Validation of AIRS/AMSU - A Water Vapor and Temperature Data With In Situ Aircraft Observations From the Surface to UT/LS From 87 N 67 S

Transcription

1 Sn Jose Stte University From the SeletedWorks of Minghui Dio June 26, 2013 Vlidtion of AIRS/AMSU - A Wter Vpor nd Temperture Dt With In Situ Airrft Oservtions From the Surfe to UT/LS From 87 N 67 S Minghui Dio, Prineton University Loyeh Jumm, Sonom Tehnology, In. Justin Sheffield, Prineton University Eri F. Wood, Prineton University Mrk A. Zondlo, Prineton University Aville t:

2 JOURNAL OF GEOPHYSICAL RESEARCH: ATMOSPHERES, VOL. 118, , doi: /jgrd.50483, 2013 Vlidtion of AIRS/AMSU-A wter vpor nd temperture dt with in situ irrft oservtions from the surfe to UT/LS from 87 N 67 S Minghui Dio, 1,2 Loyeh Jumm, 3 Justin Sheffield, 1 Eri F. Wood, 1 nd Mrk A. Zondlo 1,2 Reeived 15 Jnury 2013; revised 8 My 2013; epted 9 My 2013; pulished 26 June [1] Vlidtion of the Atmospheri Infrred Sounder (AIRS)/Advned Mirowve Sounding Unit (AMSU-A) dt set with in situ oservtions provides useful informtion on its pplition to limte nd wether studies. However, different spe/time verging windows hve een used in pst studies, nd questions remin on the vrition of errors in spe, suh s etween lnd/oen nd the Northern/Southern Hemispheres. In this study, in situ irrft mesurements of wter vpor nd temperture re ompred with the AIRS/ AMSU-A retrievls (Version 5 Level 2) from 87 Nto67 S nd from the surfe to the upper troposphere nd lower strtosphere (UT/LS). By using smller omprison window (1 h nd 22.5 km) thn previous studies, we show tht the solute perentge differene of wter vpor ( dh 2 Oper ) is ~20 60% nd the solute temperture differene ( dtemp ) is ~ K. The lnd retrievls show improvements versus Version 4 y ~5% in wter vpor onentrtion nd ~0.2 K in temperture t mr. The lnd (oen) retrievls re older nd drier (wrmer nd moister) thn the in situ oservtions in the oundry lyer, wrmer nd drier (wrmer nd moister) t the UT/LS. No signifint differenes etween hemispheres re noted. Overll, future omprisons re suggested to e done within 4 h nd 100 km in order to keep the errors from window sizes within ~10%. To onstrin the unertinties in previous vlidtion results, we show tht every 22.5 km (or 1 h) inrement in window sizes ontriutes to ~2% dh 2 Oper nd ~0.1 K dtemp inreses. Cittion: Dio, M., L. Jumm, J. Sheffield, E. F. Wood, nd M. A. Zondlo (2013), Vlidtion of AIRS/AMSU-A wter vpor nd temperture dt with in situ irrft oservtions from the surfe to UT/LS from 87 N 67 S, J. Geophys. Res. Atmos., 118, , doi: /jgrd Introdution [2] Glol mesurements of wter vpor nd temperture re ritil for ssessing limte nd wether models. The NASA Atmospheri Infrred Sounder (AIRS) nd the Advned Mirowve Sounding Unit (AMSU-A) ord the Aqu stellite provide glol mesurements of wter vpor nd temperture twie per dy sine My The AIRS/AMSU-A dt sets lso over wide rnge of vertil levels from the surfe (1100 mr) to the strtosphere (50 mr for wter vpor; 0.1 mr for temperture). With glol overge nd vertil smpling, the AIRS/AMSU-A oservtions provide unique dt set tht hs een pplied in vlidtions of multiple limte Additionl supporting informtion my e found in the online version of this rtile. 1 Deprtment of Civil nd Environmentl Engineering, Prineton University, Prineton, New Jersey, USA. 2 Center for Mid-Infrred Tehnologies for Helth nd the Environment, Prineton University, Prineton, New Jersey, USA. 3 Sonom Tehnology, In., Petlum, Cliforni, USA. Corresponding uthor: M. A. Zondlo, Deprtment of Civil nd Environmentl Engineering, Prineton University, E206A Engineering Qud., Prineton, NJ 08544, USA. (mzondlo@prineton.edu) Amerin Geophysil Union. All Rights Reserved X/13/ /jgrd nd wether forest models [Golderg et l., 2003; Pgno et l., 2004; Chhine et l., 2006; Piere et l., 2006; Fsullo nd Trenerth, 2012; Jing et l., 2012; Tin et l., 2013]. The AIRS/AMSU-A wter vpor nd temperture mesurements hve lso een used to nlyze the glol distriutions of reltive humidity nd ie supersturtion [Gettelmn et l., 2006, 2006; Khn et l., 2007, 2009; Lmquin et l., 2012]. Other nlyses sed on AIRS/AMSU-A oservtions, suh s the wter vpor nd temperture vrine sling, hve lso een used to ompre with [Khn et l., 2011] or improve [Cusk et l., 2007] su-grid sle model prmeteriztions. Besides the glol-sle nlyses, AIRS/AMSU-A dt hve een used to understnd regionl limtologies. For exmple, the AIRS/AMSU-A retrievls of lnd surfe reltive humidity nd moisture provide n essentil tool for understnding the hydrologi yles over ontinentl res, espeilly for regions with srity of dily meteorologil mesurements, suh s su-shrn Afri [Ferguson nd Wood, 2010]. [3] The vrious pplitions of AIRS/AMSU-A dt rely on the ury of the oservtions. Thus, there is n ongoing need to ssess nd improve the stellite oservtions y ompring with ground-sed or irorne mesurements. Previous vlidtions inlude omprisons with rdiosonde oservtions [Divkrl et l., 2006; MMillin et l., 2007], dropsonde oservtions [Wu, 2009], nd in situ irrft oservtions 6816

3 Tle 1. Comprison Shemes nd Dt Resolution of Previous nd Current Work Referene In Situ Dt Sptil Coverge Resolution of Averged In Situ Dt Dt Version Sptil Window Temporl Window Gettelmn et l. [2004] Airrft (PreAVE) 5 S 40 N, Texs nd Cost Ri 50 s flight dt V ox Sme lendr dy Divkrl et l. Rdiosonde Glol point mesurements Vertilly verged V4 100 km rdius 3 h [2006] to AIRS in Wu [2009] Dropsonde 16 W 34 W, 10 N 22 N Vertilly verged to AIRS in V ox 4.5 h dytime psses Lmquin Airrft (MOZAIC) 120 W 150 E, 30 S 90 N 1 min flight dt V5 Within AMSU footprint 30 min et l. [2012] ~22.5 km rdius Current work Airrft (START08 nd HIPPO Glol #1 5) deployments DIAO ET AL.: GLOBAL AIRS/AMSU-A H 2 O AND T VALATIONS 87 N 67 S, 84 W 180 W 128 E North Ameri, entrl Pifi Oen 100 s flight dt V5 Compre vrious sizes of sptil windows Compre vrious sizes of temporl windows [Gettelmn et l., 2004; Lmquin et l., 2012]. For over-lnd vlidtion, the AIRS/AMSU-A surfe ir temperture dt hve een ompred with ground meteorologil sttions [Go et l., 2008] s well s other stellite remote sensing oservtions, suh s the Advned Mirowve Snning Rdiometer for EOS [Jones et l., 2010]. Rdiosonde oservtions hve lso een ompred with the AIRS/AMSU-A Version 4 Level 1-B wter vpor nd temperture dt etween lnd nd oen from the surfe to the upper troposphere/lower strtosphere (UT/LS) [Divkrl et l., 2006]. However, these types of omprisons re sujet to two mjor hllenges: first, the stellite oservtions usully hve lower sptil nd temporl resolutions thn in situ oservtions, nd seond, euse the stellite smples reltively lrge volume of ir with different sptil response over nerly instntneous time durtion, irrft mesurements will never oinide t the ext sme time nd lotion s the stellite oservtions. To resolve the differenes in resolution, the higher-resolution in situ dt re usully verged to orser resolution efore ompring them with stellite oservtions. In ddition, ritrry seletion riteri re usully pplied to define ertin sptil nd temporl rnges so tht only the stellite oservtions within these rnges will e ompred with the in situ mesurements. Idelly, one would like to ompre the stellite oservtions with the in situ oservtions tht re s lose s possile in lotion nd time. However, this would stritly limit the numer of dt smples. Sine the dt quntity lso influenes the sttistil signifine of the finl ssessment, the omprisons usully need to lne the size of the verging window versus the numer of dt ville. In previous omprisons, different sptil nd temporl rnges hve een used. For exmple, Tle 1 shows the sizes of the sptil windows in previous studies from 22.5 km rdius [Lmquin et l., 2012] to 100 km rdius [Divkrl et l., 2006] nd the sizes of the temporl windows from 30 min [Lmquin et l., 2012] to within the sme lendr dy [Gettelmn et l., 2004]. The inherent vriility of the tmosphere ultimtely drives the sensitivity to the spe nd time verging windows. Therefore, these lrge differenes in the sptil nd temporl windows set rrier to interompre or onsolidte these previous vlidtions. Indeed, no previous study hs ddressed the sensitivities of the vlidtion results to vrious sptil nd temporl verging window sizes. [4] In ddition to the unertinties in the omprison methods, the in situ oservtions lso hve their own limittions in sptil nd temporl overge. For exmple, previous irrft oservtions used in vlidtion studies re mostly restrited to horizontl lyer, preventing full vertil profile omprison with the AIRS/AMSU-A dt set [Gettelmn et l., 2004; Lmquin et l., 2012]. The vertil profile omprisons hve minly relied on rdiosonde [Divkrl et l., 2006; MMillin et l., 2007] nd dropsonde [Wu, 2009] oservtions, yet these previous dropsonde omprisons were onfined to limited geogrphil lotions. Furthermore, the rdiosonde wter vpor mesurements in the UT/LS hve een reported to suffer from slow responses t low tempertures (e.g., 27 s response time t 40 C) nd ventiltion prolems when iing ours [Miloshevih et l., 2001; Verver et l., 2006]. In prtiulr, there were few in situ dt sets ville over the oen in the Southern Hemisphere (Tle 1, third olumn). Finlly, there hs een no previous vlidtion tht distinguishes etween the lnd nd oen omprisons or etween the Northern Hemisphere (NH) nd Southern Hemisphere (SH) for the Version 5 Level 2 AIRS/ AMSU-A dt. Thus, it is unler if the Version 5 lnd retrievl hs improved sine Version 4 [Divkrl et l., 2006; Toin et l., 2006]. The smpling limittions in the vertil nd horizontl dimensions limit our ssessment of the performnes of AIRS/AMSU-A. In order to ssess the AIRS/AMSU-A performne from glol view, more dvned dt sets re needed to provide full overge of vertil lyers from the surfe to the UT/LS, nd lso over lnd nd oen in oth hemispheres. [5] In this study, we investigted the sensitivity of vlidtion errors to sptil nd temporl windows nd ompred AIRS/AMSU-A Version 5 Level 2 dt etween lnd nd oen nd etween NH nd SH. Our nlyses re sed on the in situ mesurements of two irrft flight mpigns. The flight mpigns re the Ntionl Siene Foundtion (NSF) HIAPER Pole-to-Pole Oservtions (HIPPO) Glol mpign deployments #1 5 from 2009 to 2011 (flight trks shown in Figures 1 nd 1) [Wofsy et l., 2011] nd the NSF Strtosphere Troposphere Anlyses of Regionl Trnsport 2008 (START08) mpign (Figure 1) [Pn et l., 2010]. The omined dt sets rnge from 87 Nto67 S, over oth North Ameri nd the entrl Pifi Oen (Tle 1), nd provide lrge numer of tmospheri vertil profiles from the surfe to the UT/LS. The START08 mpign provided ~120 h flight time nd ~90 vertil trnsets ross the therml tropopuse. The HIPPO Glol mpign provided ~400 h flight time nd ~600 vertil trnsets from the surfe to the tropil UT or the extrtropil UT/LS. In prtiulr, the HIPPO Glol mpign ws designed to hieve 6817

4 Figure 1. Google mps of the (, ) HIPPO Glol mpign #1 5 deployments (yellow trks in Figures 1 nd 1) nd the () START08 mpign (green trks in Figure 1). Only flights used in this study re shown in the figure. vertil profile for every 2.2 of ltitude, whih provided unpreedented fine-grined tmospheri trnsets using in situ mesurements. The omined dt set ws ompred with the AIRS/AMSU-A Version 5 Level 2 wter vpor nd temperture t vrious pressure levels in the oth hemispheres s well s over lnd nd oen. To ddress the sensitivities of the omprisons etween AIRS/AMSU-A nd in situ oservtions with respet to different sptil nd temporl verging windows, we propose new method to test these sensitivities with vrious omintions of sptil nd temporl window sizes. Not only did we ondut the omprisons under the sme temporl nd sptil sles of previous nlyses, suh s Gettelmn et l. [2004] nd Divkrl et l. [2006], ut we lso tested the sensitivities of the vlidtion results from 1 h up to 12 h, s well s from 22.5 km up to 270 km rdii. 2. Dt Coverge nd Hndling 2.1. AIRS/AMSU Oservtions [6] AIRS nd AMSU-A re two of the six sensors ord the NASA Aqu stellite. Aqu is sun-synhronous stellite with sending nd desending orits rossing the equtor t ~1:30 A.M. lol time (desending orit) nd ~01:30 P.M. lol time (sending orit), respetively. This enles AIRS nd AMSU-A to sn the Erth two times dy overing 95% of the Erth s surfe [Chhine et l., 2006]. AIRS is high-spetrl resolution spetrometer with 2378 hnnels lrgely overing the infrred (IR) spetrum from 650 to 2665 m 1 with nominl resolving power (l/δl) of It is ross-trk snning sensor with one sn every 2.7 s, field of view (FOV) of 1.1 nd sptil resolution of 13.5 km t ndir [Chhine et l., 2006; Divkrl et l., 2006]. Beuse AMSU-A sns three times s slowly (one in 8 s) s AIRS, there re nine AIRS footprints within eh AMSU-A footprint. The AIRS/AMSU-A pir together mesures omined footprint t 45 km 45 km horizontl resolution. Three produt types re mde ville from the retrievl steps: level 1 produts (rdines nd rightness temperture), level 2 produts (geoloted, loud-lered rdines nd retrieved physil quntities in two or three dimensions, suh s moisture nd temperture), nd level 3 produts (1 1 gridded produts) [Olsen et l., 2007]. Tle 1 shows the different versions of AIRS/AMSU-A dt used in previous studies. In this study, we use Version 5 Level 2 stndrd produts, whih ontin 28 stndrd pressure levels for temperture nd 15 pressure levels for wter vpor. The produt type is denoted y the ronym AIRX2RET nd is mde ville s hdf-formtted files with dt reorded every 6 min ( grnule of AIRS/AMSU-A dt) Airrft Oservtions [7] We use the Vertil Cvity Surfe Emitting Lser (VCSEL) hygrometer onord the NSF Gulfstrem V (GV) Figure 2. GV dt distriutions in HIPPO nd START08 t different () ltitudes nd () pressures. 6818

5 . Method 1 Cirulr window. Method 2 Annulr window Figure 3. () Method 1 irulr window (yellow region) nd () Method 2 nnulr window (purple region) for seleting the AIRS/AMSU-A dt within ertin sptil nd temporl rnges round eh irrft oservtion (lk dot in the middle). irrft to mesure wter vpor from the surfe to the UT/ LS. The VCSEL hygrometer is n open-pth, ner-infrred lser working t 25 Hz [Zondlo et l., 2010]. Two sorption lines re used to mesure wter vpor: the nm line for wter vpor onentrtions from the mid-troposphere to the UT/LS nd the nm line for the lower troposphere. The mesurement hs n ury of 6% nd preision 3%. Clirtions nd interomprisons with other hygrometers hve een demonstrted in the literture [Zondlo et l., 2010]. The VCSEL mesurements were verged to 1 Hz in the HIPPO nd START08 mpigns. Temperture mesurements were reorded y Rosemount temperture proe, whih were reported t 1 Hz. Other positioning vriles used for the omprisons inlude referene stti pressure nd GPS-derived longitude nd ltitude. [8] We omined the HIPPO nd START08 flight mpigns for the vlidtion. The omined dt set overs from 87 Nto67 S, 84 W 180 W 128 E, trnseting from the surfe to the UT/LS, with 299 nd 214 flight hours over lnd nd oen, respetively. Almost ll flight hours were t the dytime, s the plne usully took off t ~10 A.M. lol time nd lnded t ~6 P.M. lol time. The ltitudinl nd vertil distriutions of the irrft oservtions re illustrted in Figure 2. Our smpling is mostly limited to Northern Ameri, nd southern hemispheri lnd is exlusively New Zelnd nd Austrli. Flight trks in HIPPO were generlly stright diretions nd took no efforts to void louds exept over deep onvetion. 3. Comprison Methods 3.1. Sptil nd Temporl Comprison Windows [9] The irrft mesurements were verged to 100 s resolution, whih is ~23 km in the horizontl nd less thn 1 km in the vertil. Therefore, in this study, one GV oservtion stnds for one dtum of the 100 s verged timeseries. The verging prepres the irrft dt into orser resolution omprle to the sle (~22.5 km) of AIRS/AMSU-A dt set. We define the solute differenes etween the horizontl lotions of AIRS/AMSU-A nd GV oservtions s ddist (in km) nd the solute temporl differenes s dtime (in hours). [10] Two methods were then used to selet the AIRS/ AMSU-A pixel grid dt whih were losely olloted with the irrft oservtions in time nd spe. The first method (Figure 3) ompres the irrft oservtions with ll the stellite oservtions within M hndnkm (M nd N re ritrry vlues). The irulr window method ws used in ll previous studies nd the M nd N vlues re shown in Tle 1. [11] The seond method, whih is novel, uses n nnulr window for omprisons (Figure 3). The region seleted y the nnulr window is the differene etween two irulr windows, whih mens tht stellite oservtions within the rnge of M to (M + M 0 ) h nd N to (N + N 0 ) km round the irrft dt were seleted for omprison. For exmple, if we hoose 6 to12 h, only the stellite dt tht hppened etween 6 nd 12 h efore or fter the in situ mesurement would e ompred. Using this method, we n quntify how the temporl nd sptil differenes etween the in situ nd stellite oservtions influene their greement for wter vpor nd temperture. [12] After seleting the AIRS/AMSU-A dt, there might e more thn one pixel grid inside the temporl nd sptil windows round one GV oservtion. We define ddist s the men vlue of ll the solute sptil differenes etween the stellite dt nd the GV oservtion. Similrly, dtime is the men vlue of ll the solute temporl differenes. The lultions of ddist nd dtime re ddist ¼ dtime ¼ X k i¼0 X k i¼0 j AIRSi GV j ; (1) k j AIRSi GV j : (2) k [13] Here AIRS i denotes the AIRS/AMSU-A pixel grid dt within the omprison window; k denotes the numer of AIRS/AMSU-A dt round one GV oservtion stisfying oth the sptil nd temporl window riteri. We note tht Tle 2. Qulity Control of AIRS/AMSU-A Temperture nd Wter Vpor Dt Temperture dt Wter vpor dt Qulity Control All unit dt (grids) within eh FOV retrieved t Pressure > PGood re deleted, s reommended y previous literture [Susskind, 2007] 1. All FOVs with Qul_H 2 O = 2 re deleted, sine it mens tht the entire wter vpor olumn or FOV is of d qulity 2. All grid dt with H 2 OMMRStdErr > 0.5 H 2 OMMRStd re deleted, s reommended previously [Olsen et l., 2007] 3. All grid dt retrieved t Pressure > PBest re deleted 6819

6 Figure 4. An exmple of sptil nd temporl differenes etween AIRS/AMSU-A nd GV irrft oservtions in HIPPO#1 RF04. The time stmp is sed on UTC (in h) on the dte tht the irrft took off. The UTC time in the next dy is leled s UTC + 1 dy. () The GV flight trk in red nd light lue, where the strt of eh olored segment represents the strt of new hour in UTC. The UTC time for the strt of eh segment of flight trk is leled in the white text ox, suh s UTC 20:00. The segment with the est o-lotion in spe nd time etween the irrft nd AIRS/AMSU-A is leled in the red text ox. Two swths re illustrted in Figure 4 with yellow nd lue olors, respetively. The UTC time of eh grnule is leled in yellow or lue squre oxes ording to the swth s olor. () The 100 s resolution time series of the verged dtime vlue for ll AIRS/AMSU-A dt within 67.5 km nd 3 h of eh GV oservtion; () similr to Figure 4 ut for ddist. ddist nd dtime re divided y the sme denomintor k sine they selet the sme AIRS/AMSU-A dt for lultion Vertil Interpoltion [14] Wter vpor nd temperture reported y AIRS/ AMSU-A were further interpolted onto the sme pressure level s the GV flight level. The interpoltion is neessry step efore the omprison, euse the vertil mismthes etween the two dt sets n led to different redings of wter vpor nd temperture. To help minimize the impt of vertil dislotions etween the two dt sets, we interpolted the stellite dt y ssuming log-sle wter vpor distriution nd liner-sle temperture distriution in the pressure profile. [15] Wter vpor ws interpolted y log 10 (wter vpor mixing rtio) = + Pressure. The oeffiients nd were determined from AIRS/AMSU-A dt t two lyers: the first lyer is where the GV flies through, nd the seond lyer is where its midlevel is the seond losest to the GV [16] As n exmple, if the GV flies t 370 mr, it is within the AIRS/AMSU retrievl lyer of mr. The lyers ove nd elow the flight level re mr nd mr, respetively. To deide whih of these two lyers should e used for the interpoltion, we ompre their midlevels nd selet the one loser to the irrft flight level. In this se, the midlevel of the mr lyer (i.e., 450 mr) is loser to the flight level (370 mr) thn the midlevel of the mr lyer (i.e., 250 mr). The reson tht we ompre the midlevels to the flight level is euse the AIRS/AMSU-A wter vpor mesurement reported t eh lyer represents the verge wter vpor onentrtion inside the lyer, even though AIRS/AMSU-A Level 2 dt re leled with the ottom level of the lyer. For exmple, the mr lyer is leled s 500 m, while the mr lyer is leled s 400 m. The interpoltion exmple for wter vpor is ð AIRS H 2 O interpolted vlue ðg=kgþ¼ 10 log 10ðAIRS H 2 O 500m Þ log 10 ðairs H 2 O 400m ÞÞ Pressure GV ð 350mr 450 mr 350mr þ log 10ðAIRS H 2 O 400m ÞÞ ; (3) flight level esides the first lyer. We only used two lyers for the interpoltion euse the more lyers we use in interpoltion, the higher hne tht one of them will not stisfy the qulity ontrol riteri, nd lso euse the interpoltion ssumption of log-sle distriution of wter vpor my not pply to lrge vertil rnge. AIRS H 2 O 500 m nd AIRS H 2 O 400 m stnd for the retrievls of the men wter vpor onentrtion t mr nd mr, respetively. [17] Temperture ws interpolted y Temperture = + d Pressure. Different from the wter vpor retrievl, the temperture retrievl represents the temperture t the ext 6820

7 d Figure 5. An exmple of wter vpor omprison for HIPPO#1 RF04. () The time series of log-sle wter vpor mss mixing rtios (g/kg) for VCSEL hygrometer (lk) nd AIRS/AMSU-A (red). () The numer of AIRS dt stisfying the sptil nd temporl seletion windows (lk line) nd the remining numer fter pplying the wter vpor qulity ontrol riteri (red line); () the time series of dh 2 Oper, i.e., the solute perentge differene of wter vpor mss mixing rtio; (d) similr to Figure 5, ut for dh 2 Oper. pressure level. Therefore, the temperture interpoltion oeffiients nd d were lulted from the two AIRS/AMSU-A levels where the GV flew in etween. For exmple, if the GV flies t 370 mr, the AIRS/AMSU-A 300 mr nd 400 mr levels would e used for the interpoltion. The lultion of the liner interpoltion onto the GV flight level is AIRS Temperture interpolted vlue ¼ AIRS T 400m AIRS T 300m Þ Pressure GV 300mr 400mr 300mr þairs T 300m ; (4) AIRS T 400 m nd AIRS T 300 m stnd for the AIRS/AMSU-A temperture retrievls t 400 mr nd 300 mr, respetively AIRS/AMSU-A Dt Qulity Control [18] After seleting the AIRS/AMSU-A dt within ertin horizontl sptil nd temporl windows round one GV point, we qulity ontrolled the seleted stellite oservtions of temperture nd wter vpor with different riteri (Tle 2). The presene of louds in the FOV of the stellite sensor n omplite the retrievl proesses [Susskind et l., 2003, 2006]. Therefore, for eh vrile, we heked the qulity of the dt t the two pressure levels eing used in the vertil interpoltions. If one of the two levels did not pss the qulity ontrol riteri, the retrievl would not e used to ompre with this GV oservtion. This filtering proess selets stellite retrievls tht re of good qulity nd lso redues the omputtionl ost y eliminting the nonqulified dt efore moving to the next step of omprison. We note tht there re higher rtios of dt t the lower ltitudes eing filtered out, sine the retrievls t lower ltitudes re more likely to e omplited y the presene of louds Definitions of Wter Vpor nd Temperture Differenes Between AIRS/AMSU-A nd GV Mesurements [19] For one GV oservtion, there might e more thn one AIRS/AMSU-A dt round it tht stisfy ll the seletion window riteri nd the qulity ontrol riteri. In the first step, the omprison ws onduted etween this GV oservtion nd eh of the AIRS/AMSU-A dt surrounding it. Then ll of these individul omprisons were verged to represent the finl omprison result for this GV point. The reson for tking the verge for ll AIRS/AMSU-A dt round one GV oservtion is to mke sure tht the finl omprison result is not hevily weighted over few GV oservtions with lrge numer of AIRS/AMSU-A dt round them. 6821

8 DIAO ET AL.: GLOBAL AIRS/AMSU-A H2O AND T VALATIONS d Figure 6. Similr to Figure 5 ut for the temperture omprison exmple of HIPPO#1 RF04. [20] The differene in wter vpor etween AIRS/AMSU-A nd the VCSEL hygrometer ws represented in two wys: the solute differene of wter vpor in perentge, defined s dh2oper (%) (eqution (5)), nd the differene of wter vpor in perentge, defined s dh2oper (%) (eqution (5)). The perentge differene ws used insted of the differene (in g/kg) to llow ssessment of the full vertil profile of wter vpor retrievl. Wter vpor onentrtions vry y 5 orders of mgnitude from the surfe to the UT/LS, nd the differene (in g/kg) t the UT/LS is orders of mgnitude Figure 7. Comprisons of () wter vpor nd () temperture for ll flights. The stterplots show the orreltion etween AIRS/AMSU-A (ordinte) nd GV dt (siss). The olor oding represents the pressure of the GV irrft from the UT/LS (purple) to the surfe (red). For wter vpor omprison, the liner fit (red line) is log10(airs) = + log10(vcsel). For temperture, the liner fit (red line) is AIRS = + (GV). AIRS denotes the whole AIRS/AMSU-A dt set. Pr denotes the Person-R vlue. When the Pr vlue is loser to 1, it stnds for stronger orreltion etween the two dt sets. 6822

9 Figure 8. Wter vpor omprisons in eh pressure lyer for ll flight dt. Different olors represent different seletion windows in spe nd time: 1 h nd 22.5 km (solid lk line), 3 h nd 100 km (dotted red line, sme sizes s Divkrl et l. [2006]), 12 h nd 67.5 km (dshed lue line, similr sizes to within 1 lendr dy nd 0.5 in Gettelmn et l. [2004]). () The dh 2 Oper vlues for ll the GV oservtions in eh pressure lyer from the surfe to the UT/LS. Error rs denote one stndrd devition for ll the dh 2 Oper inside eh pressure level. () Similr to Figure 8 ut for dh 2 Oper. () The numer of GV dt eing used for the wter vpor omprisons in eh pressure lyer. smller thn the differene t the surfe. For eh AIRS/AMSU-A nd GV omprison pir, we lulted one dh 2 Oper nd one dh 2 Oper vlue, nd the finl dh 2 Oper (or dh 2 Oper) represents the men of ll the dh 2 Oper (or dh 2 Oper) vlues for this GV oservtion. The eqution is shown s elow, for whih we use AIRS to denote the whole AIRS/AMSU-A dt set: jdh 2 Operj ¼ dh 2 Oper ¼ X i¼0 VCSEL 100% ; (5) VCSEL 100% : (5) jairs i VCSELj X AIRS i VCSEL i¼0 [21] Similrly, the differene in temperture etween the AIRS/AMSU-A nd the GV mesurement ws represented in two wys: solute temperture differene, dtemp (in Kelvin) (eqution (6)), nd temperture differene, dtemp (in Kelvin) (eqution (6)): jdtempj ¼ X X j AIRS i¼0 i GVj ; (6) i¼0 dtemp ¼ AIRS i GV : (6) [22] We ution here tht the numer of AIRS/AMSU-A wter vpor dt ( in equtions (5) nd (5)) nd the numer of temperture dt ( in equtions (6) nd (6)) used to ompre with the GV oservtion might not e the sme, sine these two vriles require different qulity ontrol riteri s shown in Tle Averged dh 2 Oper nd dtemp Vlues in Eh Pressure Level [23] After dh 2 Oper, dh 2 Oper, dtemp nd dtemp were lulted for eh GV oservtion using equtions 6823

10 Figure 9. Similr to Figure 8 ut for temperture omprison. () dtemp, () dtemp, nd () the numer of GV oservtions for temperture omprisons t eh pressure level. (5), (5), (6), nd (6), we lulted their verge vlues in eh pressure level s jdh 2 Operj in eh pressure level ¼ dh 2 Oper in eh pressure level ¼ jdtempjin eh pressure level ¼ dtemp in eh pressure level ¼ X X jdh j¼0 2Operj j ; (7) dh j¼0 2Oper j ; (7) X d X d jdtempj j¼0 j ; (8) d dtemp j¼0 j : (8) d [24] Here nd d denote the numer of GV oservtions in eh pressure level tht hve een ompred with AIRS/ AMSU-A wter vpor nd temperture dt, respetively. Agin, we note tht nd d might not e the sme, sine these two vriles require different qulity ontrol riteri. 4. Results 4.1. Comprison Exmples From HIPPO#1 RF04 [25] A typil flight exmple of HIPPO#1 RF04 is used to illustrte the geogrphil positions of the GV irrft nd the AIRS/AMSU-A FOVs round the irrft trk (Figure 4). During this flight, the irrft took off from Anhorge, Alsk, t UTC 19:30 (Anhorge lol time 11:30 A.M.) nd lnded t Honolulu, Hwii, t UTC +1 dy 03:50 (Hwii lol time 5:50 P.M.). The flight durtion ws out 8 h nd 20 min. The horizontl lotion of the GV flight trk is shown s the thik line in red nd light lue in Figure 4. The olor hnge etween light lue nd red represents the eginning of new hour in UTC. Thus, eh segment of the flight trk in light lue or red represents 1 h of flight, exept tht the first nd lst segments re not full hours. The UTC time of the eginning of eh segment is leled with white text oxes with lk rrows pointing to the strt of eh UTC hour. The red text ox for UTC + 1 dy 00:00 shows the segment of the flight where o-lotion in oth spe nd time hppened etween the GV nd stellite oservtions. The losest two swths round the GV flight trk re plotted in yellow nd lue (Figure 4). The UTC time for eh grnule is leled y yellow or lue ox, nd eh grnule is 6 min prt. Within the AIRS grnule UTC + 1 dy 00:08, the GV nd stellite dt were olloted in oth spe nd time. During this 8 h flight, only few dt smples were ville when the irrft nd the stellite oservtions hppened t the sme lotion nd time, illustrting the diffiulty of ompring the oservtions etween irrft nd stellite, s they onstntly hnge reltive positions in spe nd time. [26] The time series of the differenes in time nd spe for this exmple flight (HIPPO#1 RF04) re illustrted in 6824

11 d e f Figure 10. Influenes of sptil nd temporl window sizes on the wter vpor omprison results. ( ) Comprisons with fixed temporl window size of 0 3 h nd inresing horizontl sptil window sizes. Different olors in Figures denote km (solid lk), km (dotted red), nd km (dshed lue). Figures re vertil profiles of dh 2 Oper, dh 2 Oper, nd the numer of GV oservtions involved in the omprison t eh pressure level, respetively. (d f) Similr to Figures ut for inresing temporl window sizes t fixed sptil window of km. Different olors in Figures 10d 10f denote different temporl window sizes: 0 3 h (solid lk), 3 6 h (dsh-dotted green), nd 6 12 h (short dshed purple). All error rs denote one stndrd devition for dh 2 Oper or dh 2 Oper vlues t eh pressure level. Figures 4 nd 4, respetively. The ddist nd dtime were lulted sed on equtions (1) nd (2), respetively. Here we use the 3 h nd 67.5 km window, whih is the suset of two previous windows: 3 h nd 100 km [Divkrl et l., 2006], nd 12 h nd 67.5 km [Gettelmn et l., 2004]. This window pir is omprle to the men horizontl wind speed of 6.7 m/s in the midtroposphere [Peixoto nd Oort, 1992]. By hoosing this omintion of sptil nd temporl windows, we onsider the possile trnsport of the ir within 3 h of the flight trk t men horizontl speed of 6.7 m/s, lthough not the diretion of the trnsport whih is eyond the sope of urrent work. In Figure 4, dtime rehed zero t UTC + 1 dy 00:11, ut ddist t this time stmp ws not zero (Figure 4) euse there were more thn one AIRS/AMSU-A dt used to lulte the verge ddist within 67.5 km (eqution (1)). The dtime from UTC 20:22 to 20:54 ws disontinuous euse two different swths of AIRS/AMSU-A were used for the omprisons. During the erlier segment, the GV flight trk ws loser to the yellow grnules of AIRS/AMSU-A dt, while in the ltter segment, the GV flew into different lotion nd ws loser to the lue grnules. As the two sets of grnules were ~1.5 h prt, the verged dtime hd n rupt hnge. The grdul hnges of dtime fter UTC 21:15 represent the ontinuous hnges in irrft time. [27] We further lulted the time series for wter vpor (Figure 5) nd temperture omprisons t 100 s resolution using the 67.5 km nd 3 h window (Figure 6). The time series of dh 2 Oper (%) (Figure 5) ws lulted y eqution (5), tht is, the verge vlue of ll the solute perentge differenes etween ll the AIRS/AMSU-A dt round one GV oservtion. Similrly, Figure 5d shows dh 2 Oper (%) s lulted y eqution (5). The dh 2 Oper vlues rnge from ~10% to 160%, nd dh 2 Oper vlues rnge from 160% to 120%. Generlly, the lrger dh 2 Oper hppens t the UT/LS, not only euse slight vrition t low wter vpor onentrtions n result in lrge perentge differene ut lso due to the low sensitivity of AIRS in this region [Fetzer et l., 2008; Ling et l., 2010]. The wter vpor onentrtion of AIRS/AMSU-A in Figure 5 ws lulted y 6825

12 d e f Figure 11. Similr to Figure 10, ut for the influenes of ddist nd dtime on the temperture omprison results. tking the verge of ll qulifying AIRS/AMSU-A points efore lulting ny differenes, while the dh 2 Oper in Figure 5 (or dh 2 OperinFigure5d)wslultedytking the solute perentge differenes (or perentge differene) first nd then verging them. Similrly, the exmple of temperture omprisons is illustrted in Figure 6. dtemp nd dtemp in Figures 6 nd 6d were lulted y equtions (6) nd (6), respetively. In this exmple, dtemp rnges from ~0 to 5 K, nd dtemp rnges from ~ 3 to5k Syntheses of Wter Vpor nd Temperture Comprisons for All Flights [28] The omprisons etween AIRS/AMSU-A nd GV dt were onduted for the whole dt set using the 3 h nd 67.5 km window. Figures 7 nd 7 show the orreltion of wter vpor nd temperture, respetively, with olor oding of pressure from the surfe (red) to the UT/LS (purple). Eh AIRS/AMSU-A dtum in Figure 7 represents the men vlue of ll qulifying stellite dt round one GV oservtion. Wter vpor mixing rtio (g/kg) orreltion ws fitted y log 10 (H 2 O AIRS )= + log 10 (H 2 O VCSEL ), where the slope is nd the interept is (Figure 7). Temperture orreltion ws fitted y T AIRS = + T GV, where the slope is nd the interept is (Figure 7). The slope of the temperture liner fit is loser to 1 thn the wter vpor slope, implying etter greement etween the stellite nd irrft tempertures dt thn the wter vpor dt. The Person-R (Pr) orreltion oeffiients for wter vpor nd temperture omprisons re nd 0.997, respetively. Gettelmn et l. [2004] showed tht for the vlidtion of Version 3 Level 2 dt in the UT/LS region (pressure 500 hp), the Pr vlues of wter vpor nd temperture omprisons were 0.91 nd 0.98, respetively [Gettelmn et l., 2004, Figure 2]. In omprison, t 3 h nd 67.5 km window, our vlidtion results of pressure 500 hp show tht the Pr of wter vpor nd temperture re nd 0.976, respetively. And for pressure >500 hp, the Pr of wter vpor nd temperture re nd 0.987, respetively. Therefore, ompred with AIRS Version 3 Level 2 dt in Gettelmn et l. [2004], AIRS Version 5 Level 2 dt show etter greement with the in situ oservtions for wter vpor nd similr greement for temperture Wter Vpor nd Temperture Comprisons Using Cirulr Comprison Windows [29] The omprison results of wter vpor nd temperture in eh AIRS/AMSU-A pressure level re shown in Figures 8 nd 9, respetively. dh 2 Oper nd dh 2 Oper, nd dtemp nd dtemp represent the men vlues in eh pressure level (equtions (7) nd (7), nd (8) nd (8), respetively). The AIRS/AMSU-A pressure levels involved in the omprison re from Level #1 (1100 mr) to Level #13 (100 mr). The mximum nd minimum pressure vlues of the whole irrft dt set re 1021 mr nd 133 mr, 6826

13 Figure 12. Comprisons of wter vpor mesurements over oen (solid lk line) nd lnd (dotted red line). The omprisons re within the 3 h nd 67.5 km windows. Vertil profiles of () dh 2 Oper nd () dh 2 Oper. () The numer of GV oservtions in eh pressure level. Error rs denote one stndrd devition of dh 2 Oper or dh 2 Oper in eh pressure lyer. respetively. We note tht lthough in previous studies, wter vpor dt t pressure >300 mr were usully used for the omprison with limte models [Jing et l., 2012; Tin et l., 2013], there is still strong need to show the wter vpor retrievl unertinties in the UT/LS region sine other studies hve lso used wter vpor t mr for nlyses of reltive humidity nd ie supersturtion distriutions [Gettelmn et l., 2006; Lmquin et l., 2012]. In ddition, we do not exlude the interomprisons t low pressures (pressure < 300 mr) euse the sensitivity of AIRS is more dependent on the wter vpor onentrtion (~10 20 ppmv) [Gettelmn et l., 2004; Red et l., 2007] insted of pressure. In ft, further lgorithm improvements in the future my help to improve the sensitivity of AIRS wter vpor retrievls. Thus, understnding the edge of AIRS sensitivity t the UT/LS is importnt. To ompre with the pst vlidtion results, we hose severl sptil nd temporl windows tht hve een used in previous nlyses. In Figures 8 nd 9, we used three omintions of sptil nd temporl windows to ompre the stellite dt: (1) 3h nd 100 km (dotted red line, sme windows s Divkrl et l. [2006]), (2) 12 h nd 67.5 km (dshed lue line, similr to Gettelmn et l. [2004] within 1 lendr dy nd 0.5 ), nd (3) smller window of 1 h nd 22.5 km (solid lk line). This smll window size hs not een reported efore in previous vlidtions from the surfe to the UT/LS. Lmquin et l. [2012] pplied similr omintion of smll sptil ( 22.5 km) nd temporl ( 30 min) windows, ut were restrited to the UT/LS region. Using the smll window, we n improve the ssessments of the AIRS/AMSU-A dt y showing results when they were losely olloted with in situ mesurements. [30] In generl, s the sptil nd temporl window sizes inrese, oth dh 2 Oper nd dtemp vlues eome lrger, nd this trend hppens t lmost every pressure level. The exeption hppens in the oundry lyer ( 900 mr) nd the UT/LS ( 200 mr), where there were fewer smples. Aside from the differenes in the mount of dt, the different sensitivities to omprison windows ould lso e result of the different vrine sling hrteristis in these pressure rnges [Khn nd Teixeir, 2009]. When using the sme sptil nd temporl window sizes (3 h nd 100 km), our results show similr dh 2 Oper nd dtemp results to Divkrl et l. [2006], i.e., ~20 60% for dh 2 Oper nd ~2.5 1 K for dtemp from the surfe to the UT/LS. The dtemp t the surfe (~2.5 K in Figure 8) is slightly higher thn tht of Divkrl et l. [2006] (~1.7 K), whih ould e result of the smll numer of qulified omprisons (e.g., 11 GV oservtions) t the surfe in our dt. However, using the smller window of 1 h nd 22.5 km, the wter 6827

14 Figure 13. Similr to Figure 12 ut for dtemp nd dtemp vlues over lnd nd oen. vpor nd temperture omprison results n e improved y up to ~10% in dh 2 Oper nd ~0.5 K in dtemp ompred with using the 3 h nd 100 km window. In ddition, using the 12 h nd 67.5 km window (similr to the window size of Gettelmn et l. [2004]) in our dt resulted in up to ~10% inrese in dh 2 Oper nd ~0.3 K inrese in dtemp thn using the window of Divkrl et l. [2006]. These results imply tht the interomprisons etween different vlidtion studies need to ount for the unertinties ontriuted from the vrious window sizes Sensitivities of the Comprison Results to Sptil nd Temporl Differenes Using Annulr Comprison Window [31] The omprisons sed on Method 1 show tht oth sptil nd temporl window sizes influene the omprison results. To further quntify these influenes, we used Method 2 the nnulr window to lulte the dh 2 Oper, dh 2 Oper, dtemp, nd dtemp vlues for the stellite dt in eh nd of spe nd time round the GV oservtions. To illustrte the sensitivities to different time window sizes, we fixed the distne window t km nd ompred the AIRS/AMSU-A dt within 0 3, 3 6, nd 6 12 h of the GV dt. Similrly, we fixed the time window t 0 6 h nd mde the omprison t , , nd km. The results show tht dh 2 Oper nd dtemp vlues inrese with the sptil nd temporl window sizes, even though the numer of GV oservtions seleted y eh window is lmost the sme (Figures 10 nd 11). There is higher sensitivity of dh 2 Oper to the inresing time window sizes t ~ mr thn in the oundry lyer or ove 200 mr, nd the mximum sensitivity is round 550 mr. The mximum sensitivity t 550 mr grees with the previous finding tht the vrine sling generlly hs the highest vlue round mr t 24 S, 90 W nd 24 N, 90 W [Khn nd Teixeir, 2009]. Similrly, dtemp hs high sensitivity to inresing time window t ~ mr. Compring Figures 10 nd 10d, the inrese in dh 2 Oper from 0 3 h to 3 6 h is similr to tht from km to km, while the dh 2 Oper inrese from km to km is muh lrger thn tht from 3 6 h to 6 12 h. For temperture, the inrese in dtemp for eh sptil nd temporl window nd is similr. The vertil nd horizontl vriilities of the sensitivities of dh 2 Oper nd dtemp to sptil nd temporl windows might e result of different vrine sling hrteristis t differene regions, s well s the diurnl yle or the vertil nd horizontl wind speed distriutions, nd further investigtion is needed to determine the ext use. [32] On the ontrry, the hnges of dh 2 Oper nd dtemp vlues re not s ovious ompred with those of dh 2 Oper nd dtemp. The less ovious trend for dh 2 Oper nd dtemp my e result of the rndom distriution of stellite dt round the GV oservtion s the verging window sizes expnd. Tht is to sy, the AIRS/AMSU-A dt vlue my eome either lrger or smller thn the irrft dt 6828

15 Figure 14. Comprisons of dh 2 Oper vlues etween the NH nd SH. Solid lk line represents the omprison results over lnd, while dotted red line is for over oen. The vertil profiles of dh 2 Oper in the () NH nd () SH. The numer of GV oservtions used in the omprisons t eh pressure level in the () NH nd (d) SH. vlue when the distne nd time differenes inrese. Therefore, even though the disgreement etween eh AIRS/AMSU-A dtum nd the irrft oservtion eomes lrger s the omprison window expnds, the men of these differenes (inluding oth positive nd negtive vlues) does not hnge very muh Oen Versus Lnd Comprisons for Wter Vpor nd Temperture [33] The lnd frtion in the stellite FOV hs n influene on the qulity of AIRS/AMSU-A dt s illustrted in previous work [Divkrl et l., 2006]. Here we seprted ll the omprisons into lnd (lndfr 0.5) nd oen (lndfr < 0.5). To ensure enough omprison smples in eh tegory, we roden the window size to 3 h nd 67.5 km. The vertil profiles of dh 2 Oper, dh 2 Oper, dtemp, nd dtemp of lnd nd oen re illustrted in Figures 12 nd 13, respetively. The numer of GV oservtions used in the dh 2 Oper lultion ws the sme s tht in dh 2 Oper lultion (Figure 12), sine oth lultions were sujet to the sme seletion windows nd qulity ontrol riteri. Similrly, the numer of GV oservtions is the sme for oth dtemp nd dtemp lultions (Figure 13). At mr, the dh 2 Oper vlues over lnd re ~0 5% higher thn those over oen (Figure 12). This result suggests n improvement in the lnd retrievl in Version 5, sine previously n ~10% disgreement in wter vpor root men squre vlues were reported etween lnd nd oen vlidtions using Version 4 dt [Divkrl et l., 2006]. Yet in the oundry lyer ( 800 mr) nd the UT/LS ( 300 mr), the dh 2 Oper vlues over lnd still hve ~10% disgreement with those over oen. [34] The dh 2 Oper vertil profile in Figure 12 shows lrger disgreement (~20 50%) etween lnd nd oen oservtions ompred with the dh 2 Oper nlyses in Figure 12 in the oundry lyer ( 800 mr) nd the UT/ LS ( 200 mr). Bsed on the dh 2 Oper nlyses, the lnd (nd oen) retrievls re drier (nd moister) thn the irrft oservtions y up to ~ 30% (~20%) of dh 2 Oper in the oundry lyer nd UT/LS. Yet we note tht the lrge differenes in the oundry lyer ould prtly result from the lk 6829

16 d Figure 15. Similr to Figure 14, ut for dtemp vlues etween the NH nd SH. of smpling. Severl ftors n ontriute to the lrge differenes etween the over-lnd nd over-oen mesurements of AIRS/AMSU-A, inluding the existene of louds nd the intrinsi tmospheri vriilities. [35] An improved temperture retrievl over lnd is lso shown in Version 5 dt ompred with Version 4 dt t pressure 700 mr. The dtemp disgreement etween lnd nd oen oservtions of Version 5 is ~0.3 K t pressure 700 mr (Figure 13), while the previous results of Version 4 showed tht the disgreement in the temperture root men squre vlues etween lnd nd oen is ~0.5 K. However, lrge temperture disgreements (~1 1.5 K in dtemp ) etween lnd nd oen mesurements still exist t the lower ltitudes ( 700 mr). The dtemp profile in Figure 13 shows lrger disgreements etween lnd nd oen thn dtemp nlyses t pressure 300 mr nd 700 mr. Compred with the irrft oservtions, the lnd (nd oen) retrievls hve older (nd wrmer) temperture of ~ 0.5 to 1K (0 0.5 K) in the oundry lyer, while t the UT/LS, oth lnd nd oen retrievls hve wrmer temperture y ~0 2K (0.5 2 K). Overll, the improvements in Version 5 wter vpor nd temperture retrievls t mr show potentil pplitions of AIRS/ AMSU-A dt in studying reltive humidity, preipittion, nd loud properties over lnd Wter Vpor nd Temperture Comprison Between the NH nd SH [36] ThedH 2 Oper nd dtemp omprison results for the NH nd SH re shown in Figures 14 nd 15, respetively. Similr to the lnd nd oen nlyses, the NH nd SH omprisons were sed on the 3 h nd 67.5 km window to ensure enough omprison smples. The solid lk line nd the dotted red line represent the lnd nd oen oservtions in eh hemisphere, respetively. The irrft did not smple mny over-lnd regions in the SH, mostly over Austrli nd New Zelnd. Therefore, the lrge dh 2 Oper vlue over lnd in the SH (Figure 14) might e due to the lk of smpling. On the other hnd, the omprison results over oen show no signifint differenes in dh 2 Oper nd dtemp etween the two hemispheres Mtries of Comprison Results y Vrious Sptil nd Temporl Seletion Windows [37] To illustrte the influenes of vrious sptil nd temporl windows, we use ontour plots to show the sensitivities of wter vpor nd temperture omprisons to vrious ddist 6830

17 d Figure 16. Mtries of wter vpor omprison results ( dh 2 Oper ) sed on vrious sptil nd temporl windows. dtime inreses from 0 to 12 h y 3 h steps (y xis), nd ddist inreses from 0 to 270 km y 67.5 km steps (x xis). The enter vlues of eh sptil or temporl window nds were used for plotting the ontours (e.g., 1.5 h, 4.5 h, 7.5 h, nd 10.5 h; km, km, km, nd km). The omprisons were seprted into four pressure rnges: () Levels 1 4 of mr, () Levels 5 7 of mr, () Levels 8 10 of mr, nd (d) Levels of mr. Eh grdient is 1/20 of the full olor sle. nd dtime vlues (Figures 16 nd 17). The purpose of demonstrting the sensitivity of omprison results to ddist nd dtime is to help onstrin the unertinties oming from the ritrry seletions of sptil nd temporl windows in previous vlidtion studies. Here we test ddist from 0 to 270 km (i.e., km) nd dtime from 0 to 12 h (i.e., 12 1 h). The seletion of ddist nd dtime overs most of the temporl nd distne window sizes used in previous studies [Gettelmn et l., 2004; Divkrl et l., 2006; Lmquin et l., 2012]. The wter vpor omprisons were inned into four pressure rnges: (1) Levels 1 4 of mr, (2) Levels 5 7 of mr, (3) Levels 8 10 of mr, nd (4) Levels of mr (Figure 16). We note tht the pressure ins for temperture re slightly different from those for wter vpor, whih re (1) Levels 1 4 of mr, (2) Levels 5 7 of mr, (3) Levels 8 10 of mr,nd(4) Levels of mr (Figure 17). We use different pressure ins for wter vpor nd temperture euse temperture is retrieved extly on the leled pressure level, while wter vpor is the men vlue etween the leled pressure level nd its upper level. [38] The dh 2 Oper vlues in the ontour plots show n inresing trend from the surfe to the UT/LS, whih is onsistent with the nlyses in Figure 8. The grdient of the ontour lines is 1/20 of the full olor sle for ll ontour plots in this study. The dh 2 Oper vlues re lrger t the higher ltitudes euse slight solute hnge in wter 6831

18 Figure 17. Similr to Figure 16, mtries of temperture omprison results ( dtemp ) sed on vrious sptil nd temporl windows. Four pressure rnges were nlyzed: () Levels 1 4 of mr, () Levels 5 7 of mr, () Levels 8 10 of mr, nd (d) Levels of mr. We ution tht the pressure rnges for temperture re disontinuous nd different from those for wter vpor euse the AIRS/AMSU-A temperture dt re reported s the vlue on the leled pressure level nd the wter vpor dt re reported s the men vlue etween the leled pressure level nd its upper level. vpor onentrtion t the UT/LS n led to lrge perentge hnge. The lower tropospheri dh 2 Oper vlues lso hve smller grdient (fewer ontour lines) when ompred with the higher ltitudes, inditing tht the influenes of ddist nd dtime re less ovious t the lower troposphere. Similr nlyses on the Pr vlue of wter vpor orreltion show tht Pr vlues derese from ~ 0.9 to ~ 0.7 from the surfe to the UT/LS, suggesting weker orreltion t the UT/LS. Compred with wter vpor, the Pr vlues of temperture show nrrower rnge from ~0.9 to 1.0. [39] The sensitivity of dh 2 Oper is different etween 4h nd 4 h. For omprisons 4h, dh 2 Oper is lrgely influened y ddist from 1100 to 200 mr (Figures 16 16), nd only t mr the influene of dtime eomes more dominnt (Figure 16d). For omprisons 4h, dh 2 Oper is lrgely influened y ddist t mr (Figure 16), yet dtime shows more dominnt influene for 700 mr (Figures 16 16d). Compred with dh 2 Oper, dtemp shows similr trend t 4h, tht is, dtemp is dominntly influened y ddist nd dtime t mr nd 700 mr, respetively (Figure 17). However, dtemp shows different trend thn dh 2 Oper t 4 h, tht is, dtemp is eqully influened y ddist nd dtime from 1100 to 500 mr (Figures 17 nd 17), while t mr, the influene of ddist eomes more dominnt (Figures 17 nd 17d). The higher sensitivity of dh 2 Oper to ddist thn dtemp t mr my e result of the lrger vriility in wter vpor onentrtions with respet to topogrphil vritions. [40] To test the ltitudinl influenes on the sensitivities of dh 2 Oper nd dtemp to sptil nd temporl windows, we seleted the pressure levels of 5 10 (i.e., mr for wter vpor, nd mr for temperture) nd seprted the omprisons into four ltitudinl regions s 6832

19 Figure 18. Ltitudinl influenes on the wter vpor omprison sensitivities to ddist nd dtime. Four ltitudinl regions were nlyzed: () 10 N 10 S, () 10 N 30 N nd 10 S 30 S, () 30 N 60 N nd 30 S 60 S, nd (d) 60 N 90 N nd 60 S 90 S. The dh 2 Oper nlyses re within wter vpor Levels 5 10, whih is mr. shown in Figure 18 nd 19: (1) 10 N 10 S, (2) 10 N 30 N nd 10 S 30 S, (3) 30 N 60 N nd 30 S 60 S, nd (4) 60 N 90 N nd 60 S 90 S. For oth wter vpor nd temperture omprisons, the sensitivity to ddist reltively inreses with respet to the sensitivity to dtime, whih is onsistent with the previous finding tht the sptil vriilities of wter vpor nd temperture inrese with ltitudes [Khn et l., 2011]. Wter vpor omprisons show etter results t the deep tropis (10 N 10 S) nd the polr regions (60 N 90 N, 60 S 90 S) thn the sutropis nd midltitudes (10 S 60 S, 10 N 60 N), while temperture shows etter results t the tropis (30 N 30 S) thn the extrtropis (30 N 90 N, 30 S 90 S). Therefore, the mjor differene etween the wter vpor nd temperture sensitivities ross ltitudes is t the polr regions, where wter vpor (temperture) retrievls show smller (lrger) disgreements with in situ oservtions ompred with the other ltitudes. 5. Summry nd Implitions 5.1. Implitions for Vlidtion Studies [41] In this study, we used omprehensive dt set inluding in situ mesurements over lnd nd oen in oth hemispheres to ompre with the wter vpor nd temperture retrievls of AIRS/AMSU-A. Bsed on our dt set, the dh 2 Oper nd dtemp vlues derese y ~5 10% nd ~0.2 K, respetively, when the window sizes derese from 3h nd 100km [Divkrl et l., 2006] to 1 h nd 22.5 km. Similrly, dh 2 Oper nd dtemp derese y ~10 20% nd ~0.5 K, respetively, when the window sizes nrrow from 12 h nd 67.5 km [Gettelmn et l., 2004] to 1 h nd 22.5 km. [42] The nnulr window further quntifies the sensitivities to eh nd of ddist nd dtime. The ddist hs lrger influene on wter vpor nd temperture t lower ltitudes, whih is onsistent with previous findings of lrger sptil vriilities of temperture nd wter vpor t lower ltitudes [Khn et l., 2011]. In ontrst, t the higher ltitudes, dtime 6833