For the Level 1C dataset, several updates were made to the corrections and calibration adjustments for the six SSM/I instruments from the original SSMI(S) FCDR. This work was done to extend the Level 1C data back to SSM/I on board DMSP F08 in July of 1987 in a manner that is fully consistent with the TRMM and GPM-era radiometers shown in
Figure 2. Since this work was just completed and has not been published elsewhere, details of the updated SSM/I corrections including the cross-track bias corrections, geolocation and view angle updates, as well as the intercalibration, are provided in the following section. Note that several aspects of the original SSMI(S) FCDR including the quality control, antenna pattern corrections, updates to the spacecraft ephemeris, etc. remain unchanged.
3.1.1. Cross-Scan Biases
An analysis of the F08 to F14 SSM/I instruments [
17] found deviations near the edges of the scan, with a significant decrease in the Tb of 1.5–2.0 K near the right edge for all instruments. The reason for this decrease is due to an obstruction by the spacecraft and/or cold-space mirror partially intruding into the field-of-view. In the initial version of the SSMI(S) FCDR, a multiplicative correction for this falloff was applied based on an analysis of mean, clear-sky antenna temperatures over ocean. This correction, however, did not take into account the effect of instrument mount offsets. An offset in the spacecraft roll direction will impart an asymmetric slope in the antenna temperatures across the scan, while an offset in the spacecraft pitch direction will result in a symmetric curvature across the scan [
4]. These effects are most evident over ocean for the lower frequency (i.e., 19, 22, 37 GHz) vertically-polarized (v-pol) channels due to the significant change in Tb with view angle.
To account for these effects, differences between the mean observed and simulated cross-scan Tb were computed for each sensor and channel. The top panel in
Figure 3 shows the resulting cross-scan pattern (blue line) for the 19 GHz vertically-polarized channel on the F10 SSM/I. The results for this channel are shown, as they clearly exhibit a left to right slope across the center of the scan associated with an offset in the roll direction, as well as a slight downward curvature, corresponding to a negative pitch. In this case, a roll of −0.08 and a pitch of −0.10, based on simulated Tb, are fit to the center part of the scan (black dots) in order to avoid the edge of scan falloff. The resulting fit is shown by the solid black line. Note that this fit corresponds to a mean roll and pitch value derived using the 19v, 22v, and 37v channels over a 2-year period. For each satellite, the results were checked for consistency between two different years and for the three low-frequency v-pol channels. In all cases, variations in the derived roll and pitch values between channels/years were within a few hundredths of a degree. The final roll and pitch values derived from the cross-scan analysis are given for each sensor in
Table 2.
Once the mean roll and pitch offsets were determined, the data were reprocessed for each sensor to provide updated view angles, or Earth incidence angles (EIA), accounting for the roll and pitch offsets. Observed minus simulated Tb were then recomputed across the scan both over cold ocean scenes and over warm vegetated land scenes. The resulting cross-scan bias patterns are shown in the lower panel in
Figure 3. There are slight differences between the cold and warm results, with the most significant being a larger falloff on the right side of the scan at warmer temperatures. This is consistent with the results found for TRMM TMI [
15], indicating a fraction of the signal near the edge of scan is coming from the satellite or reflected from cold space. This means that the same approach used for TMI should work for SSM/I, which makes sense, since TMI is a modification of the SSM/I design. To correct for the edge-of-scan falloff, therefore, we use the equation given in [
15], which solves for the temperature and fraction of the signal coming from the edge-of-scan obstruction. The result is a flat line for both cold and warm scenes, and is thus not shown.
3.1.2. Geolocation Analysis
The roll and pitch offsets computed from the cross-scan Tb bias patterns only describe two degrees of freedom for the sensor mount angles. To determine offsets in the spacecraft yaw direction, as well as a potential offset in the view angle or half-cone angle, we perform a coastline analysis similar to that used for the SSMI(S) FCDR [
4]. Note that for SSM/I, the nominal off-nadir view angle is 45 degrees, so the half-cone angle offset derived here is added to the nominal value. At SSM/I frequencies, there is a large contrast between land and ocean Tb, and small offsets in yaw or half-cone angles impact the geolocation, which causes an apparent shift in the coastlines. This shift is in opposite directions for ascending versus descending passes. Creating ascending minus descending gridded Tb maps makes the coastlines stand out where the geolocation is off. Offsets for the yaw and half-cone angles are determined by minimizing the root mean square error (RMSE) of the Tb difference maps around the coastline regions. While this procedure is similar to what was done for the SSMI(S) FCDR geolocation [
4], that analysis used pre-launch estimates of the half-cone angle, or elevation offsets, from [
17]. By deriving both roll and pitch from the cross-scan Tb analysis, instead of just the roll offset as in [
4], we are able to derive both a yaw and half-cone angle offset based on the coastline analysis.
The spacecraft ephemeris calculated from two-line element (TLE) data files is used along with the pitch and roll angles from the cross-track scan Tb analysis, and the yaw and half-cone angles are allowed to vary. The yaw and half-cone angles that minimize the RMSE at the coastlines are determined for each of the seven channels and six SSM/I radiometers. The Tb data are gridded at 0.25° over a full year in order to minimize the impact of geophysical variability and sampling noise. Australia was chosen for the geolocation analysis as it has large sections of coastlines in every direction.
Figure 4 shows the derived values of the yaw and half-cone angle offsets for the F10 19v, 22v, and 37v channels for two different years, both early and late in the mission lifetime. Results for the other four channels are derived to verify consistency, but only 19v, 22v, and 37v are used to derive the final numbers, as these channels are the least impacted by geophysical variability.
Another significant difference between the SSMI(S) FCDR and the Level 1C datasets is that the pitch, roll, yaw, and half-cone angle offsets are fixed for each SSM/I sensor for the entire mission in the Level 1C, whereas the SSMI(S) FCDR dataset derived time-dependent pitch, roll, and yaw angle offsets [
4]. There is some indication from
Figure 4 that the F10 yaw offsets change between 1992 and 1996 by ~0.07°, as all three channels show this same difference. It should be noted that the DMSP F10 spacecraft is in a more eccentric orbit than the other five satellites carrying the SSM/I radiometers. This results in significantly larger altitude variations, and thus, variations in EIA of well over 1 degree for this instrument; however, the results indicate similar consistency in the derived roll, pitch, yaw and half-cone offsets over time, as for the other sensors. The other SSM/I sensors also exhibit small differences between the two years analyzed, though differences from the mean values are less than 0.05° for all cases. Also, differences or errors in the yaw offsets have no impact on the EIA, and thus the calibration, only on the calculated pixel latitude and longitude. Since the mount angles are not expected to change over time, the explanation for these time-dependent differences is likely due to changes in spacecraft attitude control or timing errors. Note that timing errors have the same impact as yaw offsets, and therefore, cannot be derived independently. While attitude errors might result in a slow change over time as the spacecraft orbit decays, it is also possible that changes could occur abruptly or more quickly than can be assessed using the coastline geolocation technique. As a result, we chose to use fixed values for the pitch, roll, yaw, and half-cone angle offsets and to consider these small time-dependent changes as a residual geolocation error, and with the exception of the yaw angle offsets, in the calculated pixel view angles.
Table 2 gives the updated pitch, roll, yaw, and half-cone angle offsets used for the Level 1C dataset.
Figure 5 shows a comparison between the SSMI(S) FCDR and Level 1C geolocation for the F10 19v channel. The ascending minus descending gridded Tb are shown for the Australian coastline, divided into three scan position groups: left (scan positions 6 to 23), center (24 to 41) and right (42 to 59). While there are only minor differences between the SSMI(S) FCDR and Level 1C geolocation at the center of the scan, there are noticeable differences between the left and right sides of the scan. Since the SSMI(S) FCDR uses pre-launch values of the half-cone offset, the derived pitch has to account for offsets in the cone angle. Pitch and half-cone angle offsets have a similar impact on the geolocation at the center of the scan, but exhibit different behavior at the scan edges [
15]. The Level 1C geolocation shows a similar pattern for the left, center, and right subsets of the scan, which confirms the accuracy of the pitch and roll values calculated from the cross-track scan Tb analysis.
3.1.3. SSM/I Level 1C Intercalibration
The SSM/I intercalibration for the Level 1C dataset uses GMI as the calibration reference, and applies a linear intercalibration adjustment based on differences over both cold ocean scenes and warm, unpolarized, vegetated land scenes. The dominant sources of calibration error are the antenna pattern spillover, emissive reflectors, and warm calibration target temperature errors, all of which can be corrected using a 2-point linear correction [
12]. Analysis of on-orbit data from deep-space calibration maneuvers for GMI showed that errors in the spillover correction alone can easily result in scene-temperature dependent errors of several Kelvin, particularly for lower frequency channels with beam efficiency values of 95% or less [
11]. Although there is no overlap between the SSM/I and GMI data records, the GMI calibration was linked back in time by first intercalibrating TMI using 1+ years of overlapping data [
16], then using TMI to intercalibrate SSM/I on board F11, F13, F14, and F15, and finally linking the F11 calibration to F10 and then to F08. As mentioned previously, the original SSMI(S) FCDR was intercalibrated to the F13 SSM/I, which was chosen for its stability and the length of the available data record, not as an absolute calibration reference. The intercalibration was also based only on radiometrically cold ocean scenes, and thus, did not account for scene-dependent temperature differences in the calibration. While this was not that unreasonable for intercalibrating the SSM/I instruments given their similarity, it is unlikely to be the case relative to a high-quality calibration reference like GMI.
Intercalibration of the F11, F13, F14, and F15 SSM/I’s was done using multiple techniques, including double differences with the GMI-calibrated TMI Level 1C [
7,
12], a vicarious cold technique [
18], and a vicarious warm-scene technique [
19]. This is basically the same approach used by the XCAL team for the intercalibration of the GPM radiometer constellation, with multiple techniques over both cold and warm scenes providing a consistency check, as well as a measure of the residual uncertainties [
12]. Details of the individual intercalibration techniques are left to the above references. For the F08 and F10 sensors, the intercalibration was somewhat more complicated, as a double difference approach with a well-calibrated instrument like TMI was not an option, and the results had to be daisy-chained between sensors to reference the results back to GMI. For F08, this meant comparing the calibration over a corresponding time period with F10, then to F11, and then to the GMI-calibration TMI Level 1C. As a result, the uncertainties for the F08 and F10 calibration are significantly larger, since the intercalibration estimates are subject to calibration uncertainties in the intermediate sensors. While the same vicarious cold [
18] and warm [
19] were used for these sensors, a single difference technique comparing simulated minus observed Tb for the sensor pairs was also used. This technique gives comparable results over ocean to the vicarious approach, but it does not work well over land, and thus, was not used for the warm scene calibration. Results from the vicarious warm-scene technique [
19], however, show good agreement with double difference results versus TMI over vegetated land. As a result, both warm-scene techniques are used for the four SSM/I sensors overlapping the TMI data record, but only results from the vicarious warm technique are used for the F08 and F10 intercalibration.
The cold and warm scene calibration offsets for the six SSM/I instruments are given, along with the associated scene temperatures in
Table 3. Given the similarity of the instruments, it is not surprising that the calibration offsets are similar, although there are differences between sensors. As with the prior Level 1C intercalibration by the XCAL team [
12], the use of multiple techniques adds confidence to the validity of the results. For the GPM constellation sensors, the XCAL approach using multiple techniques found residual uncertainties in the intercalibration offsets within 0.5 K over cold scenes (ocean) and within 1.0 K over warm scenes (vegetated land) [
12]. Results from the various techniques for the SSM/I sensors discussed here are consistent with those results. Calibration errors relative to the GMI reference standard, however, are significantly larger for the older SSM/I sensors in particular due to uncertainties in the intercalibration of the intermediate sensors, as well as potential calibration drifts that have not been fully corrected for. Given the lack of a long-term calibration reference, the best guidance for users is to consider small climate trends or changes in retrieved geophysical variables with skepticism, particularly for the pre-TRMM data record.