For pairs of Sun-synchronous polar-orbiting satellites with different altitudes, theoretically, their nadir points cross each other within a few seconds near the north and south polar regions (typically in the 70–80∘ N and 70–80∘ S latitude zones) for a certain time period (dependent on orbit altitude difference), which are referred to as SNO observations (Cao et al., 2004). When a SNO occurs, the radiometers from both satellites view the Earth at the same place and same time from different altitudes, which greatly reduces the comparison uncertainties related to the difference of satellite observational time and viewing geometries. One of the limitations of high-latitude SNO-based comparison is that the instruments cannot be compared over a full brightness temperature (BT) dynamic range. However, Aqua and SNPP satellites basically have the same equatorial crossing time (∼ 13:30 LST) but different altitudes (SNPP's altitude is 824 km and Aqua's is 705 km). Consequently, every few days, these two satellites are stacked nearly on top of each other with a wealth of coincident data through the full Aqua swath, yielding temporal and spatial coincidences varying with latitude, though with a larger time difference than high-latitude SNOs (Uprety et al., 2013). This provides ample opportunities to compare SNPP CrIS measurements with Aqua AIRS at different latitudes with a large dynamic range.

Figure 2a shows time series of SNO latitudes for CrIS–AIRS, CrIS–IASI/A, and CrIS–IASI/B. Since SNPP and Aqua satellites are both in afternoon orbits, the SNO events between SNPP and Aqua occurred every 2 days. They were most concentrated in the south and north polar regions and the tropics, while some were scattered in the mid-latitude regions. On the other hand, since MetOp-A and MetOp-B satellites are in mid-morning orbits, the SNOs between MetOp and SNPP occurred roughly every 50 days. Fortunately, once an SNO event occurs, their orbits cross each other in a number of consecutive orbits when satellites fly to the polar regions. Therefore, many SNO observations can be collected during each SNO event. It can be clearly seen in Fig. 2b, showing an enlarged plot of Fig. 2a for a SNO case between CrIS and IASI/B on 1 March 2013.

An example of a SNO event of SNPP and MetOp-A at 18:22 UTC on 27 August 2013 is presented in Fig. 3, where CrIS nadir footprints (indicated by the red circles) and IASI nadir footprints (represented by the blue circles) crossed each other almost at the same time. Since their orbits cross each other at a relatively large angle (see Fig. 3), only measurements in two nadir FORs with smallest scan angles (±1.6∘) are extracted for comparison in order to reduce the uncertainties from the orientation of matchup data. CrIS has better spatial coverage than IASI. Accordingly, two SNO cases between AIRS on Aqua and CrIS on SNPP are given in Fig. 4, including the polar SNO case occurring at 21:49 UTC and the tropical SNO case at 16:15 UTC on 30 August 2013. Different from the CrIS–IASI SNO case, the SNPP and Aqua orbits follow each other. Therefore, large parts of their swaths overlapped, while the CrIS swath is wider than AIRS. As a result, a different strategy is used to collocate CrIS–AIRS measurements. First, CrIS measurements falling in two nadir FORs are extracted. In the following steps, AIRS pixels that are overlapped with these nadir CrIS measurements are searched by checking collocation criteria. Finally, CrIS and AIRS spectra are paired together for comparison.

Table 2 summarizes the threshold values used for spatial collocation of CrIS and AIRS/IASI FOVs. In order to collocate CrIS FOVs with AIRS and IASI FOVs, we assume that (1) the FOVs of AIRS, CrIS, and IASI are approximately circular at nadir, (2) the spatial response is uniformly distributed within the FOVs, and (3) there is no variation of the spatial response between the different channels (e.g., from long-wave channels to short-wave channels). Note that in reality, the AIRS FOVs were trimmed on both sides by the blinds attached to the AIRS FOV to avoid overlapping effects between neighboring footprints (Schreier et al., 2010).

Considering their different spatial size of FOVs at nadir – 12.0 km for IASI, 13.5 km for AIRS, and 14.0 km for CrIS – we use the distance between two FOVs to pair overlapped FOVs for CrIS and AIRS/IASI. Specifically, a threshold value of 6.5 km is used for CrIS and IASI and 6.875 km is used for CrIS and AIRS. In addition, the observational time and view angle differences (see Table 2 for the specific values) are further examined to filter out the data. Note that the time difference threshold is increased to 15 min for the AIRS–CrIS tropical SNOs because of their large time differences (Uprety et al., 2013).

Finally, for a typical interferometer like CrIS or IASI, instrument FOV geometry (including beam divergence introduced by finite FOV and off-axis effects caused by its position on a focal plane) and instrument imperfections (such as various optical, mechanical, and electrical imperfections of the interferometer system during the sweep of the optical path difference) impose an apodization function on its interferogram or the so-called self-apodization. The self apodization of the interferogram results in an instrument line shape (ILS) that is shifted and stretched in frequency as well as an amplitude reduction, which has been corrected for CrIS and IASI spectra. However, radiance nonuniformity within the instrument's FOV also affects the ILS associated with each true wavenumber, which has not been taken into account in CrIS and IASI radiance products (Genest and Tremblay, 1999; Tobin et al., 2009). Moreover, inhomogeneous scenes can introduce spatial collocation uncertainties owing to the different point spread functions of CrIS and IASI/AIRS. To avoid these uncertainties, only uniform scenes are selected. We use the collocated Visible Infrared Imaging Radiometer Suite (VIIRS) pixels to check CrIS–AIRS/IASI collocated scene homogeneity by examining the standard deviation to the mean ratio of the VIIRS band M16 (with a central wavelength of 11.8 µm) radiances from the pixels falling in corresponding CrIS FOVs (Wang et al., 2013). On-board the same satellite platform, VIIRS is a whisk-broom scanning imaging radiometer, collecting visible and infrared imagery of the Earth through 22 spectral bands between 0.412 and 12.01 µm. These bands include 16 moderate resolution bands (M-bands) with a spatial resolution of 750 m at nadir, five imaging resolution bands (I-bands) with a 375 m at spatial resolution nadir, and one panchromatic day–night band (DNB) with a 750 m spatial resolution throughout the scan. VIIRS has four bands that are fully overlapped with CrIS, that is M13 (4.07 µm), M15 (10.7 µm), M16 (11.8 µm) and I5 (11.45 µm). Intercomparison between CrIS and VIIRS indicates good agreement between M16 and CrIS (Wang et al., 2012) and therefore VIIRS M16 band is used to characterize the CrIS scene uniformity. Specifically, scene uniformity of a circular area centered at CrIS FOVs with a radius of 14 km is checked. The area basically includes both collocated CrIS and IASI or AIRS FOVs and actually is the environment surrounding the collocated measurements. A spatially uniform environment surrounding the collocated measurements is desirable to compensate for minor violations of collocation and coincidental criteria as well as to reduce the uncertainties caused by navigation errors. Furthermore, it avoids a potential threat of nonuniform features (e.g., moving clouds) to the collocation. Demonstrated in Fig. 5 is the BT difference between CrIS and IASI/B at 900 cm-1 vs. the standard deviation to mean ratio of the VIIRS radiances in band M16. It shows that with a larger ratio, the spread of the BT difference becomes larger, whereas the BT differences converge when the ratio values decrease. From a practical point of view, this ratio is an effective method to constrain scene homogeneity to reduce the comparison uncertainties, specifically when a smaller value is chosen.

Similar sensitivity tests have been performed to check how the relaxed collocation criteria affect the BT differences. It is found that the FOV distance and collocation environment uniformity are the major factors that control the scattering pattern of BT differences and sample number, while time differences have small effects. For CrIS and IASI SNOs, since only observations in two nadir FORs are used, all the samples meet the threshold of angle differences. On the other hand, for CrIS and AIRS SNOs, in order to meet the angle difference threshold, the CrIS FOVs' satellite zenith angles are in the range of 0.6–3.36∘ while the collocated AIRS FOVs' zenith angles range from 0.5 to 8.0∘. In conclusion, constricting collocation criteria reduces both the sample size and the scattering of BT differences. However, as shown later, we still have enough samples to cover the instrument dynamic range.

Figure 6 gives the mean spectra of paired CrIS and IASI SNO FOVs for the SNO event shown in Fig. 3. During this SNO event, CrIS on SNPP was operated in the FSR mode. Accordingly, CrIS spectra are collocated at the full resolution mode, indicated by red lines (Chen et al., 2014). The full resolution spectra can be further processed to be the normal mode data (like current officially released CrIS SDR data) by the ADL software, which are represented by blue lines in Fig. 6. In order to clearly show the agreement between CrIS and IASI, an enlarged plot is given in Fig. 6b at the spectral range from 2200 to 2400 cm-1. However, in order to quantify the detailed differences between them, we need to convert CrIS and IASI onto common spectral grids, which are discussed below.

This part describes the method of how to convert paired CrIS–IASI and CrIS–AIRS spectra onto common spectral grids in order to quantify their radiometric differences.

After the above two steps, the CrIS–AIRS and CrIS–IASI spectra are spatially and spectrally matched together. A whole year of data in 2013 is processed based on the above steps, including CrIS–IASI/A, CrIS–IASI/B, and CrIS–AIRS. For each pair of sensors, the SNO observations are categorized into the SNO observations at the north polar region (north SNOs), those at the south polar region (south SNOs), and those in the tropics (tropical SNOs, CrIS–AIRS only).

Based on the collocation criteria, a total of 2748 pairs of spectra were collocated for CrIS and IASI-A in 2013, composed of 1274 pairs from north SNOs and 1474 pairs from south SNOs. For CrIS and IASI-B, there are a total of 2008 pairs of the SNO spectra in 2013, categorized as 952 pairs from north SNOs and 1056 pairs from south SNOs, which are fewer than CrIS–IASI/A due to the fact that IASI-B data have been distributed under operational mode since 24 April 2013. Given a sufficient number of collocations, the radiometric noise from each instrument should be canceled out and therefore should not contribute to the mean difference.

For CrIS and AIRS, due to the fact that the two instruments follow each other, each SNO event has hundreds (around 200–600) of collocated spectra that meet the collocation criteria. Consequently, the CrIS and AIRS spectra are averaged for each SNO event and then their BT differences are then computed in order to show how their BT differences vary with time. Specifically, there are 324 and 390 SNO events that occurred in the polar regions and the tropics, respectively.

Figures 9a and 10a show CrIS spectral distribution for CrIS–IASI/A and CrIS–IASI/B SNO observations, where the red lines indicate the observations for north SNOs and the blue lines represent those for south SNOs. If one examines the spectra at the atmospheric window regions (850–900 cm-1), the BT values range from 220 to 295 K in north SNOs, while they range from 190 to 270 K in south SNOs. Especially, from 04:00 to 05:30 UTC on 3 August 2013 near the north of middle Siberia (94∘ E, 72∘ N), there are 16 samples whose BT values at atmospheric windows regions are larger than 290 K. This suggests that CrIS–IASI SNO observations have a relatively large dynamic range for both CrIS–IASI/A and CrIS–IASI/B SNO samples. The mean and standard deviation of BT differences of CrIS–IASI/A are given in Fig. 9b, while those from CrIS–IASI/B are shown in Fig. 10b. At LWIR and MWIR bands, CrIS agrees well with IASI and the BT differences are less than 0.2 K, while the standard deviation ranges from 0.2 to 0.7 K dependent on different spectral absorption regions. Compared with both IASI/A and IASI/B, CrIS is slightly warmer than IASI (less than 0.2 K) at most spectral regions. However, at SWIR band, the BT differences range from 0.2 to 1.0 K and the standard deviation also increases, especially for the south SNOs. It is not surprising that BT differences are larger at the SWIR band because the radiance values in SWIR are much smaller than LWIR and MWIR bands given the same BT values. The explanation of the large standard deviation probably is due to the decreasing radiance value at the short-wave band for cold scenes for both CrIS and IASI. When one examines the mean BT differences of CrIS–IASI, a considerable sharp increase can be found at spectral line transition region (around 2400 cm-1), which are shown in both Figs. 9b and 10b. It is probably caused by two reasons. First, given a relatively large spectral resolution of 2.5 cm-1 for the CrIS SWIR band, it is challenging to achieve the same spectral accuracy as the LWIR and SWIR bands (Strow et al., 2013b). Therefore, the spectral calibration for SWIR band is not as good as LWIR and MWIR bands. Secondly, it is speculated that there is a small nonlinear response for the detectors in SWIR band, which currently are treated as linear detectors and there is no nonlinear correction for the CrIS SWIR band (D. Tobin, University of Wisconsin, personal communication, 2013). However, the investigations are not conclusive at this point and the root cause is still under investigation.

When one carefully compares CrIS–IASI/A and CrIS–IASI/B results, especially for results of the south SNO samples, we can find that CrIS–IASI/A and CrIS–IASI/B are slightly different. If one treats CrIS as a transfer radiometer and then CrIS measurements are assumed to be canceled out, the double differences of CrIS–IASI/B and CrIS–IASI/A indicate that IASI/B is slightly warmer (less than 0.1 K) than IASI/A in LWIR bands and they are well consistent with each other in MWIR bands. The observed differences between IASI-A and IASI-B in the long-wave band are suspected to be due to the differences of instruments' nonlinearity corrections. This result agrees well with the findings by D. Jouglet from CNES (Jouglet et al., 2014), who directly compared IASI/A and IASI/B spectra with 50 min difference. However, the double difference results should be carefully interpreted because two data sets were collected in different time periods and the bias from the transfer radiometer CrIS may not be totally canceled out.

In order to show how the BT differences are dependent on scene temperatures, scatter plots of CrIS–IASI BT differences vs. CrIS BTs at three individual channels (including one CO2 absorption channel, one long-wave window channel, and one water vapor channel) are shown in Fig. 11 for CrIS–IASI/A and in Fig. 12 for CrIS–IASI/B, correspondingly. Specifically, the red dots indicate the observations from north SNOs and the blue dots represent those from south SNOs, while the green lines give the linear fitting line. For all three individual channels, the slope of linear regression is very small albeit statistically significant, indicating that there are no strong scene-dependent features. This also applies to the BT differences between CrIS and IASI/B, shown in Fig. 12. However, it is necessary to point out that CrIS–IASI comparison is only limited to relatively cold scenes in the polar regions and the comprehensive assessment of CrIS–IASI radiometric consistency still needs to be extended to the low-latitude regions in the future.

The CrIS–AIRS SNO comparison results are shown in Fig. 13, including CrIS spectral distribution and the mean and standard deviation of CrIS–AIRS BT difference. They are categorized into north, south, and tropical SNOs and the global data set that combines all three SNO data sets together. In addition, the statistics of CrIS–AIRS BT differences along the 25 spectral regions are summarized in Table 3. The SNO spectral distribution clearly shows the characteristic in different climate regions if one visually inspects the dynamic range of the BT values at the atmospheric window region. In other words, the BTs from tropical SNOs (ranging from 265 to 310 K) are different from polar SNOs (ranging from 210 to 280 K). The following findings are revealed based on Fig. 13 and Table 3. First, at the 650–681.875 cm-1 and 690–710 cm-1 spectral regions in LWIR band, CrIS is warmer than AIRS for all three SNO data sets and the BT differences range from 0.026 to 0.167 K. At the other 10 spectral regions (spectral region 3–11 and 12 in Table 3) in LWIR band, CrIS and AIRS agree well and the BT differences are less than 0.1 K for north and south SNOs; however, CrIS is warmer than AIRS by ∼ 0.1 K for tropical SNOs. At the ozone absorption region of 1075–1095 cm-1 in LWIR band, CrIS is slightly colder than AIRS and the BT differences range from -0.02 to -0.161 K. Second, among six of eight spectral regions (spectral region 14–19 in Table 3) in the MWIR band, CrIS–AIRS BT differences are less than or around 0.1 K for all three SNO data sets. Only at the 1240.50–1262.00 and 1591.25–1613.00 cm-1 spectral regions, are BT differences larger than 0.15 K in some SNO data sets. Finally, in the SWIR band, CrIS is warmer than AIRS from 0.01 to 0.42 K for south and north SNOs but their agreement is much better for tropical SNOs. To combine the three SNO data sets together, Fig. 13 and Table 3 also summarize the global CrIS–AIRS BT difference statistics. Generally speaking, among 21 of 25 comparison spectral regions, the CrIS–AIRS BT differences are less than or around 0.1 K; in the remaining four spectral regions (one in the LWIR band and three in the SWIR band), the BT differences range from 0.15 to 0.21 K. Moreover, among 16 of 25 spectral regions, CrIS is slightly warmer than AIRS.

Given the fact that each SNO data set has its own unique BT distribution, it is possible to check their BT differences vs. scene BTs with a relatively large dynamic range. Four spectral regions are selected, including the CO2 absorption region (650–681.88 cm-1), the atmospheric window region (889–903 cm-1), the water vapor absorption region (1380–1410 cm-1), and the short-wave region (2190–2260 cm-1). Finally, the three SNO data sets can be simple combined together to derive the global statistics. Figure 14 gives CrIS–AIRS BT differences varying with CrIS BTs, where the red, blue, and black dots indicate the data from north, south, and tropical SNOs, respectively. Specifically, the BT differences at the CO2 absorption, water vapor absorption band short-wave regions show weak scene-dependent features. At the atmospheric window region, there is no scene-dependent feature for polar SNOs (range from 210 to 270 K), but the BT difference increase with the scene BTs for tropical SNOs when BTs are larger than 270 K. These features can be clearly revealed by the regression line shown in Fig. 14. One explanation of these scene-dependent features can be due to the imperfection of the instrument's nonlinearity correction. For AIRS and CrIS instrument, nonlinearity correction has been performed during radiometric calibration to correct the nonlinear response for the instrument detectors (Pagano et al., 2003; Tobin et al., 2013b). The deficiency of nonlinearity correction for one or both of instruments can result in this kind of scene-dependent pattern based on the previous study (Tobin et al., 2013b). Figure 15 shows the AIRS–CrIS BT difference time series at the four spectral regions. The BT difference time series from polar SNOs show weak seasonal variations with magnitude of 0.05 K in the water vapor absorption regions, 0.1 K in the CO2 absorption region, and 0.15 K in the short-wave regions. This can be explained by the scene-dependent BT difference features shown in Fig. 14. In other time series of the BT differences, there is no apparent seasonal variation and there are no long-term trends.

Due the fact that both CrIS and AIRS are on afternoon orbits, one of the advantages is that the CrIS–AIRS intercomparison can be performed globally. On the other hand, however, because of different instrument characteristics (grating spectrometer vs. interferometer), it is very hard to compare them at the finest spectral scale. Consequently, a simply average method is carried out through several boxcar-passed spectral regions. This method has been used by different groups. While the results indicate that CrIS and AIRS have small radiometric differences, it very difficult to compare them because these comparisons are performed at different comparison spectral regions. We suggest that another round of CrIS and AIRS intercomparison should be planned by choosing common comparison spectral regions by different groups in order to further advance our understanding on their radiometric differences.

For CrIS and IASI intercomparison, the results from different group are consistent with each other. Specifically, while different collocation methods and criteria are used, this study and the one by Tobin et al. (2013a, b) both identify the large BT differences at the spectral transition region of about 2400 cm-1. No matter whether CrIS and IASI are compared at finest spectral scale or boxcar-averaged spectral regions, all the studies show CrIS is slightly warmer than IASI at LWIR band. However, CrIS–IASI comparison is limited to the polar regions and thus future studies should extend the comparison into the mid-latitude and tropical regions. The double difference method relative to a common transfer target (e.g., a transfer radiometer and radiative transfer model simulations) may be feasible in future studies.

As a final note, we would like to point out that, given the fact that the radiometric differences do exist between hyperspectral instruments, the intercalibration community should be cautious to use only one hyperspectral sounder as a prime reference. A best practice should compare the broad- or narrow-band instruments against all four hyperspectral sounders to avoid biasing the results.