Determination of tropical radiation budget at high time and space resolution by combination of geostationary and polar satellite measurements.  Steven Dewitte y Natividad Manalo-Smith z Dominique Crommelynck x Jan Cornelis { July 14, 1996 Abstract

ERB(Earth Radiation Budget) data describe how the earth gains engergy, due to absorbed solar radiation ux, or looses energy, due to emitted thermal radiation ux. Since the atmospheric energetics is described by dierential equations with derivatives in space and in time, high resolution spatial and temporal ERB data should be available. ERB data derived from satellite measurements exist (ERBE(Earth Radiation Budget Experiment) [8] , ScaRaB(Scanning radiometer for Radiation Budget measurements) [9], planned: CERES(Cloud and the Earth's Radiant Energy System)) but the space/time resolution of these data is insucient to use them directly as input to atmospheric models. The purpose of this work is to combine these data with the high space/time resolution images of geostationary meteorological satellites (such as METEOSAT or the future MSG(Meteosat Second Genereation) carrying SEVIRI(Spinning Enhanced Visible and Infrared Imager) and GERB(Geostationary Earth Radiation Budget)), in order to obtain ERB data with much higher spatial  presented at 'International Conference on Tropical Climatology, Meteorology and Hydrology in memoriam F. Bultot (1924-1995), May 22-24, 1996, Brussels, Belgium y S. Dewitte is with the Royal Meteorological Institute of Belgium, Department of Aerology, Ringlaan 3, B-1180 Brussels, Belgium. Tel.: ++32-2-3730613, fax: ++32-2-3746788, email: [email protected] and with the Vrije Universiteit Brussel, Department of Electronics, Pleinlaan 2, B-1050 Brussels, Belgium. z N. Manalo-Smith is with Analytical Services and Materials, Hampton, VA 23666, Tel.: 804-827-4630, email: [email protected] x D. Crommelynck is with the Royal Meteorological Institute of Belgium, Department of Aerology, Ringlaan 3, B-1180 Brussels, Belgium. Tel.: ++32-2-3730600, fax: ++32-23746788, email: [email protected] { J. Cornelis is with the Vrije Universiteit Brussel, Department of Electronics, Pleinlaan 2, B-1050 Brussels, Belgium. Tel.: ++32-2-6292930, fax: ++32-2-6292883, email: [email protected]

1

and temporal resolution. As a testcase, we derive this data from November 1986 METEOSAT images and we compare the results with those of the ERBE instruments on the polar satellites NOAA 9 and NOAA 10. We obtain dierences close to the minimum ones obtainable according to theoretical simulations.

1 Introduction Weather prediction is based on a combination of the (partial) knowledge of processes that occur within the earth's atmosphere and on its surface, and of the (partial) measurement of the state of the earth's atmosphere and surface. In Numerical Weather Prediction (NWP) these two are combined in a Kalman lter like way [3], [4]. The knowledge is present as the internal equations of a numerical model of the atmosphere. The measurements are various kinds of meteorological observations, mainly ground based and satellite based, which are assimilated in the internal variables of the numerical model, i.e. the internal variables are updated to correspond better to the measurements. The assimilation of the measurements brings the internal variables in correspondence with reality, if the number of available independent measurements is suciently high, if they are accurate enough, and if they are suciently correlated with the internal variables. A large amount of measurements with uniform quality and high information content can be obtained using satellite data [5]. Currently mainly cloud motion winds and IR soundings are used as satellite data for NWP. A third possible candidate would be radiation budget data at high space/time resolution (at least synoptic) , whose great potential interest has been shown in [6], [7]. Radiation budget data at the top of the atmosphere derived from polar satellite measurements exist (ERBE(Earth Radiation Budget Experiment) [8] , ScaRaB(Scanning radiometer for Radiation Budget measurements) [9], planned: CERES(Cloud and the Earth's Radiant Energy System)) but the time resolution of these data is insucient1 to obtain a synoptic time sampling. A much better time resolution can be obtained with geostationary satellites (30 minutes for the current METEOSAT instruments, fteen minutes for the planned MSG(METEOSAT Second Generation) SEVIRI(Spinning Enhanced Visible and InfraRed Imager) instrument, ve minutes for the planned MSG GERB(Geostationary Earth Radiation Budget) instrument). The possibility to derive the radiation budget terms from geostationary satellite instruments has been demonstrated in [10], [11] (METEOSAT-ERB longwave term), [12] (METEOSAT-ERB shortwave term), [13], [14] (GOES longwave term) and [13], [15] (GOES shortwave term). In this paper we extend this work to METEOSAT-ERB derived radiation budget term estimates at the highest possible time/space resolution (time resolution 30 minutes, spatial resolution: longwave: 5km at nadir; shortwave: 2.5 km at nadir) and we measure an ex1

Basically a polar satellite ies only two times a day over a given point on earth.

2

perimental upper bound of the accuracy obtained at ERBE pixel level (size  50kmx50km at nadir) by intercomparison of the ERBE and METEOSAT-ERB estimates of the radiation budget terms over the ERBE pixel footprint. In section II we de ne the terms of the radiation budget of the earth, and of the satellite radiances that are used to estimate them. In section III we describe the estimation of the radiation budget terms from full resolution METEOSAT data. In section IV we compare the METEOSAT-ERB estimates with the ERBE estimates over ERBE footprints. We formulate our conclusions in section V.

2 De nitions The radiation from the sun is the prime energy source for the processes that occur within the earth's atmosphere and on its surface. Oberved from earth, the sun can be considered in good approximation as a point source. The solar constant SSOLAR is de ned as the solar ux at the mean distance earth-sun (=1 astromical unit) for overhead sun. The value of SSOLAR is varying with a chaotic eleven year cycle between 1363W=m2 and 1369W=m2 [16]. In good approximation it can be considered as a constant with value 1366W=m2 At a given point on earth we have a solar ux at the top of the atmosphere FSOLAR

SSOLAR cos( ) FSOLAR = ( rearth sun ?sun )2 a:u:

1

(1)

where rearth?sun is the earth-sun distance and sun is the solar-zenith angle (angle between the direction of the sun and the vertical, depending on position on the earth, on date of the year and on time of the day). Part of the solar ux FSOLAR is re ected back to space as the re ected shortwave ux at the top of the atmosphere FSW

R 1 d R 2 d R =2 dLsolar (; )sin()cos() = (2) FSW R05m 0R 2 0R =2  solar  0 d 0 d 0 dL (; )sin()cos() where  and  are the angular coordinates of a spherical coordinate system with z-axis corresponding to the local vertical direction. Lsolar is the spectral  radiance from solar origin re ected at the top of the earth's atmosphere. As a result we have a net shortwave ux entering the top of the atmosphere FSW;IN FSW;IN = FSOLAR ? FSW

(3) If this ux were the only non zero one at the top of the atmosphere, the mean temperature of the earth would be rising constantly. This is not the case, because all the components of the earth and its atmosphere have a certain temperature, and therefore emit some thermal radiation which results in an emitted longwave ux at the top of the atmosphere FLW . 3

R 1 d R 2 d R =2 dLearth(; )sin()cos() = FLW R050m 0 R 2 0 R =2  earth (4)  5m d 0 d 0 dL (; )sin()cos() where Learth is the spectral radiance from earthern origin at the top of  the earth's atmosphere. Lsolar nds its origin in the radiation emitted by the  sun, approximated roughly by black-body radiation with T  5000K , Learth  nds its origin in the radiation emitted by the earth, approximated roughly by black-body radiation with T  300K . Fortunately for remote sensing purposes, these temperatures are suciently dierent to have a good spectral separaearth (spectral range tion between Lsolar  (spectral range  [0m ? 5m]) and L  [5m ? 50m]). The combination of FSW;IN and FLW gives the local instantanuous radiation budget at the top of the atmosphere RB RB = FSW;IN ? FLW

(5) which can be positive (local instantaneous heating) or negative (local instantaneous cooling). The long term global annual mean of RB has to be zero to have a climate equilibrium. Its components are shown in gure 1, according to http://asd-www.larc.nasa.gov/erbe/ASDerbe.html. If we want to estimate the terms of RB from satellite instrument measurements, all we have available in general are a number of ltered radiances

Lsat;channel

Lsat;channel =

Z1 0

earth dchannel ()(Lsolar  + L )(sat ; sat )

(6)

where sat and sat are the angular coordinates with which the satellite is seen from earth and where channel () is the spectral response of the satellite instrument channel under consideration. To estimate FSW or FLW from ltered radiances two conversion are necessary, an angular one and a spectral one. We can either start with the angular conversion, or we can start with the spectral conversion. If we start with the angular conversion we get as (ideal) intermediate results the ltered uxes Fchannel

Fchannel =

Z1 Z 0

d



2 0

d

Z =

2

0

earth dchannel ()(Lsolar  +L )(; )sin()cos()

(7) If we start with the spectral conversion we get as (ideal) intermediate results the un ltered radiances Lsat;type

Lsat;type =

Z1 0

dLtype  (sat ; sat )

where type equals earth or solar. 4

(8)

For ERBE we had polar satellite instruments with three available channels with spectral responses TOTAL () (spectral range  [0:3m ? 50m]), SW () (spectral range  [0:3m ? 5m]) and LW () (spectral range  [5m ? 50m]). From these ltered radiances the ERBE secondary processing derives FLW;ERBE as estimate of FLW and FSW;ERBE as estimate of FSW . For METEOSAT we have a geostationary satellite instrument with three available channels: the visible channel with spectral response V IS () (spectral range  [0:4m ? 1:1m]) and digital count CV IS , the watervapour channel with spectral response WV () (spectral range  [5:7m ? 7:1m]) and digital count CWV , and the infrared window channel with spectral response IR () (spectral range  [10:5m ? 12:5m]) and digital count CIR .

3 Estimation of radiation budget terms from METEOSAT measurements

3.1 Longwave ux

For the estimation of FLW from METEOSAT data we use the method of [10], [11] and apply it to full resolution IR and WV data. We repeat the dierent steps here brie y. The digital counts CIR ,CWV are converted to calibrated ltered radiances LIR ,LWV using the linear calibration relations

LIR = kIR (CIR ? CIR;0 )

(9)

LWV = kWV (CWV ? CWV;0 ) (10) where the calibration coecients kIR , CIR;0 , kWV and CWV;0 are given in

the radiometer calibration reports. From the ltered radiances LIR ,LWV the ltered uxes FIR ,FWV are estimated as

FIR = a(sat )LIR (sat ) + b(sat )

(11)

FWV = c(sat )LWV (sat ) + d(sat )

(12)

a(sat ) = k1 + k2 (sec(sat ) ? 1) + k3 (sec(sat ) ? 1)2

(13)

b(sat ) = k4 + k5 (sec(sat ) ? 1) + k6 (sec(sat ) ? 1)2

(14)

c(sat ) = l1 + l2 (sec(sat ) ? 1) + l3 (sec(sat ) ? 1)2

(15)

5

d(sat ) = l4 + l5 (sec(sat ) ? 1) + l6 (sec(sat ) ? 1)2 (16) where the coecients ki , li , i = 1; : : : ; 6 are obtained as best t coecients

from radiative transfer simulations [10], [11]. From the ltered uxes FIR ,FWV the emitted longwave ux FLW is estimated as

FLW = 0 +

X Fi + X Fi 3

3

i=1

i IR

i=1

i WV

(17)

where the coecients i , i = 0; : : : ; 3 i , i = 1; : : : ; 3 are also obtained as best t coecients from radiative transfer simulations [10], [11]. Figure 2 shows the result of this procedure applied to the METEOSAT IR and WV data of 08/11/1986(dd/mm/yyyy) timeslot 18 (nominally 08:30 GMT08:55 GMT).

3.2 Re ected shortwave ux

Calibration of METEOSAT 2 visible channel The digital counts CV IS

are converted to calibrated ltered radiances LV IS using the linear calibration relation

LV IS = kV IS (CV IS ? CV IS;0)

(18) where CV IS;0 =lowest count during night=2 and where kV IS is based on the factors nominal!absolute (date) kV IS;ISCCP (date) = kVnominal (19) IS kV IS;ISCCP !absolute (date) are measured every reported in [17]. The factors kVnominal IS;ISCCP

three months by comparison of METEOSAT 2 visible channel measurements and NOAA 9 AVHHR channel 1 (spectral range  [0:55m ? 0:7m]) measurements with coinciding viewing geometry and time of acquisition. The factors !absolute (date) from 04/1985 to 04/1987, calculated from [17] are given kVnominal IS;ISCCP in table 1. 04= 07= 10= 01= 04= 07= 10= 01= 04= 1985 1985 1985 1986 1986 1986 1986 1987 1987 1:1886 1:1842 1:1702 1:1275 1:1731 1:1897 1:2397 1:1728 1:1946 !absolute (date) from 04/1985 to 04/1987 with time Table 1: Factors kVnominal IS;ISCCP step of three months.

6

!absolute (date) are almost constant over We notice that the factors kVnominal IS;ISCCP this two-year period, with a downward exception for 01/1986, and an upward exception for 10/1986. Using the four months (09/1986-12/1986) of METEOSAT ISCCP B2 data we have available, we are able to identify a METEOSAT timing problem which !absolute (10=1986) and explains coincides with the exceptionally high kVnominal IS;ISCCP it. For selected days we estimated the true time of acquisition of the METEOSAT images from the location of the day/night separation in the visible images and compared this to the nominal time of acquisition. The dierences are given in table 2.

02=09=1986 01=10=1986 08=11=1986 03=12=1986 00 100 150 100 Table 2: Dierence in estimated true acquisition time and nominal acquisition time of METEOSAT images for selected days. We get an exceptionally high dierence of 15' for 08/11/1986. If this is not !absolute (10=1986), NOAA 9 taken into account in the measurement of kVnominal IS;ISCCP measurements taken around 14:30 local time (this is the equator crossing time of the sun synchronuous NOAA 9) are believed to be corresponding to METEOSAT 2 measurements taken around 14:45 local time. Because of the lower illumination at 14:45 compared to 14:30, the obtained METEOSAT calibration factor is to high with a factor approximatively equal to

cos(2 224:5 ) 0:79 = = 1:055 cos(2 224:75 ) 0:75 !absolute (10=1986) by this factor we nd If we correct kVnominal IS;ISCCP

(20)

!absolute (10=1986))corrected = 1:2397=1:055 = 1:1750 (kVnominal (21) IS;ISCCP which is in the normal range of table 1. This leads us to the assumption that kV IS is a constant in time (i.e. the METEOSAT 2 visible channel doesn't seem to degrade in time). As best estimate we take the mean of the ISCCP measurements over the two year period.

kV IS =

P

04=87 date=04=85 kV IS;ISCCP (date) = k nominal 1:1807 V IS

9

(22)

Default spectral conversion The highest contribution to FSW is coming from the clouds in the atmosphere (albedos up to 80% compared to albedos up to 30% for clear-sky surfaces), and clouds can be modelled in good approximation 7

as grey absorbers, i.e. they do not modify the spectral distribution of the incoming solar radiation [18]. This justi es the conversion of LV IS to LSW as a simple linear conversion with a constant conversion factor equal to the one for the reference solar spectrum Lsolar;reference ().

LSW = R 1 0

R 1 dLsolar;reference()

0 L dLsolar;reference ()V IS () V IS

(23)

where V IS () is the METEOSAT VIS channel spectral sensitivity. We use the values calculated in [17].

Angular conversion Scene separation To estimate FSW from LSW it is necessary to know the angular distribution of LSW , which depends strongly on the scene type and the cloud cover [19]. This knowledge is present for discrete cloud cover classes in the analytical re ection functions of [20]. It is necessary to do a scene iden cation/separation in order to use the correct re ection function to estimate FSW from LSW . We decompose LSW as LSW = (1 ? f )  LSW;scene?lower + f  LSW;scene?upper (24) where LSW;scene?lower and LSW;scene?upper are radiances for discrete cloud cover [20] scenes and where f is a scene fraction with 0  f  1. For this

scene separation previously obtained clear-sky/cloud separation [1], [2] results are used.

Single scene angular conversion For the discrete cloud cover scenes the

uxes are related to the radiances by

FSW;scene = BDRFLSW;scene scene (sun ; sat ; )

(25)

where BDRFscene is the analytical bidirectional re ection function for the particular scene given in [20]. The analytic expression for BDRF was formulated by applying a t to the ERBE operational bidirectional re ectance model tabulations by use of parameters in the expression. For every scene, there exists a set of model coecients applicable to any combination of viewing conditions. The forms of these expressions are based on theoretical considerations. The models are smooth in terms of directional angles and satisfy reciprocity, i.e., they are invariant with respect to the interchange of the incoming and outgoing directions. 8

Scene recomposition The ux FSW is recomposed from the dicrete cloud

cover uxes using

FSW = (1 ? f )  FSW;scene?lower + f  FSW;scene?upper

(26) where f ,scene-lower and scene-upper are the same as obtained in the radiance decomposition of equation 24.

3.3 Net shortwave ux entering the top of the atmosphere

For every METEOSAT pixel FSOLAR is calculated from equation 1. FSW;IN is calculated by equation 3 from FSOLAR and FSW . Figure 3 shows the result for 08/11/1986 timeslot 18.

3.4 Radiation budget at the top of the atmosphere

For every METEOSAT pixel RB is calculated by equation 5 from FSW;IN and FLW . Figure 4 shows the result for 08/11/1986 timeslot 18.

4 Comparison of METEOSAT-ERB and ERBE estimations of radiation budget terms In this section we compare the obtained METEOSAT-ERB uxes with those obtained by ERBE. The comparison is done at ERBE pixel level. We make the comparison of METEOSAT-ERB with the ERBE instrument on board the AM satellite NOAA 10, and with the ERBE instrument on board the PM satellite NOAA 9. Figure 5 shows the METEOSAT visible image at timeslot 18 (nominally 08:30 GMT-08:55 GMT) and the corresponding NOAA 10 shortwave ux estimations. Figure 6 shows the METEOSAT visible image at timeslot 35 (nominally 17:00 GMT-17:25 GMT) and the corresponding NOAA 9 shortwave ux estimations. The METEOSAT timeslots 18 and 35 are those with the best time correspondence with the NOAA 10 and NOAA 9 measurements respectively.

4.1 Convolution of METEOSAT-ERB estimation with ERBE point spread function

To obtain METEOSAT-ERB ux estimates at ERBE pixel level FERBE(METEOSAT ?ERB) from the full resolution METEOSAT-ERB uxes FMETEOSAT ?ERB (FLW or FSW obtained in the previous section), a convolution is done with the ERBE point spread function [21] PSF (; )

9

R

)FMETEOSAT ?ERB dd FERBE(METEOSAT ?ERB) = PSF (;R PSF (; )dd

(27)

where  is the ERBE cross-track scan angle and is the ERBE along-track scan angle.

4.2 Comparison of longwave ux estimations

Figure 7 shows the METEOSAT-ERB estimation versus the ERBE estimation of the emitted longwave ux over the ERBE footprints in the northsouth region of gure 5. In the case of perfect agreement between the estimated uxes this gure would show a perfect y = x relation. In order to quantify the deviation from this perfect agreement case a linear t y = ax is performed. The best tting line is shown in gure 7. q Table 3 shows the slopes a and the RMS dierences (y ? ax)2 obtained for the four regions northnorth, southnorth, northsouth and southsouth (see gure 5) for the comparison of METEOSAT timeslot 18 (nominally 08:30 GMT-08:55 GMT) compared to NOAA 10 measurements (see gure 5).

SLOPE RMS %RMS

northnorth southnorth northsouth southsouth 0:98 0:95 0:98 0:95 8:1W=m2 5:7W=m2 6:7W=m2 8:8W=m2 3:3% 2:0% 2:7% 3:4%

Table 3: Slope of best linear t through origin and residual RMS dierence between METEOSAT-ERB and NOAA 10 ERBE emitted longwave ux estimations. The procentual RMS dierence is given relative to the RMS values over the given region of the estimated ux.

q

Table 4 shows the slopes a and the RMS dierences (y ? ax)2 obtained for the four regions northnorth, southnorth, northsouth and southsouth (see gure 5) for the comparison of METEOSAT timeslot 34 (nominally 16:30 GMT16:55 GMT) compared to NOAA 9 measurements (see gure 6). METEOSAT timeslot 34 is used instead of timeslot 35 (which corresponds better in time with the NOAA 9 measurements) because the METEOSAT IR image is not available for timeslot 35.

4.3 Comparison of shortwave ux estimations

Figure 8 shows the METEOSAT-ERB estimation versus the ERBE estimation of the re ected shortwave ux over the ERBE footprints in the northsouth region 10

SLOPE RMS %RMS

northnorth southnorth northsouth southsouth 0:96 1:00 1:00 1:00 8:7W=m2 9:9W=m2 13:4W=m2 12:0W=m2 3:7% 3:7% 5:8% 5:1%

Table 4: Slope of best linear t through origin and residual RMS dierence between METEOSAT-ERB and NOAA 9 ERBE emitted longwave ux estimations. The procentual RMS dierence is given relative to the RMS values over the given region of the estimated ux. of gure 5. As for the longwave case in the case of perfect agreement between the estimated uxes this gure would show a perfect y = x relation. In order to quantify the deviation from this perfect agreement case a linear t y = ax is performed. The best tting line is shown in gure 8. For the comparison of dierent re ected shortwave ux estimates we expect a high sensitivity to time and space correspondence of these estimates, due to the high spatial (broken clouds) and temporal (solar-zenith angle) variation of these uxes. To investigate quantitatively the importance of an accurate spatial correspondence we remake the comparison of gure 8 with and without convolution of the METEOSAT-ERB uxes with the point spread function associated to the ERBE pixel. In the case without convolution with the ERBE PSF, the METEOSAT-ERB pixel lying in the center of the ERBE footprint is taken as METEOSAT-ERB estimate for the whole footprint. To investigate quantitatively the importance of an accurate time correspondence we remake the comparison of gure 8 at two times, once at the METEOSAT timeslot nearest to the ERBE acquisition and once at one METEOSAT timeslot earlier (half an hour earlier). Table 5 shows the slopes a obtained in function of the time correspondence (best correspondence for slot 18) and of the spatial correpondence method (PSF convolution or central pixel selection).

slot17 slot18 SLOPE central pixel 0:96 1:17 PSF convolution 0:96 1:16 Table 5: Slope of best linear t through origin for METEOSAT-ERB and NOAA 10 ERBE re ected shortwave ux estimations in function of METEOSAT timeslot and of spatial correspondence method.

q

Table 6 shows the RMS dierences (y ? ax)2 obtained in function of the time correspondence (best correspondence for slot 18) and of the spatial corre11

pondence method (PSF convolution or central pixel selection).

slot17 slot18 RMS central pixel 62:1W=m2(30:6%) 63:8W=m2(31:4%) PSF convolution 38:3W=m2(18:7%) 31:5W=m2(15:4%) Table 6: Residual RMS dierence after best linear t through origin for METEOSAT-ERB and NOAA 10 ERBE re ected shortwave ux estimations in function of METEOSAT timeslot and of spatial correspondence method. The procentual RMS dierence is given relative to the RMS values over the given region of the estimated ux.

q

Table 7 shows the RMS dierences (y ? ax)2 relative to a, i.e.

q

((y=a) ? x)2 .

RMS SLOPE

slot17 slot18 central pixel 64:7W=m2(30:6%) 58:4W=m2(31:4%) PSF convolution 39:9W=m2(18:7%) 27:2W=m2(15:4%) Table 7: Relative residual RMS dierence after best linear t through origin for METEOSAT-ERB and NOAA 10 ERBE re ected shortwave ux estimations in function of METEOSAT timeslot and of spatial correspondence method. The procentual RMS dierence is given relative to the RMS values over the given region of the estimated ux.

q

Table 8 shows the slopes a and the RMS dierences (y ? ax)2 obtained for the four regions northnorth, southnorth, northsouth and southsouth (see gure 5) for the comparison of METEOSAT timeslot 18 (nominally 08:30 GMT-08:55 GMT) compared to NOAA 10 measurements (see gure 5).

northnorth southnorth northsouth southsouth SLOPE 1:40 0:99 1:16 0:93 RMS 34:5W=m2 25:3W=m2 31:5W=m2 25:3W=m2 %RMS 27:9% 14:5% 15:4% 11:5% Table 8: Slope of best linear t through origin and residual RMS dierence between METEOSAT-ERB and NOAA 10 ERBE re ected shortwave ux estimations. The procentual RMS dierence is given relative to the RMS values over the given region of the estimated ux.

q

Table 9 shows the slopes a and the RMS dierences (y ? ax)2 obtained for the four regions northnorth, southnorth, northsouth and southsouth (see gure 12

5) for the comparison of METEOSAT timeslot 35 (nominally 17:00 GMT-17:25 GMT) compared to NOAA 9 measurements (see gure 6).

northnorth southnorth northsouth southsouth SLOPE 1:17 1:02 1:14 1:09 RMS 35:5W=m2 42:0W=m2 36:1W=m2 43:3W=m2 %RMS 15:9% 20:9% 13:3% 15:5% Table 9: Slope of best linear t through origin and residual RMS dierence between METEOSAT-ERB and NOAA 9 ERBE re ected shortwave ux estimations. The procentual RMS dierence is given relative to the RMS values over the given region of the estimated ux.

5 Conclusions In the study presented in this paper we derived the radiation budget term estimates at the highest possible time/space resolution from METEOSAT images and compared the results (more speci cally the emitted longwave ux and the re ected shortwave ux), at the best possible temporal and spatial correspondence, with their ERBE estimates. We made the comparison for four separate regions and for two ERBE instruments (more speci cally the one carried by NOAA 10 and the one carried by NOAA 9). For both radiation budget terms (the emitted longwave ux and the re ected shortwave ux), for both ERBE instruments and for each region we measurement the systematic disagreement (slope of best linear t dierent from one) and the stochastic disagreement (RMS dierence around best linear t) of METEOSAT-ERB compared to ERBE. For the longwave case we nd slopes very close to one for all regions, and RMS dierences in the range 5-9 W/m2 for the NOAA 10 case, and in the range 8-14W/m2 for the NOAA 9 case. The worse results for NOAA 9 compared to NOAA 10 are most likely explained by the worse time correspondence METEOSAT-ERB/ERBE. For the shortwave case we nd slopes rather close to one (neglectable compared to the RMS dierence) for the southnorth and the southsouth region and substantially higher (not neglectable compared to the RMS dierence) for the northnorth and the northsouth region. These last regions correspond exactly to the regions where the BDRF's to be applied to the METEOSAT-ERB radiances (see equation 25) are the largest, namely the sunglint region (northsouth region in gures 5 and 6) and the region close to the day/night separation (northnorth region in gures 5 and 6). Very likely this indicates that the applied BDRF's are not accurate enough or that they are not applied accurately enough to remove the largest anisotropies completely. 13

If we believe the ERBE shortwave ux estimates to be regionally unbiased, we can correct the METEOSAT-ERB shortwave ux estimates by dividing them by the tted slopes. After this correction we nd RMS dierences METEOSATERB/ERBE in the range 24-28W/m2 for the NOAA 10 case, and in the range 30-42W/m2 for the NOAA 9 case. The worse results for NOAA 9 might be explained by the METEOSAT viewing-zenith angle, which is on the average higher for gure 5 than for gure 6. We can compare our results with the theoretical uncertainties on the radiation budget parameters found in [22], [23], [19]. If we believe [22], [19] that the angular uncertainty is the dominant source of uncertainty in a single satellite estimation of a radiation budget term (FLW or FSW ), the uncertainties on ERBE and METEOSAT-ERB estimates should be essentially the same. If moreover we assume that the errors on these two estimates are stochastically independent, then the RMS dierences we nd between METEOSAT-ERB p and ERBE estimates on FLW or FSW at ERBE footprint level should be 2 higher than the single satellite estimation uncertainties found in [23]. This leads us to estimated single satellite uncertainties of 4-7W/m2 on FLW and of 17-20W/m2 on FSW (for suciently low viewing-zenith and solar-zenith angles). The uncertainties found in [23] are 3.8W/m2 on FLW , and ranging from 5-11W/m2 for clear-sky ocean scenes to 16-26W/m2 for overcast scenes on FSW . We are rather close to this, which seems to indicate that the uncertainties on our METEOSATERB derived radiation budget term estimates are coming close to the minimum uncertainties possible using the currently known anisotropy models.

References [1] B.R. Barkstrom et. al, Earth Radiation Budget Experiment (ERBE) archival and April 1985 results, Bulletin of the American Meteorological Society, vol. 70, pp 1254-1262, 1989. [2] J.L. Monge, R.S. Kandel, L.A. Pakhomov, V.I. Adasko, , Proc. New Developments and Applications in Optical Radiometry III. Davos, Metrologia 28, pp 261-264, 1991. [3] L. Bengston, Medium-Range Forecasting{The experience of ECMWF, Bulletin of the American Meteorological Society, vol. 66 no.9, pp.1133-1146, September 1985. [4] E. Kalnay et al., The NMC/NCAR 40-Year Reanalysis Project, accepted for publication by Bulletin of the American Meteorological Society, April 1995. [5] P. K. Rao et. al, Weather Satellites: Systems, Data and Environmental Applications, American Meteorological Society, Boston, 503 pp. 14

[6] J. Van Mieghem, Radiation data needed in dynamical meteorology, Archiv fur Meteorologie, Geophysik und Bioklimatologie, Serie A: Meteorologie und Geophysik, Band 10, 4. Heft 1958 [7] J. Van Mieghem, Pour une exploration synoptique du champ du rayonnement dans l'atmosphere, Archiv fur Meteorologie, Geophysik und Bioklimatologie, Serie A: Meteorologie und Geophysik, Band 13, 2. Heft 1962 [8] J. Schmetz and Q. Liu, Outgoing longwave radiation and its diurnal variation at regional scales derived from METEOSAT, Journal of Geophysical Research, vol. 93 no. D9, pp. 11192-11204, September 1988. [9] J. Schmetz, M. Mhita and L. van de Berg, METEOSAT observations of longwave cloud-radiative forcing for April 1985, Journal of Climate, vol. 3, pp. 784-791, July 1990. [10] M. Vesperini and Y. Fouquart, Determination of Broad-Band Shortwave Fluxes from th4e Meteosat Visible Channel by Comparison to ERBE, Beitr. Phys. Atmosph., vol. 67 no. 2, pp. 121-132, 1994. [11] P. Minnis and E.F. Harrison, Diurnal Variability of Regional Cloud and Clear-Sky Radiative Parameters Derived from GOES Data. PartIII: November 1978 Radiative Parameters, Journal of Climate and Applied Meteorology, vol. 23 no. 7, pp. 1032-1051, July 1984. [12] P. Minnis, D.F. Young and E.F. Harrison, Examination of the Relationship between Outgoing Infrared Window and Total Longwave Fluxes using Satellite Data, Journal of Climate, vol. 4 no. 11, pp. 1114-1133, November 1991. [13] D. Doelling et. al, On the role of satellite-measured narrowband radiances for computing the earths's radiation balance, Seventh Conference on Atmospheric Radiation, American Meteorological Society, San Francisco, July 23-27 1990. [14] D. Crommelynck and S. Dewitte, Solar constant temporal and frequency characteristics, submitted for publication to Solar Physics, 1996. [15] W.B. Rossow, Y. Desormeaux, C.L. Brest, A. Walker, International Satellite Cloud Climatology Project (ISCCP), WMO/TD-No. 520, WCRP-77, World Meteorological Organization, October 1992. [16] D. Crommelynck and A. Jouko, Estimation of the spectral distribution using a simple algorithm, IRS' 92: Current Problems in Atmospheric Radaition, Talin, Estonia, August 1992, Deepak Publishing ISBN0-93719428-X (1993). 15

[17] B. Wielicki et. al, Mission to Planet Earth: Role of Clouds and Radiation in Climate, submitted for publication to Bulletin of the American Meteorological Society, 1995. [18] N. Manalo-Smith, W.F. Staylor and G.L. Smith, Analytic forms of bidirectional re ectance functions, Seventh Conference on Satellite Meteorology and Oceanography, Monterey, California, June 6-10, 1994. [19] S. Dewitte, E. Nyssen, D. Crommelynck, J. Cornelis , Multi stage analysis of METEOSAT images , European Symposium on Satellite Remote Sensing II, SPIE Vol. 2579, pp. 170-181, November 1995 [20] S. Dewitte, J. Cornelis, D. Crommelynck, A generic image processing tool : topdown Bayesian image analysis, presented at NATO AGARD conference Remote Sensing: A valuable source of information, Toulouse, April 1996 [21] G.L. Smith, Eects of time response on point spread function of a scanning radiometer, Applied Optics, 33, 7031-7037, Oct. 1994. [22] C. Standfuss, H.-D. Hollweg and H. Grassl, The impact of a radiation budget scanner aboard meteosat second generation on the accuracy of regional radiation budget parameters, ESA nal report, ESTEC purchase order no. 125144, 1993. [23] F.J. Diekmann, Fehler in der Szeneidenti kation aus SatellitenDaten Auswirkungen auf die Bestimmung von Strahlungshaushaltsparameters, Mitteilungen des Inst. f. Geophysik und Meteorologie d. Universitat zu Koln, Heft 59.

16

Figure 1: The components of the Earth's Radiation Budget, obtained from http://asd-www.larc.nasa.gov/erbe/ASDerbe.html.

17

0 W/m2

683 W/m2

Figure 2: Longwave ux estimated from METEOSAT IR and WV data, for 08/11/1986 timeslot 18 (nominally 08:30 GMT-08:55 GMT). 18

0 W/m2

1366 W/m2

Figure 3: Net shortwave ux estimated from METEOSAT VIS data, for 08/11/1986 timeslot 18 (nominally 08:30 GMT-08:55 GMT). 19

-683 W/m2

0 W/m2

683 W/m2

Figure 4: Radiation budget estimated from METEOSAT IR,WV and VIS data, for 08/11/1986 timeslot 18 (nominally 08:30 GMT-08:55 GMT). 20

northnorth

southnorth

northsouth

southsouth Figure 5: Left: METEOSAT visible image for timeslot 18 (nominally 08:30 GMT-08:55 GMT) of 08/11/1986, right: corresponding shortwave ux estimations from NOAA 10 ERBE instrument.

21

Figure 6: Left: METEOSAT visible image for timeslot 35 (nominally 17:00 GMT-17:25 GMT) of 08/11/1986, right: corresponding shortwave ux estimations from NOAA 9 ERBE instrument.

22

300

METEOSAT emitted longwave flux (W/m2)

280 260 240 220 200 180 160 140 120 100 140

160

180

200 220 240 260 ERBE emitted longwave flux (W/m2)

280

Figure 7: Comparison of ERBE (NOAA 10) and METEOSAT-ERB estimations of longwave ux(W=m2 ) over ERBE footprints, for METEOSAT northsouth region, for 08/11/1986 METEOSAT timeslot 18 (nominally 08:30 GMT-08:55 GMT).

23

300

320

METEOSAT reflected shortwave flux (W/m2)

500

400

300

200

100

0 0

50

100

150 200 250 300 ERBE reflected shortwave flux (W/m2)

Figure 8: Comparison of ERBE (NOAA 10) and METEOSAT-ERB estimations of shortwave ux(W=m2 ) over ERBE footprints, for METEOSAT northsouth region, for 08/11/1986 METEOSAT timeslot 18 (nominally 08:30 GMT-08:55 GMT).

24

350

400