An optimal cloud-clearing method for AIRS radiances using MODIS

Transcription

1 An optmal cloud-clearng method for AIRS radances usng MODIS Jun Chan-Y H.-L. Xuebao Tmothy J. Schmt #, and W. Paul Menzel Cooperatve Insttute for Meteorologcal Satellte Studes Unversty of Wsconsn-Madson 5 West Dayton Street Madson, WI # Offce of Research and Applcatons NOAA/NESDIS 5 West Dayton Street Madson, WI Abstract The Atmospherc Infrared Sounder (AIRS), onboard NASA s EOA Aqua satellte, provdes atmospherc vertcal temperature and mosture soundng nformaton wth hgh vertcal resoluton and accuracy for Numercal Weather Predcton (NWP). Due to ts relatvely poor spatal resoluton (3.5 km at nadr), the chance for an AIRS footprnt to be clear s small. However, the Moderate Resoluton Imagng Spectroradometer (MODIS), also onboard Aqua, provdes co-located clear radances at several nfrared (IR) spectral broad bands wth km spatal resoluton wthn many AIRS cloudy footprnts. An optmal cloud-removal or cloud-clearng (CC) algorthm s developed to retreve the AIRS clear column radances from the combned MODIS IR clear radances wth hgh spatal resoluton and the AIRS cloudy radances on a sngle footprnt bass. The AIRS cloud-removed or cloud-cleared radance spectrum s convoluted to all the MODIS IR spectral bands wth spectral response functons (SRFs), and the convoluted brghtness temperatures (BTs) are compared wth MODIS clear BT observatons wthn all successful cloud-cleared footprnts. The bas and the standard devaton between the convoluted BTs and MODIS clear BT observatons s less than 0.5 K and 0.5 K, respectvely, over both water and land for most MODIS IR spectral bands. The retreved AIRS soundngs from cloud-cleared radance spectra are compared wth the ECMWF analyss and dropsonde data for CC performance evaluaton. The AIRS cloud-cleared BT spectrum s also compared wth ts nearby clear BT spectrum, the dfference, accountng the effects due to scene non- unformty, s reasonable accordng the analyss. It s found that more than 30% of the AIRS cloudy (partly and overcast) footprnts n ths study have been successfully cloud-cleared usng the optmal cloudclearng method, revealng the potental applcaton of ths method on the operatonal processng of future GOES-R Hyperspectral Envronmental Sute (HES) cloudy radance measurements when the collocated Advanced Baselne Imager (ABI) IR data s avalable.

2 I. Introducton The Atmospherc Infrared Sounder (AIRS) ( (Aumann et al. 003) on NASA s Earth Observng System (EOS) Aqua satellte s a hgh spectral resoluton (/Δ = 00) nfrared (IR) sounder wth 378 channels. AIRS measures radances n the IR regon μm, whch can be used to estmate the vertcal profles of atmospherc temperature and water profles [] from the Earth s surface to an alttude of 40 km wth a horzontal resoluton of 3.5 km at nadr. Due to ts relatvely poor spatal resoluton, the chance for an AIRS footprnt to be completely clear s less than 0% statstcally [3]. How to get soundngs (temperature and mosture vertcal profles) under cloudy skes becomes very mportant. There are several ways for handlng clouds n the soundng retreval process; one of the most effectve ways s to perform cloud-removal or cloud-clearng (CC) on the AIRS cloudy radance spectrum. The cloud-clearng doesn t requre modelng of clouds and s, therefore, a cost-effectve method of handlng cloudy nfrared scenes. For example, use mcrowave data for AIRS cloud-clearng (Sussknd et al. 003). One of the mportant questons s how to effectvely perform cloud-removal or cloud-clearng for the AIRS cloudy footprnts, whle stll retanng the sngle footprnt soundng gradent nformaton for Numercal Weather Predcton (NWP). Smth et al. (004) combne Moderate-Resoluton Imagng Spectroradometer IR clear radances and AIRS cloudy radances for cloud-clearng usng the tradtonal sngle band N * approach. L et al. (005) has developed an optmal CC algorthm usng mult-band MODIS for N *. In ther approach, the N * s determned from MODIS IR clear radances at 9 IR bands along wth the AIRS cloudy radances. Once N * s determned, the AIRS cloud-cleared radances (CCRs) can be obtaned by smply applyng N * to the cloudy radances from the two adjacent AIRS footprnts. The optmal cloud-clearng methodology s an extenson of the tradtonal sngle band cloud-clearng technque. By combnng collocated MODIS IR clear radance observatons and the AIRS cloudy radance measurements, the clear column radances for all AIRS channels can be retreved by usng the optmal cloud-clearng method. In order to make the optmal MODIS/AIRS cloud-clearng effectve, two adjacent AIRS cloudy footprnts (a par) wll be used to retreve one AIRS cloud-cleared radance (CCR) spectrum. For each partly cloudy AIRS footprnt (defned as prncpal footprnt n the center of 3 by 3 processng area), another adjacent AIRS cloudy footprnt (defned as supplementary footprnt) wll also be used to form a par for the optmal MODIS/AIRS cloud-clearng. The CCR spectrum from the par represents the clear column radance spectrum of the prncpal footprnt snce the qualty control (QC) s based on the comparson between the MODIS IR clear brghtness temperature (BT) observatons and the convoluted AIRS cloudcleared BT spectrum wth MODIS spectral response functons (SRFs) wthn the prncpal footprnt. Therefore, the prncpal footprnt must be partly cloudy to ensure that MODIS IR clear radance observatons exst, whle the supplementary footprnt can be ether partly cloudy or overcast. Snce the 3 by 3 box moves only by sngle pxel, each cloudy footprnt wll have a chance to be prncpal footprnt n the cloud-clearng process.

3 A case from Hurrcane Isabel s used for testng the algorthm. The MODIS clear BT observatons wthn all successful cloud-cleared footprnts are used for valdatng the AIRS cloud-cleared BTs convoluted to all the MODIS IR spectral bands wth SRFs. The bas and the standard devaton between the convoluted cloud-cleared BTs and the MODIS clear BT observatons are calculated; the bas and standard devaton s less than 0.5 K and 0.5 K, respectvely, for most MODIS IR spectral bands. Comparng wth the tradtonal sngle band N * cloud-cleared BTs, the standard devaton was sgnfcantly reduced wth optmal cloud-clearng for IR mddlewave (water vapor absorpton) and shortwave spectral regons, ndcatng the advantange of the mult-band algorthm over the sngle-band approach. Ths s expected snce the clear radance used for the sngle N * algorthm comes from the longwave based. The AIRS cloud-cleared BT spectrum s also compared wth ts nearby clear BT spectrum; the dfference s reasonable, consderng the effect of scene non-unformty. It s found that approxmately more than 50% of AIRS partly cloudy footprnts (or more than 30% of AIRS cloudy footprnts ncludng overcast cloudy footprnts) n ths study can be successfully cloud-cleared usng the optmal MODIS/AIRS cloud-clearng method. These postve results reveal the potental applcaton of ths method for the operatonal processng of hyperspectral IR sounder cloudy radance measurements for atmospherc temperature, mosture and ozone soundng retreval when IR mager data s avalable, or real tme processng of the MODIS/AIRS data from drect broadcast. Usng MODIS to help AIRS sub-pxel cloud characterzaton and cloud property retreval has been nvestgated (L et al. 004a, 004b, 005). These technques that utlze AIRS and MODIS data are also applcable to the processng of data from the future operatonal Advanced Baselne Imager (ABI) [0] and Hyperspectral Envronmental Sute (HES) systems (Gurka and Schmt 004; Gurka et al. 004) on the next generaton of Geostatonary Operatonal Envronmental Satellte (GOES)-R where no mcrowave soundng measurements are avalable. II. MODIS/AIRS collocaton and AIRS cloud maskng from MODIS A. Collocaton Between MODIS and AIRS Measurements AIRS spatal coverage s provded by the scan head assembly, contanng a cross track rotary scan mrror and calbrators. The scan mrror has two speed regmes: Durng the frst seconds t rotates at 49.5 degree/second, generatng a scan lne wth 90 ground footprnts, each wth a. degree dameter feld-of-vew (FOV). Durng the remanng seconds the scan mrror fnshes the remanng 6 degrees of a full revoluton. The AIRS spatal dstrbuton s used n the collocaton between the MODIS and AIRS measurements, whch s the frst step for the optmal MODIS/AIRS cloud-clearng. The MODIS pxels wth km spatal resoluton are collocated wthn an AIRS footprnt. Several collocaton algorthms have been developed that are based on the scannng geometry of two nstruments flown on the same satellte (Nagle 998; Frey et al. 995). Wth a set of AIRS earth-located observatons, the footprnt of each AIRS observaton descrbes a fgure that s crcular at nadr, quas-ellpsodal at ntermedate scan angles, and ovular at extreme scan angles. The dameter of the AIRS footprnt at nadr s approxmately 3.5 km. Dependng on the angular dfference between the AIRS and

4 MODIS slant range vectors, a weght (ω) s assgned to each MODIS pxel collocated to AIRS: f the MODIS pxel les at the center of the AIRS oval, and zero f at the outer edge. The collocaton s modeled correctly and the algorthm provdes an accuracy better than km, provded that the geometry nformaton from both nstruments s accurate. B. AIRS Cloud Maskng Usng MODIS Once the MODIS pxels are collocated wth the AIRS fooprnts, the cloud propertes wthn the AIRS sub-pxel can be characterzed usng the MODIS cloud mask, cloud phase mask (Kng et al. 003; Platnck et al. 003), and the MODIS classfcaton mask (L et al. 003). The AIRS cloud mask, cloud phase mask, as well the cloud-layer nformaton mask can be generated from MODIS products wth km spatal resoluton (L et al. 004a). For each AIRS footprnt, a clear coverage (0 ~ ) s created by accountng for the percentage of MODIS pxels wth confdent clear and probably clear (Ackerman et al. 998) wthn the footprnt. Only confdent clear MODIS pxels wthn the AIRS footprnt are averaged for optmal MODIS/AIRS cloud-clearng; the averaged MODIS clear radance for a gven spectral band wthn an AIRS footprnt s obtaned by R clr M = np np clr, l ωl RM l= l= / ω, () l clr l where R, M s the radance of confdent clear pxel l, np s the number of confdent clear pxels wthn the AIRS footprnt, and ω l s the weght of pxel l wthn the AIRS footprnt as descrbed n secton II.A. III. Methodology of optmal MODIS/AIRS cloud-clearng R R Gven that and are the AIRS cloudy radance spectra as a functon of wavenumber from two adjacent footprnts (footprnt s the prncpal, whle footprnt s the supplementary), assume that the two adjacent footprnts have (a) the same atmospherc temperature and mosture profles, (b) the same surface skn temperatures and surface IR emssvty spectrum, (c) the same cloud-top heght, whle the only dfference s the effectve cloud emssvty ( Nε c, the product of the cloud emssvty, ε c, and the fractonal cloud coverage, N ). The cloudy radance spectra of the two adjacent footprnts can be expressed as clr ovc R = ( Nεc ) R + N εc R, () clr ovc R = ( Nεc ) R + N εc R, (3) clr where subscrpt c denotes the cloud, and R ovc and R are the clear and overcast column radance spectra, respectvely, for both footprnts and. Rewrte Eq.() and Eq.() as clr ovc clr R R = N εc ( R R ), (4) clr ovc clr R R = N εc ( R R ). (5) From Eq.(4) and Eq.(5), we have

5 R R R R clr clr = N ε N c εc = N * (6) Eq.(6) defnes N * (Smth 968), whch s ndependent of wavenumber assumng ε c = εc (.e., the cloud propertes are the same for footprnts and ). If N * s known or cc estmated, then the AIRS CCR spectrum can be retreved by R * R R N R cc = (7) * N The key queston s how to determne N *. In order to determne N *, one needs to know the clear radance at a certan wavenumber wth good accuracy. Smth et al. (004) used the MODIS IR wndow spectral band (band 3, μm) clear radance together wth the convoluted AIRS cloudy radances wth MODIS μm SRF wthn the two adjacent footprnts to determne N *, that s clr f N * 3( R ) RM 3 =, (8) clr f ( R ) R 3 M 3 clr where f s the SRF for MODIS IR band, and s the averaged observed clear radance for MODIS IR spectral band for the two adjacent AIRS footprnts (see Eq.()). Once N * s determned by Eq. (8), the AIRS clear column radance spectrum for ths footprnt par then can be retreved (see Eq.(7)). Eq.(8) s the tradtonal way to estmate N * usng a sngle IR wndow spectral band. The N * value determned from a sngle MODIS spectral band, usng dfferent MODIS IR wndow spectral bands, may result n dfferent cloud-cleared spectra. The results are also hghly dependent on the qualty control (QC) for the sngle band N * cloud-clearng. In order to obtan an optmal N * value and make the cloud-clearng result dependent on an objectve qualty control crteron based on all MODIS spectral bands, an optmal cloud-clearng methodology s developed. In the optmal MODIS/AIRS cloud-clearng, the defnton of N * n Eq.(6) s stll adopted. However, the N * value s determned by smultaneously mnmzng the dfferences between the MODIS IR clear radance observatons and the convoluted AIRS CCR spectrum wth SRFs for all 9 MODIS IR spectral bands wthn footprnt (the prncpal footprnt). The cost functon for the optmal cloud-clearng s defned as clr cc J ( N*) = [( R ( )] M f R. (9) σ Snce the CCR spectrum, J ( N*) = σ R M cc R, s a functon of N * (see Eq.(7)), Eq.(9) becomes R R N * clr [( R f ( )] M * N, (0) where σ s the radance detector nose (NEdR) for MODIS band. Gven that the SRF f s a lnear operator for MODIS IR spectral band, Eq.(0) becomes

6 J * clr ( N*) = [( R ( ) ( )] M f * R + f R *. () σ N N Therefore, the frst order dervatve of J (N * ) wth respect to N * can be derved from Eq.() as * * J ( N ) clr N = [ RM f ( R ) + f ( R )][ f ( R ) f ( R )] * * * * N ( N ) σ N N () The soluton of N * * should mnmze the cost functon J ( N ) ; therefore, * J ( N ) = 0. (3) * N N * then can be analytcally solved from Eq.(3) as clr [ f ( R ) RM ][ f ( R ) f ( R )] * σ N =. (4) clr [ f ( R ) RM ][ f ( R ) f ( R )] σ Once N * s determned by Eq.(4), the cloud-cleared AIRS radance spectrum, arbtrarly located at the locaton of the prncpal footprnt, can be retreved by applyng N * to Eq. (7). In ths study, the 9 MODIS IR spectral bands, 4, 5, 8, 30, 3, 3, 33, and 34 are used. Note that MODIS IR spectral bands 0, 3 and 7 are not used due to the convoluton error ntroduced by the spectral gaps n the AIRS measurements, MODIS IR spectral band s not used due to the large detector nose, whle MODIS IR spectral IR bands 35 and 36 are not used due to ther SRF calbraton error (Tobn et al. 004). Fgure shows the MODIS SRFs overlayed wth an AIRS BT spectrum. Snce AIRS does not have an 8.5 μm spectral regon channel, MODIS band 9 s excluded n the determnaton of N *. The advantage of Eq.(4) s that the multple MODIS spectral bands are weghted n the N * calculaton. There are two weghtng factors, the detector nose accounts for the observaton error n the N * calculaton, whle the radance contrast between the two adjacent footprnts accounts for the cloud heght effect n the N * calculaton. For example, f the par contans low clouds, the MODIS mddle and upper level bands 33 and 34 wll have less weght because f ( R ) f ( R ) s small; however, when the par contans hgh clouds, bands 33 and 34 wll nfluence the N * calculaton. Besdes the cloud heght effect, cloud fracton s another factor contrbutng to the spectral weghtng that maxmzes the nformaton content of the optmal cloud-clearng approach. IV. Applcaton of optmal MODIS/AIRS cloud-clearng MODIS and AIRS data from Hurrcane Isabel are used n the study. The AIRS cloud mask s derved from the MODIS cloud mask wth km spatal resoluton. A fractonal coverage (0 ~ ) s derved to represent clear (0) to cloudy () for a gven AIRS footprnt. The AIRS footprnts or felds-of-vew (FOVs) then can be classfed as fully cloudy (overcast), partly cloudy, and clear. Only AIRS footprnts classfed as partly N

7 cloudy are used for cloud-clearng. The upper left panel of Fgure shows the AIRS BT mage of channel 560 ( cm - ) on 7 September 003 (AIRS granule 84); the presence of cold clouds s ndcated by the blue (cold) colors. The lower left panel shows the AIRS coverage of clear, partly cloudy and overcast cloudy footprnts derved from the MODIS cloud mask wth km spatal resoluton (see upper rght panel). For comparson, the MODIS cloud phase mask wth km spatal resoluton s also ncluded (see lower rght panel). It can be seen that most partly cloudy footprnts contan water clouds. Note that the color bars have dfferent defntons n the four panels. Fgure 3 shows the dagram of the prncpal (center) footprnt and ts 8 surroundng supplementary footprnts. The steps of optmal cloud-clearng are: Step : For each partly cloudy AIRS footprnt (prncpal footprnt l ), fnd ts nearby cloudy footprnt n any drecton (maxmum eght nearby cloudy footprnts) k (k =,, 3,, 8); * Step : For each par ( l, k), calculate N ( k), ( k =,, 3,, 8) usng Eq.(4); Step 3: Calculate R cc (k) ( k =,, 3,, 8) from Eq. (7) Step 4: Calculate clr cc RES( k) = [( RM f ( R ( k))] σ Step 5: Fnd k m, that makes RES( k m ) = mn RES( k). Step 6: Apply qualty control to the selected R cc ( k m ), and calculate TBRMS = I [ T ( R b clr M ) T b ( f ( R cc ( km )))], (5) where I s the total number of MODIS spectral bands used for QC, T b s the functon that converts the radance to BT for MODIS spectral band. The optmal cloud-clearng s successful only when TBRMS < 0.5 K, otherwse the cloud-clearng for ths prncpal footprnt s rejected. The same 9 MODIS IR spectral bands used for N * determnaton (see Eq.(4)) are also used also for the QC. The CCR radance spectrum R cc ( k m ) s the fnal clear column radance spectrum for the cloudy footprnt l (the prncpal footprnt). The footprnt ndex l starts from the frst footprnt to the last footprnt of the AIRS granule. In order to make accurate CCRs, qualty control must be appled to R cc ( k m ). If the root mean square (RMS) dfference between the MODIS IR clear BT observatons and convoluted AIRS cloud-cleared BTs (from R cc ( k m ) ) wth MODIS SRFs wthn the prncpal footprnt s greater than 0.5 K (as defned n Eq.(5)), the CCR spectrum s then rejected. The optmal cloud-clearng method s appled to all AIRS footprnts that are partly cloudy. The optmal cloud-clearng wll not be performed f the number of MODIS clear pxels s less than 0% of the total number of MODIS pxels wthn ths partly cloudy AIRS footprnt. Fgure 4 shows BT mages of MODIS band 8 (7.3 μm) convoluted from the AIRS clear footprnts (upper left panel) and the one convoluted from AIRS clear plus successful cloud-cleared footprnts (upper rght panel); the MODIS clear

8 BT observatons wth km spatal resoluton are also shown (lower rght panel). It can be seen that cloud-cleared footprnts fll many areas where clear AIRS radances are not avalable, especally over Florda and Cuba, whch llustrates that soundngs are acheved over Florda and Cuba and ther nearby oceanc areas from AIRS cloudy radance measurements wth help of hgh spatal resoluton of MODIS. In addton, the CCRs are also avalable over land areas wthn the states of Alabama and Tennessee. The lower left panel of Fgure 4 shows the percentages of clear AIRS footprnts, AIRS footprnts wth optmal cloud-clearng successful (CC-S) (pass the qualty control), AIRS footprnts wth optmal cloud-clearng fal (CC-F), and overcast AIRS footprnts (cloud-clearng wll not be performed). It can be seen that.6% of AIRS footprnts (each AIRS granule contans 35 scan lnes, each scan lne has 90 footprnts) are successfully cloud-cleared, or more than 30% of AIRS cloudy footprnts are successfully cloud-cleared. Snce only partly cloudy footprnts are used for cloud-clearng attempts, the success rate s more than 50% n ths partcular case. The tradtonal sngle band N * cloud-clearng method s also appled; MODIS spectral band 3 ( μm) s used for N * determnaton whle the 9 MODIS IR spectral bands, 4, 5, 8, 30, 3, 3, 33, and 34 are used for qualty control and a smlar success rate of optmal cloud-clearng s acheved. Fgure 5 shows the bas and standard devaton between the MODIS clear BT observatons and convoluted AIRS cloudclearng BTs from all footprnts wth cloud-clearng successfully performed. Results from tradtonal sngle band N * cloud-clearng approach and optmal cloud-clearng method are shown n the fgure. For optmal cloud-clearng method, the cloud-clearng bas for MODIS bands, 3, 5, and 30 through 34 are very small (less than 0.5 K); the bas for bands 4 and 8 are slghtly larger but stll less than 0.5 K. However, bas for bands 0, 7, 35 and 36 are relatvely large (greater than.0 K) due to the convoluton bas [9] mentoned above. The AIRS poppng channels and/or the AIRS channel gap n the spectral regon cause the convoluton bas. MODIS bands 35 and 36 mght also have an SRF calbraton bas. Those bases are removable provded that the relable estmates are avalable. The standard devaton s very small (less than 0.5 K) for almost all the MODIS IR spectral bands (only band 7 s slghtly larger than 0.5 K), whch ndcates the good agreement between the MODIS clear BT observatons and AIRS cloud-cleared BTs. For the tradtonal sngle band N * cloud-clearng approach, the bas between the MODIS clear BT observatons and AIRS cloud-cleared BTs s smlar to that from the optmal cloud-clearng method; however, the standard devatons for MODIS shortwave bands 0 through 3 are worse than that of the optmal cloud-clearng method. Tradtonal sngle band N * approach s better n lowgwave IR wndow regon (μm) than the optmal cloud-clearng method because t uses MODIS μm spectral band for N * calculaton. The two methods have smlar performance for the other mddleand longwave spectral bands except for spectral band 3 for whch the optmal approach performs better. Although the success rate of cloud-clearng s smlar between the two methods, optmal cloud-clearng results are much closer to the MODIS clear observatons for MODIS shortwave spectral bands. In the tradtonal sngle band N * cloud-clearng method, only band 3 ( μm) s used for N * calculaton, and IR shortwave spectral effects mght not be fully accounted for n dervng N *. Wth the optmal cloud-clearng method, 9 MODIS IR spectral bands are used smultaneously to balance the N *

9 spectrally, therefore, cloud-clearng results should be optmal when compared wth the MODIS clear observatons. W. L. Smth (personal communcaton) suggested usng two N * s n cloud-clearng (e.g., one N * from MODIS spectral bands, 4, and 5 for AIRS IR shortwave, whle the other N * from MODIS spectral bands 8, 30, 3, 3, 33, and 34 for AIRS IR mddlewave and longwave). Two N * s should mnmze the effects of coregstraton errors and perform better than one N *. Fgure 6 shows the scatterplot between the MODIS IR clear BT observatons and the convoluted AIRS cloud-cleared BTs wth the tradtonal sngle band N * cloudclearng approach and the optmal cloud-clearng method for MODIS spectral bands (upper left), 5 (upper rght), 7 (lower left) and 3 (lower rght). Approxmately 400 AIRS partly cloudy footprnts are ncluded n the plot. It can be seen that the optmal cloud-clearng method creates less scatter and has a larger correlaton wth MODIS IR clear observatons for MODIS IR spectral band (3.95 μm). Comparng the AIRS cloud-cleared BT spectrum and ts nearby clear footprnt BT spectrum also helps to evaluate the performance of cloud-clearng [3]. The upper and mddle panels of Fgure 7 show the bas and standard devaton between the AIRS cloudcleared BT spectra and ther nearby clear footprnt BT spectra over water where the atmosphere and surface are assumed homogenous between the two clear adjacent footprnts. It can be seen that the standard devaton of AIRS cloud-cleared BTs s less than K for most spectral regons. Part of the standard devaton s due to the atmospherc non-homogenety dfference between the two adjacent footprnts. The lower panel of Fgure 7 shows the root mean square dfference (RMSD) between the twoadjacent clear AIRS footprnt pars over water of the entre granule. It can be seen that the BT dfference can be K n the shortwave spectral regon due to the non-unformty of the atmospherc and surface between the two adjacent clear AIRS footprnts. Therefore, the actual standard devaton of the AIRS cloud-cleared BT should be much smaller than that shown n the mddle panel of Fgure 7. Fgure 8 shows the MODIS cloud mask wth km spatal resoluton supermposed to the AIRS footprnts from a small area over Florda (see the box on the upper left panel of Fgure ), the footprnts ndcated by arrows are the prncpal and ts supplementary cloudy footprnts used for optmal cloud-clearng, and ts nearby clear footprnt for comparson. The lower panel of Fgure 8 shows the optmal cloud-cleared BT spectrum (red lne) from the two adjacent cloudy BT spectra (black and green lnes), as well as the nearby clear BT spectrum (blue lne) (lower panel). It can be seen that the optmal cloud-cleared AIRS BTs are much warmer than those of the two adjacent cloudy footprnts, ndcatng the successful removal of cloud effects n the optmal cloud-clearng. The cloud-cleared BT spectrum s also reasonably close to ts nearby clear BT spectrum, dfferences n wndow regon mght be caused by the non-unformty of surface skn temperature n the two footprnts.

10 V. Dscussons Unlke usng AMSU n AIRS cloud-clearng, an algorthm for AIRS cloudclearng usng MODIS data s developed. Snce both the mager and sounder measure radances at the same IR spectral regons, a drect relatonshp between an mager IR radance and a sounder IR radance spectrum for a gven mager spectral band provdes unque advantages for mager/sounder cloud-clearng. The advantages of mager/sounder cloud-clearng are: () t s easy to fnd mager clear pxels wthn the sounder footprnt whch s crtcal for the N * calculaton and the qualty control on CCRs; () cloudclearng can be acheved on a sngle footprnt bass (hence mantanng the spatal gradent nformaton); and (3) mager IR clear radances provde tropospherc atmospherc nformaton that enhances the effectveness of cloud-clearng for IR sounder cloudy radances. However, there are also lmtatons: () the cloud-clearng can only be done wth partly cloudy footprnts, () N * has to be constant for the whole IR spectrum, and (3) the surface and the atmospherc profle must be homogeneous wthn the par of two adjacent footprnts. The cloud-clearng mght fal when one of the above physcal assumptons fals. For example, n the presence of ce clouds, the cloud-clearng mght gve spectrally nconsstent results due to N * not beng constant n the IR spectral regon. In addton, when the radance contrasts n the IR wndow regon s too small, N * tends to be and the cloud-clearng can amplfy nose rather than remove the cloud effects on the measurements. However, all these falures wll be fltered by the qualty control procedure (see Eq.(5)). The qualty control s very mportant to assure that the fnal CCRs representng the clear part of the prncpal footprnt ft the MODIS IR clear radance observatons n all IR spectral regons smultaneously. Certan factors wll also affect the cloud-clearng results. The operatonal MODIS cloud mask wth km spatal resoluton used for determnng the MODIS clear IR radances wthn the prncpal footprnt s crtcal for the success of the cloud-clearng. Only confdent clear MODIS pxels should be selected to determne the MODIS IR clear radances. The daytme MODIS cloud mask should be more accurate than the nghttme mask because of the use of vsble (VIS) and near IR (NIR) bands. MODIS/AIRS collocaton s also very mportant to assure the success of the cloud-clearng; mscollocated MODIS clear pxels wth the AIRS footprnt may result n addtonal cloudclearng errors. A relable algorthm s necessary to provde good mager/sounder collocaton. The tme dfference between the mager and sounder s an ssue n cloudclearng. Snce MODIS and AIRS are n the same spacecraft, the tme dfference between MODIS and AIRS for a gven AIRS footprnt s small enough to assure the same clouds and atmospherc measurements are observed. The convoluton error due to the AIRS spectral gaps or bad channels that are excluded n the calculaton s another error source for MODIS/AIRS cloud-clearng. Tobn et al. [9], [0] have estmated the bas due to the AIRS spectral gaps or bad channels; they found that wndow regon bands, 3 and 3 have less bas, longwave CO bands 33 throught 36 have mean bases ncreasng to K for band 36, and water vapor bands 8 and 7 have mean bases of 0.48 and.05 K respectvely. Those bases have not been taken nto account yet when applyng MODIS SRFs. The SRF and gap ssues could be mtgated f a forward model for both nstruments s used [0]. There are also other sources of error for MODIS/AIRS

11 cloud-clearng such as MODIS and AIRS calbraton errors that need to be consdered when nterpolatng and usng the MODIS/AIRS CCRs. VI. Conclusons and future work In ths paper, optmal cloud-clearng for sounder cloudy radances usng mager IR clear radances has been successfully demonstrated by usng AIRS and MODIS. About 30% of AIRS cloudy footprnts (or 50% of the partly cloudy footprnts) are successfully cloud-cleared wth the help of MODIS hgh spatal resoluton data. In the optmal mager/sounder cloud-clearng, the mager provdes a cloud mask for sounder footprnts whle the multspectral mager IR provdes clear radance observatons to synergstcally determne N* and to be used as qualty control. The followng conclusons can be drawn from ths study: () Optmal mager/sounder cloud-clearng usng mult-spectral mager IR spectral bands has advantages over the tradtonal sngle band N * cloud-clearng approach. () MODIS IR spectral bands, 4, 5, 8, and 30 through 34 are used to determne N * and qualty control; more than 30% cloudy footprnts (or more than 50% of partly cloudy footprnts) are successfully cloud-cleared wth the help of MODIS. The success rate s above 50%. (3) The convoluted AIRS CCRs are compared wth MODIS IR clear radance observatons. The bas s less than 0.5 K for most MODIS IR spectral bands, whle the standard devaton s less than 0.5 K for almost all the MODIS IR spectral bands. (4) CCRs are compared wth ther nearby clear AIRS radances. The standard devaton s wthn K for most AIRS channels. Part of the standard devaton s due to the natural (atmospherc and surface) varablty of the two clear adjacent footprnts. Ths work s very effectve for cloud-clearng the AIRS footprnts contamnated by water clouds. The approach could be employed on GOES-R wth HES sounder and ABI (e.g., ABI 3.9, 6.5, 7.0, 7.4, 8.5, 9.73, 0.35,.,.3 and 3.3 μm bands are crtcal for HES/ABI cloud-clearng). Future work wll focus on cloud-clearng for footprnts wth mutl-layer clouds. For example, employng mult-layer cloud-clearng approach developed by Sussknd et al. [] and usng MODIS clear observatons for hgh spatal resoluton cloud-clearng could be of great potental References Aumann, H. H., M. T. Chahne, C. Gauter, M. D. Goldberg, E. Kalnay, L. M. McMlln, H. Revercomb, P. W. Rosenkranz, W. L. Smth, D. H. Staeln, L. L. Strow, and J. Sussknd, 003: AIRS/AMSU/HSB on the Aqua msson: desgn, scence objectves, data products, and processng systems. IEEE Trans on Geosc. and Remote Sensng, vol. 4, pp Ackerman, S. A., K. I. Strabala, W. P. Menzel, R. A. Frey, C. C. Moeller, and L. E. Gumley, 998: Determnaton of clear sky from clouds wth MODIS, J. Geophys. Res., vol. 03, No. D4, pp

12 Frey, R. A., S. A. Ackerman, and B. J. Soden, 995: Clmate parameters from satellte spectral measurements. Part I: Collocated AVHRR and HIRS/ observatons of spectral greenhouse parameter, J. Clm., vol. 9, pp Gurka, J. J., and T. J. Scmht, 004: Baselne nstruments planned for the GOES-R seres, 3 th Conference on Satellte Meteorology, Norfolk, VA, 0-3 September 004 (preprnts). Paper P.5, Amercan Meteorologcal Socety, Boston, MA. Gurka, J. J., T. J. Schmt, and R. R. Reynolds, 004: Hghlghts from the second GOES users' conference: Recommendatons for the GOES-R seres, Internatonal Conference on Interactve Informaton and Processng Systems (IIPS) for Meteorology, Oceanography, and Hydrology, 0th, Seattle, WA, - 5 January 004 (preprnts), paper P., Amercan Meteorologcal Socety, Boston, MA. Kng, M. D., W. P. Menzel, Y. J. Kaufman, D. Tanre, B. Gao, S. Platnck, S. A. Ackerman, L. A. Remer, R. Pncus, and P. A. Hubanks, 003: Cloud and aerosol propertes, precptable water, and profles of temperature and water vapor from MODIS. IEEE Transactons on Geoscence and Remote Sensng, vol. 4, pp L, J., W. P. Menzel, Z. Yang, R. A. Frey, and S. A. Ackerman, 003; Hgh spatal resoluton surface and cloud type classfcaton from MODIS mult-spectral band measurements, J. Appl. Meteorol., vol. 4, pp L, J., W. P. Menzel, F. Sun, T. J. Schmt, and J. Gurka, 004a: AIRS sub-pxel cloud characterzaton usng MODIS cloud products, J. Appl. Meteorol. Vol. 43, pp L, J., W. P. Menzel, W. Zhang, F. Sun, T. J. Schmt, J. J. Gurka, and E. Wesz, 004b: Synergstc use of MODIS and AIRS n a varatonal retreval of cloud parameters, J. Appl. Meteorol., vol. 43, pp L, J., H.-L. Huang, C.-Y. Lu, P. Yang, T. J. Schmt, H. We, E. Wesz and W. P. Menzel, 005: Retreval of cloud mcrophyscal propertes from MODIS and AIRS, J. Appl. Meteorol. (n press). L, J., C.-Y. Lu, H.-L. Huang, T. J. Schmt, X. Wu, W. P. Menzel, and J. J. Gurka, 005: Optmal Cloud-Clearng for AIRS radances usng MODIS, IEEE Trans. Geoscence and Remote Sensng, 43, Nagle, F. W., 998: The Assocaton of dsparate satellte observatons, Proceedng of the Second Symposum on Integrated Observng Systems, 6 January 998, Phoenx, AZ, Amercan Meteorologcal Socety, Boston, MA, P Platnck, S., M. D. Kng, S. A. Ackerman, W. P. Menzel, B. A. Baum, J. C. Red, and R. A. Frey, 003: The MODIS cloud products: algorthms and examples from Terra. IEEE Transactons on Geoscence and Remote Sensng, vol. 4, pp Schmt, T., J. M. M. Gunshor, W. Paul Menzel, J. L, S. Bachmeer, and J. J. Gurka, 005: Introducton the next-generaton advanced baselne mager (ABI) on Geostatonary Operatonal Envronmental Satelltes (GOES)-R, Bull. Amercan Meteorol. Socety (n press). Smth, W. L., 968: An mproved method for calculatng tropospherc temperature and mosture from satellte radometer measurements, Mon. Wea. Rev. vol. 96, pp

13 Smth, W. L., D. K. Zhou, H.-L. Huang, Jun L, X. Lu, and A. M. Larar, 004: Extracton of profle nformaton from cloud contamnated radances, Workshop on Assmlaton of hgh spectral resoluton sounders n NWP, 8 June July 004, ECMWF, Shnfeld Park, Readng, p Sussknd, J., C. D. Barnet, J. M. Blasdell, 003: Retreval of atmospherc and surface parameters from AIRS/AMSU/HSB data n the presence of clouds, IEEE Trans. Geosc. and Remote Sensng, vol. 4, pp Tobn, D. C., H. E. Revercomb, C. C. Moeller, R. O. Knuteson, F. A. Best, W. L. Smth et al., 004: Valdaton of Atmospherc InfraRed Sounder (AIRS) spectral radances wth the Scannng Hgh-resoluton Interferometer Sounder (S-HIS) arcraft nstrument, Remote Sensng of Clouds and the Atmosphere IX, Proceedngs of SPIE 557, 3-5 September 004, Maspalomas, Gran Canara, Span. Fgure. The MODIS SRFs overlayed wth an AIRS BT spectrum. The unts of the abscssa and ordnate are wavenumber and brghtness temperature, respectvely.

14 Fgure. The AIRS BT mage of channel 560 ( cm -, approxmately μm) on 7 September 003 (upper left panel), the MODIS cloud mask wth km spatal resoluton (upper rght panel), the AIRS coverage of clear, partly cloudy and overcast cloudy footprnts (lower left panel), and the MODIS cloud phase mask wth km spatal resoluton (lower rght panel).

15 3 5 8 CC R 3 CC5 R CC 8 R R CC R CC 7 7 CC 6 CC CC 4 R R R 4 6 Fgure 3. The dagram of the prncpal footprnt and ts 8 surroundng supplementary footprnts n the optmal cloud-clearng procedure.

16 6.35% 30.75% 0.30%.60% Clear CC-S CC-F Full Cloud Fgure 4. The BT mages of MODIS band 8 (7.3 μm) convoluted from the AIRS clear footprnts (upper left panel), the one convoluted from AIRS clear plus successful cloudcleared footprnts (upper rght panel), the MODIS clear BT observatons wth km spatal resoluton (lower rght panel), and the percentages of clear AIRS footprnts, AIRS footprnts wth vartatonal cloud-clearng successful (CC-S), AIRS footprnts wth optmal cloud-clearng fal (CC-F), and the overcast AIRS footprnts.

17 Fgure 5. The bas and standard devaton between the MODIS clear BT observatons and convoluted AIRS cloud-clearng BTs from all footprnts wth both the optmal cloudclearng and the sngle band N * cloud-clearng successful.

18 Fgure 6. The scatterplot between the MODIS IR clear BT observatons and the convoluted AIRS cloud-cleared BT wth the tradtonal sngle band N * cloud-clearng approach and the optmal cloud-clearng method for MODIS spectral bands (upper left), 5 (upper rght), 7 (lower left) and 3 (lower rght).

19 Fgure 7. The bas (upper panel) and standard devaton (mddle panel) between the AIRS cloud-cleared BT spectra and ther nearby clear footprnt BT spectra along wth the RMSD between the two-adjacent clear AIRS footprnt pars over the water of the granule.