Special Sensor Microwave Imager (SSM/I) Intersensor Calibration Using a Simultaneous Conical Overpass Technique

1. Introduction

The history of environmental satellite measurements now spans several decades, which is relatively sufficient to form a climate data record (CDR) for climate-related studies. Based on the National Research Council’s definition (NRC 2004), a CDR is a time series of measurements of sufficient length, consistency, and continuity to determine climate variability and change. The Special Sensor Microwave Imager (SSM/I) series provide one of the longest time series of satellite microwave measurements, from July 1987 to the present. These observations are being followed with a similar sensor, the Special Sensor Microwave Imager/Sounder (SSMIS), which will continue to operate for at least the next decade. This long record of consistent measurements from multiple similar sensors is extremely important in generating CDRs for climate change research and analysis. However, the long-term multiple SSM/I measurements are not accurate enough to be directly applied in climate-related studies. The uncertainty caused by simply stitching multiple SSM/I datasets together as a CDR arises from instrument offsets, instrument degradation, signal interference, satellite orbital drift, missing data, etc. In addition, when these Defense Meteorological Satellite Program (DMSP) instruments were originally designed, the instrument calibration was mainly focused on their weather and environment applications, and their long-term performance stability has not been thoroughly assessed to date. Therefore, different SSM/I sensors have to be carefully calibrated to a reference satellite or a stable reference system in order to produce consistent and high quality CDRs for climate analysis and reanalysis. The calibrated SSM/I datasets computed by Remote Sensing Systems (RSS) are now available online (http://www.ssmi.com/ssmi). Another set of the calibrated SSM/I data computed by the research group led by Prof. C. Kummerow at Colorado State University (CSU) can also be found online (http://rain.atmos.colostate.edu).

Previous studies based on the Microwave Sounding Unit (MSU), which is carried on board the National Oceanic and Atmospheric Administration (NOAA) polar-orbiting satellites, show that the atmospheric tropospheric temperature warms up within a range of 0.08–0.21 K decade⁻¹ (Christy et al. 2003; Grody et al. 2004; Wentz and Schabel 1998; Prabhakara et al. 2000; Zou et al. 2006, 2009). The MSU intersensor calibration also has a prominent impact on the temperature climate trend not only by increasing our confidence in the estimation of climate trends, but also by improving the consistency of temperature trends at the surface and in the troposphere (Zou et al. 2009; Mears and Wentz 2005). The most recent study indicates a trend of 0.2 K decade⁻¹ for the global tropospheric temperature (Zou et al. 2009).

Intersensor calibration between SSM/I and SSMIS is also progressing well and is geared toward climate applications. The SSM/I anomaly due to the antenna field-of-view intrusion by the spacecraft and the glare suppression system was successfully corrected (e.g., Colton and Poe 1999). The SSM/I measurements are calibrated with respect to the radiative transfer simulations over oceans (Hilburn and Wentz 2008) and the well-calibrated SSM/I data reduce the discrepancy between the observed precipitation trend and the climate model prediction (Wentz et al. 2007). Recently, the observational anomalies of the first SSMIS on board the F-16 satellite were investigated and found to be caused by solar illumination on the SSMIS warm calibration target and antenna reflector emission (Kunkee et al. 2008; Yan and Weng 2008, 2009).

Several approaches have been commonly used for satellite intersensor calibrations, including 1) intercomparison between satellite and ground-based measurements, 2) comparison with clear-sky radiative transfer model simulations, 3) analysis of two overlapping sensor measurements at nearly simultaneous temporal and spatial locations, and 4) matching up the statistical properties of two sensor measurements at selected spatial scales. The simultaneous nadir overpass (SNO) technique was developed at NOAA/National Environmental Satellite, Data, and Information Service (NESDIS) (Cao et al. 2007) and has been used for MSU and AMSU temperature retrievals (Zou et al. 2006; Iacovazzi and Cao 2007, 2008; Iacovazzi et al. 2009). A similar simultaneous conical overpass (SCO) technique is developed for conically scanning instruments and the preliminary results show that the SCO calibration scheme can effectively remove biases between SSM/I or SSMIS sensors (Yan and Weng 2006, 2008; Weng et al. 2009; Yang et al. 2009).

For DMSP satellites, the intersensor calibration can be performed at either antenna temperature (T_a) or brightness temperature (T_b). The data record of satellite T_b measurements is called the sensor data record (SDR). Here, T_a is the effective blackbody temperature of the radiance on the feedhorn, while T_b is the calibrated effective blackbody temperature of the radiance on the antenna reflector. The conversion of T_a to T_b is performed to conduct the antenna pattern correction (CPA) in correcting the incomplete radiometric coupling between the reflector and feedhorn and the cross-polarization coupling between channels and sidelobe contamination (Colton and Poe 1999). It is recommended that the calibration of the radiometric measurements during the National Polar-orbiting Operational Environmental Satellite System (NPOESS) era should be conducted at T_b level. Therefore, this study presents the process involved with the SSM/I SDR intersensor calibration technique. The impacts of this scheme on the SSM/I-derived total precipitable water (TPW) and surface precipitation trends are also investigated.

2. Methodology and dataset

The SSM/I measurements are available at five frequency channels with vertical (V) and horizontal (H) polarization, except at the water vapor channel: that is, 19 (V, H), 22 (V), 37 (V, H), and 85 (V, H) GHz. The DMSP SSM/I datasets have been archived at Colorado Sate University (CSU) and NOAA/National Climatic Data Center (NCDC). The 1987–92 SSM/I datasets are in the Wentz format and the Wentz decoding procedures were applied (Wentz 1988, 1991, 1993), while the 1993–2006 SSM/I data are the NOAA temperature data record (TDR) products. The description of the TDR products can be found in the NOAA documentation (Akunuri et al. 2009; NOAA/NCDC 2009). The 1987–99 SSM/I data from CSU and the 2000–06 datasets from NOAA are used in this study. The T_a to T_b conversion is based on the method discussed in Colton and Poe (1999).

Figure 1 displays the time series of rain-free monthly mean T_b at 37V GHz from valid SSM/I measurements over the 60°S–60°N oceanic areas. It is obvious that SSM/I instruments provide a continuous measurement averaged over the oceanic region since July 1987; however, brightness temperature trends from different sensors are quite different. The T_b biases of all overlapped SSM/I sensors shown in the bottom panel of Fig. 1 indicate that the bias varies with time and different SCO pairs by as much as ±1.2 K, while their mean absolute intersensor T_b bias against the F-13 is 0.39 K. This large bias among different sensors demonstrates that the SSM/I SDRs and their derived environmental data records (EDRs) from the existing calibrations may not be suitable for climate studies. Thus, the calibration efforts for all SSM/I sensors must be conducted in order to generate unbiased and high quality CDRs for climate change analysis and trend studies.

Fortunately, the SSM/I measurements during their overlapping periods provide an alternative way of checking their consistency and of selecting a reference frame for the SSM/I intersensor calibrations. Figure 2 shows the local equatorial crossing times of the SSM/I sensors at their ascending nodes. It is evident that F-13 and F-14 have more overlapping time periods with other SSM/I sensors. Either one is naturally an ideal candidate as a reference satellite. However, F-13 has the longest data record and the smallest change in equatorial crossing time due to satellite orbit drift. Although we cannot prove that F-13 is absolutely accurate as the reference satellite, we use F-13 as the reference for the SSM/I cross-sensor calibration because of the stable F-13 T_b time series in Fig. 1 and its water vapor products (see Fig. 11). In addition, a sensitivity study was conducted in the SSM/I T_a intersensor calibration process using F-13 as the reference satellite, in which we assumed that F-13 was not taken as the absolute reference; instead, an F-13 T_a bias was optimized to minimize the mean absolute intersensor bias (not discussed in this paper). Results show that the optimized F-13 T_a bias is small so that the choice of F-13 as the reference satellite in the SSM/I intersensor calibration is reasonable. Therefore, F-13 is finally selected as the reference satellite. However, we plan to investigate the stability of the F-13 sensor and its calibration in our continued efforts to improve the SSM/I-based CDRs, once the information is available.

The SSM/I measurements from a pair of the DMSP satellites are matched when they are simultaneously overpassing local areas, typically at high latitudes. These measurements are called the SCO pairs defined in details in section 3c. They are supposed to be identical if the sensors with the same incidence and azimuth angles are all well calibrated. Figure 3 shows the locations of the SCO pairs over the North and South Poles used in this study. However, a bias between two different SSM/I sensors normally exists due to many factors such as instrument calibration, instrument degradation, sources of interference to signals, satellite orbital drift, and incidence and azimuth angles. The bias between the SCO pairs should be removed in order to generate consistent SDRs, EDRs, and CDRs. We do not explicitly correct any possible error due to different sensor incidence and azimuth angles, although this bias should be very small in this study due to the SCO pair selection procedure to be discussed in section 3c. The data quality control of the SCO pixels described in section 3c is one of the key procedures in the calibration process. A double-difference technique (DDT) will be applied for SSM/I datasets if there is no direct interception between F-13 and another satellite A (S_A). In this case a third satellite (S_B), which has abundant SCOs with both F-13 and S_A, will be used as a transfer radiometer to intercompare F-13 and S_A. In this study, as discussed in section 3c, F-14 is applied as the transfer radiometer between F-13 and F-15 for SCO over a water surface. It should be pointed out that the SCO-based bias correction scheme could not remove the biases that have a spatial variation.

SSM/I measurements from the F-08, F-10, F-11, F-13, F-14, and F-15 satellites during 1987–2006 are applied in this study, while only SSM/I data during 1990–2006 are used in the trend analysis. The SSM/I-based TPW and precipitation retrievals are derived from the NOAA heritage algorithms (Weng and Grody 1994; Ferraro et al. 1996). Since the Tropical Rainfall Measuring Mission (TRMM) rainfall products have been regarded by the science community as the most accurate precipitation measurements from satellite remote sensing (Yang and Smith 2008; Yang et al. 2008; Wolff and Fisher 2008, 2009; Yamamoto et al. 2008; Kummerow et al. 2000), three official TRMM rain products are utilized to demonstrate the impacts of the SSM/I SDR calibration on rain retrievals. These TRMM rain products are 2A12, the TRMM Microwave Imager (TMI) rain-only algorithm (Olson et al. 2006; Yang et al. 2006; Kummerow et al. 1996, 2001); 2A25, the TRMM precipitation radar (PR) rain-only algorithm (Iguchi et al. 2000); and 2B31, the combined TMI–PR rain algorithm (Haddad et al. 1997). In addition, the multisatellite blended TRMM rain algorithm, 3B42 (Huffman et al. 2007), is included in the analysis. The 2A12 rain product includes the TRMM orbital instantaneous rain retrievals at the TMI footprint scale of 10 km × 10 km, while both the 2A25 and 2B31 rain products are at the TRMM PR pixel resolution of 5 km × 5 km. The 3B42 rain product is at 3-h ¼° grid resolution. A significance test on the trend analysis is conducted following the method discussed by von Storch and Zwiers (1999).

In addition, only the TRMM rain data that temporally match up with the SSM/I measurements at the orbital 0.5° × 0.5° scale are used in the rain assessment study. Thus, the TRMM TMI swath is applied for 2A12, while the TRMM PR swath is used for both 2A25 and 2B31. To mitigate the sampling issue, we conduct the assessment and intercomparison study for the monthly rainfall with the TRMM products at 5° × 5° grid resolution.

3. SSM/I SDR intersensor calibration

b. RADCAL beacon interference on F-15 22V GHz

RADCAL, a system of instruments on board the DMSP F-15 spacecraft, consists of redundant C-band transponders with unique antennas, Doppler transmitters operating at 150 and 400 MHz, and deployable antennas. The purpose of the RADCAL C-band transponder–antennas is to provide a signal source for ground-based C-band radar interrogation and tracking, while the primary purpose of the Doppler transmitters and antennas is to determine satellite positions for comparison with the radar data. A secondary purpose of the Doppler systems is to support the Coherent Electromagnetic Tomography (CERTO) experiment. RADCAL has been operational since 14 August 2006 and considerably affects the F-15 SSM/I and Special Sensor Microwave Temperature-2 (SSMT-2) sensor data. In particular, the 150-MHz beacon produces considerable increases in brightness temperatures at 22 GHz (G. A. Poe et al. 2006, unpublished manuscript).

Since the RADCAL beacon interference on F-15 22V-GHz channel is steady, an initial correction algorithm is applied to remove the interference. Figure 5 exhibits the SSM/I F-15 T_a error at 22V GHz due to the RADCAL beacon interference on 30 August 2006 as a function of scan position for the ascending and descending nodes (Yan and Weng 2006). The errors are based on the mean differences between the global SSM/I observations and the radiative transfer model simulations under oceanic cloud-free conditions. A polynomial function is applied to fit the error curves; then, this error is subtracted from the raw T_a at 22V GHz:

View Expanded

View Expanded

In Eq. (1), X is the scan position, and the coefficients of a₀, a₁, a₂, and a₃ are given in Table 1. The fitted lines are overlapped in Fig. 5 to clearly show the fitting applied for this study. This correction of the F-15 RADCAL beacon interference is applied before the scan-angle bias correction.

However, G. A. Poe et al. (2006, unpublished manuscript) indicate an increased RADCAL interference during Earth shadow, resulting in an approximately 1.3-K increase in the vicinity of the maximum interference at beam position 25. This effect is apparent for the descending orbits that enter Earth’s shadow for part of the year due to Earth’s elliptical orbit and leads to a clear impact on EDRs, especially TPW, during Earth shadow. This Earth shadow effect on TPW is small for global or tropical ocean averages, while it could create a notably artificial spatial variation at regional scales, especially when the time in Earth shadow increases considerably during summer 2007. In addition, the RADCAL interference also affects the F-15 85-GHz channels at a lesser degree. Since only 4 months of F-15 datasets affected by the RADCAL interference were involved in this study, their impact on the 85-GHz channels should not substantially change the results from this study. Therefore, we do not discuss the influence of the interference during Earth shadow and its impact on the 85-GHz channels in this paper. However, we plan to further investigate these issues through our continued efforts to improve the SSM/I and SSMIS data quality at NOAA/NESDIS.

c. SCO technique for linear SDR intersensor bias correction

The physical principle of the SCO technique in intersensor calibration studies is primarily based on the assumption that simultaneous measurements at a location from two different sensors of the same design should be highly correlated. If one sensor is regarded as a reference, the other can be calibrated to this reference. The skill of the SCO technique requires minimization of the measurement differences caused by noninstrumental factors. Thus, the SCO differences between two different sensors are primarily due to instrumental errors, which should be removed during the post launch calibration processes.

Many experiments with different SCO constraints are conducted for an optimal result. All possible SCO pairs are first quality controlled for the same orbital node (e.g., ascending or descending) and similar pixel positions. A spatial distance (Δd) of 3 km between the SCO pair is used to ensure that at least the footprints of two SSM/I instruments are overlapped by 75%–90%. A reasonable time difference (Δt) between two sensors is required to lead to a reliable analysis. Figure 6 displays the mean bias, standard deviation, and the SCO pair samples against different time criterion at 22V GHz between F-13 and other SSM/I sensors over water surfaces. It is evident that the bias and standard deviation vary slightly with different time criteria, except for the F-13/F-11 bias, which increases dramatically when the time difference is less than 2 min due to the increase in the uncertainties with limited samples. Similar results exist for other channels. Over land (figure omitted), the SCO bias difference is very small with the different time criteria (0.5, 1, and 2 min). However, the bias increases considerably when the time difference is greater than 5 min. Thus, in general, the 2-min criterion is used and the SCO samples are enough for the analysis of the intersensor bias correction. In addition, the samples over inhomogeneous background conditions are eliminated by applying the standard deviation (σ) of nine neighboring pixels surrounding a candidate SCO pair. The SCO pair is taken as a good one if σ is less than 2 K for a homogeneous surface. Features similar to those for the water surface are found for SCO pairs over ice and coastal surface types, except that σ less than 5 K is used for coastal areas. Finally, the absolute T_b difference (|ΔT_b|) of an SCO pair should be less than 10 K. Only about 1% (0.06%) of the oceanic (continental) SCO pairs were excluded with this criterion.

After careful analysis of the SCO pairs using F-13 as a reference satellite, the criteria for collecting high-quality SCO pairs are defined as follows:

View Expanded

In addition, the same orbital nodes and similar scan positions (less than three scan positions difference) are also required. The bias distribution with the SCO time differences is inspected to eliminate any potentially large bias caused by small samples of the SCO pairs that were not well distributed.

All SCO pixels are sorted into four categories based on surface type (i.e., water, land, ice, and coast) to avoid the contamination caused by mixing these surface types and were carefully analyzed accordingly. As an example, Fig. 7 presents the SCO T_b bias [T_b(F-13) − T_b(F-14)] at 37V GHz against their interception time difference over the water background. It is evident that 2059 SCO pixels are well distributed around the exact interception time (Δt = 0) between F-13 and F-14. The statistical mean bias is −0.58 K with a standard deviation of 0.60 K. Since it is difficult to find many exactly simultaneous measurements between any two SSM/I sensors, the statistical mean bias of the SCO pixels with nearly simultaneous measurements is a reliable choice for obtaining the bias between these two sensors. Similar processes are conducted for all other SSM/I channels to estimate the intersensor bias coefficients between F-13 and other SSM/I satellites.

Due to the limited SCO pixels between F-13 and F-15 over water surfaces, a DDT approach is utilized to estimate the T_b bias, that is, using F-14 (which has better overlaps with both F-13 and F-15) as a transfer radiometer to connect them so that subtraction of the T_b bias between F-15 and F-14 from that between F-13 and F-14 results in the T_b bias_F-13−F-15 because the F-14 effect is literally cancelled out from the double difference (T_b bias_F-13−F-15 = T_b bias_F-13−F-14 − T_b bias_F-15−F-14). Please be advised that the DDT can only be applied when there are insufficient good-quality SCO pixels between two SSM/I sensors so that a third sensor must be used as a transfer radiometer. Finally, no bias correction is applied with F-08 because there are no reliable SCO pixels between F-08 and F-13 and there are insufficient F-08–F10 match-ups to apply a DDT. Additional efforts are planned at NOAA/NESDIS to produce a calibrated F-08 dataset in the near future. The final intersensor bias correction coefficients and standard deviations for F-10, F-11, F-13, F-14, and F-15 using F-13 as the reference satellite are listed in Table 2.

4. Calibration impacts on SSM/I SDR

With the continuously increasing time span of environmental satellite measurements, the satellite remote sensing derived EDRs are becoming more important in climate-related studies. Since the climate trends of meteorological parameters are small compared to the natural variability in the system, improvements in satellite remote sensing datasets will be important for use in quantifying climate trends. After implementation of the SDR intersensor calibration procedure as discussed in section 3, the trends from all SSM/I SDR time series are now more consistent.

Figure 8 shows the time series of SSM/I oceanic rain-free monthly mean intersensor calibrated T_b and their intersensor bias at the 37V-GHz channel. When compared to Fig. 1, it is apparent that the new time series has a considerably improved consistency among these SSM/I sensors, with the intersensor bias reduced dramatically. The variation of the intersensor biases with time and different overlapping SSM/I sensors is only around ±0.5 K, indicating a 58% decrease of the bias after intersensor calibration. We use the mean absolute bias to summarize how calibration offsets affect the consistency of the monthly T_b time series among different sensors. This parameter is the average absolute monthly T_b biases, which highlights the spread, as opposed to the grand-average offsets captured by a simple average of the monthly biases. The mean absolute bias after calibration is only 0.20 K, indicating a mean absolute bias deduction of 49%. Similar results are found for other SSM/I channels. The mean absolute T_b bias against F-13 before and after the SDR intersensor calibration and the percentage change for each channel are summarized in Table 3. The averaged deduction of the mean absolute bias with the SDR calibration is about 30%. Therefore, these results demonstrate that the newly developed SSM/I SDR intersensor calibration scheme is very useful and can substantially improve the consistency of the SSM/I SDR time series with dramatically reduced intersensor biases that have a considerably smaller temporal variation.

The improved consistency of the SSM/I SDRs should lead to a more reliable SDR trend analysis. Since F-08 is not included in the processing of the SSM/I SDR intersensor bias corrections using F-13 as the reference satellite, F-08 is not involved in the trend analysis. Thus, the trend analysis throughout out this paper is only based on SSM/I measurements from the time period 1990–2006. Figure 9 presents the time series of the SSM/I oceanic rain-free monthly T_b before and after intersensor SDR calibration at the 37V-GHz channel. The SSM/I T_b time series before calibration presents a standard deviation (σ) of 0.49 K and a linear trend of −0.12 K decade⁻¹ at the 2.5% significance level. After calibration, the monthly T_b time series has the same mean value, but much smaller σ (0.32) and larger trend (−0.32 K decade⁻¹) at the 0.1% significance level. Similar analyses are conducted for other channels and the results are summarized in Table 4. It is obvious that SDR intersensor calibration using F-13 as the reference satellite will not substantially change the averaged monthly T_b, but will lead to a more consistent time series among multiple SSM/I sensors with a mean reduced σ of 21.4% for all channels. The most important impact of this calibration is its role in changing the T_b trend considerably at every SSM/I channel; that is, the trend magnitudes are reduced by 37.1%, 72.2%, 74.1%, and 77.3% for channels 19V, 37H, 85V, and 85H GHz, respectively, while they increase by a larger percentage at 19H, 22V, and 37V GHz because the original trend was small.

5. Impacts of SSM/I intersensor calibration on TPW and precipitation

6. Discussion and conclusions

A new NOAA/NESDS SSM/I SDR intersensor calibration scheme is developed for climate studies. This calibration scheme is built on the SCO technique, which collects observations at the same time and location from two SSM/I sensors that have an overlapping time period. The SCO intersensor calibration scheme requires an SSM/I sensor to be the reference radiometer so that other sensors can be calibrated against this reference sensor. We have selected F-13 as a reference sensor because of its most stable local equatorial crossing time and because it has the most interceptions with other satellites.

The SSM/I scan-angle-dependent bias is analyzed by comparing the mean oceanic rain-free T_b at each scan position against its central position. Results show a maximum bias of −2.5 K for high-frequency channels and −1.75 K for low-frequency channels near the end of the scan due to the glare suppression system, and a notable bias of −0.5 K near the beginning of the scan caused by the SSM/I spacecraft intrusion. This scan-angle-dependent bias varies with the orbital ascending and descending nodes, as well as the seasons. In addition, a statistical method is developed in an early effort to remove the RADCAL beacon interference on the F-15 22V-GHz channel. Both the scan-angle-dependent bias and the 22V-GHz channel interference have to be removed prior to any SSM/I SDR intersensor bias corrections. The impacts of the RADCAL interference on the 85-GHz channel and its increased interference during the Earth shadow period are not discussed in this study. This issue will be considered in our continued calibration efforts.

Because of the important influence of surface background on satellite passive microwave radiance, the surface inhomogeneity has a huge impact on the collected SCO pixels. To eliminate maximally the impacts of the surface inhomogeneity on the SCO pixels, a tight quality control procedure was developed so that the bias of the SCO pixels for two overlapped SSM/I sensors would be mainly caused by the sensor differences. In addition to the same satellite orbital nodes and similar scan positions, the SCO criteria are set as Δd ≤ 3 km, Δt ≤ 2 min, σ ≤ 2 K, and |ΔT_b| ≤ 10 K for three surface categories (i.e., water, land, and ice surface types), while σ ≤ 5 k is used for coastal situations. Notably, when there are not enough high quality SCO pixels between two sensors, a third satellite that has good SCO pixels with both of these two sensors is utilized as the transfer radiometer to create the desired SCO pixels (This process is normally referred as the DDT.). Therefore, the current NOAA/NESDIS SSM/I SDR intersensor calibration scheme has two major components: 1) removal of the scan-angle-dependent bias and the RADCAL beacon interference error on F-15 22V GHz after 15 August 2006 and 2) removal of the intersensor bias against the reference satellite (F-13) over different surface types. (Because of the lack of useful SCO pixels against F-13, F-08 is not applied in the SDR intersensor calibration.)

Results demonstrate that the newly developed SSM/I SDR intersensor calibration scheme substantially improves the consistency of the monthly mean oceanic rain-free T_b time series for all sensors, in terms of 58% reduced maximum intersensor bias, 30% declined mean absolute bias, and 21% decreased standard deviation. In addition, as we expected, the intersensor calibration has little impact on the mean rain-free oceanic monthly T_b; however, it displays a dramatic influence on the climate trend of the monthly T_b series. Because of the consistency of the T_b series for all SSM/I sensors, the SDR’s climate trend becomes more reliable. The SDR’s climate trend over ocean varies from −0.70 to 0.36 K decade⁻¹ before the intersensor calibration and from −0.44 to 0.09 K decade⁻¹ after the calibration, resulting in a percentage change of 37% at the 19V-GHz channel, and more than 70% at other channels.

The TPW and precipitation are taken as examples to demonstrate the impacts of the SSM/I SDR intersensor calibration on the associated EDRs and CDRs. The F-14 SSM/I measurements during December 2006 were applied to illustrate the differences in the precipitation retrievals before and after the intersensor calibration. The NOAA heritage rain algorithm is used in this analysis. The SSM/I SDR intersensor calibration leads to a systematic deduction of oceanic precipitation with an overall change of −2.7% and a systematic continental rainfall increase with a mean difference of 1.7%. The resultant oceanic precipitation retrievals after the intersensor calibration improve their comparisons with the matched official TRMM rain products from the TMI, PR, and TMI–PR combined algorithms by reduced biases of 14%, 9%, and 15%, respectively.

The most important impact on TPW and precipitation from the SSM/I SDR intersensor calibration is that the resultant monthly TPW and precipitation show a dramatically improved consistency among all sensors, indicating that the EDRs and CDRs with the intersensor-calibrated SSM/I measurements are more suitable for climate-related studies. In general, the maximum TPW intersensor bias from before to after the calibration changes from −1.5 to ±0.1 mm for the global ocean and from −3 to ±0.5 mm for the tropical ocean, resulting in 75% and 20% deductions of the mean absolute intersensor bias, respectively. The corresponding TPW standard deviations are also decreased by 40% and 21%. The TPW climate trends after the intersensor calibration are 0.34 mm decade⁻¹ (or 1.59% decade⁻¹) for the global ocean and 0.63 mm decade⁻¹ (or 1.39% decade⁻¹) for the tropical ocean, showing trend decreases of 38% and 54% from before the calibration, respectively. The TPW trend with the intersensor-calibrated SSM/I measurements are in a good agreement with previously published results.

The impacts of calibration on precipitation retrievals are relatively small in comparison with TPW but are still important. The standard deviation of the monthly oceanic precipitation is generally reduced by 10%, and mean absolute rainfall intersensor biases are also decreased by 20.6%, 15.7%, 6.5%, and 0% for ocean, land, global, and tropical ocean datasets, respectively.

Overall, this study provides robust evidence that the cross-sensor calibration has some impact, resulting in more consistent SDRs, EDRs, and CDRs. The SSM/I SDR intersensor calibration can potentially improve climate-trend studies.