ATBD for EUMETSAT Pre- Operational GSICS Inter-Calibration of Meteosat-IASI

Transcription

1 ATBD for EUMETSAT Pre- Operatonal GSICS Inter-Calbraton of Doc.o. : Issue : va Date : ovember 0 WBS : EUMETSAT Eumetsat-Allee, D-6495 Darmstadt, Germany Tel: Fax: EUMETSAT The copyrght of ths document s the property of EUMETSAT.

2 Document Change Record Issue / Revson Date DC. o Summary of Changes v Orgnal based on Pre-Operaton ATBD for SEVIRI- IASI, extended to nclude specfc detals for Rapd Scan Servce MSG data and Meteosat Frst Generaton. Incorporatng the followng: EUM/MET/TE/09/0775 SEVIRI-IASI ATBD EUM/MET/TE//09 MVIRI-IASI ATBD va 0-- Updated n preparaton for MSG-3 Also lmted ncdence angle to <35 n.c.v. Page of 47

3 Table of Contents 0 Introducton EUMETSAT s Inter-Calbraton Algorthm Subsettng... 6.a. Select Orbt Fnd Collocatons... 9.a. Collocaton n Space... 0.b. Concurrent n Tme....c. Algnment n Vewng Geometry... 3.d. Pre-Select Channels... 5.e. Plot Collocaton Map Transform Data a. Convert Radances b. Spectral Matchng c. Spatal Matchng... 3.d. Vewng Geometry Matchng e. Temporal Matchng Flterng a. Unformty Test b. Outler Rejecton c. Auxlary Datasets Montorng a. Defne Standard Radances (Offlne) b. Regresson of Most Recent Results c. Bas Calculaton d. Consstency Test e. Trend Calculaton f. Generate Plots for GSICS Bas Montorng Flow Summary of Steps 5 AD 6 for SEVIRI-IASI GSICS Correcton a. Defne Smoothng Perod (Offlne) b. Calculate Coeffcents for GSICS ear-real-tme Correcton c. Calculate Coeffcents for GSICS Re-Analyss Correcton d. Re-Calculate Calbraton Coeffcents Page 3 of 47

4 0 ITRODUCTIO The Global Space-based Inter-Calbraton System (GSICS) ams to nter-calbrate a dverse range of satellte nstruments to produce correctons ensurng ther data are consstent, allowng them to be used to produce globally homogeneous products for envronmental montorng. Although these nstruments operate on dfferent technologes for dfferent applcatons, ther nter-calbraton can be based on common prncples: Observatons are collocated, transformed, compared and analysed to produce calbraton correcton functons, transformng the observatons to common references. To ensure the maxmum consstency and traceablty, t s desrable to base all the nter-calbraton algorthms on common prncples, followng a herarchcal approach, descrbed here. Ths algorthm s defned as a seres of generc steps revsed at the GSICS Data Workng Group web meetng (ovember 009): ) Subsettng ) Collocatng 3) Transformng 4) Flterng 5) Montorng 6) Correctng Each step comprses a number of dscrete components, outlned n the contents. Each component can be defned n a herarchcal way, startng from purposes, whch apply to all nter-calbratons, buldng up to mplementaton detals for specfc nstrument pars:. Descrbe the purpose of each component n ths generc data flow.. Provde dfferent optons for how these may be mplemented n general.. Recommend procedures for the nter-calbraton class (e.g. GEO-LEO). v. Provde specfc detals for each nstrument par (e.g. SEVIRI-IASI). The mplementaton of the algorthm need only follow the overall logc so the components need not be executed strctly sequentally. For example, some parts may be performed teratvely, or multple components may be combned wthn a sngle loop n the code. 0. EUMETSAT s Inter-Calbraton Algorthm Ths document forms the Algorthm Theoretcal Bass Document (ATBD) for the ntercalbraton of the nfrared channels of SEVIRI on the Geostatonary (GEO) Meteosat Second Generaton and MVIRI on Meteosat Frst Generaton satelltes wth the Infrared Atmospherc Soundng Interferometer (IASI) on board LEO Metop satelltes. Ths document refers only to mplementaton verson submtted as a canddate Pre-Operatonal GSICS product, prevously descrbed as v0.4 n EUMETSAT [009]. Page 4 of 47

5 MO Level Data REF Level Data Orbtal Predcton. Subsettng Subset MO Data Collocaton Subset REF Data Archve ~ month Colloc. Crtera. Collocatng Collocated Data Archve ~ month SRFs, PSFs, Transformaton 3. Transformng Comparson Data Archve ~ year Masks, flags, 4. Flterng Analyss Analyss Data Archve ~ year 5. Montorng 6. Correctng 7. Dagnosng Bas Montorng MO Lvl Data Users Correcton Coeffs Products GSICS Correcton Reports Re-Cal Data Fgure : Dagram of generc data flow for nter-calbraton of montored (MO) nstrument wth respect to reference (REF) nstrument Page 5 of 47

6 . SUBSETTIG Acquston of raw satellte data s obvously a crtcal frst step n an nter-calbraton method based on comparng collocated observatons. To facltate the acquston of data for the purpose of nter-comparson of satellte nstruments, predcton of the tme and locaton of collocaton events s also mportant. MO Level Data REF Level Data Orbtal Predcton. Subsettng Subset MO Data Subset REF Data Archve ~ month Fgure : Step of Generc Data Flow, showng nputs and outputs. MO refers to the montored nstrument. REF refers to the reference nstrument. Page 6 of 47

7 .a. Select Orbt.a.. Purpose We frst perform a rough cut to reduce the data volume and only nclude relevant portons of the dataset (channels, area, tme, vewng geometry). The purpose s to select portons of data collected by the two nstruments that are lkely to produce collocatons. Ths s desrable because typcally less than 0.% of measurements are collocated. The processng tme s reduced substantally by excludng measurements unlkely to produce collocatons. Data s selected on a per-orbt or per-mage bass. To do ths, we need to know how often to do nter-calbraton whch s based on the observed rate of change and must be defned teratvely wth the results of the nter-calbraton process (see 5.f)..a.. General Optons The smplest, but neffcent approach s tral-and-error,.e., compare the tme and locaton of all pars of fles wthn a gven tme wndow..a.. Infrared GEO-LEO nter-satellte/nter-sensor Class For nter-calbratons between geostatonary and sun-synchronous satelltes, the orbts provde collocatons near the GEO Sub-Satellte Pont (SSP) wthn fxed tme wndows every day and nght. In ths case, we adopt the smple approach outlned above. We defne the GEO Feld of Regard (FoR) as an area close to the GEO Sub-Satellte Pont (SSP), whch s vewed by the GEO sensor wth a zenth angle less than a threshold. Wu [009] defned a threshold angular dstance from nadr of less than 60 based on geometrc consderatons, whch s the maxmum ncdence angle of most LEO sounders. Ths corresponds to ±5 n lattude and longtude from the GEO SSP. The GEO and LEO data s then subset to only nclude observatons wthn ths FoR wthn each nter-calbraton perod. Mathematcally, the GEO FoR s the collecton of locatons whose arc angle (angular dstance) to nadr s less than a threshold or, equvalently, the cosne of ths angle s larger than mn_cos_arc. We chose the threshold mn_cos_arc 0.5,.e., angular dstance less than 60 degree. Computatonally, wth known Earth coordnates of GEO nadr G (0, geo_nad_lon) and granule centre P (gra_ctr_lat, gra_ctr_lon) and approxmatng the Earth as beng sphercal, the arc angle between a LEO pxel and LEO nadr can be computed wth cosne theorem for a rght angle on a sphere (see Fgure 3): Equaton cos GP ( ) cos( gra _ ctr _ lat) cos( geo _ nad _ lon gra _ ctr _ lon) If the LEO pxel s outsde of GEO FoR, no collocaton s consdered possble. ote the arc angle GP on the left panel of Fgure 3, whch s the same as the angle GOP on the rght panel, s smaller than the angle SPZ (rght panel), the zenth angle of GEO from the pxel. Page 7 of 47

8 Ths means that the nstrument zenth angle s always less than 60 degrees for all collocatons. S P 6.63R e G E Z P G R e O Fgure 3: Computng arc angle to satellte nadr and zenth angle of satellte from Earth locaton.a.v. Specfcs for Pre-Operatonal For Meteosat, the GEO FoR ncludes all data wthn ±5 lat/lon of the SSP. All IASI data wthn ths area shall be collected from every overpass each 4 h perod, begnnng 00:00:00 UTC. The IASI data wthn ths overpass s then geographcally subset to only nclude data wthn ths smaller GEO FoR by applyng tme flterng. The mean observng tme wthn each subset IASI orbt shall be extracted and stored. The subset Meteosat mages shall be extracted wth equator crossng tmes closest to the mean observaton tme wthn each subset IASI orbt. Page 8 of 47

9 . FID COLLOCATIOS A set of observatons from a par of nstruments wthn a common perod (e.g. day) s requred as nput to the algorthm. The frst step s to obtan these data from both nstruments, select the relevant comparable portons and dentfy the pxels that are spatally collocated, temporally concurrent, geometrcally algned and spectrally compatble and calculate the mean and varance of these radances. Subset MO Data Subset REF Data Colloc. Crtera. Collocatng Collocated Data Archve ~ month Fgure 4: Step of Generc Data Flow, showng nputs and outputs Page 9 of 47

10 .a. Collocaton n Space.a.. Purpose The followng components of ths step defne whch pxels can be used n the drect comparson. To do ths, we frst extract the central locaton of each nstruments pxels and determne whch pxels can consdered to be collocated, based on ther centres beng separated by less than a pre-determned threshold dstance. At the same tme we dentfy the pxels that defne the target area (FoV) and envronment around each collocaton. These are later averaged n 3.c. The target area s defned to be a lttle larger than the larger Feld of Vew (FoV) of the nstruments so t covers all the contrbutng radaton n event of small navgaton errors, whle beng large enough to ensure relable statstcs of the varance are avalable. The exact rato of the target area to the FoV wll be nstrument-specfc, but n general wll range to 3 tmes the FoV, wth a mnmum of 9 'ndependent' pxels..a.. General Optons Where an nstrument s pxels follow fxed geographc coordnates, t s possble to used a look-up table to whch dentfy pxels match a gven target s locaton. Ths s the most effcent and recommended opton where avalable (often for geostatonary nstruments)..a.. Infrared GEO-LEO nter-satellte/nter-sensor Class The spatal collocaton crtera s based on the nomnal radus of the LEO FoV at nadr. Ths s taken as a threshold for the maxmum dstance between the centre of the LEO and GEO pxels for them to be consdered spatally collocated. However, gven the geometry of the already subset data, t s assumed that all LEO pxels wthn the GEO FoR wll be wthn the threshold dstance from a GEO pxel. The GEO pxel closest to the centre of each LEO FoV can be dentfed usng a reverse look-up-table (e.g. usng a McIDAS functon)..a.v. Specfcs for Pre-Operatonal The GEO pxel closest to the centre of each IASI FoV s dentfed usng a reverse look-uptable (e.g. usng a McIDAS functon). The IASI FoV s defned as a crcle of km dameter at nadr..a.v.. Specfcs for Pre-Operatonal MVIRI-IASI The MVIRI FoV s defned as square pxels wth dmensons of 5x5km at SSP. An array of 3x3 MVIRI pxels centred on the pxel closest to centre of each IASI pxel are taken to represent the collocaton target area correspondng to the IASI FoV.a.v.. Specfcs for Pre-Operatonal SEVIRI-IASI The SEVIRI FoV s defned as square pxels wth dmensons of 3x3km at SSP. An array of 5x5 SEVIRI pxels centred on the pxel closest to centre of each IASI pxel are taken to represent the collocaton target area correspondng to the IASI FoV. Page 0 of 47

11 .b. Concurrent n Tme.b.. Purpose ext we need to dentfy whch of those pxels dentfed n the prevous step as spatally collocated are also collocated n tme. Although even collocated measurements at very dfferent tmes may contrbute to the nter-calbraton, f treated properly, the capablty of processng collocated measurements s lmted and the more closely concurrent ones are more valuable for the nter-calbraton..b.. General Optons Each pxel dentfed as beng spatally collocated s tested sequentally to check whether the observatons from both nstruments were sampled suffcently closely n tme.e. separated n tme by no more than a specfc threshold. Ths threshold should be chosen to allow a suffcent number of collocatons, whle not ntroducng excessve nose due to temporal varablty of the target radance relatve to ts spatal varablty on a scale of the collocaton target area see Hewson [009]..b.. Infrared GEO-LEO nter-satellte/nter-sensor Class The tme at whch each collocated pxel of the GEO mage was sampled s extracted or calculated and compared to for the collocated LEO pxel. If the dfference s greater than a threshold of 300s, the collocaton s rejected, otherwse t s retaned for further processng. Equaton : LEO _ tme GEO _ tme < max_ sec, where max_sec300s The problem wth applyng a tme collocaton crtera n the above form s that t wll often lead to only a part of the collocated pxels beng analysed. As the GEO mage s often clmatologcally asymmetrc about the equator, ths can lead to the collocated radances havng dfferent dstrbutons, whch can affect the results. A possble soluton to ths problem s to apply the tme collocaton crtera to the mean tmes at whch the collocated GEO and LEO pxels were sampled. Ths would ensure ether all or none of the pxels wthn each overpass are consdered to be collocated n tme..b.v. Specfcs for Pre-Operatonal The tme at whch each collocated pxel of the Meteosat mage was sampled s approxmated by nterpolatng between the sensng start and end tme gven n the meta data, accordng to the scan lne number. Ths s compared to the sample tme gven n the IASI Level c dataset..b.v.. Specfcs for Pre-Operatonal MVIRI-IASI The tme at whch each collocated pxel of the MVIRI mage was sampled s approxmated by nterpolatng between the sensng start and end tme gven n the meta data, accordng to the scan lne number, whch ncrements lnearly from, just below the South Pole to 500, Page of 47

12 just above the orth Pole. If the dfference s greater than a threshold of max_sec900 s, the collocaton s rejected, otherwse t s retaned for further processng..b.v.. Specfcs for Pre-Operatonal SEVIRI-IASI n Full Dsk Imagng mode SEVIRI s full dsk magng mode starts scannng from scan lne, just below the South Pole to 37, just above the orth Pole n a perod of 74.4 s. If the SEVIRI-IASI samplng tme dfference s greater than a threshold of max_sec300s, the collocaton s rejected, otherwse t s retaned for further processng..b.v.3. Specfcs for Pre-Operatonal SEVIRI-IASI n Rapd Scannng Servce In Rapd Scannng Servce mode, SEVIRI scans only 464 lnes n a perod of 37.8 s, coverng If the SEVIRI-IASI samplng tme dfference s greater than a threshold of max_sec300s, the collocaton s rejected, otherwse t s retaned for further processng. The actual threshold appled n the prototype code s 800 s however, as Meteosat-7 completes a full dsc scan n ths perod, the standard devaton of tme dfferences of 63 s s closer to the value expected from a unform samplng dstrbuton wthn the ±900 s, so that value s used n ths analyss. Page of 47

13 .c. Algnment n Vewng Geometry.c.. Purpose The next step s to ensure the selected collocated pxels have been observed under comparable condtons. Ths means they should be algned such that they vew the surface at smlar ncdence angles (whch may nclude azmuth and polarsaton as well as elevaton angles) through smlar atmospherc paths..c.. General Optons Each pxel dentfed as beng spatally and temporally collocated s tested sequentally to check whether the vewng geometry of the observatons from both nstruments was suffcently close. The crteron for zenth angle s defned n terms of atmospherc path length, accordng to the dfference n the secant of the observatons zenth angles and the dfference n azmuth angles. If these are less than pre-determned thresholds the collocated pxels can be consdered to be algned n vewng geometry and ncluded n further analyss. Otherwse they are rejected..c.. Infrared GEO-LEO nter-satellte/nter-sensor Class The geometrc algnment of thermal nfrared channels depends only on the zenth angle and not azmuth or polarsaton. Equaton 3: cos cos ( geo _ zen) ( leo _ zen) < max_ zen The threshold value for max_zen can be qute large for wndow channels (e.g., 0.05 for 0.8 μm channel) but must be rather small for more absorptve channels (e.g., <0.0 for 3.4 μm channel). However, unless there are partcular needs to ncrease the sample sze for wndow channels, a common threshold value of max_zen0.0 s recommended for all channels. Ths results n collocatons beng dstrbuted approxmately symmetrcally about the equator mappng out a characterstc slanted hourglass pattern. Another aspect of vewng geometry algnment s azmuth angle. Smlar zenth angle assures smlar path length; addtonal requrement of smlar azmuth angle assures smlar lne-ofsght. Lne-of-sght algnment s relevant for IR spectrum n certan cases. For nfrared wndow channels, land surface emsson durng daytme may be ansotropc [Mnns et al. 004]. For shortwave IR band (e.g., 4 μm), azmuth angle algnment s requred durng daytme when solar radaton s consderable. It s, therefore recommended that ntercalbraton over land and n ths band are lmted to nght-tme only cases at the expense of lmtng the dynamc range of the results..c.v. Specfcs for Pre-Operatonal The uncertanty analyss [EUMETSAT, 00] suggested there would be no ncrease n overall uncertanty by adoptng max_zen0.05. However, subsequent testng of the proposed relaxaton of the geometrc collocaton threshold [EUM/MET/REP//063] showed small, but sgnfcant changes to the relatve bases (up to 0.05 K) were produced by ths change. Page 3 of 47

14 .c.v.. Specfcs for Pre-Operatonal MVIRI-IASI It s therefore recommended that max_zen0.0 for MVIRI n full dsc scannng mode, but relaxed to max_zen0.05 for Rapd Scannng Servce mode to ncrease the number of collocatons..c.v.. Specfcs for Pre-Operatonal SEVIRI-IASI Full Dsk Imagng Mode It s therefore recommended that max_zen0.0 for SEVIRI n full dsk magng mode. Furthermore, t was found that collocatons wth large ncdence angles can cause the regresson to generate erroneous results occasonally for unknown reasons. Therefore, we lmt the maxmum ncdence angle to 35..c.v.3. Specfcs for Pre-Operatonal SEVIRI-IASI Rapd Scannng Servce A relaxaton of max_zen0.05 s recommended for Rapd Scannng Servce mode to ncrease the number of collocatons. Furthermore, t was found that collocatons wth large ncdence angles can cause the regresson to generate erroneous results occasonally for unknown reasons. Therefore, we lmt the maxmum ncdence angle to 35. Page 4 of 47

15 .d. Pre-Select Channels.d.. Purpose Only broadly comparable channels from both nstruments are selected to reduce data volume..d.. General Optons Ths selecton s based on pre-determned crtera for each nstrument par..d.. Infrared GEO-LEO nter-satellte/nter-sensor Class Only the channels of the GEO and LEO sensors are selected n the thermal nfrared range of 3-5µm..d.v. Specfcs for Pre-Operatonal Select only the water vapour and thermal nfrared channels of Meteosat. Select all channels for IASI..d.v.. Specfcs for Pre-Operatonal MVIRI-IASI Select MVIRI s operatonal nfrared and water vapour channels: WV- and IR-..d.v.. Specfcs for Pre-Operatonal SEVIRI-IASI Select SEVIRI s nfrared channels: 3.9, 6., 7.3, 8.7, 9.7, 0.8,.0, 3.4 μm. Fgure 5: Example radance spectra measured by IASI (blue), expressed n brghtness temperature (K) and Spectral Response Functons of SEVIRI channels 3- from rght to left (red/green). Page 5 of 47

16 .e. Plot Collocaton Map.e.. Purpose When nterpretng the nter-calbraton results t s often helpful to vsualse the dstrbuton of the source data used n the comparson..e.. General Optons Ths can be acheved by producng a map showng the dstrbuton of collocaton targets..e.. Infrared GEO-LEO nter-satellte/nter-sensor Class The map s produced showng all the GEO-LEO pxels meetng the collocaton crtera every day. These ponts are overlad on a background mage from an nfrared wndow channel of the GEO nstrument. Ths allows the dstrbuton of cloud to be vsualsed and consdered n the nterpretaton of the results..e.v. Specfcs for Pre-Operatonal An mage s produced of radance of the IR0.8 channel of SEVIRI or the IR channel of MVIRI over the GEO FoR on a fxed radance scale runnng from 80 mw/m/st/cm- ( whte) to 40 mw/m/st/cm- (black). The poston of the centre of all IASI FoVs s over-plotted on ths mage n grey and those pxels meetng the collocaton crtera are over-plotted n red, as shown n Fgure 7. Fgure 6: Example collocaton map, follow nset legend. Page 6 of 47

17 3. TRASFORM DATA In ths step, collocated data are transformed to allow ther drect comparson. Ths ncludes modfyng the spectral, temporal and spatal characterstcs of the observatons, whch requres knowledge of the nstruments characterstcs. The outputs of ths step are the best estmates of the channel radances, together wth estmates of ther uncertanty. Collocated Data SRFs, PSFs, 3. Transformng Comparson Data Archve ~ year Fgure 7: Step 3 of Generc Data Flow, showng nputs and outputs Page 7 of 47

18 3.a. Convert Radances 3.a.. Purpose Convert observatons from both nstruments to a common defnton of radance to allow drect comparson. 3.a.. General Optons The nstruments observatons are converted from Level.5/b/c data to radances, usng pre-defned, publshed algorthms specfc for each nstrument. 3.a.. Infrared GEO-LEO nter-satellte/nter-sensor Class Perform comparson n radance unts: mw/m /st/cm -. 3.a.v. Specfcs for Pre-Operatonal IASI data are converted to radances usng the publshed algorthm [EUMETSAT, 008a]. 3.a.v. Specfcs for Pre-Operatonal MVIRI-IASI Radances are converted to brghtness temperatures followng Tjemkes [005]. 3.a.v. Specfcs for Pre-Operatonal SEVIRI-IASI The Meteosat radance defnton applcable to each level.5 dataset, descrbed by EUMETSAT [00], s used, accountng for the nstrument s Spectral Response Functons [EUMETSAT, 006]. Page 8 of 47

19 3.b. Spectral Matchng 3.b.. Purpose Frstly, we must dentfy whch channel sets provde suffcent common nformaton to allow meanngful nter-calbraton. These are then transformed nto comparable pseudo channels, accountng for the defcences n channel matches. 3.b.. General Optons The Spectral Response Functons (SRFs) must be defned for all channels. The observatons of channels dentfed as comparable are then co-averaged usng pre-determned weghtngs to gve pseudo channel radances. A Radatve Transfer Model can be used to account for any dfferences n the pseudo channels characterstcs. The uncertanty due to spectral msmatches s then estmated for each channel. 3.b.. Infrared GEO-LEO nter-satellte/nter-sensor Class For hyper-spectral nstruments, all SRFs are frst transformed to a common spectral grd. The LEO hyperspectral channels are then convolved wth the GEO channels SRFs to create synthetc radances n pseudo-channels, accountng for the spectral samplng and stablty n an error budget. R Equaton 4: ν ν Φν dν R GEO Φ ν dν ν where R GEO s the smulated GEO radance, R ν s LEO radance at wave number ν, and Φ ν s GEO spectral response at wave number ν. In general LEO hyperspectral sounders do not provde complete spectral coverage of the GEO channels ether by desgn (e.g. gaps between detector bands), or by subsequent hardware falure (e.g. broken or nosy channels). The radances n these gap channels shall be accounted by one of the followng technques: Tahara and Kato [009] defne vrtual channels named gap channels to fll the spectral gaps and ntroduce the spectral compensaton method by constraned optmzaton. The gap channels to fll the AIRS spectral gaps (AIRS gap channels) are defned by 0.5 cm - ntervals, and are characterzed by a unque SRF, whose shape s a Gaussan curve wth a sgma of 0.5 cm -. The gap channels to extend the IASI spectral regon (IASI gap channels) are defned by the same ntervals (0.5 cm - ) and SRFs as the IASI level c channels. The radances of the mssng channels are calculated by regresson analyss usng radatve transfer smulated radances wth respect to the eght atmospherc model profles as explanatory varables. K calc sm Equaton 5: log I c0 + ck log I, k ( hyper and gap channels), k calc sm where I s the calculated radance of the hyper channel, I, k s the smulated radance of the hyper channel wth respect to the atmospherc model profle k, c k ( k, K, K ) are regresson coeffcents, and K s the number of the atmospherc model profles. Eq. 6 ntroduces logarthm radances as response and explanatory varables n order to ncrease Page 9 of 47

20 fttng accuracy and avod calculaton of negatve radance. The regresson coeffcents c k are ndependent of the hyper channels, and are generated for each scan poston of the hyper sounder. c are obtaned by the least-square method applyng a set of valdly observed k obs radances I n place of calc I to Eq. 6, obs sm Equaton 7: { c k} arg mn log I c0 + ck log I, k. obs exst( I ) k Once the regresson coeffcents c k are computed, the radances of the mssng channels can be calculated by Eq. 6. It mght be possble to apply the observed radances of all hyper channels to Eq. 7 to compute c k and then calculate the radances of all mssng channels at once. However, ths yelds a large fttng error n practce. In nter-calbraton applcaton, the coeffcents c k are computed for each broadband channel spectral regon. Eq. 6 and Eq. 7 use sm the smulated radances I. For the radance smulaton, ths study uses the followng eght,k atmospherc model profles:. U.S. standard wthout cloud,. U.S. standard wth opaque cloud wth tops at 500 hpa alttude, 3. U.S. sta ndard wth opaque cloud wth tops at 00 hpa alttude, 4. Tropcal wthout cloud, 5. Tropcal wth opaque cloud wth tops at 500 hpa alttude, 6. Tropcal wth opaque cloud wth tops at 00 hpa alttude, 7. Md-lattude summer wthout cloud, 8. Md-lattude wnter wthout cloud. These profles nclude not only clear sky condtons but also cloudy condtons because Eq. 6 should be applcable under any weather condtons. As for radatve transfer code, the lneby-lne code LBLRTM (Clough et al., 995) verson. s used wth the HITRA004 spectroscopy lne parameter database (Rothman et al., 003) ncludng the AER updates verson.0 (AER Web page). The emssvtes of the surface and clouds are assumed to be one. The beneft of ths spectral compensaton method s that t does not requre radatve transfer computaton to be run n nter-calbraton operaton. Ths not only speeds up the computaton but also prevents super channel radance computaton from ntroducng bases contaned n radatve transfer code and atmospherc state felds. 3.b.v. Specfcs for Pre-Operatonal IASI channels are assumed to be spectrally stable and contguously sampled wth a spectral resoluton of 0.5 cm -. 3.b.v.. Specfcs for Pre-Operatonal MVIRI-IASI MVIRI s SRFs for WV- and IR- channels, publshed by EUMETSAT [005] are nterpo lated onto IASI s spectral grd n wavenumber-space usng a blnear nterpolaton. Any negatve responses n the nterpolated SRFs are set to zero. Page 0 of 47

21 3.b.v.. Specfcs for Pre-Operatonal SEVIRI-IASI SEVIRI s SRFs for an operatng temperature of 95K, publshed by EUMETSAT [006] are nterpolat ed onto IASI s spectral grd n wavenumber-space usng a blnear nterpolaton. Any negatve responses n the nterpolated SRFs are set to zero. The radanc e mssng from IASI s coverage of SEVIRI IR3.9 channel s also estmated followng the constraned optmzaton approach descrbed above [Tahara and Kato, 009], usng coeffcents specfed theren specfcally for SEVIRI-IASI. These gap channels to extend the IASI spectral regon ( IASI gap channels) are defned by the same ntervals (0.5 cm - ) and SRFs as the IASI level c channels. Page of 47

22 3.c. Spatal Matchng 3.c.. Purpose The observatons from each nstrument are transformed to comparable spatal scales. Ths nvolves averagng all the pxels dentfed n as beng wthn the target and envronment areas. The uncertanty due to spatal varablty s estmated. 3.c.. General Optons The Pont Spread Functons (PSFs) of each nstrument are dentfed. The target area and envronment around t were specfed n. ow the pxels wthn these areas are dentfed and ther radances are averaged and ther varance calculated to estmate the uncertanty on the average due to spatal varablty, accountng for any over-samplng. 3.c.. Infrared GEO-LEO nter-satellte/nter-sensor Class The target area s defned as the nomnal LEO FoV at nadr. The GEO pxels wthn target area are averaged usng a unform weghtng and ther varance calculated. The envronment s defned by the GEO pxels wthn twce the radus of the target area from the centre of each LEO FoV. 3.c.v. Specfcs for Pre-Operatonal As above, where the IASI FoV s defned as a crcle of km dameter at nadr. 3.c.v.. Specfcs for Prototype MVIRI-IASI The MVIRI FoV s defned nomnally as square pxels wth lengths of 5km at SSP. These are assumed to be constant across collocaton doman. The target area s defned by arrays of 3x3 MVIRI pxels closest to centre of each IASI FoV, as shown n Fgure 9. Ths s somewhat larger than the sze of the IASI FoV at nadr, but smaller at the extremes of ts scan. The envronment s defned by arrays of 5x5 MVIRI pxels on the same centre. 3x3 MVIRI pxels (5km grd at SSP) defne Target Area + IASI FoV km dameter near nadr 5x5 MVIRI pxels (5km grd at SSP) defne Envronment Fgure 8: Defnton of Target Area as 3x3 MVIRI pxels to spatally match an IASI FoV. 3.c.v.. Specfcs for Prototype SEVIRI-IASI The SEVIRI FoV s defned nomnally as square pxels wth lengths of 3km at SSP. These are assumed to be constant across collocaton doman. The target area s defned by arrays of 5x5 SEVIRI pxels closest to centre of each IASI FoV, as shown n Fgure 9. Ths s Page of 47

23 somewhat larger than the sze of the IASI FoV at nadr, but smaller at the extremes of ts scan. The envronment s not defned, as t s not used n further analyss. The envronment s defned by a array 9x9 SEVIRI pxels, centred on the IASI FoV. SEVIRI pxels (3km grd at SSP) defnng one Target Area + IASI FoV km dameter near nadr Fgure 9: Defnton of Target Area as 5x5 SEVIRI pxels to spatally match an IASI FoV. Envronment s defned by array of 9x9 SEVIRI pxels centred on IASI FoV. Page 3 of 47

24 3.d. Vewng Geometry Matchng 3.d.. Purpose Despte the collocaton crtera descrbed n.c, each nstrument can measure radance from the collocaton targets n slghtly dfferent vewng geometry. It may be possble to account for small dfferences by consderng smplfed a radatve transfer model. 3.d.. General Optons Dfferences n vewng geometry wthn t he collocaton crter a descrbed n.c are assumed to be neglgble and gnored n further analyss. Although t may be pos sble to account for small dfferences by consderng smplfe d a radatve transfer model, ths has not been mplemented at ths tme. 3.d.. Infrared GEO-LEO nter-satellte/nter-sensor Class Dfferences n vewng geometry wthn the collocaton crtera descrbed n.c are assumed to be neglgble and gnored n further analyss. 3.d.v. Specfcs for Pre-Operatonal As above. Page 4 of 47

25 3.e. Temporal Matchng 3.e.. Purpose Dfferent nstruments measure radance from the collocaton targets at dfferent tmes. The mpact of ths dfference can usually be reduced by careful selecton, but not completely elmnated. The tmng dfference between nstruments observatons s establshed and the uncertanty of the comparson s estmated based on (expected or observed) varablty over ths tmescale. 3.e.. General Optons Each nstrument s sample tmngs are dentfed. 3.e.. Infrared GEO-LEO nter-satellte/nter-sensor Class Only the GEO mage closest to the LEO equator crossng tme s selected. The tme dfference between the collocated GEO and LEO observatons s neglected and the collocaton targets are assumed to be sampled smultaneously, contrbutng no addtonal uncertanty to the comparson. 3.e.v. Specfcs for Pre-Operatonal As above. Page 5 of 47

26 4. FILTERIG The collocated and transformed data wll be archved for analyss. Before that, the GSICS nter-calbraton algorthm reserves the opportunty to remove certan data that should not be analyzed (qualty control), and to add auxlary data that wll add further analyss. For example, t may be useful to ncorporate land/sea/ce masks and/or cloud flags to better classfy the results. Comparson Data Masks, flags, 4. Flterng Analyss Data Archve ~ year Fgure 0: Step 4 of Generc Data Flow, showng nputs and outputs. Page 6 of 47

27 4.a. Unformty Test 4.a.. Purpose Knowledge of scene unformty s crtcal n reducng and evaluatng nter-calbraton uncertanty. To reduce uncertanty n the comparson due to spatal/temporal msmatches, the collocaton dataset may be fltered so only observatons n homogenous scenes are compared. 4.a.. General Optons The approach adopted n ths verson s not to reject collocatons based on a threshold of scene varablty, but to use scene varances as weghtngs n the regresson of collocated radances. Comparatvely, the threshold opton has the theoretcal dsadvantage of subjectvty but practcal advantage of substantally reducng the amount of data to be archved. Recent analyss [Tobn, personal communcaton, 009] also ndcates that the threshold opton s always suboptma l compared to the weght opton. 4.a.. Infrared GEO-LEO nter-satellte/nter-sensor Class The varance of the radances of all the GEO pxe ls wthn each LEO FoV s calculated n 3.c. 4.a.v. Specfcs for Pre-Operatonal An opton s ncluded to reject any targets where the standard devaton of the scen e radance s >5% of the standard radance (see 4b). Ths s only used to clean-up the scatterplots showng the regresson of collocated radances to prevent ponts wth large error bars (hgh spatal varablty) domnatng the plots generated n 5.b, as they have a neglgble mpact on the regresson coeffcents. Page 7 of 47

28 4.b. Outler Rejecton 4.b.. Purpose To prevent anomalous observatons havng undue nfluence on the results, outlers may be dentfed and rejected on a statstcal bass. Small number of anomalous pxels n the envronment, even concentrated, may not fal the unformty test. However, f they appear only n one sensor s feld of vew but not the other, t can cause unwanted bas n a sngle comparson. 4.b.. General Optons The radances n the target area are compared wth those n the surroundng envronment, and those targets whch are sgnfcantly dfferent from the envronment (3) may be rejected. For a normally dstrbuted populaton of sze, mean M, and standard devaton S, the dfference between a sngle sample and M has the probablty of ~68% to be less than S, ~95% to be less than S, and so forth. Smlarly, the dfference between the mean of n samples and M has the probablty of ~68% to be less than S/n[(-n)/(-)], ~95% to be less than S/n[(-n)/(-)], and so forth. Ths property s used to test whether the collocaton area s an outler for the otherwse unform envronment: n S n Equaton 6: R Gaussan( 3) M n n where R s radance from ndvdual pxel, n s the number of samples, and Gaussan s a threshold. The probablty that the rejected sample s an outler s 68% f Gaussan, 95% f Gaussan, and more than 99% f Gaussan3. 4.b.. Infrared GEO-LEO nter-satellte/nter-sensor Class The mean GEO radances wthn each LEO FoV are compared to the mean of ther envronment. Targets where ths dfference s >3 tmes the standard devaton of the envronment s radances are rejected. 4.b.v. Specfcs for Pre-Operatonal As above. Page 8 of 47

29 4.c. Auxlary Datasets 4.c.. Purpose It may be useful to ncorporate land/sea/ce masks and/or cloud flags to allow analyss of statstcs n terms of other geophyscal varables e.g. land/sea/ce, cloud cover, etc. It may also be possble to estmate the spatal varablty wthn the LEO FoV from collocated AVHRR observatons from the same LEO satellte. 4.c.. General Optons ot yet mplemented. 4.c.. Infrared GEO-LEO nter-satellte/nter-sensor Class ot yet mplemented. 4.c.v. Specfcs for Pre-Operatonal ot yet mplemented. Page 9 of 47

30 5. MOITORIG Ths step ncludes the actual comparson of the collocated radances produced n Steps -4, the producton of statstcs summarsng the results to be used n the Correctng step, and reportng any dfferences n ways meanngful to a range of users. Analyss Data 5. Montorng Bas Montorng Fgure : Step 5 of Generc Data Flow, showng nputs and outputs. Page 30 of 47

31 5.a. Defne Standard Radances (Offlne) 5.a.. Purpose Ths component provdes standard reference scene radances at whch nstruments ntercalbraton bas can be drectly compared and convenently expressed n unts understandable by the users. Because bases can be scene-dependent, t s necessary to defne channelspecfc standard radances. More than one standard radance may be needed for dfferent applcatons e.g. clear/cloudy, day/nght. Ths component s carred out offlne. 5.a.. General Optons The standard radances should be calculated for each channel a pror usng a Radatve Transfe r Model (RTM) based on a standard atmospherc profle and surface condtons. The reference radance should be calculated at nadr, at nght for IR channels or at a gven solar angle (for vs/nr channels), n a 976 US Standard Atmosphere, n clear skes, over the sea wth a SST+5C and wnd speed (7m/s), usng some standard RTM, accountng for the SRF of each channel. Ths has the advantages of beng ndependent of any nstrument bases and provdes standard radances aganst whch we can compare the nstruments relatve bases derved from a number of dfferent nter-calbraton technques. 5.a.. Infrared GEO-LEO nter-satellte/nter-sensor Class A above. 5.a.v. Specfcs for Pre-Operatonal As above. 5.a.v. Specfcs for Prototype MVIRI-IASI The calculaton of standard radances s mplemented drectly, usng RTTOV-9, gvng the followng results for the IR channels MVIRI on both Meteosat-7: Channel WV IR T bstd (K) a.v.. Specfcs for Pre-Operatonal SEVIRI-IASI The calculaton of standard radances s mplemented drectly, usng RTTOV-9, gvng the followng results for the IR channels SEVIRI on both Meteosat-8 and -9: Ch (μm) T bstd (K) Page 3 of 47

32 5.b. Regresson of Most Recent Results 5.b.. Purpose Regresson s used as the bass of the systematc comparson of collocated radances from two nstruments. (Ths comparson may also be done n counts or brghtness temperature.) Regresson coeffcents shall be made avalable to users to apply the GSICS Correcton to the montored nstrument, re-calbratng ts radances to be consstent wth those of the reference nstrument. Scatterplots of the regresson data should also be produced to allow vsualsaton of the dstrbuton of radances. Regressons also allow us to nvestgate how bases depend on varous geophyscal varables and provdes statstcs of any sgnfcant dependences, whch can used to refne correctons and allows nvestgaton of the possble causes. Such nvestgatons should be carred out offlne and may result n future refnements to the ATBD. 5.b.. General Optons The recommended approach s to perform a weghted lnear regresson of collocated radances. The nverse of the sum of the spatal and temporal varance of the target radance and the radometrc nose provde an estmated uncertanty on each dependent pont, whch s used as a weghtng. (Includng the radometrc nose ensures that very homogeneous targets scenes where all the pxels gve the same radance do not have undue nfluence on the weghted regresson.) Ths method produces estmates of regresson coeffcents descrbng the slope and offset of the relatonshp between the two nstruments radances together wth ther uncertantes, expressed as a covarance. The problem of correlaton between the uncertantes on each coeffcent may be reduced by performng the regresson on a transformed dataset for example, by subtractng the mean or reference radance from each set. The observatons of the reference nstrument, x, and montored nstrument, y, are ftted to a straght lne model of the form: Equaton 7: ˆ ( x) y a + bx We assume an uncertanty assocated wth each measurement, y, s known and that the dependent varable, x s also known. To ft the observed data to the above mod el, we mnmse the ch-square mert functon: y a bx Equaton 8: χ ( a, b) Ths can be mplemented followng the method descrbed n Secton 5. of umercal Recpes [Press et al., 996], whch s mplemented n the POLY_FIT functon of IDL, yeldng the followng estmates of the regresson coeffcents: Page 3 of 47

33 Page 33 of 47 Equaton 9: Equaton 0: x x y a, x x y x x x y x y x b, ther uncertantes: Equaton : a x, x x Equaton : b x x, and ther covarance: Equaton 3: ( ), cov x x x b a. 5.b.. Infrared GEO-LEO nter-satellte/nter-sensor Class ter-calbratons are repeated daly usng only nght-tme LEO overpasses. Collocatons are he nverse th sum of the spatal and temporal varance of target radances and er radometrc nose level n the regresson. (The ncluson of the radometrc nose ensures produced to allow vsualsaton of the dstrbuton of dances, followng the example shown n Fgure. In weghted by t e th the weghts never become nfnte due to collocaton targets wth zero varance.) Scatterplots of the regresson data should also be ra

34 Fgure : Example scatterplot showng regresson of collocated radances, followng legend. 5.b.v. Specfcs for Pre-Operatonal Implement as above. The range of ncdence angles was mplctly extended to <40 by changng the FoR constrants. Inter-calbratons are attempted every day (although only ~½ of cases contan collocatons). The temporal varance s assumed to be equal to the spatal varance, so ts contrbuton to the weghtng s multpled by and added n quadrature to the radometrc nose. The radometrc nose for IASI s assumed to be neglgble when averaged over all channels wthn the SRF of each Meteosat channel. 5.b.v.. Specfcs for Pre-Operatonal MVIRI-IASI The radometrc nose on each pxel for the MVIRI channels s gven by [Schmetz et al. 00]: Channel WV IR Met-7 ose [K] b.v.. Specfcs for Pre-Operatonal SEVIRI-IASI The radometrc nose on each pxel for the SEVIRI channels s gven by [EUMETSAT, 007] for example, for ambent calbratons at 95K. For MSG-3 values are taken from Table 36 of EUM/MSG/REP//045. Channel IR MSG- ose [K] MSG- ose [K] MSG-3 ose [K] Page 34 of 47

35 5.c. Bas Calculaton 5.c.. Purpose Inter-calbraton bases should be drectly comparable for representatve scenes and convenently expressed n unts understandable by the users. Because bases can be scenedependent, they are evaluated here at the standard radances defned n 5.a. 5.c.. General Optons Regresson coeffcents are appled to estmate expected bas, Δ ŷ( x STD ) ( ) Eq yˆ x STD, and uncertanty,, for standard radances, accountng for correlaton between regresson coeffcents. uaton 4: ( STD ) STD STD notng that y STD x STD and Δˆ y x a + bx y, Equaton 5: ( x ) + x + cov( a, b)x ŷ STD a b STD The results may be expressed n absolute or percentage bas n radance, or brghtness temperature dfferences. STD 5.c.. Infrared GEO-LEO nter-satellte/nter-sensor Class Bases and ther uncertantes are converted from radances to brghtness temperatures for vsualsaton purposes. 5.c.v. Specfcs for Pre-Operatonal The defnton of effectve radance s used n the converson to brghtness temperatures [EUMETSAT, 008b]. Page 35 of 47

36 5.d. Consstency Test 5.d.. Purpose The most recent results are tested for statstcal consstency wth the prevous tme seres of results. Users should be alerted to any sudden changes n the calbraton of the nstruments, allowng them to nvestgate potental causes and reset trend statstcs calculated n 5.e. The consstency test may be performed n terms of regresson coeffcents or bases. 5.d.. General Optons The bases calculated for standard radances from the most recent collocatons are compared to the statstcs of the bases trends calculated n 5.e from prevous results. If the most recent result falls outsde the 3- (99.7%) confdence lmts estmated from the trend statstcs, an alert should be rased. Ths alert should trgger the Prncple Investgator to check the cause of the change and reset the trends by ssung a trend reset. Equaton 6: y yˆ ( ) yˆ x ( x ) Gaussan ( 3) 5.d.. Infrared GEO-LEO nter-satellte/nter-sensor Class As above. 5.d.v. Specfcs for Pre-Operatonal As above. Page 36 of 47

37 5.e. Trend Calculaton 5.e.. Purpose It s mportant to establsh whether an nstrument s calbraton s changng slowly wth tme. It s possble to establsh ths from a tme-seres of nter-comparsons by calculatng a trend lne usng a lnear regresson wth date as the ndependent varable. Only the porton of the tme seres snce the most recent trend reset s analysed, to allow for step changes n the nstruments calbraton. 5.e.. General Optons The tme seres of bases evaluated at standard radances can be regressed aganst the tme (date) as the ndependent varable. The lnear regresson can be weghted by the calculated uncertanty on each bas. The regresson coeffcents ncludng uncertantes (and ther covarances) are calculated by the least squares method descrbed n 0. In ths case, the varables, x and y are tme seres of Julan dates and radance bases estmated n 5.c for each orbt snce the most recent trend reset, respectvely. 5.e.. Infrared GEO-LEO nter-satellte/nter-sensor Class As above. 5.e.v. Specfcs for Pre-Operatonal As above. Page 37 of 47

38 5.f. Generate Plots for GSICS Bas Montorng 5.f.. Purpose The results should be reported quantfyng the magntude of relatve bases by ntercalbraton. Ths should allow users to montor changes n nstrument calbraton. 5.f.. General Optons Plots and tables of relatve bases and uncertantes for standard radances should be produced. These may show the evoluton of the bases and ther dependence on geophyscal varables. These all results should be uploaded to the GSICS Data and Products server, and made avalable from the GPRC s approprate nter-calbraton webpage. 5.f.. Infrared GEO-LEO nter-satellte/nter-sensor Class Plots should be regularly updated showng the relatve brghtness temperature bases for the standard radances n each channel as tme seres wth uncertantes. The trend lne and monthly mean bases (and ther uncertantes) should be calculated from these tme seres, followng the example n Fgure 3. Ths allows the most recent result to be tested for consstency wth the seres of prevous results. If sgnfcant dfferences are found operators should be alerted, gvng them the opportunty to nvestgate further. Fgure 3: Example of tme seres plot showng relatve bas of IR3.4 channel of Meteosat-9 and IASI at reference radance followng nset legend. 5.f.v. Specfcs for Pre-Operatonal The GSICS Bas Montorng plots are now generated separately usng a dedcated tools, whch reads data drectly from the netcdf fles contanng the GSICS Correctons. So ths step s only needed for nternal montorng. Page 38 of 47

39 FLOW SUMMARY OF STEPS 5 AD 6 FOR SEVIRI-IASI Analyss Data Latest Archve Tme Seres of Analyss Data Analyss Data 5b. Regresson 5. Montorng Regresson Coeffcents 5a. Defne Standard Radances 5c. Calculate Bases Latest Bases Standard Radances Archve 6. Correctng Bas Tme Seres 5e. Trend Calc. 5d. Consstency Test Trend Stats 6a. Defne Smoothng Perod Trend Resets (+ Alert) 5f. Plot Tme Seres of Bases Smoothng Perod 6b. Smooth Results Bas Montorng Products Correcton Coeffs Fgure 4: Summary of Recommended Data Flow wthn Steps 5 and 6 for SEVIRI-IASI Page 39 of 47

40 6. GSICS CORRECTIO Ths fnal step of the algorthm s to calculate the GSICS Correcton, allowng the calbraton of one nstrument s observed data to be modfed to become consstent wth that of the reference nstrument. The form of the GSICS Correcton wll be defned offlne and can be nstrument specfc. However, applcaton of the correcton reles on the Correcton Coeffcents suppled by the nter-comparsons performed n the prevous steps of the algorthm from the Analyss Data. Analyss Data 6. Correctng Correcton Coeffs Tme Seres of Inter-calbraton Regresson Coeffcents MO Lvl Data Satellte/Instrument/ Ch Date/Tme Geometry Radances/Counts Products GSICS Correcton e.g. Look-Up Table, FORTRA subroutne, ew calbraton coeffcents, Re-Cal Data Corrected Radances Wth Uncertantes Users Fgure 5: Step 6 of Generc Data Flow, showng nputs and outputs, and llustratng schematcally how the correcton could be appled by users. Page 40 of 47

41 6.a. Defne Smoothng Perod (Offlne) 6.a.. Purpose It s possble to combne data from a tme seres of nter-comparson results to reduce the random component of the uncertanty on the fnal GSICS Correcton. (See 6.a). However, ths requres us to defne representatve perods over whch the results can be smoothed wthout ntroducng bas due to calbraton drfts durng the smoothng perod. Ths perod can be defned by comparng the observed rate of change of nter-comparson results wth a pre-determned threshold, based on the requred or achevable accuracy. In general, ths defnton s performed offlne as t requr es an n-depth analyss of the nstruments relatve bases and consderaton of lkely explanatory mechansms. However, t could also be fnetuned n near real-tme. The followng descrbes the general approaches that should be mplemented. 6.a.. General Optons In 5.e., tme seres of radance bases are regressed aganst date as the ndependent varable. dδˆ yref Ths yelds an estmate of the rate of change of bas wth tme,, whch can be dt compared to the threshold Δy max to determne the smoothng perod, τ s : dδyˆ REF Equaton 7: τ s Δymax dt 6.a.. Infrared GEO-LEO nter-satellte/nter-sensor Class As above. 6.a.v. Specfcs for Pre-Operatonal As above. 6.a.v.. Specfcs for Prototype MVIRI-IASI The threshold value s taken to correspond to the typcal uncertanty on the nter-comparson, whch s equvalent to Δy max 0. K. The MVIRI channel wth the hghest rate of change s IR-, dδyˆ REF where 0.K / month. dt Ths yelds the followng smoothng perods: τ s 4.5 days for the ear Real-Tme Correcton τ s 9 days for the ear Real-Tme Correcton (to match the orbtal repeat cycle of Metop) Page 4 of 47

42 6.a.v.. Specfcs for Pre-Operatonal SEVIRI-IASI As above, where the threshold value s taken to correspond to the typcal uncertanty on the daly nter-comparsons, whch s equvalent to Δy max 0.05 K. The SEVIRI channel wth the hghest rate of change s IR3.4, where d Δyˆ REF K / month. dt Ths yelds the followng smoothng perods: τ s 4.5 days for the ear Real-Tme Correcton τ s 9 days for the ear Real-Tme Correcton (to match the orbtal repeat cycle of Metop) Page 4 of 47

43 6.b. Calculate Coeffcents for GSICS ear-real-tme Correcton 6.b.. Purpose In order to reduce the random component of the uncertanty on the GSICS Correcton, t s necessary to combne data from a tme seres of nter-comparson. The regresson process descrbed n 5.b s repeated usng all the collocated radances obtaned over the smoothng perod defned n 6.a. The resultng regresson coeffcents (and uncertantes) provde the Correcton Coeffcents used as nput to the GSICS Correcton. These regresson coeffcents are then used to evaluate the Standard Bas (also wth uncertantes) at a set of Standard Radances. The correcton coeffcents and standard bases are suppled n a netcdf format [defned at 6.b.. General Optons All the collocaton data wthn the smoothng perod before and ncludng the current date s combned and the regresson of 5.b repeated on the aggregate dataset. Ths approach ensures all data s used optmally, wth approprate weghtng accordng to ts estmated uncertanty. 6.b.. Infrared GEO-LEO nter-satellte/nter-sensor Class As above. 6.b.v. Specfcs for Pre-Operatonal Implement as above, usng a smoothng perod t-4d to t-0 (where t s the current date). Page 43 of 47