9.3 VALIDATION OF THE CERES SHORTWAVE MEASUREMENTS OVER DESERT AND CLOUD SCENES Chris Currey and Richard Green NASA Langley Research Center, Hampton, Virginia 23681- 219 9 1. INTRODUCTION 5.6% lower than the corresponding ERBE monthly albe- dos. Angular distribution models are used to convert The Clouds and the Earth's Radiant Energy System unfiltered radiances to albedos. Shortwave anisotropy (CERES) broadband shortwave channel, 0.3 - 5 pm, varies with solar zenith angle, viewing zenith angle, azi- measures atmosphere and surface reflected solar radia- muth angle, scene type (e.g. ocean, land, and desert), tion. CERES provides an in-flight tungsten lamp to mon- and cloud condition (fraction, optical depth, particle size itor the shortwave radiometric stability over the lifetime of and phase). By restricting the data to nadir views only, the instrument. Initial Tropical Rainfall Measurement we can compare different data sets without modeling the Mission (TRMM) tungsten calibration results indicate a anisotropy. Results within the remainder of this paper 0.6% change over the first six months of mission life are based on instantaneous nadir measurements. (Priestley et al. 1998). The three channel intercompari- Bright radiometrically stable desert targets (Rao et son technique and the solar diffuser calibration results ai. 1997), ice sheets (Loeb 1997), and clouds (Vermote show shortwave stabilities of 0.2% and 0.3% for the and Kaufman 1995) have been used to calibrate instru- same time period (Priestley et al. 1998). Current belief is ments and monitor sensor degradation. Clear-sky iso- that the tungsten lamp, not the radiometer, has drifted. tropic albedo, or reflectance, is defined as Radiometers, optics, and onboard calibrators exposed to the harsh space environment often degrade. Indepen- dent validation studies based on bright Earth viewing measurements are necessary to support the in-flight fil- where, I is the unfiltered radiance ( Wm-2sr-' ) measured tered calibration results. at the spacecraft, E, is the solar constant (1365 Instrument measurements of filtered radiances are Wm-2), 8, is the solar zenith angle, and d is the - converted to unfiltered radiances using the spectral cor- Earth-Sun distance in astronomical units (0.98 1.02). rection algorithm (Green and Avis 1996). The spectral In general, we will make use of the least squares correction algorithm corrects for the imperfect spectral regression = A+Bx, where 7 is the estimate of response of the instrument, thus providing an estimate of reflectance at solar zenith 2. This point estimate of the solar radiance reflected back to space. Unfiltered reflectance calculated at ? = 40°, referred to as R4, is radiances from different instruments may then be com- the point of least variance. R4 is the primary statistic pared. This paper compares broadband shortwave for target spatial and temporal testing as well as reflec- results for CERES, Earth Radiation Budget Experiment tance comparison from different instruments. The esti- (ERBE), and Scanning Radiometer for Radiation Budget mate of the variability about the regression line is (ScaRaB).- ScaRaB data is from the Meteor 3 spacecraft -!-c00 (Mar. '94 Feb. '95). Validation Earth targets provide a 2 . . A 2 a, = (yi-A-Bxi) check of the spectral correction coefficients, and provide n-2 i=l a means for cross-calibrating different datasets. and the point estimate variance is - - 2. FORMULATION Staylor (1993) used monthly clear-sky albedos from the Sahara desert (20'- 30°N, -'0 30°E), to determine the stability of the first two years of ERBE shortwave 3. DESERT REFLECTANCE RESULTS data. Results have been extended to include ERBE data - The Empty Quarter portion of the Saudi Desert (20'- over the entire five year lifetime (Mar. '85 Feb. '90). 22.5'N, 50' - 52.5'E), approximately 260 km by 280 km, The monthly dispersion ( (T /X ) is 3.4%. If we correct for consists mostly of sand dunes and sand seas. The lack seasonal variation, the monthly dispersion reduces to of moisture and saline nature of the sand cause the 1.0%. The CERES Sahara monthly albedos average region to be virtually free of vegetation. Clear nadir scenes are selected based on the ERBE scene identifi- * Corresponding author address Chris Currey, Mail Stop cation algorithm. Figure 1 shows the reflectance for 423, NASA Langley Research Center, Hampton, Virginia clear, partly cloudy, mostly cloudy, and overcast scenes 23681- 219 9; e-mail: j.c.currey@1 arc.nasa.gov. for five years of ERBE data from the Earth Radiation 4. CLOUD REFLECTANCE RESULTS Budget Satellite (ERBS). Clear is 0-5% cloud cover, and partly cloudy is 550% cloud cover. Misclassification Deep convective clouds are evaluated as validation between clear and partly cloudy is not considered a targets for broadband shortwave data. Clouds, unlike problem since there is minimal difference between deserts, are not limited to small uniform geographical reflectance regressions for 25" I 8,I52.5". Note the regions. Reflectance is brighter and less susceptible to large increase in reflectance for mostly cloudy and over- atmospheric scattering and absorption due to the high cast scenes. The Empty Quarter was clear 84% of the altitudes (>lo km). Measurement scenes are screened time; less than 5% of the scenes were mostly cloudy or based on black body brightness temperature thresholds overcast. using Planck's law: (4) 0.5 0 CLR 0 PC + MC 0.4 x OVR where C, = 3.74~10W~m -*, C, = 14387 mK, h + denotes the wavelength interval, and I, represents the 5 ++ xclr = 84.41 radiant energy emitted at the given temperature and 't: %pc -7.73 j xmc = 4.81 wavelength. Filtered radiance thresholds for 205 K, oa xon I0.W 215 K, and 225 K are calculated by applying instrument 5 spectral responses to the black body radiances and inte- grating over 8 - 12 pm ,5- 100 pm , and 10.5 - 12.5 pm 0.2 for the CERES, ERBE, and ScaRaB chmnels, respec- tively. To ensure independent sampling, reflectances are calculated per cloud system, which are typically sepa- 0.1 rated by more than 1000 km. Cloud systems must con- 20 30 40 50 Solar Zenith tain two or more nadir measurements. They are sorted based on the standard deviation, or dispersion Figure 1 : Sensitivity to Scene Identification (D = o h), of radiances within the system. Clouds over land and ocean are combined since no statistical differ- To determine the temporal stability of the Empty ence in ,R was detected. Quarter, scenes are segregated into four seasons. Table 1 presents the reflectance linear regressions Nadir radiances are averaged for each satellite over- for cosOo over 0" IO, I8 0". The number of cloud sys- pass. Seasonal estimates of R, vary from 0.188 to tems (n), point estimates ( ,R ) with standard deviations 0.192. The maximum uncertainty of ,R from (3) is in parentheses, and regression uncertainties ( ) are (T, 1.4%, with Student's t tests revealing no significant sta- calculated for each dispersion (D) and brightness tem- tistical differences. perature (TB). ERBE cloud reflectances were calculated To determine spatial homogeneity, the Empty Quar- from four years of ERBS data (Jan. - Aug., '86 - '89). ter is subdivided into four 1.25' quadrants; overpass These dates were chosen to match the available CERES nadir radiances are averaged for each quadrant. Quad- data. Cloud reflectance tends to increase as dispersion rant estimates of ,R vary from 0.187 to 0.193. The decreases. For the 205 K threshold, ,?I increases from maximum quadrant uncertainty of ,R is 0.7-% . Stu- 0.7638 to 0.7755 as dispersion varies from 0.20 to 0.02. dent's t- tests show one quadrant (21.25' 22.5'N, Reflectance also increases as temperature decreases. 51.25' 52.5'E) to be significantly different than the For 010.07, reflectance increases from 0.6896 to other three quadrants. This quadrant is removed from 0.7671 as temperature decreases from 225 K to 205 K. further analysis. The number of uniform clouds that meet each dispersion The ERBE reflectance of the Empty Quarter is criteria decreases from 215 K to 225 K. Cloud systems determined using all data from the three spatially uniform with temperature thresholds less than 205 K have the quadrants for 25" IO, I5 2.5". R, is 0.192, and from lowest variability, o, < 0.03, over the full solar zenith (3) the standard deviation is 0.4%. Using the same tech- range 0" I Bo I 80". - nique, eight months (Jan. '98 Aug. '98) of CERES The temporal stability of bright uniform 205 K clouds Empty Quarter measurements produce an ,R of (D I 0.04), is determined by processing each year of 0.178 f1.2%. Thus, CERES reflectance is 7% lower ERBE data independently. Figure 2 shows the mea- than ERBE, and statistically different. No attempt was sured reflectance versus the case, for each year. The made to calculate the point estimate reflectance using mean yearly reflectance point estimate (R,) is ScaRaB data due to an insufficient number of clear sam- 0.774f0.3%. The largest deviation of 0.5% occurs in ples collected over the nine month dataset. 1988. We conclude that ,R for uniform cold clouds with T, I 205 K is extremely stable over time. Table 1 : ERBE Deep Convective Cloud Reflectance T, I2 05 K T, 5 215 K TB 5 225 K I To compare reflectances from different instruments, to the same months as CERES. Reflectance is calcu- CERES nadir field of view (FOV) 10 km measurements lated for the solar zenith range 0" I8, I80 " and cloud are averaged to match ERBE 40 km and ScaRaB 60 km dispersion D 20.07. CERES 40 km point estimate (aa= nadir FOV sizes. For ERBE, this amounts to averaging reflectance 0.7518M0.2%)is 2% darker than two scans of 9 measurements each centered at nadir. ERBE (ria = 0.7671 kO.2%). CERES 60 km point esti- For ScaRab we used a 3x13 average. Table 2 shows mate reflectance (ria= 0.7487M0.4%i)s 4% darker the final results comparing CERES, ERBE, and ScaRaB than ScaRaB ( Ra = 0.7762 M.4%). scanners. Increasing FOV size decreases the standard error (os), yet decreases the number of detected uni- 5. SUMMARY form clouds. Scans are averaged thus decreasing the number of available FOVs, and 10 km measurements Earth targets have been identified for validation of within the enlarged FOV are required to be overcast. To the CERES broadband shortwave channel. Using five exclude seasonal effects, ERBE and ScaRaB are limited Table 2. Reflectances for T, I205 K and D I0.0 7 0.9 Samples 0.8 E 0.7 0.8 0.1 0.3 0.5 0.7 0.9 cos ,e, Figure 2: ERBE Cloud Reflectance Temporal Stability years of ERBE scanner data, the Empty Quarter (2.5' 7. REFERENCES region) within the Saudi Desert is spatially and tempo- rally characterized. Empty Quarter reflectance for the Green, R., L. Avis, 1996: Validation of ERBS Scanner first eight months of CERES is defined with a standard Radiances, J. Atmos. Ocean. Tech., 13, 851- 862. deviation of 1.2%. CERES desert reflectance is 7% Loeb, N., 1997: In-flight calibration of NOAA AVHRR vis- lower than ERBE. ible and near-IR bands over Greenland and Antarc- The desert as a calibration site is limited by its small tica, Int. J. Remote Sens., 18,477-490. size, possible greening, and overhead atmospheric scat- Priestley, K., R. Lee, B. Barkstrom, S. Thomas, K. Thorn- tering and absorption. These limitations are overcome hill, J. Paden, D. Pandey, R. Wilson, H. Bitting, and with the coldest deep convective clouds. High bright G. Smith, 1998: Radiometric stability of the CERES clouds with temperatures less than 205 K are found to scanning thermistor bolometer radiometers, Proc, be numerous and consistent to define CERES reflec- SPIE, 3439,303-31 0. tance with a standard deviation less than 0.4%. These Rao, C. R., J. Chen, J. Sullivan, N. Zhang, and W. Wang, clouds are 2% darker than ERBE, and 4% darker than 1997: Vicarious calibration of meteorological satel- ScaRaB. Measured radiances must be unfiltered with lite sensors in the visible and near-infrared regions spectral correction coefficients based on preflight instru- of the spectrum, Proc. SPIE, 311 7, 320-331. ment calibration spectral responses. Uncertainties in the Staylor, E, 1993: Stability of the Earth Radiation Budget spectral correction process may account for the differ- Experiment Scanner Results for the First Two Years ences between CERES, ERBE, and ScaRaB. The fact of Multiple Satellite Operation, J. Atmos. Ocean. that deserts and clouds give different results supports Tech., 10, 827-832. the possibility of a spectral problem. Vermote, E., Y. Kaufman, 1995: Absolute calibration of AVHRR visible and near-infrared channels using 6. ACKNOWLEDGEMENTS ocean and cloud views, Int. J. Remote Sens., 16, 2317 -2340. The authors thank Martial Haeffelin for processing ScaRaB data, Stephanie Weckmann for processing CERES and ERBE monthly data, and Jean Philippe Duvel, ScaRaB Principal Investigator, for providing the ScaRaB data and window channel conversion routine.