Validation of Clouds and Earth Radiant Energy System instruments aboard the Terra and Aqua satellites

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110,, doi: /2004jd004776, 2005 Validation of Clouds and Earth Radiant Energy System instruments aboard the Terra and Aqua satellites Z. Peter Szewczyk SAIC, Hampton, Virginia, USA G. Louis Smith National Institute for Aerospace, Hampton, Virginia, USA Kory J. Priestley NASA Langley Research Center, Hampton, Virginia, USA Received 15 March 2004; revised 10 June 2004; accepted 16 June 2004; published 19 January [1] A comparison of unfiltered radiances measured by Clouds and Earth Radiant Energy System (CERES) instruments (FM1 and FM4) operating on two different platforms, Terra and Aqua satellites, is presented. Data for the comparison were collected at orbital crossings in July and August 2002 and June Using a special scanning mode, viewing geometries of the instruments were matched to provide a large data set for comparing all three channels. In addition, the data collected over Greenland were used for a more stringent test of the consistency of the shortwave radiances. Statistics are computed for different scene types, and a confidence test is applied to compiled averages to show the consistency of 1% between measurements taken from the two different platforms. Results of the unfiltered radiance comparison are based on Edition2 of the FM1 and FM4 ES8 (Earth Radiation Budget Experiment like) data product. Citation: Szewczyk, Z. P., G. L. Smith, and K. J. Priestly (2005), Validation of Clouds and Earth Radiant Energy System instruments aboard the Terra and Aqua satellites, J. Geophys. Res., 110,, doi: /2004jd Introduction [2] The primary goal of NASA s Earth Observing System (EOS) is long-term climate observations. There have been two decades of continuous monitoring of Earth s energy budget to detect climate changes and anomalies. Several different instruments and platforms have been involved in this mission over the years [Smith, 2000]. A great effort has always been put into ensuring that there are time overlaps between these instruments so that radiance validation can be performed for the consistency and continuity of observations. [3] A Clouds and Earth Radiant Energy System (CERES) instrument is a scanning thermistor bolometer designed to measure reflected solar radiation and outgoing longwave radiation emitted from Earth for radiation budget studies [Wielicki et al., 1996]. Since 1998, CERES instruments have played an important role in the EOS mission with the launch of the Proto-Flight Model (PFM) on board the Tropical Rainfall Measuring Mission (TRMM) satellite [Smith et al., 2004]. Two CERES instruments, Flight Model 1 and 2 (FM1 and FM2), have been operating on board the Terra satellite since the beginning of In May 2002, two additional CERES Copyright 2005 by the American Geophysical Union /05/2004JD instruments (FM3 and FM4) were put in service on board the Aqua satellite. The Terra and Aqua satellites are in Sun-synchronous, polar descending and ascending orbits, respectively. [4] The comparison of the shortwave radiances from the CERES instruments aboard the TRMM and Terra spacecrafts was reported by Szewczyk [2002]. Measurements were compared when the satellite ground tracks crossed within 15 min, and the instruments viewing angles (azimuth and zenith) were matched within a prescribed tolerance. Once the matched data were extracted, a statistical analysis was performed on both data distributions. It was assumed that the distributions were statistically independent, and to reduce spatial noise, averaging was performed on a 1 1 (latitude, longitude) grid. The general approach followed the work of Haeffelin [2001]. Validation of the CERES FM1 and FM2 instruments aboard the Terra spacecraft was reported by Lee et al. [2000] and Barkstrom et al. [2000]. [5] In order to compare the CERES instruments on the Terra and Aqua satellites, a new scanning experiment was quickly designed and implemented just one month after the Aqua instruments had gone into service. The Terra and Aqua orbits cross at 70 N at local noon and 70 S at local midnight. By rotating the FM1 aboard the Terra and the FM4 aboard the Aqua so as to scan in the same plane at the orbit crossing point, both instruments view the same 1of10

2 Table 1. A CERES Data Record Showing Earth Viewing Portions of a Full Scan View Scan Position Space 1 1:65 Earth 66:265 Space 2 266:330 Space 3 331:395 Earth 396:595 Space 4 596:660 locations from the same directions, thus measuring the same radiances. The validation data were acquired in July and August 2002 when the Sun illumination of the Northern Hemisphere was sufficient to give large reflected shortwave radiances. Preliminary validation results using this data set were reported by Szewczyk et al. [2003]. The experiment was also repeated in June During that time, the two CERES instruments, FM1 and FM4, were scanning in the same plane in the vicinity of nodes of their Sun-synchronous orbits. The scanning experiment could only last 90 s for the daytime and 90 s for the nighttime data acquisition. Despite the fact it covered only about 3% of all data collected per orbit, it was designed to allow rigorous validation of measurements. The major objective of this work is to establish the relation between radiances measured by the FM1 and FM4 instruments. This paper describes a method for the validation data acquisition and the results of comparing radiances. 2. CERES Radiances and Operation Modes [6] A CERES scanner has three channels, a total ( mm), a shortwave (0.3 5 mm), and a longwave window (8 12 mm). The shortwave and window channels measure directly corresponding radiances. Broadband longwave radiances are produced by subtracting the shortwave radiances from the total channel. A full sweep of the scanner lasts 6.6 s producing 660 samples with a sampling rate of 100 Hz. Samples corresponding to Earth- and space-viewing portions of a scan are shown in Table 1. [7] The CERES instrument looks into the space (space 1 and 4) on the anti-sun side of an orbit, and the sensors cool there to set the output for zero radiances. On the opposite side of a scan (space 2 and 3), the instrument also looks at an internal calibration source to monitor and verify its stability to 0.1%. Earth-viewing samples represent radiation of a footprint that measures about 16 km by 32 km at the nadir. [8] There are two basic operational modes of a scanner, namely (1) a cross-track (XT) scan where the scanning plane is perpendicular to the satellite velocity vector, and (2) a rotational azimuth plane scan (RAPS) where the scanning plane orientation changes at a constant rate. A programmable azimuth plane scan (PAPS) mode is a variant of the RAPS mode introduced to increase sampling over an Earth target or to satisfy conditions for the scanning geometry. In this mode, the instrument head is rotated so that Earth target lies in the scanning plane or the scanning plane orientation follows a prescribed schedule. The head orientation, the instrument relative azimuth, is adjusted as a satellite moves along its trajectory and is kept constant for a certain number of scans. The duration of each steady interval is a function of a relative azimuth tolerance deemed sufficient for a given application. 3. Scanning Experiment [9] Planning tools that enable predictions for the instrument head orientation, or the PAPS mode, reside on the CERES Web site ( pps_index.html) for special operations. Calculations are based on a satellite ground track prediction file available for seven days in advance. The satellite location is given in one-min intervals in this file and is then interpolated to the duration of a single scan or 6.6 s. In this experiment, the FM1 and FM4 orientations are kept constant for about 90 s as the satellites fly about the orbital intersections. All calculations are updated daily; however, there is a 48-hour turnaround latency for commanding a CERES instrument into the PAPS mode as required by the agreement for nominal mission operations. There are several factors taken into account when designing the scanning experiment for the CERES validation Time Constraints [10] The Terra and Aqua satellites are in 81.8 and 98.2 Sun-synchronous orbits, respectively flying in opposite directions. Their respective equatorial crossing times are 10:30 AM and 1:30 PM. Therefore they arrive at the nodes (approximately 70 N and 70 S latitude) of their intersecting orbits about 15 min apart. It is clear then that the vicinity of orbital crossings presents an opportunity for comparisons provided that the viewing geometries are matched for each instrument. The northern crossing is over the Arctic around noon, and the southern crossing is over the Antarctic around midnight. To compare the shortwave channels, it is best to use the maximum insolation occurring around the northern summer solstice on 20 June. The scanning experiment was carried out in July and August 2002, just a few weeks after the Aqua launch, when the Sun illumination of the northern node was still better than average. The experiment was carried out for three weeks at the best time of the year in 2003, namely around the Northern Hemisphere summer solstice on 20 June Geometry Constraints [11] The scanning plane orientation with respect to the Sun of an instrument depends on the scanning head orientation with respect to the orbital plane and the satellite heading. In the vicinity of a node the separation between scanning planes is about 46 when both instruments are in a normal, cross-track scanning mode. By rotating the head of each instrument, the scanning planes can coincide and produce measurements suitable for the comparison. A CERES instrument scans across Earth, and an elevation angle of the scanner determines the viewing zenith angle of each footprint. It is assumed that the viewing zenith angles should be within 10 of each other to produce a valid footprint (viewed from the same direction) for comparing all three channels. The reflective light of each footprint changes with its orientation to the principal (solar) plane; therefore, the relative azimuth (RAZ) angle difference should be less than 20 to compare shortwave measurements. By aligning scanners, the RAZ condition is satisfied for almost all observa- 2of10

3 Figure 1. FM1 scanning pattern at each orbital crossing. Shortwave radiances measured on 10 July 2002 are shown. See color version of this figure at back of this issue. tions. Finally, since the solar zenith angle is the same for each instrument, shortwave measurements do not need to be normalized with respect to each other Site Constraints [12] The best comparison scene location is Greenland, since its interior is covered by an ice sheet three kilometers thick. The picture of Greenland as a plane parallel sheet of ice may not be too realistic however, because there may be features such as moguls, crevasses, etc. In this case, there are shadows on one side and extra insolation on the sunlit side, creating a more complex scene. However, the Greenland site still appears to be the most homogenous scene in this part of the Northern 3of10

4 Figure 2. FM4 scanning pattern at each orbital crossing. Shortwave radiances measured on 10 July 2002 are shown. See color version of this figure at back of this issue. Hemisphere. Therefore this site is used to verify the measurement consistency in the most ideal of conditions. The remainder of the terrain at 70 N is the Arctic Ocean, which has much meltwater over the ice leading to more complex reflective properties. There is also a strip of Northern Siberia and Canada with bare land and some vegetation exposed at this time of the year. Therefore radiance measurements for various scenes are also collected to be able to compare the spectral responses of both instruments Viewing Direction Constraints [13] The best comparison viewing direction is in the plane normal to the principal plane, which is designated as the 4of10

5 Watts per square meter per steradian Figure 3. Scan planes over Greenland of FM1 orbiting south are shown. They are orthogonal to the Sun at local noon. See color version of this figure at back of this issue. minor plane. There are far fewer features in this plane for the plane parallel radiative transfer, as forward or backward scatter occur in the principal plane. Effects of shadows can be reduced by scanning in the minor plane or the vertical plane perpendicular to the principal plane. Scanning in any other plane views the sunlit side features and also the shadows more than scanning in the minor plane. For this reason, the scanners on Terra and on Aqua are rotated to have the scan axis in the principal plane, i.e., due south as the spacecrafts cross the orbital plane intersection at noon. The principal plane rotates only slightly as the spacecrafts move over nodes of their orbits, so the instrument head is kept at the same relative azimuth during the comparison data collection period of about 90 s. Although the dynamic range of the nighttime data is minimal, the same procedure is followed for the Antarctic intersection to facilitate data 5of10

6 Watts per square meter per steradian Figure 4. Scan planes over Greenland of FM4 orbiting north are shown. They are orthogonal to the Sun at local noon. See color version of this figure at back of this issue. processing. Figures 1 and 2 illustrate the data collection that meets aforementioned objectives. At each orbital crossing FM1 and FM4 are rotated to scan in the minor plane, and this scanning experiment is repeated for the duration of the validation campaign. Figures 3 and 4 show details of this scanning pattern implemented by FM1 and FM4 over Greenland. The scanning plane orientation is shown to be orthogonal to the solar plane, which is oriented precisely South at the local noon Spatial Noise Constraints [14] Daytime scenes are very dynamic in their nature, affected by cloud cover, terrain, and the direction and the time of observations. Measurements of even seemingly 6of10

7 Table 2. Statistical Analysis of Unfiltered Radiances for July 2002 Radiance m FM4 m diff m diff,% s diff N orbx a Test SW Lwday Lwnight Wnday 5.6/mm Wnnight 3.0/mm Table 4. Statistical Analysis of Unfiltered Radiances for June 2003 Radiance m FM4 m diff m diff,% s diff N orbx a Test SW Lwday Lwnight Wnday 5.6/mm Wnnight 3.1/mm homogeneous scenes may have a wide range, but in general the cloud coverage at a given geolocation is fairly constant within a 20-min time span. Since Terra crosses a node only 15 min before Aqua does, the time differential is of lesser importance. However, the spatial variations are much more pronounced; therefore, averaging over gridded data is used to reduce the dependence of radiance measurements on the spatial noise. The size of a 1 1 grid box changes with the latitude, but in the vicinity of orbital intersections it remains almost constant at 110 km by 40 km. For an average of a grid box to be valid, it is required that at least 20 footprints lie in it or that at least 75% of its area is covered by footprints. If these conditions are satisfied, then an average is computed as a measurement to be compared. 4. Terra-Aqua Validation Campaign [15] The Terra-Aqua campaign for the validation data acquisition has already taken place twice. It is intended as an annual activity to continuously monitor the performance of CERES instruments. The first campaign began just a few weeks after Aqua was launched and ran from 4 July 2002 to 22 August The second campaign was executed in June Each day could produce validation data from up to 30 orbital crossings, and roughly half of all the data were collected for the nighttime longwave and window radiances in the southern hemisphere. Greenland-only data are confined to the region of 68 N to72 N and 30 W to48 W. Collecting such a large amount of data enabled the data processing to have greatly reduced spatial noise and improved statistical significance of differences. Meaningful statistics were computed even when all measurements were divided into subsets using the Earth Radiation Budget Experiment (ERBE) like scene identification [Wielicki and Green, 1989]. 5. Statistics [16] The design of the scanning experiment involving two instruments is such that radiances are sampled in the same manner. It means that the direction of the remote sensing of a given scene is the same for both instruments, and the difference in the time of observations is negligible. CERES Table 3. Statistical Analysis of Unfiltered Radiances for August 2002 Radiance m FM4 m diff m diff,% s diff N orbx a Test SW Lwday Lwnight Wnday 5.3/mm Wnnight 3.0/mm instruments sample the same radiation and therefore, its mean value over a grid box should be the same. Thus the mean can be used as a parameter to be analyzed for its statistical equivalence. [17] Radiance measurements by FM1 and FM4 collected over a given period of time form two sets of values for the analysis. A sample in each set is an average value of a real radiance measured by each instrument for a given grid box. Since each grid box has two averages associated with it, an analysis of their difference provides information about the discrepancy between the measurement sets. If that difference is zero over the entire set of paired grid box averages, one could say that both instruments measurements are consistent. In practice, a set of these differences is likely to follow a normal distribution. In such a case, a standard statistical measure of the consistency of a finite set is the a confidence test based on the Student s t distribution. The test gives an interval for the mean that is likely to contain any new average computed when more data have become available. This likelihood, a, is typically set to 95%. [18] One can argue that in order to ensure statistical independence of measurements, data collected during each orbital crossing should be lumped together. A sample would then be an average computed for all the data collected in close proximity of time and space. This approach would produce one average for each orbital crossing per instrument, and each sample would then be statistically independent. A set of such differences between the two instruments averages could be analyzed as in the other approach. Both approaches are being used in this work and the results are reported in the next section. 6. Results [19] The Terra-Aqua (FM1-FM4) validation campaign was run for almost two months in 2002 and three weeks in This presented an opportunity to investigate the consistency of measurements with time. Therefore the statistical analysis of data is done for each month separately. Tables 2, 3, and 4 show the comparison of radiance measurements for daytime and nighttime longwave and window, and also daytime shortwave radiances. The longwave and shortwave radiances are in [Wm 2 sr 1 ], and Table 5. Statistical Analysis of Unfiltered Shortwave Radiances Over Greenland Period m FM4 m diff m diff,% s diff N orbx a Test July August June of10

8 Figure 5. Scatterplots of daytime longwave, daytime window, and shortwave radiances for clear ocean, clear desert, and overcast scene types are shown. the window radiances are in [Wm 2 sr 1 mm 1 ]. The 95% confidence interval for averages, a test, is computed by assuming that the average radiance (m) for each orbital crossing is one statistically independent sample. The differences (m diff ) are reported as FM4 FM1. The tables also contain the spread of the differences (s diff ) and the number of orbital crossings (N orbx ) for each month. [20] The tables show very consistent measurements between instruments with differences under 1%. The shortwave radiances are slightly ( 0.4 ± 0.1%) smaller when measured by the FM4 for all-sky data, and the difference changes slightly with time ( 0.9 ± 0.1%) for June The daytime longwave radiances measured by FM4 are slightly greater, 0.7 to 0.9 ± 0.0%, for the same time period, but due to a very small value of the standard variation, the 95% confidence interval is zero. Only the nighttime window channel shows a relative difference of 1%, but its absolute value is an order of magnitude smaller than those of the other channels. It is important to note that the results of these comparisons of unfiltered radiances are based on FM1 8of10

9 Figure 6. shown. Scatterplots of nighttime longwave and window radiances for the overcast scene type are data taken from an ES8 (ERBE-like) Edition2 data product and FM4 data from an ES8 Edition2 data product. As discussed in a previous section, measurements of shortwave radiances over Greenland play an important role in verifying the consistency of both instruments. Table 5 shows the comparison of shortwave radiances. For the very bright scene, the relative difference is 0.2% and 0.1% for July and August 2002 with the tolerance of 0.4%. This difference changes to 0.6% for June 2003 with the tolerance of 0.2%. The confidence interval is smaller in this case as the standard deviation is smaller than in the other two cases. On the basis of differences shown in Table 5, one can simply state that both instruments measure the same radiance if a source is fairly uniform and homogenous. However, a small drift has been observed between the FM1 and FM4 shortwave portion of the total channel that will have to be further monitored and corrected, if needed. [21] A series of scatterplots is presented to visualize the differences between each grid box for all the validation data. Each plot represents the data collected for different scene types, which follow the ERBE-like scene classification. In order to show the consistency of the spectral response, clear sky and overcast scenes are plotted. Figure 5 shows scatterplots of daytime longwave, daytime window, and shortwave for clear ocean, clear land, and overcast conditions. These plots contain all the data collected during the validation campaign of 2002 and Figure 6 shows scatterplots of nighttime data collected over Antarctica for overcast conditions. Figure 7 shows a scatterplot of shortwave radiances collected over Greenland. A distinction is made in the plot between data collected in July 2002, August 2002, and June Closing Remarks [22] It has been shown that the experiment in which the CERES scanners, FM1 on Terra and FM4 on Aqua were scanning in the plane perpendicular to the solar plane provided a set of valuable validation data. By repeating it over an extended period of time, a large amount of validation data was collected. This in turn allowed the computation of significant statistics to support the claim that the radiance measurements have consistency within 1% regardless of the scene type. Moreover, the Greenland-only data indicate that the consistency is better than 1% when a source of radiation is more uniform and homogenous. [23] Several standard statistical measures have been computed to compare and validate unfiltered radiances. They indicate that measurements of FM1 and FM4 belong to the same distribution. The a test shows that there is the 95% Figure 7. A scatterplot of shortwave radiances over Greenland collected during 2002 and 2003 campaigns. Data include clear snow and overcast conditions. 9of10

10 likelihood that radiances measured by each instrument are within 1% as was established for various scene types. It is important to underscore the fact that the validation effort involving CERES instruments constitutes a basis for establishing the consistency in long-term Earth s energy budget measurements. [24] Acknowledgment. This work was supported by the Earth Science Enterprise and the NASA Langley Research Center under the contract NAS to Science Applications International Corporation (SAIC) and the National Institute for Aerospace. References Barkstrom, B. R., B. A. Wielicki, G. L. Smith, R. B. Lee, K. J. Priestley, T. P. Charlock, and D. P. Kratz (2000), Validation of CERES/TERRA data, Proc. SPIE, 4169, Haeffelin, M. (2001), Intercalibration of CERES and ScaRaB Earth radiation budget datasets using temporally and spatially collocated radiance measurements, Geophys. Res. Lett., 28, Lee, R. B., III, K. J. Priestley, B. R. Barkstrom, S. Thomas, A. Al-Hajjah, J. Paden, D. K. Pandey, R. S. Wilson, and G. L. Smith (2000), Terra Spacecraft CERES flight model 1 and 2 sensor measurement precisions: Ground to flight determinations, Proc. SPIE, , Smith, G. L. (2000), Critical overview of radiation budget estimates from satellites, Adv. Space Res., 24, Smith, G. L., B. A. Wielicki, B. R. Barkstrom, R. B. Lee, K. J. Priestley, T. P. Charlock, P. Minnis, D. P. Kratz, and N. G. Loeb (2004), Clouds and Earth Radiant Energy System (CERES): An overview, Adv. Space Res., 33, Szewczyk, Z. P. (2002), A recovery and comparison of shortwave radiances measured by CERES instruments operating on TRMM and Terra satellites, Proc. SPIE, 4882, Szewczyk, Z. P., G. L. Smith, and K. J. Priestley (2003), Validation of CERES instruments aboard the Terra and Aqua satellites, Proc. SPIE, , Wielicki, B. A., and R. N. Green (1989), Scene identification for ERBE radiative flux retrieval, J. Appl. Meteorol., 28, Wielicki, B. A., et al. (1996), Clouds and the Earth s Radiant Energy System (CERES): An Earth observing system experiment, Bull. Am. Meteorol. Soc., 77, K. J. Priestley, NASA Langley Research Center, Hampton, VA 23681, USA. (k.j.priestley@larc.nasa.gov) G. L. Smith, National Institute for Aerospace, Hampton, VA 23681, USA. (g.l.smith@larc.nasa.gov) Z. P. Szewczyk, SAIC, Hampton, VA 23666, USA. (z.p.szewczyk@ larc.nasa.gov) 10 of 10

11 Figure 1. FM1 scanning pattern at each orbital crossing. Shortwave radiances measured on 10 July 2002 are shown. 3of10

12 Figure 2. FM4 scanning pattern at each orbital crossing. Shortwave radiances measured on 10 July 2002 are shown. 4of10

13 Figure 3. Scan planes over Greenland of FM1 orbiting south are shown. They are orthogonal to the Sun at local noon. 5of10

14 Figure 4. Scan planes over Greenland of FM4 orbiting north are shown. They are orthogonal to the Sun at local noon. 6of10