppm accuracy specification prior to optimization.

Table 1. Relative Frequency Differences in ppm Observed With In-Orbit Data From CCAST Using TVAC Focal Plane Positionsa
LW: Relative to FOV 5LW: Absolute minus FOV 5
  1. aThese differences are relative to FOV 5 only. FOV pattern same as in Figure 1. Table 1(LW: Relative to FOV5) uses the relative frequency calibration algorithm, while Table 1 (LW: Absolute minus FOV5) is derived from the absolute frequency calibration algorithm by subtracting the FOV 5-measured frequency error from the other FOVs.

4.2 Absolute Frequency Calibration

[46] The second calibration method compares clear scene observed radiances to radiances computed from 3 h ECMWF forecast/analysis model data, providing both absolute and relative calibration. A CrIS-specific version of the AIRS fast forward model, called SARTA [Strow et al., 2006], was used to generate the simulated radiances. Simulated radiances are generally not accurate enough for the observed minus computed approach used for relative calibration. Instead, we Fourier interpolate the computed spectrum to a fine-frequency scale and compute the cross-correlation between the observed and computed spectra, varying the frequency scale of the computed spectra until we find a peak in the cross-correlation. The SDR and computed radiances are Hamming apodized when using this approach. Very small residual errors in the CrIS SDR processing can distort the sinc response function, leading to spectral calibration errors that are largely suppressed with a Hamming apodization.

[47] This process is still quite sensitive to errors in the computed spectra. We simulated the cross-correlation frequency calibration approach using IASI data. Since the IASI data has higher spectral resolution, it is easier to use for frequency calibration. We established the truth calibration using the IASI data as is and then converted the IASI data to the CrIS spectral resolution to simulate frequency calibration for CrIS. The best results were achieved using spectral regions with a very flat baseline, i.e., regions not overly sensitive to the atmospheric lapse rate. The best region for band 1, longwave, was the window region from 775 to 825 cm−1, consisting mostly of water vapor lines and avoiding the O3 region that is hard to simulate. The best nominal region for band 2, midwave, was from 1489 to 1578 cm−1, a region of strong water vapor lines. Simulations showed that absolute calibration of CrIS using band 3, the shortwave, was not possible, with errors far greater than the specification of 10 ppm. This is due to the low spectral resolution in the shortwave.

[48] The CrIS cross-track viewing geometry introduces Doppler shifts to the observed radiances due to the rotation of the Earth. Doppler shifts reach a maximum for the highest scan angles at the equator, where the atmospheric motion relative to the instrument FOR is greatest. The CrIS scenes used here for frequency calibration were restricted to scan angles of ± 20°, which restricts any scene Doppler shifts to a maximum of ±1 ppm at the equator, depending on scan direction and descending versus ascending node. In addition, the daily averages of clear ocean tropical scenes used here will contain an antisymmetric mix of Doppler shifts, which will average to zero. Comparisons of daily averaged spectral calibration between ascending and descending nodes, for example, agree to within 0.04±0.18 ppm, indicating that the Doppler effect is indeed antisymmetric versus scan angle and averages out of our analysis.

[49] Table 1 (LW: Absolute minus FOV 5) shows results for the longwave absolute frequency calibration, where we have subtracted out the FOV 5 result (0.5 ppm) in order to directly compare the absolute calibration with the relative calibration results shown in Table 1(LW: Relative to FOV 5). Again note that these results were derived from radiances produced by the CCAST test bed algorithm. The relative ppm errors for band 1 agree quite well using these two different techniques, which provides some validation of the absolute frequency calibration approach, since the relative frequency calibration approach is more accurate. The FOV 5 absolute calibration offset of 0.5 ppm, which almost solely due to errors in Neon calibration, suggested that there was no need to change the Neon calibration from the TVAC value.

4.3 Calibration Corrections for IDPS Code

[50] The CrIS spectral calibration was subsequently repeated using the updated focal plane positions derived in section 4.1. We evaluated SDRs produced by both CCAST and by the off-line version of the operational IDPS software called ADL, described in Han et al. [2013], this issue. The Neon calibration was not changed from its preflight value based on the analysis presented in section 4.2. A full day's worth of radiances were produced (25 February 2012) using both ADL and CCAST, and the frequency calibration re-done using the new focal plane parameters.

[51] The frequency calibration derived from the ADL radiances, however, produced a different result than found using CCAST-generated radiances. Frequency errors for FOVs 1–4 and 6–9 using the ADL data were within 0.1 ppm of the data analyzed from CCAST. However, the FOV 5 frequencies were 1.4 and 2.2 ppm lower for bands 1 and 2 using ADL. This result was not consistent with the other eight FOVs and was assumed to be some kind of error or bug in the operational SDR software, which treats FOV 5 as a special case. (Subsequently, several bugs were found in the IDPS SDR software for FOV 5 of similar sign and magnitude, analysis by Han et al. [2013].

[52] In order to achieve a uniform frequency calibration among all FOVs, with the existing operational SDR software, the positions of FOVs 1–4 and 6–9 were moved slightly to create the same frequency calibration errors exhibited by FOV 5. This ensures a consistent frequency calibration among all nine FOVs for bands 1 and 2. No changes were made to band 3 (SW). The effect of these adjustments on the absolute frequency calibration were deemed too small to consider with only 1 day of data; therefore, the Neon calibration was not changed.

5 Post-Launch Validation

[53] Minor changes were made to the CrIS on-board digital filter coefficients in April 2012, to reduce variation in instrument response with sweep direction. Thus, we started detailed post-launch validation in mid-April 2012 and have since examined 1 year of data through April 2013.

5.1 Relative Frequency Calibration

[54] The relative frequency calibration for all three bands has been monitored for over 1 year. Table 2shows the yearly average of the IDPS SDR relative frequency calibration. These results exhibit extremely uniform spectral calibration among all FOVs for all bands, at a level far below NWP requirements, and generally below several tenths of a ppm. The standard error for these measurements are 0.003, 0.006, and 0.019 ppm, respectively, for the LW, MW, and SW spectral bands, indicating that the detector focal plane alignment relative to the interferometer optical axis is extremely stable. However, in subsequent processing, the relative calibration can probably be improved by further adjustment of the focal plane geometry based on the ppm errors shown in Table 2.

Table 2. FOV ppm Errors Relative to FOV5 for All Three Bands After In-Orbit Calibration of the Focal Plane Positions for Each FOVa
  1. aThese are averages over 1 year and use the IDPS official SDR product. The standard errors (2 σ) are relatively independent of FOV and are equal to 0.003, 0.006, and 0.019 ppm for the LW, MW, and SW, respectively.−0.23−0.12−0.31−0.06−0.41−0.68

5.2 Absolute Frequency Calibration

[55] The absolute frequency calibration results for SDRs created with the ADL/IDPS software and the final, corrected, focal plane positions are shown in Table 3. All errors are at the 1 ppm level or lower. These data are derived from the 25 February 2012 observations. The FOV 5 calibration error of −0.6 ppm was again considered too small to force a change in the Neon calibration, which has been left at the value determined during TVAC testing.

Table 3. Absolute ppm Frequency Calibration Errors for the Longwave Band Using ADL/IDPS SDRs Processed With the Operational Focal Plane Positionsa
  1. aThe nominal disagreement between the present Neon calibration and these observations would be the FOV 5 result of −0.6 ppm.

[56] The IDPS SDR algorithm only updates parameters dependent on the metrology laser wavelength if it changes by more than 2 ppm, as measured by the Neon calibration system. At that point, the processing stops and new apodization correction matrices are generated based on the new value of the metrology laser. Unfortunately, the actual metrology laser wavelength used to generate these tables is presently not reported in any of the official CrIS output files. (This omission was fixed on July 10, 2013.) Consequently, since it is not possible to know the exact metrology wavelength used to produce the observed CrIS radiance spectra to better than 2 ppm, the official SDR spectra cannot, in turn, be used to calibrate the Neon lamp to better than 2 ppm.

[57] The accuracy and stability of the CrIS frequency calibration is examined here using the absolute calibration approach discussed in section 4.2. Clear, ocean tropical scenes of near-nadir observations (scan angles within 12° of nadir) are subsetted from the SDRs and individually matched to the ECMWF forecast/analysis. Daily averages of the observed and computed spectra are generated separately for descending and ascending orbits. The computed spectra are Fourier interpolated and then cross-correlated with the observed spectra for a variety of frequency offsets, as described earlier. The frequency offset that produces the largest cross-correlation is the frequency calibration error of the observed spectra.

[58] We present here only descending orbit results, although as stated earlier descending and ascending frequency calibrations agree almost perfectly. Figure 9 plots the observed frequency calibration error over 1 year derived from both band 1 longwave and band 2 midwave, showing only the FOV 5 results. The relative calibration among all nine FOVs remains unchanged since February 2012. Superimposed on this graph is the relative variation of the metrology laser wavelength as measured by the Neon calibration system (and not necessarily used by the SDR algorithm). This curve has been offset by 0.6 ppm in order to agree with the observed band 1 calibration offset observed in late February 2012.

Figure 9.

Comparison of the Neon subsystem frequency calibration versus calibration using the upwelling radiances. The Neon has been offset by −0.61 ppm (see text). The sharp spikes in late June and mid-December 2012 are due to NPP spacecraft issues, not CrIS malfunctions. The upwelling calibration is for the daily average of FOV5, descending orbit using clear tropical ocean scenes. Note the band 2 midwave frequency calibration agrees very well with the band 1 calibration.

[59] The frequency calibration derived from bands 1 and 2 agrees very well, differing by only 0.19 ppm on average. They both track the Neon calibration (metrology laser) quite accurately, with a peak-to-peak variability of about 2 ppm over 1 year. The spikes in the metrology laser wavelength are due to SNPP spacecraft issues, not CrIS malfunctions, and reflect true variations in the metrology laser wavelength. Changes in the metrology laser wavelength of greater than 2 ppm in mid-December should have triggered one or more recomputations of the apodization correction matrices within the SDR algorithm. Since the SDR output files do not contain the metrology laser wavelength currently used in the apodization correction matrices, it is difficult to determine the exact metrology laser wavelength in use once the system restabilizes.

[60] The Neon calibration system is clearly working very well and is tracking the shifts in the metrology laser that we also can derive from the observed radiances. This suggests that a smoothed version of the Neon calibration could be used to update the metrology laser wavelength used in the SDR algorithm more often and not wait for a 2 ppm shift.

[61] The cause of the 2 ppm peak-to-peak variability in the frequency calibration of the upwelling radiance seen in this figure is unknown. We do find that these frequency shifts are highly correlated with one of the interferometer baseplate temperatures, denoted “oma1”. Figure 10 plots the Neon calibration curve on the same graph as the oma1 temperature, showing a high degree of correlation between these two curves. A linear regression of the Neon calibration versus the oma1 temperature provides an extremely accurate estimate of the Neon calibration, as seen by the green fit residual in Figure 11. Starting in March 2013, this regression fit does appear to deteriorate by about 0.25 ppm. Note that even a 2 ppm error only translates into 0.04 K for CrIS Hamming-apodized radiances and these errors can be further reduced by a factor of 4× if a reprocessing algorithm such as CCAST is used that follows the Neon lamp calibration, unlike the IDPS algorithm.

Figure 10.

Comparison of the Neon calibration (as shown in Figure 9) and the “oma1” interferometer baseplate temperature.

Figure 11.

Fitting error (green) between the Neon calibration and a linear regression equation relating the “oma1” temperatures to the Neon calibration. Note the small shift in the regression solution starting in late February 2013.

[62] The root cause for the variability in the SDR frequency calibration is unknown but can be explained by a ±16 millikelvin variation of the laser metrology diode junction temperature (±1 ppm wavelength variation). Although the laser diode heat sink is thermally controlled to ±5 millikelvin tolerance, this is not the same as controlling the diode junction temperature. It is suspected that electrical wiring to the diode may be altering the junction temperature slightly due to thermal conduction via wiring to the OMA baseplate. When Neon calibration is used to determine laser metrology wavelength during normal operations, this thermally induced error is eliminated.

[63] Figure 12 shows the daily difference between the metrology laser wavelength as measured by the Neon lamp versus those measured using the upwelling SDR radiances, with a smoothed version of this difference shown in red. Again, the metrology laser wavelength has been shifted to agree with the measured frequency shifts in February 2012. This figure suggests that the SDR algorithm has indeed possibly used three different values for the metrology laser in the past year. Shifts are evident in June 2012 and after the December 2012 event. At this time we cannot identify why the metrology laser wavelength used in the SDR algorithm might have changed in the June time frame. However, there is no concrete evidence for any shifts in the Neon calibration system.

Figure 12.

The difference between the Neon calibration and upwelling calibration shown in Figure 9. This indicates that the CrIS SDR algorithm changed the apodization correction operator during the December 2012 event, as would be expected for a Neon shift of more than 2 ppm. We are uncertain as to the cause of the late May, early June shift. Data are missing in this time period due to loss of ECMWF model forecast files.

5.3 Residual Spectral Signatures

[64] Numerous other sources can contribute to CrIS radiance errors. However, spectral calibration errors will generate a well-defined pattern in radiance spectra. One way to examine error patterns in the SDR radiance product is to compare the bias versus NWP-computed spectra for boxcar spectra versus Hamming-apodized spectra.

[65] Figure 13 shows the double difference of the NWP boxcar-apodized bias minus the Hamming-apodized bias or

display math(4)

using yearly averages tropical clear ocean spectra taken from the IDPS SDR radiance product. Spectrally smooth CrIS radiometric errors and simulations errors (NWP and radiative transfer errors) are removed from the boxcar biases with this approach. The sensitivity of the Hamming radiances to spectral calibration errors is about 10× lower than the boxcar radiances; thus, signatures of spectral calibration errors will largely remain in the double differences. What also remains are CrIS sinc ILS errors, specifically deviations in the observed ILS relative to a boxcar ILS.

Figure 13.

Boxcar minus Hamming brightness temperatures averaged over a large set of clear tropical ocean scenes. Also plotted is the change in brightness temperature biases versus NWP, for a 1 ppm metrology laser offset.

[66] For reference, the brightness temperature error induced by a 1 ppm frequency calibration error is shown at the bottom of Figure 13. In general, the patterns in the DD plots do not match the 1 ppm error pattern, although there is some similarity in FOVs 3, 5, and 9 in the longwave with a magnitude similar to the 1 ppm error plot. For clarity a zoom of these results for FOV 9 is shown in Figure 14.

Figure 14.

Zoom of Figure 13 for the LW and MW bands showing only FOV9.

[67] Excessive ringing is seen in the shortwave, which is easily visible in the boxcar bias spectra. Most interesting is the out-of-family patterns seen in the FOV 5 DD in all three bands but especially in the longwave and shortwave. This suggests that the existing SDR algorithm is not treating FOV 5 correctly. As previously mentioned [Han et al., 2013], small coding/algorithm errors have recently been uncovered in the operational SDR code for FOV5.

[68] This discussion is not meant to be a definitive study of possible excessive ringing in the SDRs radiances. Note that almost all applications apodize the spectra to lower the sinc side lobes, which lowers the errors shown in these figures by roughly a factor of 10×. Our main intent is to highlight the fact that the SDR sinc ILS radiances do not exhibit any clear signs of frequency calibration errors above the 1 ppm level.

6 Conclusions

[69] The CrIS spectral calibration uncertainty is well below the 10 ppm requirement. Relative frequency calibration among FOVs is well below 1 ppm, small enough that NWP assimilation systems can treat different FOVs as a single system for bias correction, at least in terms of frequency errors. The absolute spectral calibration of the operational CrIS SDRs archived at NOAA CLASS and used by various weather services is estimated to be in the 2–3 ppm range for bands 1 and 2 based on the data shown in Figure 9 and is limited primarily by the operational processing software that only updates the spectral calibration coefficients if the Neon calibration system sees a change in the metrology laser wavelength of more than 2 ppm. The metrology laser appears to vary slightly in response to the instrument thermal environment with a yearly variation of approximately 2 ppm peak-to-peak. This work has shown that the CrIS Neon calibration system accurately tracks these small metrology laser drifts. Thus, a processing system that closely tracks the Neon calibration lamp should provide CrIS radiances with relative frequency calibration to better than 0.5 ppm with an estimated absolute accuracy of ∼1 ppm, based on analysis of the upwelling radiance spectra versus computed radiances.


[70] This work was supported by the NOAA Joint Polar Satellite System Office under grant number NA11NES4400001 and by the NASA Suomi NPP Science Team grant NNX11AK78G. We also would like to acknowledge the many people at ITT Exelis who contributed to the successful spectral calibration of CrIS. Also, Breno Imbiriba, Paul Schou, and Sergio De Souza-Machado at UMBC contributed to the CrIS clear subsetting and the computation of simulated radiances. Finally, we thank Dan Mooney of MIT Lincoln Labs for sharing his encyclopedic knowledge of CrIS and for providing us with software for reading the CrIS data packets.