Biases in Earth radiation budget observations 2. Consistent scene identification and anisotropic factors

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 101, NO. D16, PAGES 21,253-21,263, SEPTEMBER 27, 1996 Bases n Earth radaton budget observatons 2. Consstent scene dentfcaton and ansotropc factors Qan Ye and James A. Coakley Jr. College of Oceanc and Atmospherc Scences, Oregon State Unversty, Corvalls Abstract. Smple threshold scene dentfcaton methods are developed to reduce the effects of errors n scene dentfcaton on the ansotropy of reflected and emtted radances nferred from Earth Radaton Budget Experment (ERBE) scanner observatons. The ERBE maxmum lkelhood estmate (MLE) scene dentfcaton s assumed to be accurate for nadr felds of vew. Varous combnatons of neghborng ERBE scanner felds of vew at nadr are used to determne the populaton of cloud scene types as a functon of feld of vew sze. Longwave and shortwave thresholds are then determned for each of the ERBE solar zenth, satellte vew zenth, and relatve azmuth angular bns so that the populaton of cloud scene types at a partcular satellte vew zenth angle s consstent wth the feld of vew sze at the partcular satellte vew zenth angle. Dfferences between the ansotropy of reflected sunlght and emtted longwave radaton obtaned usng the new scene dentfcaton method and that obtaned usng the ERBE MLE method show that the ERBE radatve fluxes have satellte vew zenth angle dependent bases. Thresholds are also developed for cloud scene dentfcaton wth felds of vew that are constructed to have a constant sze wth satellte vew zenth angle. The angular dependence of reflected sunlght and emtted longwave radaton for scenes dentfed wth these thresholdshow lttle dependence on feld of vew sze. Ths lack of dependence s a necessary condton for usng scannng radometer data to obtan radatve fluxes. 1. Introducton surrounded by felds of vew that contan broken clouds and In part 1 [Ye and Coakley, ths ssue] a potental vew was thus lkely to have more cloud contamnaton than a feld zenth angle dependent bas n Earth radaton budget of vew dentfed as clear at the lmb. At the lmb the feld of observatons, lke those made by the Earth Radaton Budget vew s large. It was shown that large regons dentfed as Experment (ERBE), was descrbed. In ERBE the radatve beng clear were often surrounded by regons that were also fluxes were derved by frst dentfyng wthn each scanner clear. Consequently, large regons were lkely to be less cloud feld of vew the cloud scene type: clear, partly cloudy, mostly contamnated. Evdently, these large clear felds of vew were cloudy, or overcast. The dentfcaton was made on the bass parts of vast regons that were cloud free throughout. Smof the shortwave and longwave radances observed for the larly, overcast scenes at nadr were more lkely than overcast feld of vew [Welck and Green, 1989]. An ansotropc scenes at the lmb to be contamnated by breaks n the cloud. factor assocated wth that scene type was then used to convert Because clear scenes reflect sunlght and emt radaton that the radances to radatve fluxes. The vew zenth angle vares more strongly wth angle than that for overcast scenes, dependent bas arses because of (1) vew zenth angle the effect of the clusterng of clouds on certan spatal scales dependent errors n scene dentfcaton, (2) the nonunform and the ERBE nverson method causes clear scenes to appear spatal dstrbuton of clouds, and (3) the growth of the feld of less ansotropc than they should and overcast scenes to vew sze from nadr to lmb. appear more ansotropc than they should. Consequently, the The nonunform spatal dstrbuton of clouds contrbutes to ansotropy of the radatve felds obtans an apparent lnk to the bas because regons that are the sze of an ERBE scanner the sze of the regon beng vewed. Because the feld of vew feld of vew at nadr -(40 km) 2 have relatvely large sze for the ERBE scanner grows from nadr to lmb, the frequences of clear and overcast condtons at the expense of ERBE fluxes are lkely to have a satellte vew zenth angle relatvely nfrequent occurrences of partly cloudy and mostly dependent bas. cloudy condtons. Larger regons have smaller frequences of In part 1 such a bas was found for clear ocean scenes but clear and overcast condtons and larger frequences of partly not for the other scene types. Instead, t was found that the cloudy and mostly cloudy condtons. In part 1 t was shown ERBE maxmum lkelhood estmate (MLE) scene dentfcathat a scene dentfed as beng clear at nadr was often ton method [Welck and Green, 1989] often ncorrectly dentfed partly and mostly cloudy scenes at the lmb as beng SNow at CIRES, Unversty of Colorado. Boulder. Copyrght 1996 by the banercan Geophyscal [Jnon. overcast. The dependence of the frequency of scene type on feld of vew sze was determned from observatons for the nadr felds of vew by combnng the felds of vew to obtan observatons for regons of varous szes. The ERBE scene dentfcaton scheme was assumed to be correct for the nadr felds of vew. The frequency of overcast scenes at the lmb Paper number 96JD õ-0227/96/96JD-01! ,253

2 21,254 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY was found to be larger than expected gven the sze of the lmb feld of vew. Conversely, the frequences of partly cloudy and evdence that the new scene types have the cloud cover assocated wth the ERBE scene types: clear, 05Ac<0.05; mostly cloudy scenes at the lmb were smaller than expected. partly cloudy, 0.05<Ac< 0.5; mostly cloudy, 0.5<_Ac< 0.95; Because of the ncorrect dentfcaton the ansotropy of reflected and emtted radaton could not be determned for partly cloudy, mostly cloudy, and overcast scenes. Here the ERBE scene dentfcaton errors found n part 1 and overcast, 0.955Ac51. Nevertheless, thresholds were consstent wth the features of the ERBE MLE method. Larger shortwave reflectvtes and smaller emtted longwave radances were assocated wth larger cloud cover fractons. are avoded by developng a new method for dentfyng cloud In addton to the FFOV threshold method a second set of scene types from the ERBE scanner radances. The new thresholds was developed usng the "constant sze feld of method produces populatons of scene types at each satellte vew" (CFOV) observatons ntroduced n part 1. CFOV vew zenth angle that are more consstent wth the feld of vew sze at the partcular angle. The new method s based on observatons were constructed by groupng neghborng ERBE scanner felds of vew so that the sze of the regon beng smple thresholds appled to the shortwave and longwave vewed was almost nvarant wth satellte vew zenth angle. radances obtaned wth the ERBE scanner. Ansotropc For the CFOV observatons, longwave and shortwave factors for reflected sunlght and emtted radaton are derved based on the radances obtaned wth ths new scene dentfcaton. Comparsons of the ansotropc factors wth those thresholds were adjusted so that the populaton of scene types wthn an ERBE satellte vew zenth angle bn was equvalent to the populaton obtaned for the CFOV observatons at derved wth the ERBE MLE scene dentfcaton method nadr. Lke the FFOV threshold dentfcaton method, the show the effects of scene dentfcaton errors and cloud CFOV dentfcaton method removes the growth of apparent clusterng on the angular dstrbuton of reflected sunlght and cloud cover from nadr to lmb found wth the ERBE MLE emtted longwave radaton. The ansotropc factors derved scene dentfcaton scheme. In addton, the CFOV observawth the new scene dentfcaton are shown to be relatvely tons reduce effects due to the spatal scales on whch clouds nsenstve to feld of vew sze, a necessary condton for obtanng radatve fluxes from scanner observatons usng methods lke those employed by ERBE. congregate. Because of ther constant sze, the CFOV observatons should be subject to relatvely constant cloud and clear contamnaton at all satellte vew zenth angles. Fgure 1 llustrates the procedures used to derve the long- 2. Analyss Methods wave and shortwave radance thresholds for a partcular satellte vew zenth angle. The fgure llustrates the dvson The ERBE MLE scene dentfcaton method was assumed of the longwave-shortwave radance par doman for a to provde the correct dentfcaton for felds of vew at nadr. partcular Sun-scene-satellte geometry. The crosses gve the The dependence of the populaton of cloud scene types on feld of vew sze was determned by smulatng nadr felds of means of the longwave and shortwave radances assocated wth clear, partly cloudy, mostly cloudy, and overcast scenes vew of the szes that matched those assocated wth the used n the ERBE MLE scene dentfcaton method. The ERBE satellte vew zenth angle bns. The requred ellpses are meant to llustrate the standard devatons and groupngs were descrbed n part 1. The averages of the lnear correlatons for the longwave and shortwave radances shortwave and longwave radances for the scanner felds of based on the values used n the ERBE MLE scene vew that contrbuted to a smulated feld of vew were taken dentfcaton method. The fgure shows a hypothetcal case n to be the radances for the smulated feld of vew. The scene whch the ERBE MLE method ncorrectly dentfes some type was based on the nomnal cloud cover derved from the mostly cloudy felds of vew as overcast and some clear felds cloud scene types of the contrbutng scanner felds of vew. The assgnment of scene type was also descrbed n part 1. As n part 1, the observatons were from the Earth Radaton Budget Satellte (ERBS) scanner for the months of Hot September, October, and November Only ocean and overcast scene types were consdered, as the purpose of ths study was to determne whether the ansotropy of the radances obtaned wth the new scene types showed reduced senstvty to feld of vew sze. Whle the range of scene types was lmted, they consttuted approxmately 70% of the observatons obtaned wth the ERBS scanner. Thus the fndngs presented here are lkely to prove relevant for all scene types. The feld of vew sze dependence of the populaton of scene types was used to derve two sets of thresholds. For the frst set, thresholds were adjusted so that the populaton of scene types assocated wth the feld of vew sze at a partcular angle were dentcal to that for the same sze feld of vew at nadr. These thresholds are referred to as the "full resoluton feld of vew" (FFOV) thresholds. Snce the thresholds were based on the ERBE scene dentfcaton results, the scene types derved by usng the thresholds were taken to be the same as ther ERBE counterparts: clear, partly cloudy, mostly cloudy, and overcast. There s, however, no Cold Dark Shortwave Radance Brght Fgure 1. Dagram llustratng procedures used to obtan the longwave and shortwave radance thresholds based on frequences of occurrence derved from smulated nadr observatons.

3 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY 21, SZA= VZA=O.O-15.0 AZA O ' I... /"' / ' ' x Ct'EAR'SKY' ' J,'/ o PARTLY CLOUDY J, ' / /, MOSTLY CLOUDY oøøo,' / + ' - % ;'o,,,/, -, O_o?...?,,. oo ½ OOo/O :/' ' ß * :--... ooo ',o/*., **:** ** ' ** *,, ß ß, o c9, * / * **,* o /o, /;_L..,.?.,_,...,,_.,.*_...,* **. *, %4-- * +4- ß?', *.. -, J * +, _, * + * CFOV THRESHOLD / ; --- FFOV THRESHOLD SHORTWAVE RADIANCE + + Fgure 2. An example showng the effect of full resoluton of vew were determned, the boundary separatng partly and feld of vew (FFOV) and constant sze feld of vew (CFOV) mostly cloudy felds of vew was fxed by adjustng both the thresholds on scene dentfcaton. The arrows n the fgure shortwave and the longwave thresholds so that the boundary pont to observatons whch were dentfed as ether clear sky remaned normal to the lne connectng the means of shortor overcast by the Earth Radaton Budget Experment wave and longwave radances assocated wth the partly and (ERBE) maxmum lkelhood estmate (MLE) scene dentf- mostly cloudy felds of vew dentfed by the ERBE MLE caton but were redenttled by the CFOV threshold method. scene dentfcaton method. The fnal poston of the Radances are n Wm -2 sr -. boundary was determned by the frequences of occurrence obtaned from the smulated nadr observatons. of vew as partly cloudy. In ths case, the populaton of overcast and clear scenes has to be reduced accordng to the frequences of occurrence derved from the smulated nadr observatons. The followng procedures were used to obtan the desred populaton of scene types. Frst, thresholds were adjusted to dstngush between partly cloudy and clear scenes and between mostly cloudy and overcast scenes. For overcast scenes, an ntal longwave threshold was arbtrarly set at the mean value plus two standard devatons of the longwave radances for the felds of vew dentfed as beng overcast by the ERBE MLE scene dentfcaton method. Smlarly, an ntal shortwave threshold was arbtrarly set at the mean value mnus two standard devatons of the shortwave radances. If based on the populaton of scene types expected for the sze of the feld of vew at the partcular satellte vewng zenth angle, the ntal thresholds dentfed too many felds of vew as overcast, then the ntal longwave radance threshold was reduced (moved downward n Fgure 1), and the ntal shortwave radance threshold was ncreased (moved rght n Fgure 1). The rato of the change n the longwave radance to the change n the shortwave radance was set equal to the slope of the longwave-shortwave radance relatonshp derved from a lnear leastsquares ft for the scanner felds of vew dentfed as beng overcast by the ERBE MLE scene dentfcaton method. The step sze used for the shortwave radance threshold was taken to be 1 Wm -2 sr - for the overcast-mostly cloudy boundary and 0.1 Wm -2 sr - for the clear-partly cloudy and partly cloudy-mostly cloudy boundares. Smlar adjustments were made for the thresholds separatng the clear and partly cloudy felds of vew. After the radance boundares for overcast and clear felds Fgure 2 shows an example of the determnaton of FFOV and CFOV thresholds for the ERBE angular bn n whch the solar zenth angles range from 25.8 ø to 36.9 ø, the vew zenth angles range from 0.0 ø to 15.0 ø, and the azmuthal angles range from 9.0 ø to 30.0 ø. The markngs gve the MLE scene dentfcaton: clear (crosses), partly cloudy (crcles), mostly cloudy (astersks), and overcast (pluses). Felds of vew dentfed as ether clear or overcast by the MLE method but redenttled by the CFOV threshold method as partly and mostly cloudy are ndcated by arrows. For the angular bn shown n the fgure, as was often the case, the CFOV thresholds are more restrctve than the FFOV thresholds for clear and overcast scenes. Table 1. Frequences of Occurrence for Smulated Full Resoluton ERBE Scanner Observatons Satellte Vew Zenth Angle Frequency of Occurrence, % Bn Number Feld of Angular Vew Partly Mostly Range, Sze Clear Cloudy Cloudy deg 104 km 2 Ocean Ocean Ocean Overcast The frequences were obtaned from ERBS scanner observatons for September, October, and November ERBE, Earth Radaton Budget Experment; ERBS, Earth Radaton Budget Satellte.

4 21,256 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY 50.0 CLEAR SKY 50.( PARTLY CLOUDY K o... N<. FFOV OBSERVATIONS -o CFOV OBSERVATIONS 37.5 z u I,I n/ o O o 20 8O 0 5o.o 0 MOSTLY CLOUDY 50.( I,I I, o VIEWING ZENITH ANGLE (DEG) Fgure 3. Frequences of occurrence for FFOV and CFOV observatons dentfed usng the FFOV thresholds. The results are based on Earth Radaton Budget Satellte (ERBS) scanner observatons for September, October, and November Results not ncluded n the analyss. As descrbed n part 1, these procedures were used to nsure the statstcal ndependence of Table 1 lsts the frequences of occurrence obtaned from each sample. Dfferences shown n Fgure 3 between the the smulated felds of vew at nadr. The frequences of occurrence decreased for clear and overcast scenes and FFOV and the CFOV frequences for the largest satellte vew zenth angle are typcal of the samplng errors. A second ncreased for partly and mostly cloudy scenes as the feld of cause of the resdual varaton s that the FFOV thresholds do vew sze grew from the -(40 km) 2 for the satellte vewng not allow for effects due to cloud clusterng. Scenes dentfed zenth angle bn at nadr to -(115 km) 2 for the satellte wth the FFOV thresholds are lkely to have more cloud vewng zenth angle bn at the lmb, whch here was taken to be at 64 ø. Clear and overcast condtons occurred about 20% contamnaton at nadr than at the lmb. Consequently, the of the tme at nadr but only about 10% of the tme at the lmb. CFOV frequences for clear scenes at small satellte vew Correspondngly, partly and mostly cloudy condtons zenth angles are lkely to be hgher than those for large vew occurred 60% of the tme at nadr but about 80% of the tme at the lmb. The frequenceshown n Table 1 were used to establsh the FFOV and CFOV thresholds descrbed n the zenth angles. The same trend s expected for the overcast scenes. The varatons shown n Fgure 3 for the CFOV observatons, however, appear to be wthn errors due to prevou secton. samplng. Fgure 3 shows the frequences of occurrence obtaned by Fgure 4 shows the frequences of occurrence for the four applyng the FFOV thresholds to FFOV (sold lnes) and cloud categores dentfed wth the CFOV threshold method CFOV (dashed lnes) observatons. For the FFOV observafor FFOV (sold lne) and CFOV (dashed lne) observatons. tons the frequences of occurrence showed the expected trends Compared wth the results obtaned wth the FFOV thresholds wth satellte vewng zenth angle: the frequences of clear (Fgure 3), almost constant frequences of occurrence are and overcast felds of vew decreased wth ncreasng vew obtaned for the CFOV observatons. For the FFOV observazenth angle and the frequences of partly and mostly cloudy tons the CFOV thresholds produce frequences of occurrence felds of vew ncreased. For the CFOV observatons the near nadr for clear ocean and overcast felds of vew that are frequences for all scene types were relatvely constant. The smaller than those produced by the FFOV thresholds. The varatons n the frequences were caused n part by samplng. smaller frequences are expected because for clear and overcast scenes the CFOV thresholds are often more restrctve The frequences to whch the thresholds were adjusted (Table 1) were obtaned usng all felds of vew. The thresholds, than the FFOV thresholds, as was, for example, the case however, were obtaned for felds of vew that were separated shown n Fgure 2. Scenes near nadr, whch the FFOV by at least 40 scan lnes for a gven scene type, solar zenth, thresholds dentfy as clear or overcast, are often n the mdst satellte zenth, and relatve azmuth angular bn. In addton, of broken clouds. Wth the CFOV thresholds, some of these bns for whch there were fewer than eght observatons were felds of vew are dentfed as partly or mostly cloudy.

5 _ YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE IX) AND ANISOTROPY 21, CLEAR OCEAN 50.0 PARTLY CLOUDY OCEAN e,... o MOSTLY CLOUDY OCEAN 50.0 I O 4O 6O 8O 50. : FFOV OBSERVATIONS > <,--- CFOV OBSERVATIONS z I I I 2O 4O 6O 0 I I I 8O 0 2O 4O 6O 8O VIEW ZENITH ANGLE (DEGREE) Fgure 4. Frequences of occurrence for FFOV and CFOV observatons dentfed wth the CFOV threshold dentfcaton method. The results are based on ERBS observatons for September, October, and November 1986 SZA MLE - FFOV THRESHOLD I I I I x CLEAR I KY I SW MEAN o PARTLY CLOUDY - * MOSTLY CLOUDY - + MLE - FFOV THRESHOLD LW MEAN _..-""...,4. w w n w L -20 %/' k, "o _,,' '-.,., '.. - '"C'.,.,,......, '.. _ "..., I I I I I I I o o -,/ / ß _,, ' / -_ I I I 4o z w w 20 MLE - CFOV THRESHOLD t I I I I I I SW MEAN MLE - CFOV THRESHOLD - LW MEAN _ - t j,', / ',5 - /',,/ I / / / //.'/ ,0 I I I I I I o 4o VIEW ZENITH ANGLE (DEGREE) Fgure 5. Percent dfferences n the mean radances for scenes dentfed by the ERBE MLE, FFOV, and CFOV methods. The means obtaned usng the threshold methods were subtracted from those obtaned usng the ERBE MLE method. ERBS scanner observatons for September, October, and November 1986 were used. The results presented n the fgure are azmuthally averaged. They are for solar zenth angles between 25.8 ø and 36.9 ø.

6 21,258 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY Clearly, the mean radances, standard devatons, and correlatons between shortwave and longwave radances for the scene types dentfed wth the FFOV and CFOV * FFOV THRESHOLD thresholds dffer from those obtaned wth the ERBE MLE 0.65 o....e CFOV THRESHOLD scene dentfcaton. Fgure 5 shows an example of the percent dfferences n the means of the shortwave and longwave radances. The dfferences are obtaned by subtractng 0.60 the radances obtaned usng the threshold methods from those obtaned usng the ERBE MLE scene dentfcaton method. For scenes dentfed as clear by the FFOV thresholds, the means of the reflected shortwave radances are smaller than 0.55 those for the scenes dentfed as clear by ERBE MLE method. The correspondng means of the longwave radances are larger. Conversely, for scenes dentfed by the threshold 0.50 methods as overcast, the means of the reflected shortwave radances are larger and the means of the longwave radances are smaller. Snce clouds generally have hgher reflectvtes 0.45, I, I, I, than the ocean background and are at lower temperatures, the clear felds of vew dentfed by the threshold methods would appear to have less cloud contamnaton than those dentfed VIEW ZENITH ANGLE (DEGREE) by the ERBE MLE method. Lkewse, the overcast scenes Fgure 6. Fractonal cloud cover obtaned wth the ERBE would appear to be less contamnated by breaks n the cloud. MLE method (heavy sold lne), the FFOV thresholds (sold These dfferences were, of course, forced by the desgn of the lne), and the CFOV thresholds (dashed lne). Observatons threshold methods. The desgn was to reduce the growth n are from the ERBS scanner for September, October, and the number of overcast scenes wth satellte vew zenth angle obtaned wth the ERBE MLE scene dentfcaton method. Compared wth the radances obtaned wth the FFOV November thresholds, the reflected shortwave radances obtaned wth the CFOV thresholds are smaller for clear and partly cloudy ocean scenes. The shortwave radances are larger for mostly fractonal cloud cover derved from both threshold methods, usng the nomnal cloud fractons for the ERBE scene type, were relatvely constant wth ncreasng vew zenth angle cloudy ocean and overcast scenes. The longwave radances compared wth the ncrease obtaned wth the ERBE MLE obtaned wth the CFOV threshold method are larger than scene dentfcaton method. those obtaned wth the FFOV threshold method for clear Usng the procedures descrbed n part 1, FFOV and ocean scenes. Evdently, clear ocean scenes dentfed wth CFOV pseudoansotropc factors were obtaned for the FFOV the CFOV threshold method are less cloud contamnated than and CFOV thresholds. Fgure 7 shows dfferences n the those obtaned wth the FFOV thresholds. The reflectvtes of ansotropc factors obtaned wth the FFOV shortwave and overcast scenes dentfed wth the CFOV thresholds are longwave thresholds appled to the FFOV and CFOV obsergenerally hgher than those for scenes dentfed by the FFOV vatons. The fgure shows the percent dfferences (FFOVthreshold method and the emtted longwave radances are CFOV) for the solar zenth angle bns whch had the largest lower. Evdently, the overcast scenes dentfed wth the numbers of satellte zenth and azmuth angle bns for whch CFOV threshold method are less contamnated by breaks n the clouds. the dfferences n the ansotropc factors were sgnfcant at the 90% confdence level (shaded regons). Procedures for Because the shortwave-longwave radance pars assocated determnng the confdence levels were also descrbed n part wth the threshold scene types no longer match those of the ERBE scene types, the conventonal denttes, clear, partly cloudy, etc., are used here only to smplfy the dscusson. No clam s made that the thresholds dd a better job of 1. Fgure 8 shows the same dfferences but for the solar zenth angle bns whch had the smallest numbers of satellte zenth and azmuth angle bns n whch the dfferences were statstcally sgnfcant. Compared wth smlar result shown dentfyng clear scenes or overcast scenes. The thresholds n part 1 obtaned wth the ERBE MLE scene dentfcaton, smply dvded the longwave-shortwave radance doman nto regons n whch certan specfed percentages of the felds of vew were to be found. In makng these dvsons, t was assumed that each regon of the longwave-shortwave doman was occuped by a dstnct scene type. Ths assumpton, whle common to all threshold methods, was not tested. The goal the dfferences between the FFOV and the CFOV ansotropc factors were reduced both n magntude, when the dfferences were statstcally sgnfcant, and n the number of bns found to have statstcally sgnfcant dfferences. For bns n whch large percentage dfferences occurred, the dfferences lacked statstcal sgnfcance. In addton, compared wth ther was to determne whether dvdng the longwave-shortwave ERBE counterparts, the dfferences obtaned wth the radance doman so that the populaton of scene types was threshold methods appeared to be more randomly dstrbuted solely a functon of feld of vew sze would lead to an angular n the vewng zenth and azmuth angle doman. Table 2 dependence for reflected sunlght and emtted longwave gves the percentage of the number of angular bns whch had radaton that, unlke the ERBE observatons, was dfferences whch were sgnfcant at the 90% confdence ndependent of feld of vew sze. The FFOV and CFOV thresholds were appled to the ERBS scanner observatons. Fgure 6 shows that the level. The percentage s that of the total number of bns for whch there were at least eght statstcally ndependent observatons as descrbed n part 1. Clearly, wth the FFOV

7 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY 21,259 PERCENT DIFFERENCE (FFOV - CFOV) CLEAR SKY OCEAN PARTLY CLOUDY OCEAN g MOSTLY CLOUDY OCEAN Fgure 7. Percent dfferences between the FFOV and the CFOV pseudoansotropc factors for the solar zenth angle bns whch had the largest number of satellte vew zenth and relatve azmuth angle bns for whch the dfferences were sgnfcant. The scenes were dentfed wth the FFOV thresholds appled to both the FFOV and the CFOV observatons. The solar zenth angles for each scene type are gven n the fgure. The radal axs s for vew zenth angle. The polar axs s for the relatve azmuth angle. The ncrement of the contours s 2.5%. Regons n whch the dfferences were postve and sgnfcant at the 90% confdence level are shaded. Those that were negatve and sgnfcant are hatched. Table 2. Percentage of Angular Bns Whch Had Dfferences n the CFOV and FFOV Ansotropc Factors that Were Sgnfcant at the 90% Confdence Level Partly Scene Cloudy Mostly Identfcaton Clear Ocean Cloudy Method Ocean Shortwave Ocean Overcast ERBE MLE FFOV threshold CFOV threshold Longwave ERBE MLE FFOV threshold CFOV threshold The percentages are for ERBS scanner observatons for September, October, and November CFOV, constant sze feld of vew; FFOV, full-resoluton feld of vew; MLE, maxmum lkelhood estmate.

8 21,260 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY PERCENT DIFFERENCE (FFOV - CFOV) CLEAR SKY OCEAN PARTLY CLOUDY OCEAN 84., ,3 75 5! ! ! ! MOSTLY CLOUDY OCEAN Fgure 8. Same as Fgure 7 but for the solar zenth angle bns whch had the smallest number of satellte vew zenth and relatve azmuth angle bns for whch the dfferences were sgnfcant at the 90% confdence level. SZA CLEAR SKY PARTLY CLOUDY ß ---, FFOV MODEL 1.2 e--o CFOV MOD o -- o MOSTLY CLOUDY Q _ VIEWING ZENITH ANGLE (DEGREE) Fgure 9. Azmuthally averaged shortwave ansotropc factors for FFOV and CFOV observatons obtaned wth the FFOV thresholds. The azmuthally averaged ansotropc factors were averaged for all solar zenth angle bns.

9 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY 21,261 PERCENT DIFFERENCE (FFOV - CFOV) CLEAR SKY OCEAN PARTLY CLOUDY OCEAN MOSTLY CLOUDY OCEAN Fgure 10. Same as Fgure 7 but the FFOV and CFOV pseudoansotropc factors were constructed from scenes dentfed wth the CFOV thresholds. thresholds these percentages were smaller than those obtaned dentfcaton showed less evdence for a vew zenth angle wth the ERBE MLE. Fgure 9 shows the azmuthally averaged ansotropy for the dependent bas. Because of cloud clusterng and the growth of the feld of FFOV and CFOV observatons obtaned wth the FFOV vew sze from nadr to lmb, a varyng degree of cloud or threshold scene dentfcaton method. As was done n part 1 the azmuthally averaged ansotropc factors were averaged for all solar zenth angles. Because of the effects of cloud clear contamnaton was expected n the FFOV threshold scene dentfcaton. Even though the FFOV threshold method appeared to reduce effects due to errors n scene clusterng, the FFOV observatons dentfed as clear were dentfcaton obtaned wth the ERBE MLE scene expected to have more cloud contamnaton than the CFOV observatons and the overcast scenes were expected to have more clear contamnaton. Consequently, for clear skes the CFOV observatons were expected to show a hgher degree of ansotropy, and for overcast condtons they were expected to dentfcaton method, as ndcated by the nearly constant fractonal cloud cover at all satellte vew zenth angles (Fgure 6), the derved ansotropy stll appeared to be a.functon of spatal scale. The resdual dependence was due to cloud clusterng. Clear and overcast felds of vew were show a hgher degree of sotropy than the FFOV observatons sometmes extracted from the mdst of broken cloudy felds. showed. Fgure 9 shows that for clear ocean scenes the CFOV observatons had a constant sze feld of vew from average ansotropc factors for the CFOV observatons were nadr to lmb; so the degree of cloud and clear contamnaton smaller than those for the FFOV observatons at nadr and n the CFOV observatons was expected to reman constant larger at the largest vew zenth angle. For overcast scenes from nadr to lmb. AS a result, the CFOV thrcsholds ths trend was reversed. obtaned for the CFOV observatons were expected to reduce Dfferences n the azmuthally averaged ansotropc factors the spatal scale dependence of the derved ansotropc factors. were reduced from the levels of 2-10% found n part 1 for FFOV and CFOV ansotropc factors were constructed the E- E MLE scene dentfcaton to about 0.5% overall. wth the CFOV threshold scene dentfcaton. Fgure 10 The ansotropc factors derved usng the FFOV threshold shows the percent dfferences (FFOV - CFOV) for the solar scene dentfcaton method showed less spatal scale zenth angle bns whch had the largest doman of sgnfcant dependence than those obtaned usng the ERBE MLE scene dfferences at the 90% confdence level (shaded regons). dentfcaton method. Consequently, compared wth the Fgure 11 shows the same dfferences but for the solar zenth E RBE MLE scene dentfcaton, the FFOV threshold scene angle bns whch had the smallest doman of sgnfcant

10 2,262 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY PERCENT DIFFERENCE (FFOV - CFOV) CLEAR SKY OCEAN PARTLY CLOUDY OCEAN MOSTLY CLOUDY OCEAN $ Fgure 11. Same as Fgure 8 but the FFOV and CFOV pseudoansotropc factors were constructed from scenes dentfed wth the CFOV thresholds. dfferences. Compared wth Fgures 7 and 8, the domans of sgnfcant dfferences n Fgures 10 and 11 are smaller. The azmuthally averaged ansotropc factors for the FFOV and scene type was used to assgn the combned felds of vew to one of the ERBE cloud scene types. Threshold scene dentfcaton methods were developed to gve the populaton of CFOV observatons dentfed wth the CFOV thresholds are scene types at a gven satellte vew zenth angle shown n Fgure 12. Agan the azmuthally averaged correspondng to the feld of vew sze at the partcular angle. ansotropc factors were averaged for all solar zenth angles. These scene dentfcaton methods reduced the angular As expected, for overcast scenes the CFOV observatons were dependence of the scene dentfcaton errors obtaned wth the more sotropc than were the FFOV observatons. The ERBE MLE scene dentfcaton method. The thresholddfferences, however, were smaller than those shown n Fgure 9. Table 2 also shows that the percentages were smaller than derved scene types, however, are acknowledged to be defned by the thresholds rather than by some characterstc feature those obtaned wth the FFOV thresholds for both the such as fractonal cloud cover. Nevertheless, the thresholds shortwave and the longwave radances. were desgned to mmc procedures commonly used to obtan cloud cover from satellte magery. The thresholds were set so 4. Conclusons Errors n scene dentfcaton coupled wth the spatal scales on whch clouds congregate gve rse to a feld of vew that clear scenes had the largest longwave radances and the smallest reflected shortwave radances; overcast scenes had the largest reflected shortwave radances and the smallest longwave radances. sze dependence for the ansotropy of emtted and reflected The scene dentfcaton method based on thresholds that radances. In addton, as was noted n part 1, the ERBE were desgned to recover the populaton of scene types MLE scene dentfcaton method appears to have dentfed commensurate wth the areas of each of the ERBE satellte many mostly and partly cloudy scenes at the lmb as overcast. vewng zenth angle bns s referred to as the full resoluton Ths tendency was revealed by assumng that the ERBE MLE feld of vew (FFOV) threshold method. Felds of vew scene dentfcaton method was correct for nadr observatons dentfed wth ths method produced reflected and emtted and then determnng the relatonshp between the populaton of scene type and the feld of vew sze from the nadr observatons. Felds of vew of varous szes were constructed by radances wth angular dependences that were relatvely nsenstve to feld of vew sze. Isolated clear felds of vew drawn from the mdst of partly cloudy scenes, however, combnng neghborng felds of vew at nadr. A composte tended to exhbt effects due to cloud contamnaton, and

11 YE AND COAKLEY: RADIATION BUDGET, CONSISTENT SCENE ID AND ANISOTROPY 21,263 SZA CLEAR SKY PARTLY CLOUDY ß --, FFOV MODEL CFOV MODEL O vlostly CLOUDY I VIEW ZENITH ANGLE (DEGREE) Fgure 12. Azmuthally averaged shortwave ansotropc factors for FFOV and CFOV observatons obtaned wth the CFOV thresholds. The azmuthally averaged ansotropc factors were averaged for all solar zenth angle bns. overcast scenes drawn from the mdst of mostly cloudy regons lkewse showed sgns of breaks n the clouds. Evdence for these types of contamnaton came from comparsons wth felds of vew dentfed wth constant sze feld of vew (CFOV) thresholds n whch the sze of the feld of vew was set equal to the sze of the ERBE scanner feld of vew at a vew zenth angle of about 64 ø. The feld of vew sze dependence n the ansotropy of reflected and emtted radances revealed here coupled wth the ncrease n feld of vew sze wth satellte vew zenth angle from nadr to lmb should produce satellte vew angle dependent bases n the radatve fluxes obtaned by ERBE. Whle the magntudes of the bases are undoubtedly small (2-10%, dependng on satellte vew zenth angle), the results presented here ndcate that they are statstcally sgnfcant even for a sample sze as short as 3 months. Acknowledgments. We thank Bruce Barkstrom for stmulatng our curosty concernng the possble relatonshp between feld of vew sze and the ansotropy of the observed radances. We also thank Bruce Welck and Rchard Green for ther nsght and numerous valuable suggestons. Ths work was supported n part by NASA ERBE NAS and by NASA CERES NAG-l References Welck, B.A., and R.N. Green, Cloud dentfcaton for ERBE radatve flux retreval, J. Appl. Meteorol., 28, , Ye, Q., and J.A. Coakley Jr., Bases n Earth radaton budget observatons, 1, Effects of scanner spatal resoluton on the observed ansotropy. J. Geophys. Res., ths ssue. ' J'.A. C0akley Jr. (co.tespondng author), College of Oceanc and Atmosplerc Scences, Oregon State Unvcmty, Corvalls, OR Q. Ye, CIRES, Unversty of Colorado; Boulder, CO (Receved December 7, 1994; revsed December ' accepted Aprl 1, 1996.)