Radio Occultation Retrieval of Atmospheric Profiles from the FORMOSAT-3/COSMIC Mission: Early Results

Six identical micro-satellites of the FORMOSAT-3/COSMIC (Formosa Satellite #3 and Constellation Observing System for Meteorology, Ionosphere and Climate: FS-3/C) mission were successfully launched on 14 April 2006. The FS-3/C mission provides the first satellite constellation for monitoring global weather using the Global Positioning System (GPS) radio occultation (RO) technique. The mission’s primary scientificific goal is to obtain near-real time profiles of the bending angle and refractivity in the neutral atmosphere and in the ionosphere. In April, 2007 the FS-3/C mission provide about RO soundings of 2000 atmospheric vertical profiles per day in a nearly uniform distribution around the globe. The lowest altitude penetration for more than 80% of RO soundings reached below 1 km in altitude. Most soundings have penetration below 800 m altitude in the equatorial region and below 200 m altitude in polar regions. The quality and accuracy of the RO sounding profiles are in good agreement with the CHAMP (CHAllenging Minisatellite Payload) RO soundings and direct measurements using dropsondes. The FS-3/C RO sounding observations are used to support operational global weather prediction, climate monitoring and research, space weather forecasting, and ionosphere and gravity research.


INTRODUCTION
Six identical micro-satellites comprising the FS-3/C (Formosa Satellite #3 and Constellation Observing System for Meteorology, Ionosphere and Climate) mission were successfully launched into low-earth orbit from Vandenberg Air Force Base, California on 14 April 2006 (Cheng et al. 2006).The FS-3/C mission provides the first satellite constellation for monitoring global weather using the Global Positioning System (GPS) radio occultation (RO) technique.The mission is a Taiwan-US collaborative project jointly carried out by the National Space Organization (NSPO) in Taiwan and the University Corporation for Atmospheric Research (UCAR) in the United States.During a deployment period of ~16 months, each satellite will rise to its final orbit at an altitude of ~800 km, an orbit plane inclination angle of 72 degrees, and a separation angle between the neighboring orbit planes of 30 degrees longitude.The satellites are expected to operate for 5 years.The mission's scientific goal is to obtain near real time profiles of a bending angle, refractivity, temperature, pressure and water vapor in the neutral atmosphere and electron density in the ionosphere.The observations will be used to support operational global weather prediction, climate monitoring and research, space weather forecasting, and ionosphere and gravity research (Rocken et al. 2000;Wu et al. 2005a).
There are three science payloads on each satellite: a GPS Occultation Experiment (GOX) payload, a Tiny Ionospheric Photometer (TIP), and a Tri-Band Beacon (TBB) payload as shown in Fig. 1.The GOX payload is used to obtain radio occultation (RO) sounding of vertical profiles of refractivity which is a func tion of tem per a ture, pres sure and wa ter va por of the lower at mo sphere and elec tron den sity in the ion o sphere.The TIP and TBB are used to ob tain in for mation of the ion o spheric elec tron den sity.In this pa per we report the sta tus of the GOX and the qual ity of the at mo spheric data.We also pres ent a com par i son of RO sound ing data with dropsonde mea sure ments of at mo spheric pro files made during an In ten sive Ob ser va tion Pe riod (IOP) cam paign in 2006.
The GOX re ceiver mea sures the prop a ga tion time of radio sig nals from GPS sat el lites to FS-3/C sat el lites.As ra dio waves pass through the at mo sphere, they are re fracted and slowed, with the de gree of bend ing re lated to the pres sure, tem per a ture and wa ter va por in the at mo sphere and elec tron den sity in the ion o sphere (Fig. 2).From the raw phase and am pli tude mea sure ments, we can de duce ver ti cal pro files of bend ing an gles at the ray peri gee point.With the as sump tion of lo cal spher i cal sym me try, a ver ti cal pro file of refractivity can be re trieved us ing the Abelian in ver sion tech nique.The ver ti cal pro files of bend ing an gles or refractivity can be assim i lated into nu mer i cal weather pre dic tion mod els, yielding in for ma tion on tem per a ture, pres sure and wa ter va por.Ver ti cal pro files of tem per a ture and mois ture can be de duced from refractivity pro files with the use of an cil lary in for mation.Fur ther in for ma tion on the RO tech nique and the FS-3/C mis sion were pre sented in a spe cial TAO is sue (2000).
The GOX makes use of an ad vanced sig nal track ing tech nique known as open-loop (OL) track ing.With OL track ing, the RO soundings will pen e trate deep into the lower tro po sphere at all lat i tudes and lon gi tudes ob serv ing plan e tary bound ary layer heights and struc ture and providing valu able in for ma tion on low-level mois ture (Sokolovskiy et al. 2006a).
In the FS-3/C mis sion goal, the GOX is de signed to provide over 2500 at mo spheric soundings ev ery 24 hours in a nearly uni form dis tri bu tion around the globe as shown by green dots in Fig. 3.This is con sid er ably higher than the num ber of pres ent world-wide daily ra dio sonde ob ser vations (~1500 soundings from ~850 sta tions) as shown by gray dots in Fig. 3, which are con cen trated mostly over land ar eas.The RO soundings' high ver ti cal res o lu tion will com - ple ment the high hor i zon tal res o lu tion of con ven tional in frared and mi cro wave sat el lite soundings.The GOX will also ob tain the To tal Elec tron Con tent (TEC) along the GPS ray-path and the ver ti cal pro file of the elec tron den sity.The ~2500 elec tron den sity pro files be tween 90 and 800 km will de fine the ion o spheric struc ture, scin til la tion and elec tron den sity ir reg u lar i ties.
In this pa per we re port early re sults of RO soundings of ver ti cal at mo spheric pro files of the FS-3/C mis sion.In Section II we de scribe the sta tus of early RO soundings and com par i son with CHAMP re sults (Wickert et al. 2001(Wickert et al. , 2006)).In Sec tion III we de scribe the re sults of an In ten sive Ob ser va tion Pe riod (IOP) cam paign to cross-val i date the accu racy of refractivity pro files of the FS-3/C RO soundings by com par ing with refractivity pro files ob tained by us ing the at mo spheric mea sure ment from the dropsonde ex per i ment dur ing an IOP cam paign.In Sec tion IV we dis cuss the pos sible er ror source of the FS-3/C RO sound ing pro files.A summary is given in Sec tion V.

RA DIO OCCULTATION SOUNDINGS
Since the FS-3/C sat el lites were launched, more than 400000 at mo spheric ver ti cal pro files and 600000 ion o -spheric ver ti cal elec tron den sity pro files had been ex tracted from the ra dio occultation (RO) mea sure ments since 30 April 2007.The FS-3/C RO sound ing data pre sented in this pa per was down loaded from the Tai wan Anal y sis Cen ter for COS MIC (TACC: http://tacc.cwb.gov.tw).Cur rently, TACC RO soft ware is sup ported by UCAR, and the data pro cess ing soft ware will be iden ti cal to CDAAC's.
The num ber of atmPhs files (open di a monds) and the num ber of atmPrf files (solid cir cles) per day are plot ted in Fig. 4. The num ber of atmPhs files means the num ber of RO events ob served, and the num ber of atmPrf files means the num ber of RO events that could be re trieved to pro vide atmo spheric pro files.The atmPrf files in clude an in dex for data qual ity, and only pro files marked as good data are included in the num ber count.The good data pro files are deter mined by the high and low bound aries in al ti tude, phase fluc tu a tion in GPS sig nals, and the re frac tion bias with the at mo spheric model.The num ber of RO re trieval mea surements of ver ti cal pro files per day has been in creas ing with time dur ing the de ploy ment phase of the FORMOSAT-3 satellite or bits.The num ber of RO sound ing pro files is about 1800/day for the neu tral at mo sphere and 3000/day for the ion o sphere by April, 2007.How ever, the to tal daily num ber of GPS ex cess phase mea sure ments by the GOX in - stru ments has also been in creas ing as shown by the open di amonds in Fig. 4, in di cat ing that the un suc cess ful RO retrieval events were also in creas ing.The RO sound ing profiles are less than the num ber of GPS ex cess mea sure ments be cause some GPS ex cess mea sure ments could not be retrieved to ob tain RO sound ing pro files.The in crease in the to tal daily RO events from July, 2006 is mainly as so ci ated with the up load of a new OPEN LOOP firm ware (ver sion 4.3) for the sat el lite GOX sys tem one-by-one from 13 July 2006 for sev eral days and the daily RO event num ber took a large in crease in Au gust.There is a con sid er able in crease in the atmPhs data due to the new firm ware up load, but the percent age that atmPhs files were re trieved to ob tain atmPrf files did not in crease.It means the firm ware could track more ris ing RO data, but the data qual ity is not good enough to be re trieved to ob tain at mo spheric pro files.The firm ware needs to be im proved fur ther.
The white curves in Fig. 5 show the FORMOSAT-3 satellite or bits on 23 April 2007.The or bits of three sat ellites (FM#1, FM#3, and FM#4) are nearly su per im posed (the left white curves in Fig. 5).Two sat el lites (FM#1 and FM#3) are still in a tem po rary park ing or bit (at ~515 km al titude), FM#4 was just raised to the mis sion or bit and started to sep a rate from the FM#1 and FM#3 sat el lites due to different pre ces sion rate at dif fer ent al ti tude.The or bital planes of other sat el lites (FM#2, FM#5, and FM#6) were raised much ear lier and are sep a rated by 30 de grees in lon gi tude.The white dots in Fig. 5 show the lo ca tions of the at mospheric RO sound ing pro files in one day with 1405 RO events, which shows the FS-3 sat el lites' ca pa bil ity to observe at mo sphere glob ally, es pe cially over oceans and po lar re gions.With the FS-3 sat el lite or bit con stel la tion on 23 April 2007, the RO re trieval events were con cen trated in two time pe ri ods per day as shown in the lower panel of Fig. 6 which shows the hourly RO re trieval event num ber ver sus time for events within cir cles with a 2500 km ra dius cen tered at 120 de grees Lon gi tude and at zero de grees lat i tude (the dot ted and dashed line), at 25 de grees lat i tude (solid line), and at 50 de grees lat i tude (dot ted line), re spec tively.Af ter the de ploy ment of all six sat el lite or bits is fin ished, the RO  events will spread out more evenly as shown in the up per panel (ob tained by sim u la tion) in Fig. 6.
Dur ing the first few months af ter the FS-3 sat el lite launch the sat el lite or bits were very close to each other and stay al most in an or bital plane.It was the best time to ver ify that the GOX equipments on all FS-3 sat el lites all func tioned prop erly and con sis tently.When all sat el lites were close to each other, it was more prob a ble for all six sat el lites to ob tain RO re trieval pro files for the same area and at about the same time.Fig ure 7 shows the re sults of one such case on 7 May 2006.All six FS-3 sat el lites re ceived GPS sig nals that had very close paths from the same GPS sat el lite.There fore, if the max i mum hor i zon tal sep a ra tion dis tance be tween RO pro files (GPS ray tan gent points) is less than 100 ki lo me ter and the max i mum RO re trieval time dif fer ence is less than two min utes, these RO events would be com pared in this statis tics of Fig. 7.The right panel of Fig. 7 shows the dif ference be tween the refractivity pro file (la beled as C001) obtained by the FM#1 sat el lite and the refractivity pro files (labeled as C002 -C006) ob tained on 7 May 2006 by the other five FORMOSAT-3 sat el lites FM#2 -FM#6, re spec tively.The refractivity ob tained by dif fer ent FORMOSAT-3 sat ellites is quite sim i lar and their dif fer ences are larger be low 3 km lat i tude.Shown in the right panel of Fig. 7 are the dry tem per a ture pro files (la beled as C001 -C006) ob tained by all six FORMOSAT-3 sat el lites.Ex cept for the FM#6 sat ellite, the pro files of dry tem per a tures and refractivity from all other sat el lites are in good agree ment for al ti tudes above five ki lo me ters.The dry tem per a ture ob tained by the FM#6 sat el lite is larger than oth ers be cause the FM#6 GOX receiver has an older ver sion of firm ware than other FS-3 satel lites and it did not al low the FM#6 sat el lite's an tenna to track well with the GPS sig nal Dopp ler shift at higher al titudes.The prob lem hap pens some times in ris ing RO events.This prob lem was cor rected af ter the GOX firm ware of all six FS-3 sat el lites was up dated around 12 July 2006.
To eval u ate the qual ity of FS-3 GOX in stru ments, we per formed sta tis ti cal anal y sis of the GOX RO data for the pe riod from 16 April to 30 Sep tem ber 2006.The RO events ob tained by the FM#2 -FM#6 sat el lites were cho sen for analysis if they sat is fied the con di tions that the RO pro file locations (tan gent points) of FM#2 -FM#6 sat el lites were within 100 km in the hor i zon tal di rec tion from the RO profile po si tion of FM#1 sat el lite lo ca tion and the RO events were mea sured within one hour from the FM#1 RO measure ment.In all, about 20000 RO events sat isfy the constraints.In the left panel of Fig. 8, the solid line shows the bias of the frac tional refractivity dif fer ence [100 * (C002 -C001) / C001, etc.] rel a tive to the FM#1 refractivity pro files (C001), the dot ted lines show the stan dard de vi a tion of the frac tional refractivity dif fer ence rel a tive to the bias, and the dash line shows the num ber of RO events for refractivity com par i son ver sus dif fer ent al ti tude.Note that the num ber of RO events de creases as the low est al ti tude pen e tra tion of Fig. 6.Hourly RO re trieval event num ber ver sus time.The lower panel shows the num ber of re trieval events from 23 -29 April 2007.The up per panel shows the num ber of re trieval events ob tained by sim u la tion as sum ing that all six FORMOSAT-3 sat el lites are in the planned fi nal or bits.The dash-dot ted (solid, dot ted) curves are the hourly event num ber within a cir cle area cen tered at 0 (25, 50) de gree Lat i tude and 120 de gree Lon gi tude with a ra dius of 2500 km.
RO sound ing pro files de creases.The bias shows that the refractivity pro files ob served by dif fer ent FS-3 sat el lites are con sis tent among each other with a small vari a tion be tween 2 and 5 km al ti tude.Al though the bias is very small, the standard de vi a tion be low 5 km al ti tude is some what large with a value of 6% at around 1.5 km al ti tude.The large stan dard de -  vi a tion be low 5 km al ti tude may be caused by com plex at mospheric con di tions or the as sump tions made in the re trieval scheme such as the spher i cal sym me try of the at mo sphere.
To eval u ate the qual ity of FS-3 RO soundings rel a tive to the CHAMP mea sure ments, we per form a sta tis ti cal anal y sis us ing data for the pe riod from 1 Jan u ary to 10 May 2007.The CHAMP RO sound ing data are down loaded from the Tai wan Anal y sis Cen ter for COS MIC (TACC: http://tacc .cwb.gov.tw).The FS-3 RO events were cho sen for the analysis if the FS-3 sound ing lo ca tions (GPS ray tan gent points) were within 100 km in a hor i zon tal di rec tion from the CHAMP sound ing lo ca tions and the FS-3 RO events were mea sured within one hour from the CHAMP RO measure ment.In to tal, there are over 800 FS-3 RO events which sat is fied the re quire ments.In the right panel of Fig. 8 the solid line shows the bias of the frac tional dif fer ence of FS-3 refractivity pro files rel a tive to the CHAMP refractivity profiles, and the dot ted lines show the stan dard de vi a tion of the frac tional dif fer ence rel a tive to the bias and the dashed line shows the num ber of RO events for refractivity com par i son ver sus dif fer ent al ti tude.The bias (CHAMP -COS MIC) between FS-3 and CHAMP refractivity pro files are con sis tent.The Stan dard de vi a tion is as large as in the lower panel of Fig. 8.There is a peak around 12 km, which is due to some bad RO pro files with their bot tom al ti tude at ~12 km.The rea son for these bad RO pro files is still not clear ne ces si tating more study to un der stand this fea ture.
Note that the CHAMP RO re trieval pro files were obtained by us ing the CLOSE LOOP tech nique.The OPEN LOOP scheme al lows RO re trieval even when the re ceived GPS sig nal is weak or is in ter fered by the multi-path ef fects, but the CLOSE LOOP tech nique al lows RO re trieval only when the re ceived GPS sig nal is rea son ably clear.The detailed de scrip tion of the dif fer ent func tions be tween CLOSE LOOP and OPEN Loop has been dis cussed by Sokolovskiy et al. (2006b).Thus, when a CHAMP RO sound ing can reach be low 5 km, its RO soundings should have as good a qual ity as the FS-3/C RO soundings.
Be fore the de vel op ment of the OPEN LOOP scheme for RO re trieval of at mo spheric pro files, the low est al ti tude of ver ti cal pro files of the re trieved at mo spheric quan ti ties was too high, and the abil ity to re trieve at mo spheric quan ti ties be low 5 km al ti tude was quite lim ited [see for ex am ple, the re sults of CHAMP (Beyerle et al. 2006) and MicroLab-1 satellites (Gurvich et al. 2006)].Adopt ing the Open Loop technique de vel oped by the Jet Pro pul sion Lab o ra tory, the FORMOSAT-3 GOX re ceiv ers are able to re trieve at mospheric pro files down to quite low al ti tudes.Fig ure 9 shows the global dis tri bu tions of the low est al ti tude pen e tra tion of the RO sound ing pro files ob tained by FORMOSAT-3 sat ellites (135312 pro files, in all) and the CHAMP sat el lite (9512 pro files in to tal) for the pe riod from 1 Jan u ary to 10 May 2007.Over land, the pro file pen e tra tion is lim ited by orography.Thus, we sep a rate the RO sound ing pro files into two groups: one group is over the ocean and the other group is over the land.In Fig. 9, the up per and lower pan els show the dis tri bu tion of the low est al ti tude pen e tra tion of the RO re trieval pro files of the land group ob tained by the FS-3 and CHAMP sat el lites.Here the al ti tude is mea sured from the earth sur face.We note that they are mostly be low 0.5 km in the south ern po lar re gion.In most other land re gions the low est al ti tudes are all be low 1 km.Those with low est al titude above 1 km are mostly lo cated in moun tain ous ar eas such as Hi ma la yas, Ti betan high lands, and An des moun tains in the South Amer ica be cause high moun tains pre vent GPS RO sig nals with a lower peri gee point al ti tude from be ing de tected.More over, the low est al ti tude pen e tra tion of CHAMP RO soundings is sig nif i cantly higher than the FS-3 RO soundings in an equa to rial re gion.Note that be cause of the rel a tively high in cli na tion of the FS-3 sat el lite or bits, the num ber of RO sound ing events in an equa to rial re gion is much less than the RO event num ber in an high lat i tude region.We see that the low est re trieval al ti tude pen e tra tion is lower in a south ern po lar re gion and is higher in an equa torial re gion in com par i son with other ocean ar eas for both FS-3 and CHAMP data.Also, the low est al ti tude pen e tra tion of CHAMP RO soundings is sig nif i cantly higher than the FS-3 RO soundings in equa to rial and south ern mid-lat i tude re gions.Note that the pen e tra tion of CHAMP RO soundings can reach be low 2 km mainly from a higher lat i tude.
Fig ure 10 shows the al ti tude ver sus the per cent age of RO sound ing events that pen e trate to or be low the al ti tude value for the FORMOSAT-3 and CHAMP RO sound ing events.The RO sound ing events cov ered in com put ing the sta tis tics are ob tained from 1 Jan u ary to 10 May 2007.In the left panel, the al ti tude is mea sured from the earth's sur face and the sta tis tics are com puted for to tal events and for events over oceans for FORMOSAT-3 and CHAMP sep a rately.We see that more than 80% of the RO events (in which 50% of the RO events oc cur over oceans and 30% of RO events over land ar eas) have the low est al ti tude pen e tra tion equal or below 1 km for the FS-3 events, but only 50% (30% are over oceans and 20% are over land ar eas) for the CHAMP events.The right panel of Fig. 10 shows the ac cu mu lated per cent age of RO events ver sus al ti tude for RO sound ing events over land ar eas with the al ti tude mea sured from the earth's sur face (la beled as ES) and the sea level (la beled as SS) se p a rately.
To un der stand the ef fect of the spe cific hu mid ity on the low est al ti tude pen e tra tion of the RO sound ing pro files, we pres ent, in Fig. 11, the sta tis tics of the low est al ti tude pen etra tion ver sus lat i tude for the FS-3 and CHAMP RO sounding events dis cussed con cern ing Fig. 10.In Fig. 11 the left panel was ob tained from the FORMOSAT-3 RO events and the right panel was ob tained from the CHAMP RO events.The solid line rep re sents the low est al ti tude pen e tra tion for a me dian num ber of events, i.e., 50% of the sound ing pro files have the low est al ti tude be low the al ti tude given by in dicated by a solid curve.Sim i larly, the dashed line rep re sents the first quartile num ber of lower al ti tude pen e tra tion events, and the dash-dot ted line rep re sents the third quartile num ber of events.For the me dian num ber of events the low est altitude pen e tra tion is 0.3 km in the north ern po lar re gion, 0.1 km in the 70 th de gree lat i tude in the south ern po lar region, 0.5 km in a mid-lat i tude re gion, and about 1 km in an equa to rial re gion.The re sults for CHAMP data are shown in the right panel of Fig. 11, and for the me dian num ber of events the low est al ti tude pen e tra tion is 0.5 km in a po lar region and 2 -3 km in an equa to rial re gion.The re sults clearly show that us ing the Open Loop tech nique the FS-3 RO events have lower al ti tude pen e tra tion than CHAMP events which were ob tained us ing the Closed Loop tech nique.More over, for the third quartile num ber of events, the low est Fig. 10.The al ti tude ver sus the per cent age of RO sound ing events that have the low est al ti tude pen e tra tion equal or be low the al ti tude.The FORMOSAT-3 and CHAMP RO events cov ered in the sta tis tics are ob tained from 1 Jan u ary to 10 May 2007.In the left panel the al ti tude is mea sured from the earth sur face and the sta tis tics are com puted for the to tal events and for events over oceans for FORMOSAT-3 and CHAMP, sep a rately.The right panel shows the sta tis tics for events above land ar eas with an al ti tude mea sured from the earth sur face (la beled as ES) and the sea level (la beled as SS), re spec tively.al ti tude pen e tra tion is about 1.6 km in an equa to rial re gion for the FS-3 events and much lower than 3.8 km for CHAMP events.To find the cor re la tion of the low est al ti tude pen e tration of RO events with re gard to wa ter va por we also show the spe cific hu mid ity dis tri bu tion in al ti tude and lat i tude (the con tour value is shown by a color bar) along the zero de gree lon gi tude, which is ob tained by av er ag ing the spe cific humid ity over the month of March from 1968 to 1996 based on the NCEP (Na tional Cen ter for En vi ron men tal Pre dic tion) model.Ex clud ing the RO data in the po lar re gion, we find that the low est RO al ti tude pen e tra tion is roughly pro portional to the spe cific hu mid ity.Thus, our re sult con firms the com mon per cep tion that the pen e tra tion of RO is highly related to mois ture dis tri bu tion.Both the low est RO al ti tude and the NCEP spe cific hu mid ity are larger in a low lat i tude re gion.Thus, the wa ter va por is an im por tant fac tor in de termin ing the low est al ti tude pen e tra tion of GPS RO re trieval of at mo spheric ver ti cal pro files.

THE IN TEN SIVE OB SER VA TION PE RIOD (IOP) CAM PAIGN
In or der to per form cross data val i da tion be tween the GPS RO sound ing pro files and di rect mea sure ments of atmo spheric pro files, we car ried out an In ten sive Ob ser va -tion Pe riod (IOP) cam paign in 2006.Spe cif i cally we performed dropsonde mea sure ments and com pared the di rect mea sure ment re sults with FS-3 RO re trieval pro files.The com par i son be tween the FS-3 RO sound ing pro files and the ra dio sonde mea sure ments have been re ported by Wang and Lin (2007) and will not be re viewed here; how ever, com par i sons be tween dropsonde and FS-3 soundings have not been re ported, thus we re port case stud ies sup ported by the IOP cam paign dur ing ty phoons in this pa per, as found in Figs. 12 and 13, re spec tively.The left panel of Fig. 12 shows the lo ca tions of the dropsonde mea sure ments (open cir cles marked as D1 and D2) and the FS-3 RO events (solid cir cles marked as C1 and C2) per formed on 29 Oc tober 2006.The dis tances be tween the dropsonde mea surements and the FS-3 RO events were at about 100 km with a time dif fer ence of about one hour.The left fig ure in the right panel of Fig. 12 shows a com par i son be tween the refractivity pro files ob tained from at mo spheric pa ram e ters mea sured by drop sondes and the refractivity pro files obtained by the clos est FS-3 RO re trieval.The refractivity obtained from the drop sonde mea sure ments was com puted us ing the em pir i cal for mula given in Eq. ( 1) by Anthes et al. (2000), where the refractivity, N, is re lated to at mospheric pa ram e ters by which N = 77.6P/T + 3.73 ´ 10 5 e/T 2 , where T is the tem per a ture in Kel vin, P and e are the to tal pres sure and wa ter va por pres sure in mbar, re spectively.The time sam pling of the dropsonde is 0.5 sec onds.Based on the dropsonde in stru ment spec i fi ca tion pro vided by the man u fac turer, Va isala Corp., the dropsonde has a 0.2°C and 2% ac cu racy on its THERMOCAP ca pac i tive tem per a ture sen sor and H-HUMICAP thin film ca pac i tor hu mid ity sen sor in di vid u ally.Here, we use the AS PEN (At mo spheric Sound ing Pro cess ing EN vi ron ment) software de signed by NCAR to pro cess the orig i nal dropsonde pro file for our data study.The ver ti cal res o lu tion of tem pera ture and hu mid ity data from the dropsonde is 10 m on aver age.The tem per a ture sen sor is made of fine plat i num  wire with a re flec tive coat ing to min i mize heat ing from sun light and the re ten tion of wa ter drop lets.The dropsondes used around the cir cu la tion of ty phoons (Wu et al. 2005b) have rea son able sound ing pro files for weather diag no sis and model as sim i la tion.We found the dropsonde works well un der gen eral and se vere weather con di tions.The mid dle fig ure in the right panel of Fig. 12 shows the dif fer ence in refractivity (solid line is D1 -C1 and dot ted line is D2 -C2) be tween the FS-3 RO refractivity and the dropsonde mea sure ments.The max i mum dif fer ence in refractivity is about 10 be low 4 km al ti tude ex cept near the ground for D2 -C2.The right fig ure in the right panel of Fig. 12 shows the ab so lute value of the dif fer ence in refractivity di vided by the FORMOSAT-3 refractivity val ues (the solid line is 100 * |(D1 -C1) / C1| and the dot ted line is 100 * |(D2 -C2) / C2|).We see that the dif fer ence is about 5% or less on av er age, but is larger (about 8%) at about 3 -4 km al ti tude.Thus, based on these com par i sons we conclude that the FS-3 RO re trieval pro files are rea son ably accu rate even dur ing a se vere weather con di tion.Due to a lim ited bud get only a few dropsonde flights were car ried out.Also, the FS-3/C RO data den sity is still quite low with an av er age dis tance be tween neigh bor ing pro files at ~400 km when con sid er ing one day's RO data.Thus, only 4 cases with rea son ably close co-lo ca tions are re ported in this section.How ever, more com par i sons should be car ried out to reach a more de fin i tive con clu sion.
In or der to check the ac cu racy of GPS RO sound ing pro files in se vere weather con di tions, we per formed cross data val i da tion be tween the GPS RO sound ing pro files and di rect mea sure ments of at mo spheric pro files dur ing the Kamei ty phoon on 22 July 2006.The left panel of Fig. 13 shows the lo ca tions of two dropsonde mea sure ments (open cir cles marked as D1 and D2) and one FS-3 RO event (solid cir cle marked as C1).The dis tance be tween the dropsonde mea sure ments and the FS-3 RO event is about 100 km, and the time dif fer ence is about one hour.The left fig ure in the right panel of Fig. 13 shows the com par i son be tween the refractivity pro files ob tained from the at mo spheric pa ram eters mea sured by dropsondes (via the em pir i cal for mula) and the FS-3 RO refractivity pro file.The mid dle fig ure in the right panel of Fig. 13 shows the dif fer ence in refractivity (solid line is D1 -C1 and dot ted line is D2 -C1) be tween the FS-3 RO refractivity and that ob tained from the dropsonde wea ther mea sure ments.The max i mum differ ence in refractivity is less than about 3. The right fig ure in the right panel of Fig. 13 shows the ab so lute value of the dif fer ence in re fractivity di vided by the FORMOSAT-3 refractivity val ues (the solid line is 100 * |(D1 -C1) / C1| and the dot ted line is 100 * |(D2 -C1) / C1|).We see that the per cent age dif fer ence is about 3% or less ex cept near the ground where the per cent age dif fer ence is about 5%.Thus, we con clude that the FS-3 RO sound ing pro files are quite ac cu rate.

ER ROR ANAL Y SIS
The well-de tailed RO re trieval method had been published in many pa pers, es pe cially by the UCAR COS MIC group (Kursinski et al. 2000;Kuo et al. 2004).Here, we simply in tro duce the RO data pro cess ing method to ex plain the source of an RO er ror more clearly.The data pro cess ing in cludes sev eral main steps.First, we de rive the Dopp ler shift from the RO ob ser va tions.Then, we cal cu late the bending an gle, and then de rive the re frac tion (n) pro files by the fol low ing equa tion: (1) where a(a) is the bend ing an gle, a is the im pact pa ram e ter which is equal to the re frac tion multi ra dius (r) from the earth cen ter, the sub script of "Top " in di cates the top boundary that RO could ob serve or would be cho sen by optimization pur pose.The refractivity N is cal cu lated by N = 10 6 (n -1) (2) The pres sure P(z) could be de rived by the fol low ing grav ity force bal ance equa tion: (3) where R is the mix value be tween the dry spe cific con stant [Rd: 461.52 J / (Kg×K)] and the wa ter va por spe cific constant [Rv: 461.52 J / (Kg×K)].In the ini tial pro cess ing in RO re trieval, the con tri bu tion of wa ter va por is ig nored and R will be set to equal to Rd.The de pend ence of refractivity on pres sure (P in mbar), tem per a ture (T in Kel vin) and va por pres sure (e in mbar) is taken from the em pir i cal for mula: (4) The dry tem per a ture in atmPrf files is de rived by ig nor ing the wa ter va por pres sure in the above equa tion.In or der to ob tain the cor rect tem per a ture, the above for mula is rewrit ten as (5) where T on the right hand side of the equa tion is taken as the dry tem per a ture.Thus, the dry tem per a ture would be lower than a true tem per a ture in clud ing the ef fect of wa ter va por.Kuo et al. (2004) men tioned that the mag ni tude of a bending an gle and the neu tral at mo sphere-re lated vari a tions decrease ex po nen tially with al ti tude, thus, the mag ni tude of noise re mains roughly con stant and over shad ows the sig nal above a cer tain al ti tude.In or der to re duce the er ror prop a ga tion from high to low al ti tude af ter the Abel in ver sion, the UCAR group uses mod els and cost func tions to op ti mize the bend ing angle.The op ti mal bend ing an gle pro file is (Kuo et al. 2004): The sta tis tics is cal cu lated for ev ery 10-de gree in ter val in lat i tude.Ac cord ing to the re sults of atmPrf re trieved by the UCAR COS MIC group, the weight ing func tion w guess in Eq. ( 7) is zero be low 20 km in al ti tude, and the ob served bend ing an gle is used to de rive the at mo spheric parameter pro files.The left panel of Fig. 14 shows the mean bias ra tio (MBR in %) ver sus lat i tude (global) and the bias is very small at 30 km al ti tude.MBR be comes big ger (~2%) around a south ern po lar re gion at 35 km al ti tude, and at 40 km al ti tude MBR has two peaks (~1%) on both sides of the equa tor and also big ger bias in a north ern po lar re gion.The re sults show that the bias above 20 km al ti tude is larger than zero and means that most ob served bend ing an gles are larger than the guessed bend ing an gle.A larger bias also means that a higher weight ing of the guessed bend ing an gle is used.The right panel of Fig. 14 shows the stan dard de via tion of MBR (std.MBR in %) ver sus lat i tude at five dif ferent al ti tudes (20, 25, 30, 35, and 40 km) for an RO event bias.The solid line pres ents the RO num ber (scale is on the right ver ti cal axis) for the sta tis tics.The std.MBR is very small (< 0.3%) be low 25 km and be comes big ger at a higher al ti tude.At a higher al ti tude above 25 km, the std.MBR has a local peak at ~15 de gree lat i tude in the north ern hemi sphere, but be comes big ger in both north ern and south ern po lar re gions.
Ac cord ing to the in te gral for mula found in Eq. ( 1), the bias at an up per al ti tude would in flu ence the ac cu racy of refrac tion.For tu nately, the er ror in flu ence de creases with the square root of (a 2 -a o 2 ) which is larger for a re frac tion at lower al ti tude.The RO re sults at lower al ti tude should be more ac cu rate if the wa ter va por does n't ex ist.Above ~25 km, we should know how much the guessed bend ing an gle is weighted.In Eq. ( 4), the refractivity and pres sure could be re - trieved from Eqs. ( 2) and (3), re spec tively.There are still two un known pa ram e ters (tem per a ture and wa ter va por pres sure) that should be solved.The wa ter va por pres sure is ig nored to ob tain dry tem per a ture in atmPrf, the other in forma tion is then added to de rive both wa ter va por pres sure and wet tem per a ture as given in Eq. ( 5).The 1D VAR re trieval of wet tem per a ture and wa ter va por was dis cussed by Engeln et al. (2003).The 1D VAR re trieval method is used when wa ter va por ex ists, so it is al ways used at lower al ti tude.Be cause the true value of R in Eq. ( 3) is re placed by Rd, it er a tion should be per formed to im prove pres sure and wet tem per ature and wa ter va por si mul ta neously and the ac cu racy of the model is crit i cal for ob tain ing ac cu rate wa ter va por pres sure.
The other source of er ror in the RO re trieval scheme used in the pres ent stage of FS-3 GOX op er a tion is the use of the Abel trans form (Kursinski et al. 2000) and the as sumption that the at mo spheric refractivity has a spher i cal sym metry with out hor i zon tal gra di ent and de pends only on al ti tude.The cor rec tion due to the earth oblateness can be largely com puted by re de fin ing the cen ter and ra dius of the sphere at the lo cal RO event (Syndergaard 1998).How ever, the dis tribu tions of the at mo spheric tem per a ture, pres sure and wa ter va por are not spher i cally sym met ric and the refractivity varies in al ti tude, lat i tude and lon gi tude and changes with different weather sys tems.The refractivity gra di ent in the hor i zon tal di rec tion is mainly due to a hor i zon tal vari a tion of wa ter va por be low 5 km al ti tude and is the main source of error when the RO data is re trieved.The hor i zon tal refractivity gra di ent in the ion o sphere would also in flu ence the ac cu racy of RO pro files es pe cially above 30 km.Because the refractivity above 30 km al ti tude is very small, a small er ror in that al ti tude re gion might cause large de vi a tion rel a tive to the true refractivity pro files.More over, the RO peri gee point of the GPS ray at dif fer ent height is not fixed in lat i tude and lon gi tude as shown in Fig. 2. The up per panel of Fig. 15 shows the hor i zon tal span of lat i tude and lon gi tude pro jec tion of all GPS RO re trieval ver ti cal pro files ob tained on 20 May 2007.Each hor i zon tal span cov ers the peri gee point pro jec tion of the RO ver ti cal pro files from the low est al ti tude to 20 km above the low est al ti tude.The left panel of Fig. 15 shows the per cent age of events ver sus the hor i zon tal span and about 50% of the RO events have hor i zon tal span be tween 1 and 2 de grees, which cor re sponds to a dis tance of 120 to 240 km on the earth sur face.Thus, the hor i zon tal span of the ver ti cal pro files can be the ma jor source of er ror in the RO re trieval scheme.This er ror should be ad dressed as a future im prove ment for the RO re trieval scheme.

SUM MARY
In this pa per we have re ported the sta tus of the FOR -MOSAT-3/COS MIC mis sion and ex am ined the qual ity of the GPS ra dio occultation re trieval of refractivity pro files ob tained by the GOX re ceiv ers aboard the FORMOSAT-3/ COS MIC sat el lites.We have found that the qual ity and accu racy of FS-3/C and CHAMP RO data are sim i lar, but the at mo spheric RO soundings ob tained by the FS-3 sat el lites in gen eral pen e trate to a lower al ti tude than CHAMP's.The lower at mo spheric pro files are very im por tant in an a lyz ing wa ter va por data.We have also ex er cised an In ten sive Obser va tion Pe riod (IOP) cam paign to per form cross data val ida tion of GPS RO refractivity pro files against refractivity ob tained by the ra dio sonde (Wang and Lin 2007) and dropsonde mea sure ments of at mo spheric pa ram e ters via the empir i cal for mula be tween refractivity and at mo spheric parameters.Al though only four col lo cated cases are pre sented, the refractivity ob tained by GPS RO is quite con sis tent with refractivity com puted from the dropsonde mea sure ments, which im plies the good qual ity of the FS-3 mea sure ments.This in di cates that the GPS RO re trieval of ver ti cal re - fractivity pro files ob tained by the FS-3/C sat el lites is quite ac cu rate and of high qual ity.In April, 2007 the to tal daily num ber of at mo spheric RO soundings reached about 1800 -2000.By the end of 2007 the or bit de ploy ment of all six FORMOSAT-3 sat el lites should be com plete, and we ex pect the to tal daily num ber of RO soundings to reach about 2500, dis trib uted nearly uni formly around the globe.
The FS-3/C weather data has been dis trib uted to ma jor weather re search cen ters world-wide in near-real time.We ex pect that if a dem on stra tion of near-real time op er a tional use of RO sound ing ob ser va tions can be per formed, FS-3/C will com ple ment other Earth ob serv ing sys tems and improve global weather anal y ses and pre dic tion.Be cause the FS-3/C mis sion will pro vide un prec e dented amount of RO sound ing data (~2000 or more daily RO soundings dis tributed rather uni formly over the globe in clud ing the oceans and po lar re gions where there are in suf fi cient weather data), the FS-3/C mis sion can po ten tially in au gu rate an age of oper a tional GPS sound ing for weather fore cast ing, cli mate anal y sis and re search, ion o spheric mon i tor ing, and a suite of re lated earth sci ence pur suits.

Ac knowl edge ment
The most part of this ar ti cle was done when I worked at NSPO, and was fin ished when I am Post-Doc at Boston col lege.

Fig. 1 .
Fig. 1.A FS-3/C sat el lite and its sci ence pay loads: GPS Ra dio Occultation Ex per i ment (GOX), Tiny Ion o spheric Pho tom e ter (TIP) and Tri-Band Bea con (TBB).The occultation and POD (Pre ci sion Or bit De ter mi na tion) an ten nas are in the x-z plane and are in clined to na dir and ze nith, re spec tively.The TIP port and the TBB an tenna point to the na dir di rec tion.The so lar pan els open up in the y-z plane af ter the launch of sat el lites.

Fig. 3 .
Fig. 3.Over 2500 RO soundings (gray dots; each dot rep re sents the lo ca tion of RO sound ing) are ob tained by GOX re ceiv ers each day.Black dots indicate the lo ca tion of the pres ent world-wide ~850 ra dio sonde sta tions.

Fig. 4 .
Fig. 4. The to tal num ber of at mo spheric RO sound ing pro files (atmprf, solid cir cles) per day ob tained by FORMOSAT-3 sat el lites ver sus date.The open di a monds (atmPhs) show the to tal num ber of GPS ex cess phase mea sure ments that can be com puted from the GPS sig nals re ceived by FORMOSAT-3 sat el lites.

Fig. 5 .
Fig. 5. Or bits (white curves) of FORMOSAT-3 sat el lites on 29 April 2007.The white dots in di cate the RO sound ing lo ca tions in one day.

Fig. 7 .
Fig. 7.The left panel shows the dif fer ence be tween the refractivity pro file ob tained by the first FORMOSAT-3 sat el lite (la beled as C001) and those ob tained by the other five FORMOSAT-3 sat el lites (la beled as C002 -C006) on 7 May 2006.The right panel shows the dry tem per a ture pro files obtained by all FORMOSAT-3 sat el lites.All six sat el lites re mained very close to each other on 7 May 2006 which is about 3 weeks af ter the launch of FORMOSAT-3 sat el lites.

Fig. 8 .
Fig. 8.The left panel shows the bias and stan dard de vi a tions from the sta tis tics for the frac tional dif fer ence [i.e., 100 * (C002 -C001) / C001, etc.] between the refractivity pro file ob tained by the first FORMOSAT-3 sat el lite (C001) and those ob tained by the other FORMOSAT-3 sat el lites (C002 -C006) when their pro file lo ca tions are within 100 km in hor i zon tal di rec tion from the first sat el lite and the RO events are mea sured within one hour from the first sat el lite RO mea sure ment.The events cov ered in the sta tis tics were ob tained from 16 April to 30 Sep tem ber 2006.The dashed curve shows the to tal num ber of RO events used in com put ing the bias and stan dard de vi a tions.The com par i son be tween the CHAMP soundings and the FORMOSAT-3 RO soundings is shown in the right panel with the same cri te ria.The events were ob tained from 1 Jan u ary to 10 May 2007.

Fig. 9 .
Fig. 9.The left and right pan els show the dis tri bu tion of the low est al ti tude pen e tra tion (in di cated by col ors) of RO soundings above land ar eas obtained by FORMOSAT-3 sat el lites and CHAMP, re spec tively.The al ti tude is mea sured from the earth sur face for both cases.

Fig. 11 .
Fig. 11.The low est al ti tude pen e tra tion ver sus lat i tude for RO events ob tained from 1 Jan u ary to 10 May 2007.The left and right pan els are ob tained from the FORMOSAT-3 and the CHAMP RO events, re spec tively.The solid line rep re sents the me dian value of the low est al ti tude pen e tra tion, the dashed line rep re sents the first quartile value, and the dash-dot ted line rep re sents the third quartile value.Also shown is the wa ter va por spe cific humid ity dis tri bu tion in al ti tude and lat i tude (con tour value is shown by color bar).The spe cific hu mid ity is ob tained by av er ag ing over the month of March from 1968 to 1996 at 0 de gree Lon gi tude based on the NCEP model.

Fig. 12 .
Fig. 12.The left panel shows the po si tions of two FORMOSAT-3 RO events (solid cir cles marked as C1 and C2) and two dropsonde events (open circles marked as D1 and D2) ob tained from the air plane op er a tion on 29 Oc to ber 2006.The left fig ure in the right panel shows the refractivity in log arith mic scale ver sus al ti tude for the FORMOSAT-3 events (C1 and C2) and dropsonde events (D1 and D2).The mid dle fig ure in the right panel shows D1 -C1 (solid line) and D2 -C2 (dot ted line).The right fig ure in the right panel shows 100 * |(D1 -C1) / C1| (solid line) and 100 * |(D2 -C2) / C2| (dotted line).

Fig. 13 .
Fig. 13.The left panel shows the po si tions of a FORMOSAT-3 RO event (red solid cir cle, marked as C1) and sev eral dropsonde events dur ing the Kamei ty phoon on 22 July 2006.The left fig ure in the right panel shows the refractivity in log a rith mic scale ver sus al ti tude for the FORMOSAT-3 event (C1) and two dropsonde events (open cir cles, marked as D1 and D2).The mid dle fig ure in the right panel shows D1 -C1 (solid line) and D2 -C2 (dot ted line).The right fig ure in the right panel shows 100 * |(D1 -C1) / C1| (solid line) and 100 * |(D2 -C2) / C2| (dot ted line).
a guess (a) is cal cu lated by mod els, the weight ing functions w obs (a) and w guess (a) are de ter mined by the bend ing angle ob ser va tion qual ity.The model for a guess (a) is based on the at mo spheric NCEP in near-real time data pro cess ing, and the at mo spheric ECMWF model in post data pro cess ing.We com pare the ob served bend ing an gles and the optimal bend ing an gles in the atmPrf files ob tained by the UCAR COS MIC group by us ing the bend ing an gle data from Day 200, 2006 to Day 289, 2006, and the to tal RO num ber is 102954.Fig ure 15 shows the Mean Bias Ra tio (MBR) in the left panel and the stan dard de vi a tion in the right panel of MBR (std.MBR) at dif fer ent al ti tude ver sus latitude for 90 days, and the MBR for mula is given by (7) and std.MBR is cal cu lated by (8)

Fig. 14 .
Fig. 14.The left panel shows the mean bias ra tio (in %) ver sus lat i tude from Day 200, 2006 to Day 289, 2006 at three dif fer ent al ti tudes (30, 35, and 40 km).The right panel shows the stan dard de vi a tion (in a per cent age) of RO event bias.

Fig. 15 .
Fig. 15.The up per panel shows the global dis tri bu tion of the hor i zon tal span of the ver ti cal pro file of each FORMOSAT-3 RO event ob tained on 20 May 2007.The left panel shows the per cent age of the FORMOSAT-3 RO events ver sus the hor i zon tal span of the ver ti cal pro files.