Introduction

As part of the National Aeronautics and Space Administration’s (NASA) Earth Science Enterprise (ESE) program, science investigations of earth radiances are designed to define and model natural climate changes as well as man’s impact upon climate. The Clouds and the Earth’s Radiant Energy System (CERES) measurements constitute one of many datasets derived from several spacecraft sensors. To allow cross comparisons with different ESE datasets, the CERES absolute radiometric calibration is tied to the International Temperature Scale of 1990 (ITS’90). To maintain traceability to ITS’90, the CERES sensor responses are monitored both prior and subsequent to launch using built-in, in-flight calibration sources in consonance with supplemental vicarious validation studies. CERES measurements are the focus of science investigations designed to measure 1) top-of-the-atmosphere (TOA) earth radiation budget components, 2) radiative forcing properties of clouds, 3) surface radiation budget, and 4) divergence throughout the atmosphere (Wielicki et al. 1996). On 27 November 1997, the CERES Proto-Flight Model (PFM) instrument package was launched on the NASA Tropical Rainfall Measuring Mission (TRMM) spacecraft. A National Space Development Agency of Japan (NASDA) launch vehicle placed the TRMM spacecraft into a low-inclination 35°, 350-km altitude orbit. Nominal scientific data collection operations commenced on 1 January 1998.

On 1 September 1998 nominal mission operations were suspended due to the failure of an onboard voltage converter to properly regulate its line voltage. The suspect converter regulates a +28-V signal arriving from the spacecraft supply bus to a constant +15-V supply for the CERES data acquisition assembly (DAA) electronics. A NASA Tiger Team, formed in September 1998, identified possible causes for the regulator degradation and assessed the risk of continued sensor operations. Chapman and Bockman (1999) have demonstrated the survivability of the DAA electronics under these types of overvoltage conditions. The CERES Science Team identified the top scientific priority as obtaining overlapping measurements with the CERES Flight Models 1 and 2 that will be launched in late 1999 aboard NASA’s Terra spacecraft. Based upon the Tiger Team’s recommendations and the Science Team’s goals, the Proto-Flight Model was operational on a limited basis from September 1998 until the launch of NASA’s Terra spacecraft. These operations have been in support of 1) special field campaigns whose goals are to measure surface and atmospheric radiation such as INDOEX in March of 1999 and Nauru99 from 15 June to 15 July 1999, 2) intercomparisons with both NASA’s Earth Radiation Budget Experiment (ERBE) nonscanner instrument on the Earth Radiation Budget Satellite spacecraft and the French Scanner for Radiation Budget (ScaRaB) instrument on the Russian Resurs spacecraft, and 3) radiometric stability checks of the CERES PFM sensor responses. It is anticipated that full operational status will be restored to the Proto-Flight Model in January 2000.

CERES instrument

Each CERES instrument package consists of a scanning thermistor bolometer sensor assembly, an elevation axis drive system, an azimuth axis drive system, and associated electronics. The CERES instruments have been designed, manufactured, and tested by TRW Inc.’s Space and Electronics Group, Spacecraft and Technology Division of Redondo Beach, California. Each flight model contains three radiometric sensors that measure the earth-reflected solar radiance in the 0.3–5.0-μm spectral region, earth-reflected and earth-emitted radiances in the 0.3–>100-μm spectral region, and earth-emitted longwave radiances in the 8–12-μm spectral region. The three sensors are coaligned and mounted on a spindle that scans about the elevation axis. This assembly may also be commanded to simultaneously rotate around the azimuth axis at rates of between 4° and 6° s⁻¹. The flight models have a mass of less than 45.6 kg and consume less than 50 W of power dining nominal flight operations.

Each sensor consists of a telescope containing a forward baffle, Cassegrain optics, and a thermistor bolometer detector module assembly composed of active and reference detectors. The f/1.8 Cassegrain optics module has 18- and 8-mm diameter spherical silvered primary and secondary mirrors. In the shortwave and window sensor units, filters are located before the secondary mirror “spider” (the name applied to the three-legged structure that supports the secondary mirror) and in front of the active bolometer detector. The shortwave sensor filter pair consists of 1.02- and 0.51-mm-thick fused, waterless quartz elements (Si0₂), while the 8–12-μm window filter system consists of a 1-mm-thick zinc sulfide and a 0.5-mm-thick cadmium telluride filter element. At nadir, the 1.3° × 2.6° sensor field of view (small dimension is in the elevation scan direction) corresponds to a geographical footprint of approximately 10 km for the TRMM spacecraft orbital altitude of 350 km.

In the detector module assembly, the active and reference detectors are mounted on separate aluminum heatsinks that are in thermal contact via a 100-μm sputtered indium interface and actively maintained at a constant temperature of 38°C. This design simplifies thermal control by maintaining the assembly at temperatures greater than the surroundings allowing energy deposited on the detectors to diffuse to surrounding structure.

A black layer on the active detector absorbs the target scene energy and converts it into sensible heat causing a measurable change of the electrical resistance in the thermistor layer of the active detector. Each thermistor is a sintered semiconductor material with a high negative coefficient of electrical resistance, B. The electrical resistance R can be represented as a function of the local temperature T by (Astheimer 1983):

View Expanded

where R_o is the electrical resistance in ohms (Ω) at the reference temperature T_o (K). The active bolometer responds to thermal energy from the desired scene while the reference detector compensates for variations in the sensor thermal environment. Both detectors are in adjacent arms of a Wheatstone bridge and the output signal is passed through a low-noise preamplifier low-pass filter before entering a four-pole Bessel filter, after which it is sampled.

Data reduction algorithms

CERES radiance measurements are filtered measurements represented theoretically by

View Expanded

where S_λ is the spectral response of a given radiometric channel (total, window, or shortwave), I_λ is the spectral radiance incident to the instrument aperture, and I^j_f is the radiance absorbed by the detector. For a Cassegrain optical system such as CERES, S_λ may be represented as

S_λρ²_λτ_λα_λ(3)

where ρ_λ represents the spectral reflectance of each of the two silver mirrors, τ_λ is the spectral transmission of any optical filters, and α_λ is the spectral absorptance of the absorber layer of the detector. Priestley et al. (1998) presents the procedure of determining S_λ for the CERES sensors. Figure 1 displays the spectral response functions utilized in the CERES Proto-Flight Model Edition 1 ERBE-like data products.

In order to preserve continuity with ERBE long-term datasets, the initial CERES data reduction algorithms have followed the procedures developed for ERBE. The filtered radiance (W m⁻² sr⁻¹) measured by each sensor unit can be expressed by (Lee et al. 1998)

View Expanded

where

t_k+1t_kt,

and Δt is equal to 6.6 s. In (4) τ represents the time lag between the instantaneous optical pointing vector and point spread function centroid as determined prelaunch by Paden et al. (1998) and verified postlaunch by Currey et al. (1998), o(t) represents the scan position dependent offset term in digital counts as determined by Thomas et al. (1998), m(t) the instrument output signal in digital counts at time t, and m(t_k) the mean instrument output signal when viewing cold space at the start of a given scan, k. By subtracting the instrument output signal of cold space, m(t_k), from earth-viewing samples, the CERES instrument obtains a highly accurate measurement of terrestrial and solar energy flows relative to a near-zero (2.7 K) constant temperature blackbody source (Turner 1993). The radiometric gain term A_V converts changes in the thermal state of the detecting element, represented by digital counts, into an equivalent absorbed radiance. The gain term A_S properly accounts for any drift in the absolute detector signal between subsequent space looks. Theory demonstrates that the value of the A_S coefficient is equal in magnitude, but opposite in sign to that of A_V. The remaining gain terms A_H, A_D, and A_B remove potential contamination of detector output due to both thermal and electrical contamination originating from varying heatsink temperatures, T_H, digital-to-analog converter (DAC) bridge balance voltages, V_D, and detector bias voltages, V_bias.

The thermal and electrical design of the CERES Proton-Flight Model instrument has reduced both the variations and sensitivity to variations, of the heatsink temperature, T_H, bias voltage, V_bias, and DAC bridge balance voltage, V_D, to levels that are negligible during normal operations. Thus (4) reduces to the first two terms on the right-hand side for nominal operational scenarios. The current effort presents the stability of both the A_V coefficient and spectral response S_λ for the CERES PFM instrument.

Ground characterization testing revealed that the sensor’s radiometric output m(t) is affected by a small transient at the ∼1% level. Modeling studies demonstrate that this transient is caused by diffusion of thermal energy from the active to the compensating bolometer through the heatsink. In the data processing system, a numerical filter is used to account for this transient. Since the transient consists of a single slow mode whose time constant is on the order of 300 ms, its response υ(t) can be computed at time t in terms of its previous level at time (t − δt), and the sensor output voltage m(t) can be characterized by the following recursive equation:

υtp_oυtδtp₁mt(5)

The slow mode is subtracted from the sensor output voltage to yield the transient-corrected sensor output voltage u(t) as

View Expanded

The filter weights p_o and p₁ are given by

View Expanded

where 1/ζ is equal to the characteristic time of the slow mode and c is the response of the slow mode to a unit step input. Values of ζ and c for the CERES PFM instrument may be seen in Table 1. To assess the stability of the prelaunch values listed in Table 1, on 19 March 1999 the PFM instrument was placed in an operational mode in which the radiometric channels were allowed to stare first at cold space and then at the internal calibration sources with an approximately 2-s dwell at each. Analysis of this data demonstrated no detectable changes from the prelaunch values.

The A_V and A_S terms are determined by regressing sensor counts, heatsink temperature measurements, bias voltage levels, and DAC voltage levels against the calculated filtered radiances, respectively. Ground-based radiometric calibration of the CERES Proto-Flight Model instrument was conducted using a state-of-the-art Radiometric Calibration Facility (RCF) designed and built by TRW (Jarecke et al. 1991). Among the many features of the RCF are unique longwave and shortwave calibration systems. The primary infrared standard is a narrow-field blackbody (NFBB) whose temperature knowledge is traceable to the ITS’90 via seven platinum resistance thermometers (PRTs). A unique shortwave reference source consisting of an internal optical bench and filter wheel, integrating sphere, and quartz–tungsten–halogen lamp source provides stable, well-characterized radiometric sources in 14 narrow bands between 0.4 and 2.0 μm. A cryogenically cooled transfer active cavity radiometer allows traceability of calibration in the 0.4–2.0-μm region to ITS’90 (Folkman et al. 1994). Radiometric gain values, as determined during the ground-based radiometric calibration, may be seen in Table 1.

On-orbit calibration monitoring strategies include the use of onboard calibration equipment and vicarious validation studies. Onboard calibration equipment consists of an internal calibration module (ICM) and a solar calibration assembly. Vicarious validation studies include a three-channel intercomparison that allows an empirical measure of the relative magnitude and stability of the spectral response functions for both the shortwave channel and shortwave portion of the total channel.

Postlaunch calibration monitoring strategies

On-orbit calibrations are designed to monitor and quantify traceability to the ground-measured radiometric gain terms that will be revised only if changes larger than 0.5 percent in the longwave region, or 1 percent in the shortwave region are detected and verified. Vicarious validation studies using earth radiance measurements may either verify or dismiss sensor response changes indicated by ICM and solar calibrations. Detailed analyses of the first 19 months of in-flight calibrations and validation studies are completed.

Vicarious validation studies

Although the ICM and solar calibrations provide the primary measurements of radiometric stability, it is also necessary to perform ancillary empirical validation studies using measured earth radiances to either support or refute results determined from the onboard equipment. Green and Avis (1996) have developed vicarious calibration validation techniques that encompass interchannel comparisons. Kratz et al. (2000, manuscript submitted to J. Geophys. Res.) has also developed methodologies to compare theoretical atmospheric radiative transfer models (Kratz and Rose 1999) to validate instrument response. Since only a single CERES instrument has been launched to date, modified versions of the Green and Avis interchannel comparisons have been applied to the CERES PFM instrument.

Three-channel intercomparison

The three-channel intercomparison is based upon inherent redundancy in the three CERES channels. Assuming an appropriate narrow- to broadband conversion may be determined to convert the 8–12-μm window channel into a “synthetic” longwave channel (i.e., 5–>100 μm) shortwave radiances may then be determined by either the shortwave channel or the difference between the total and synthetic longwave channel. Similarly, the longwave radiance may also be determined by subtracting the shortwave channel from the total channel, and the total radiance may be determined by summing the shortwave and synthetic longwave channel.

The narrow- to broadband conversion is accomplished by regressing the nighttime window channel measurements against the nighttime total channel broadband radiance for the same footprints. Because there is potentially a large amount of uncertainty in estimating the broadband longwave radiance from a narrow window radiance we have limited this analysis to nadir footprints comprised entirely of deep convective clouds (DCCs) whose brightness temperatures are less than 215 K. Since DCCs exist only at high altitudes (>10 km), we have minimized uncertainties due to atmospheric scattering and water vapor absorption, and the resulting relationship is statistically robust. Daytime longwave measurements for DCCs are then obtained by both the narrow- to broadband conversion of window channel measurements and by subtracting shortwave channel measurements from total channel measurements. By regressing the difference of these two daytime longwave radiance values against the daytime filtered shortwave measurements, both inconsistencies and changes in the relationship between the estimated spectral response of the shortwave channel and shortwave portion of the total channel may be seen.

Since the spectral weighting of solar energy reflected off of DCCs is a weak function of solar zenith angle, uncertainties in unfiltering the shortwave radiances are minimized. Implicit assumptions in this method are that emitted longwave radiance is independent of solar zenith angle and that the algorithms used in determining the spectral unfiltering coefficients are applied consistently to both the shortwave channel and shortwave portion of the total channel.

It can be shown (see the appendix) that the three-channel intercomparison technique empirically assesses the relationship between the true spectral response functions of the total and shortwave channels and our best estimates of these functions as determined during the radiometric ground calibrations. It does not assess the absolute accuracy of either channel. Any disagreement between truth and estimate may be quantified by

View Expanded

where Δ is the daytime longwave difference; I^sw_f is the daytime filtered shortwave measurement; and â^sw, â^sw/tot, and â^lw/tot are our estimates of the true spectral unfiltering coefficients a^sw, a^sw/tot, and a^lw/tot.

From (A9) in the appendix we have shown that the filtered shortwave radiance as measured by the total channel may be estimated from the filtered shortwave measurement as

View Expanded

For CERES, b̂^sw ≈ b̂^sw/tot ≪ Î^sw/tot_f so that from (10) we get

View Expanded

Solving the integral in (11) yields a value of 1.12 for DCC. A negative error from (9) would imply the right-hand side of (11) is too small and thus either Ŝ^sw_λ is low or Ŝ^sw/tot_λ is high. A positive error would imply the opposite. Since the shortwave CERES data products are derived from shortwave channel measurements only, errors in Ŝ^sw/tot_λ will not affect these products. We have no indication from this validation test whether Ŝ^sw_λ or Ŝ^sw/tot_λ is incorrect. Moreover, even if it was known which estimate was in error, there is no information of the error as a function of λ since our tests were on the integrals of S_λ over λ.

Results

Lifetime radiometric stability of the CERES PFM sensors as determined with the ICM may be seen in Fig. 2. Flight radiometric gain values were determined from the postenvironmental ground-based radiometric calibrations conducted by TRW in September 1995. Data from December 1997 and later correspond to on-orbit internal calibration procedures where the contamination doors are open. All data points in Fig. 2 are referenced to the final three internal calibration sequences conducted during ground calibrations. Ground to initial on-orbit traceability of the absolute calibration is 0.13, 0.14, and 0.26 percent for the total, window, and shortwave channels with 95% confidence bounds of 0.05, 0.07, and 0.01 percent. Values were determined by comparing results of the final three internal calibrations conducted during ground calibrations to the mean value determined using the ICM for the period of 27 December 1997–27 January 1998 for the total and window channels and 27 December 1997–6 January 1998 for the shortwave channel.

On-orbit radiometric stability for the window and longwave portion of the total channel as determined by analyzing measurements from the ICM blackbodies may also be seen in Fig. 2. Regression analyses of the on-orbit ICM gains demonstrates the total channel sensor remained stable at the 0.03-percent level (<0.1 W m⁻²) with a 0.10-percent 95% confidence bound. The window channel sensor remained stable at the 0.22-percent level (<0.15 W m⁻²) with a 0.17-percent 95% confidence bound over the first 8 months of science data collection (January–August 1998). Data subsequent to August 1998 corresponds to short (less than 24 h) operations in support of field campaigns and intercalibrations with the ScaRaB instrument. The CERES detectors require thermal stability levels that are not fully realized in the 24 h subsequent to activation. The resulting bias variations of approximately 0.1–0.5 percent in the window channel measurements during the September 1998 to July 1999 time frame are visible in Fig. 2. These bias values are consistent with the known theoretical relationship between radiometric gain and detector substrate temperature for thermistor bolometer detectors (Astheimer 1983). Archived data products for these operations have the production strategy “Transient_Ops” in their naming convention and the bias is documented in CERES Data Quality Summaries, which data users must sign prior to ordering any CERES products.

Figure 2 also shows a drift in the SWICS lamp radiance level of approximately 1 percent over the same dataset from March 1998 to July 1999 as measured by the shortwave sensor. Figure 3 displays both the broad- and narrowband stability levels of the SWICS tungsten lamp as measured by the shortwave sensor and SiPD suggesting that the change was outside the SiPD’s spectral bandpass. Ground characterization testing of lamp specimens demonstrated relative instability at the 1-percent level. Clouse (1991) shows the responsivity of the SiPDs is known to degrade as a result of gamma radiation damage in the space environment, and Lee et al. (1993) demonstrated that for ERBE this degradation of the SiPD responsivity was as much as 6 percent over 5 yr. If the SiPD responsivity were to decrease simultaneously with a brightening of the lamp, the resultant time series of narrowband SiPD measurements could appear stable as in Fig. 3.

On-orbit radiometric stabilities for the shortwave and shortwave portion of the total channels determined from solar calibrations utilizing the MAM assemblies have been described in detail by Wilson et al. (1998) and may be seen in Fig. 4. Regression analyses show gain stability in the shortwave and shortwave portion of the total channels of better than 0.07 percent (<0.1 W m⁻²) and 0.28 percent (<0.7 W m⁻²), respectively, with 95% confidence bounds of 0.17 and 0.34 percent over the time period of January 1998–July 1999. Larger variability in the total channel measurements is the result of not correctly removing the longwave components with the current algorithms. This may be seen when the solar calibration events are divided between sunrise and sunset events. During the sunrise events, the MAM baffles heat rapidly while during sunset events the MAM baffles cool rapidly. Algorithm errors manifest themselves as biases between the two types of events and thus increase the apparent overall variability in the total channel measurements.

The DCC scene-type used in the three-channel intercomparison occurs in less than 1 percent of the total earth-viewing measurements. This necessitates that the analysis be done on a monthly basis in order to obtain a rigorous statistical relationship in the narrow- to broadband conversion of the window channel measurement. The three-channel intercomparison analysis has been completed for the months of January–August 1998.

The three-channel intercomparison validation study provides insight not only of the relative relationship between the shortwave channel and shortwave portion of the total channel, but also the stability of this relationship over time. Table 2 presents the error, as calculated from Eq. (9), in this relationship on a monthly basis. There appears on average a summed total inconsistency in the estimates of the integrated spectral response functions, Ŝ_λ, of the shortwave channel and shortwave portion of the total channel of 0.66 percent with a 0.023 95% confidence bound. This total error may be completely attributable to either channel or some combination of each. There is no indication from this analysis as to the distribution. If attributed to the shortwave channel, it would suggest that our estimate of the shortwave spectral response function, Ŝ^sw_λ, is too low and that CERES shortwave data products are high by approximately 0.66 percent (∼0.5 W m⁻² TOA globally averaged flux). Conversely, if the error is attributed to the total channel, then our estimate of the total spectral response function is too high in the shortwave region by almost the same amount, and the CERES shortwave data products are fine. Regardless of which channel the error is attributed to, the derived longwave products are low by the same amount (∼0.5 W m⁻² TOA flux). These errors are being investigated and will be addressed with the release of the Edition 2 CERES ERBE-like data products in early 2000.

Regression of the monthly errors over the 8 months demonstrates a measured relative stability of 0.14 percent (0.15 W m⁻² LW TOA flux) with a 0.17-percent confidence bound. Thus the ability of the CERES instruments to detect TOA flux changes at the 0.25-percent level has not been compromised. These results agree with the findings of the solar calibrations for both the shortwave and total channels, as well as the internal blackbody calibration of the total channel. Table 3 summarizes the on-orbit radiometric stabilities of the CERES PFM radiometric channels as measured by these various calibration and validation techniques.

Conclusions

On 27 November 1997, the CERES Proto-Flight Model (PFM) instrument package was launched on the NASA Tropical Rainfall Measuring Mission (TRMM) spacecraft. A National Space Development Agency of Japan. (NASDA) launch vehicle placed the TRMM spacecraft into a low-inclination 35°, 350-km altitude orbit. Analysis of the 19 months of on-orbit internal calibration and vicarious validation studies indicate that the ground-based radiometric calibrations that are tied to ITS’90 have been successfully carried into orbit to within 0.13, 0.14, and 0.26 percent for the total, window, and shortwave channels, respectively. Additionally, these analyses have indicated that on-orbit radiometric stability has remained at levels of better than 0.14, 0.22, and 0.14 percent for the total, window, and shortwave channels. In global average TOA flux, these levels correspond to magnitudes of less than 0.3, 0.2, and 0.15 W m⁻².

The usefulness of ancillary vicarious validation studies to support or refute primary performance monitoring techniques has been demonstrated. Although the ICM and solar calibration measurements will remain the primary source of information regarding absolute radiometric performance, novel validation techniques such as the three-channel intercomparison remain necessary to completely characterize the performance envelope of space-based remote sensing instruments.