China's meteorological satellite program was initiated in the 1970s. Over the past 40 years, 13 meteorological satellites have been developed. The polar and geostationary series satellites, known as Fengyun-1(3)(FY1(3)) and Fengyun-2(FY2), respectively, are incurrent operation. They are ranked by the World Meteorological Organization as being part of the WorldWeather Watch space component. After the launchof these satellites, the ground station has responsibility for the management and utilization of the satellitesystem, including data receiving, processing, and application(Xu et al., 2006). This study introduces innovations in the data processing algorithm for the Chinese FY meteorological satellites.

The radiometer onboard the FY2 meteorologicalsatellite senses atmospheric and earth objects remotelyin pixel format. The obtained data are then processedat the ground station. Data processing includes imagenavigation, radiation calibration, and product generation and assimilation. Image navigation yields information on the location of the observed objects. Radiation calibration indicates the amount of radiationreceived from the observed objects by the satellite. Data assimilation helps to improve numerical weatherprediction. This study uses these three measures todescribe the innovations in the FY data processing algorithm.

Section 2 introduces the FY2 image navigationalgorithm(Lu et al., 2008). As a geosynchronous meteorological satellite, FY2 makes its observations with5-km infrared(IR)or 1. 25-km visible resolution and angular resolutions of 140 and 35 µrad at an orbit36000 km above the earth's surface. Although it iscalled a geostationary satellite, in reality, the position and altitude of the satellite vary. The FY2 imagenavigation algorithm does not use the traditional l and mark matching technique. In fact, a time series of theearth's disk center-line count and the angle subtendedat the satellite by the sun and earth provide accurateinformation on the altitude and misalignment parameters of the satellite. With this information, pixel-levelimage navigation and minor animation shifts in thegeostationary orbit may be addressed.

Section 3 introduces the radiometric calibrationalgorithm for the FY2 geostationary meteorologicalsatellite, which contains two main components, i. e., inner-blackbody(IBB)calibration and lunar calibration for in-orbit conditions. In particular, the diurnalvariation in the frequency of the radiometric responsein the IR band and the annual systematic variationof the IBB calibration are solved and successfully applied in currently operating satellites 2D, 2E, and 2F. The accuracy of the calibration results is high, reaching levels better than 1 K.

Section 4 introduces the satellite instrument parameters on-orbit optimizer(SIPOn-Opt). It is widelyassumed that meteorological satellite on-orbit instruments are consistent with their design, that the instrumental performance parameters on-orbit are consistent with the values measured before the satelliteis launched, and that even if certain differences exist, there is no effective way to calculate such values. AnSIPOn-Opt for the polar orbit meteorological satellite was developed to optimize the true state of theinstrument parameters on-orbit with respect to theobservational constraints, and to help diagnose and correct the dominant observing system biases in thesatellite data preprocessing system. When applyingSIPOn-Opt to the FY3 sounding instruments, the FY3data are much improved compared with data from itscounterparts, the European and the U. S. polar orbitmeteorological satellites, and the forecast skill of thenumerical weather prediction model is also improved(Lu et al., 2011a, b, 2012). When the SIPOn-Opt isapplied to meteorological satellite instruments(MSU and AMSU-A)from Europe and the U. S., the quality and effectiveness of their data in applications over thepast 40 years are also improved(Lu et al., 2014).

Section 5 summarizes these major innovations inthe data processing algorithm for FY satellite dataprocessing. 2. Image navigation algorithm for FY2 geosyn-chronous meteorological satellites2. 1 The earth disk location in the south-north direction

An ideal spin-stabilized geostationary meteorological satellite should have a round orbit on the equatorial plane with its spin axis parallel to the earth'srotation axis. When those conditions are satisfied, theimages observed are nominal without any animationshifts. In reality, these conditions are never satisfied. The real orbit and altitude show a minor departurefrom the nominal ones. In a geostationary orbit, however, these minor deviations in orbit and altitude havean innegligible influence on image navigation.

A time series of the earth's disk center-line counts, namely the north-south movement of the earth'sdisk in the image during 7-8 June 2006 is shown inFig. 1a. The movement appears as a simple sinusoidal function. The behavior of the line counts is wellrepeated. When the image origins are placed at thefirst line and column and then rendered as an imageanimation, the earth's disk shifts in the north-southdirection one cycle each day. When the image originsare placed at the earth's disk center and then renderedinto an image animation, the earth's disk swings bothclockwise and counterclockwise one cycle each day. InFig. 1b, the sinusoidal function is simulated with thedata during 7-8 June 2006(black dots) and is then extended to 9 June 2006(hollow dots). The good overlapof the hollow dots on the extension of the sinusoidalcurve shows that the earth's disk center-line count ispredictable.

Fig. 1. Time series of the earth's disk center-line count for FY2C during(a)7-8 and (b)7-9 June 2006. The ordinateis the earth's disk center-line count(increasing downward), the abscissa is the time(UTC). Black dots represent theprevious earth disk center-line counts for 7-8 June 2006 on which the simulation and extension are based, the curve isthe simulation and extension of the earth disk center-line counts, and hollow dots are the future observations of the earthdisk center-line counts for 9 June 2006, which are independent of the extension of the curve. [From Lu et al., 2008]

Figure options

The phenomena described above reflect the impact of the satellite altitude on the imaging process. Viewed from a spin geosynchronous satellite, theearth's disk location passes through a diurnal cycle. This phenomenon is schematically illustrated in Fig. 2, which presents the FY2 observation geometry. Inan ideal situation, the spin axis of the satellite is parallel to the earth's axis of rotation. In practice, thissituation is never achieved. As shown in Fig. 2, at0600(1800)UTC, the satellite looks up(down)at theearth, and the earth's disk is in the downward(up-ward)side of the image. Around these two moments, the earth's disk center is displaced mostly in the north-south direction, and the images turn only slightly ina day. At 0000(1200)UTC, the earth's disk is minimally displaced, but the scan lines deflect the greatestamount, and the turning of the images is greatest ina single day. At 0000(1200)UTC, the satellite viewsthe earth with its spin axis turning clockwise(coun-terclockwise), while the image turns counterclockwise(clockwise).

Fig. 2. FY2 observation geometry. N is the Arctic, E isthe earth's center, and S is the Antarctic. The line linkingthe Arctic and Antarctic is the earth's rotation axis. Thespin axis of the satellite is also shown. [From Lu et al., 2008]

Figure options

Observation vectors can be measured on the observation images, as shown in Fig. 3, where E is theearth center and S is the satellite. On the line from Sto E, make a normal plane "image" through F with origin O. F is at the central column. SO is perpendicularto the plane "image". The trajectory of the scan linesacross the plane "image" forms the observation image. The 1250th scan line passes through C. C is the cross-point of the 1250th line and the central column. Assume vector S_E is from S pointing to E and vector S_Yis the spin vector of the satellite. On the plane consisting of vectors S_E and S_Y, there is another vector, S_Z, perpendicular to the plane "image". Vector S_Z passesthrough S and O. The angle subtended at satellite Sby vectors S_E and S_Z can be measured by θ+ρ+π/2. Then, the observation equation is established as follows:

Click Browse formula

Fig. 3. The satellite coordinate SYstem. E is the earth center and S is the satellite, SY matches the spin axis of thesatellite, and the SZX plane is the spin plane of the satellite. The earth center E and the satellite spin axis make up theSYE plane. The SYE plane crosses the satellite spin plane(SZX)to form the Z axis. Normally, the axis does not extendtoward the earth center but is in the plane defined by the Earth center and the satellite spin axis SY. The SX axis isdefined by SY ×SZ = SX. Take a point O along the line extended from the vector SZ. Make a plane IMAGE O, whichis perpendicular to SZ. The earth is projected on the IMAGE plane to form the observation image. On the projectedplane IMAGE, the earth center E is at F. The 1250th scan line crosses the central column of the plane IMAGE at C, which is the center of the observation image. [After Hambrick and Phillips, 1980]

Figure options

Equation(1)with 0 pitch misalignment ρ was originally expressed by Hambrick and Phillips(1980; abbreviated as HP80 hereafter). In Eq. (1), S_E, θ, and ρ can be measured on the observation image. Thespin vector(altitude)of the spin satellite, S_Y, is thensolved from Eq. (1).

In addition to the altitude of the satellite, misalignment of the visible IR spin-scan radiometer(VISSR)from the satellite should also be considered. Figure 4 is a schematic diagram of this misalignment. For an ideal spin meteorological satellite, the 1250thline should scan out a plane. In reality, due to themisalignment shown in Fig. 4b, the 1250th line scansout a cone. The angle between the cone and the spinplane is the ρ component of the misalignment. It canbe verified that the ρ component of the misalignmentis related to the deviation of the image in the north-south direction and is also the solution of Eq. (1). With Eq. (1), the earth disk location in the south-north direction is fixed.

Fig. 4. Schematic diagrams demonstrating the misalignment of the FY2 visible IR spin-scanning radiometer(VISSR). (a)Difference between the actual VISSR and the ideal; (b)impact of roll misalignment on the imaging process; (c)misalignment of the roll component; (d)misalignment of the yaw component; and (e)misalignment of the pitch component. [From Lu et al., 2008]

Figure options

Spin geosynchronous meteorological satellites usea sun sensor installed on the side face of the satellite tocontrol the pointing of the VISSR to the earth and toalign the scan lines. Based on an accurate sun position and the angle subtended at the satellite by the sun and earth(β), individual scan lines are aligned, resampled, and registered at the ground station, and observationimages are assembled. Figure 5 is a schematic diagramof the β angle geometric formulation. In Fig. 5a, S_Yis the spin axis of the satellite, S_earth is the vectorpointing from the satellite to the earth's center, and S_sun is the vector pointing from the satellite to thesun. Here, we define the following three planes, i. e., the S plane consisting of the spin axis and the sun, theE plane defined by the spin axis and the earth, and theP plane passing through the satellite and perpendic-ular to the satellite spin axis. Note that S_Y × S_sun and S_Y ×S_earth are perpendicular to the vectors S_sun and Searth projected at the satellite spin plane, respectively. The β angle can be written as follows:

Click Browse formula

Fig. 5. A schematic diagram illustrating the β angle. (a)Three-dimensional view, (b)plane view after local midnight, and (c)plane view after local noon. [From Lu et al., 2008]

Figure options

Here, β is a significant parameter for observation byspin-stabilized geosynchronous meteorological satellites. In Eq. (2), vectors S_sun and S_earth are related tosatellite position and S_Y to altitude. When the satellite position and altitude are well predicted, angle βcan be accurately calculated. The value thresholdsof β are from 2π to 0. At a time near local midnight, when the sun, earth, and spin axis of the satellite sharea common plane, β is given a value of 2π. The value ofβ decreases monotonically until a time near the nextlocal midnight when it approaches 0, as shown in Figs. 5band 5c. With the addition of Eq. (2), the navigation model becomes complete, and a fully automaticsolution of the equations is realized. 2. 3 Completely closed coordinate and parameter systems

To express the basic principle for FY2 image navigation, the above equations have been simplified. Inpractice, within a 25-min time period during whichFY2 takes a full disk image of the earth, the observation object(earth), the observation tool(satellite), and the sun that is used to align the satellite scan linesto point them at earth, are all in motion. Accurate image navigation requires the knowledge of the positionof these three objects, the altitude of the satellite, and the misalignment of the VISSR relative to the satellite. Completely closed coordinate and parameter systemsare essential.

The parameters are defined and applied in the coordinate systems with a clear geometric meaning. Inthe defined coordinate systems, the parameters maynot be conservatory with time. Thus, the parametersmust be defined according to the conservatory coordinate systems. The navigation equations should besolved by using the conservatory coordinate systems. To make the transformation from the defined to theconservatory coordinate systems, a series of intermediate coordinate systems are necessary. Correct parameter definition and transformation among different coordinate systems are essential for solving imagenavigation equations. 2. 4 Image navigation results

Figure 6 shows observation images overlaid withlatitude-longitude grids, coastlines, and other geographical features from FY2C visible images at 0456UTC 8 June 2006. The resolution for the visible channel is four times better than that of the IR channel. The coast lines overlaid on the images are predictedones. Figure 6a is a full disk image, while the othersare local section images with raw visible resolution. To show detail, the image pixel size in the local section images is zoomed to 10 times larger than the geographical feature size. These images clearly show agood overlap between images and grids.

Fig. 6. Full-resolution FY2C visible images with superimposed coastal lines showing the FY2 image navigation results. (a)Full disk image of FY2C at 0456 UTC 8 June 2006. (b)-(e)Image sections of FY2C at 0456 UTC 8 June 2006. [Adapted from Lu et al., 2008]

Figure options

Traditionally, there are two main calibrationmethods operationally utilized for FY2, i. e., vicarious in-situ calibration using simultaneous observations from space-borne and on-ground instruments and inter-calibration of observations between different on-orbit sensors. Vicarious in-situ calibration hashigh accuracy for validation but a relatively small dynamic range of the target's radiation. Inter-calibrationis a stable method but is highly dependent on the spatial matching procedure. In addition, the target earthis inevitably influenced by the atmospheric radiationcorrection in both the in-situ and the inter-calibrationmethods. Therefore, there is a growing dem and for anew calibration method that has stable performancein its radiation characteristics to overcome the limitations of the existing methods.

The photometric stability of the lunar surface canreach 10^-9 per year, and is widely used as the referencesource for space-borne sensor calibration in free space. Outside China, lunar calibration in the visible band has been conducted, but IR band lunar cali- bration israrely reported. The main difficulty lies in modelingthe nonthermal, nonuniform, and nongray body characteristics of the lunar surface for IR lunar calibration(Guo et al., 2012). Moreover, results from astronomiccalculations indicate that lunar calibration in the IRband cannot completely meet the high frequency requirements of on-orbit operational calibration. In fact, it is merely useful for the systematic error adjustmentof a space-borne sensor's radiometric response. A typical on-orbit lunar image observed by FY2F is shownin Fig. 7a, whereas Figs. 7band 7c show images aftercompensation for the relative movement between themoon and the satellite. These images are ultimatelyused for on-orbit lunar calibration in the IR band.

Fig. 7. Typical on-orbit lunar images observed by FY2F. (a)Raw lunar image in the VIS band, (b)lunar image in theIR1 band after compensating for its movement relative to the satellite with spatial oversampling, and (c)lunar image inthe IR1 band after compensating for its movement relative to the satellite with spatial ordinary sampling.

Figure options

In ordinary blackbody calibration, when describing the incident radiance to the sensors, it is criticalthat the blackbody's temperature and its distributionfield be measured as accurately as possible. However, the lunar surface temperature cannot be measured inreal time. The temperatures of targets illuminated bythe sun on the lunar surface are usually higher than300 K, and using the existing lunar surface temperature estimation model, with its accuracy of about 1. 5-2. 0 K, is unacceptable for real radiometric calibrationin the IR bands. Therefore, one of the kernel-leveltechnologies in FY2's operational calibration is the introduction of the equivalent dual-band emissivity ratio, which is independent of the temperature of thelunar surface's uniform targets, based on the modeling of the observed radiance from the moon. This ratio establishes a quantitative relationship between theobserved radiances from different thermal IR bands. The lunar calibration equation between two IR bandsis further deduced as follows:

Click Browse formula

where(a_IR1)_moon and (a_IR2)_moon are the two constantsrelated to the spectral response as well as the dynamicrange, (b_{IR1& IR2})_moon is a variable determined by theon-orbit synchronous lunar observations in both theIR1 and IR2 bands, c_IR1 and c_IR2 are the two calibration slope parameters to be solved in the IR1 and IR2 bands, respectively, and f_moon(·)is a known realfunction varying with the calibration slope. By using the results from in-lab calibration as well as introducing the relationships of the calibration parametersbetween different bands, the calibration slopes can beaccurately solved with Eq. (3). 3. 2 Absolute radiometric calibration with IBBcalibration

Some theoretical analyses indicate that the calibration slope, which is the most important parameterto be determined in radiometric calibration, is mainlydominated by the normalized detectivity(D*). Todate, with the continuous development of the HgCdTedetector technique, photoconductive and photovoltaicdetectors are used in most meteorological satellitesworldwide. D* values between 3. 5 and 12. 5 µm canreach the order of 10¹⁰ to 10¹¹ Hz^1/2W^-1, finally realizing perfect background limitation detection performance. With this means, the variation in the instru-ment's background radiation will inevitably influenceD¤ and directly cause variation in the calibration slopeparameter.

By using the FY2E satellite as an example, Fig. 8shows the temperature variation in the VISSR's front-optics, represented by the primary mirror(red curve), the secondary mirror(blue curve), and the temperature variation in the VISSR's after-optics representedby the calibration mirror(green curve), where themaximal temperature difference approaches 20 K. Itcan be easily understood that the variation in background radiation due to the changing of environmental temperature should appear as an annual feature. Thus, the suitable radiometric reference and its calibration method should have the capability to calibratediurnal variations in the sensor's radiometric response. This is the key issue for achieving high accuracy calibration in IR bands for the FY2 satellite.

Fig. 8. Temperature variations of the main optical elements of the FY2E satellite during 2012.

Figure options

Figure 9 illustrates the theoretical contrast between the typical full-path blackbody(BB)calibration and the partial-path BB calibration(or the socalled inner-blackbody calibration; IBBC). For bothfull-path and partial-path BB calibrations, there areordinarily two steps, i. e., blackbody and cold-spaceobservations, the difference between which is the validBB observation. In particular, for full-path BB calibration, the switch between the two steps is realizedby the rotation of the scanning mirror by about 90°, where both the front-optic and after-optic radiant influences are included, and the difference as a net BBobservation can be directly used to determine the calibration parameter. For IBBC, however, although theIBB can be observed by rotating the calibration mirror, the radiometric contribution of the calibrationmirror is added to the IBB observation without thecontribution of the front-optics. Therefore, the traditional full-path BB calibration method cannot besimply employed to deal with IBB observation. Thekey issue is how to accurately estimate the radiometric contributions of the front-optics as well as thecalibration-optic component(FCC). For the operational calibration of FY2 satellites, we established asingle-temperature IBBC method with a radiometriccontribution estimation model for the FCC, wheretelemetric information from multiple optical components specific infinite element modeling and in-lab calibration results are considered. The IBBC equation isgiven by Eq. (4).

Click Browse formula

Fig. 9. Contrast between typical full-path and partial-path blackbody calibrations for optical remote sensing instruments.

Figure options

where j represents a real FCC, i. e., primary mirror, secondary mirror, folding mirror, calibration mirror, and delay lens for FY2; g_i(T_j)is the modeled finite-element expression for the jth FCC; k_j is a coefficientdetermined by the material and shape of the corresponding FCC; kequal is the equivalent IBB incidentradiance, which can be estimated with in-lab results; and vIBB is the IBB observation expressed as voltage. 3. 3 Main achievements of radiometric calibration of the FY2 geostationary meteorological satellite

Figure 10 shows a real lunar calibration result forboth the IR1 and IR2 bands of FY2E by using thelunar observations at 1000 UTC 1 January 2012. Themetrics of the horizontal and vertical ordinates in Figs. 10band 10c are voltage and radiance, respectively, and R2 approaches 0. 99 with perfect performance. The calibration slopes, varying with the environmentaltemperatures before and after the satellite eclipses forboth FY2E and FY2F, are shown in Figs. 11 and 12, respectively. Specifically, the horizontal coordinates ofFigs. 11a and 11b are time. The vertical coordinateof Fig. 11a is temperature, and the vertical coordinate of Fig. 11b is slope(unit: W m^-2 sr^-1 µm^-1V^-1). In Fig. 11a, the green, blue, black, and redlines represent the temperatures of the delay lens, theprimary mirror, the secondary mirror, and the calibration mirror, respectively. In Fig. 11b, the red, blue, black, and green lines represent the calibration slopesof IR1-IR4. Noticeably, during the satellite's eclipse, the calibration slopes of all the IR bands(IR1-IR4)are characterized by great diurnal variability. In theIR1 band, for example, the relative diurnal variationof the calibration slope can reach 3%. This equals a 3-K error when observing a target with a temperature of300 K without any modification. Moreover, after thespring satellite eclipse, the calibration slope of all theVISSR's IR bands rapidly increases day by day and reaches a peak around the summer solstice. This pattern is closely related to the increase in the satellite'stemperature and the decrease in its cooling capacity. It is clear that the FY2 satellite's operational calibration results better describe the time-varying featuresof the calibration parameters for the on-orbit VISSR, which is the baseline of the proposed more highly accurate calibration of the inner-blackbody corrected bylunar emission(CIBLE)method(Guo et al., 2013). In Fig. 13, observations from the Infrared Atmospheric Sounding Interferometer(IASI)sensor underthe framework of the global space inter-calibration system(GSICS)is used as a reference, and the calibration bias is evaluated at the high temperature segment(290 K)for the FY2E IR1 band after April 2013(Note:the CIBLE method has been used for FY2E since 27March 2013). As shown in Fig. 13a, the total biasshows pretty good to be less than 1 K. Meanwhile, asshown in Fig. 13b, when the two high spectral sensors, i. e., the IASI and the Atmospheric Infrared Sounder(AIRS), are selected for reference, the relative differ-ence in the results remains about zero, indicating thatthe assessment results of Fig. 13a are reliable.

Fig. 10. Lunar calibration results of the FY2E satellite at 1000 UTC 15 January 2012. (a)Earth-disk image withmoon in the IR1 band, (b)lunar calibration curve in radiance for the IR1 band, (c)earth-disk image with moon in theIR2 band, and (d)lunar calibration curve in radiance for the IR2 band.

Figure options

Fig. 11. Calibration slope's diurnal variations in environmental temperature in the IR1-IR4 bands for FY2F on 16March 2012. (a)Temperature variations of the main optical elements and (b)calibration slope's diurnal variations inthe IR1-IR4 bands.

Figure options

Fig. 12. Calibration slope's annual variations in environmental temperature in the IR1-IR4 bands of FY2F between April and July of 2012. (a)Temperature variationsof the main optical elements and (b)blackbody counts and calibration slope's annual variations in the IR1-IR4 bands.

Figure options

Fig. 13. Calibration accuracy evaluation results from theGSICS method for the FY2E IR1 band. (a)Calibrationbias for the condition of the target's brightness temperature(BT)equal to 290 K and (b)BT double biases analysisfor FY2E with IASI and AIRS sensors.

Figure options

Fig. 14. The error terms, considered in the sensitivity study, that affect the departures(observed minus simulatedbrightness temperatures).

Figure options

The SIPOn-Opt is designed to obtain more accurate performance parameters of the instrument inorbit. The variational inversion method is adopted inthe optimization. The cost function J is the core of thealgorithm, Δv₀ is the channel central frequency, bd isthe band width, stopband is the passbandstop band, ΔT_max is the nonlinearity, k is the antenna main beamefficiency, srf is the spectral response function, c is theabnormal cold space, w is the abnormal warm load, and a is the absorption line measurement error. Interms of the relevant parameters of universal instruments in orbit, Eq. (5)represents the cost function.

Click Browse formula

where m(Δv₀; bd; stopband; ΔT_max; k; srf; c; w; a)isthe mean of the observation minus the simulation and s(Δv₀; bd; stopband; ΔT_max; k; srf; c; w; a)is the st and ard deviation of the observation minus the simulation. These are the functions of the performance parameters of the instrument in orbit. The dynamicranges of the mean and st and ard deviations σ_m and σ_sreflect the accuracy of the numerical prediction model. Errors by various instruments on-orbit propagate according to their own physical mechanisms(Lu et al., 2011b).

In terms of the FY3A/microwave temperaturesounder(MWTS), after a sensitivity evaluation, wefound that errors in the measured channel frequency and the nonlinearity of the instrument are the mainsources of the observation system errors. Therefore, the cost function(Eq. (5))can be simplified as

Click Browse formula

Fig. 15. Flow chart of the SIPOn-Opt, where v is the passband central frequency, bd is the passband width, stopband is the passbandstop band, ΔTmax is the nonlinearity, k is the antenna e±ciency, srf is the spectral response function, cis the abnormal cold space, w is the abnormal warm load, and a is the absorption line measurement error.

Figure options

Most payloads in the first of China's second-generation polar-orbiting meteorological satelliteswere being launched for the first time. FY3A wasthe first in China's history to have quantitative assimilation ability. Four of FY3A's 11 instruments areof particular importance to numerical weather prediction, especially the MWTS. We take FY3A/MWTSas an example to demonstrate how the SIPOn-Optdetects and analyzes the dominant errors of the instrument, retrieves the satellite's performance parametersin orbit, corrects the observing system errors, and improves data quality. FY3A/MWTS is also assimilatedby the world-leading ECMWF numerical forecast system to improve the model's forecast accuracy. After calculating and analyzing the sensitivity of the errors shown in Fig. 14 to the root-mean-square error between the observed and simulated values, wefound the measurement errors of central frequency and nonlinearity to be the dominant sources of errors in FY3A/MWTS. More accurate parameters ofcentral frequency and nonlinearity were then retrievedthrough SIPOn-Opt. Figure 16 shows global mapsof the fit between the observed and simulated BTsof FY3A MWTS in comparison with that of the European Meteorological Operational satellite program-A/Advanced Microwave Sounding Unit-A(MetOp-A/AMSU-A). The BTs of FY3A MWTS are obtainedthrough using more accurate channel frequencies and nonlinearity corrections including the designed centralfrequency, the measured frequency in the lab beforelaunch, and nonlinearity parameters through variational optimization. When taking the satellite observation as a reference for comparison with the simulation results, if the observation physically fits the simulation well, the departure of the observation minussimulation follows a Gaussian distribution with a zeromean and a certain root-mean-square error.

Fig. 16. Maps of first-guess BT departures(K)for MWTS channels 2-4(from left to right)by using(a)design-specifiedpassbands, (b)prelaunch-measured passbands, (c)optimized passbands, (d)passbands following nonlinearity correction, and (e)the corresponding equivalent MetOp-A first-guess departure maps. The spots at the base of the histogramsindicate the mean first-guess departure. [From Lu et al., 2011b]

Figure options

Taking FY3A/MWTS Channel 4 as an example(column 3 in Fig. 16), FY3A/MWTS Channel 4 and MetOp-A/AMSU-A Channel 9 have the same instrument design configuration. There is a 0. 5-h differencebetween the two satellites' overpass of the same location, and the observing objects are concentrated inan atmosphere of 100 hPa. In this case, the departures between the observations and simulations of thetwo should be consistent, but there is a large difference in the designed central frequency, as shown inFig. 16. We can see that this departure has beenimproved after calculation with the lab measurementparameters before the satellite is launched, but it isstill large. After analyzing the sensitivity of the errors, we found that the data quality and the departurewere affected by the measurement error of the channel central frequency and the nonlinear effect. Themeasurement frequency bias is because of the use ofGunn diode frequency lock technology. We theorizethat the channel frequency locks a vibration within aresonator. This locked channel frequency is affectedby the resonator's environment. The atmospheric environment in the lab is different from the vacuum environment in outer space, and this leads to their differentmeasurement channel frequencies. By comparing thedeparture histograms, we see that the departure is improved after the frequency bias is corrected, but thereis still a positive bias of about 1 K. Satellite remotesensing uses radiometers to measure an object, assuming a linear relationship between the electric signal and the object's energy. It then adopts a two-point linearcalibration. In fact, there is a certain nonlinear effectin radiometers. The simulated departure can be comparable to that of the AMSU-A when the nonlineareffect is also corrected.

In summary, Figs. 16a-d demonstrate that departures of MWTS channels 2-4 are subject to a Gaussi and istribution of zero mean and a smaller root-mean-square error after the frequency bias and nonlineareffect are corrected. The result is more comparable to(even better than)the results of the AMSU-A(Fig. 16e). When corrected FY3A/MWTS data are appliedto the ECMWF assimilation/prediction system, prediction skill is improved by 1% in the Southern Hemisphere(Lu et al., 2012).

The correction algorithm of the frequency measurement error and the nonlinear radiative effect wereimplemented in the operational preprocessing systemof the FY3A/MWTS at the National Satellite Meteorological Center(NSMC)of the China MeteorologicalAdministration(CMA)in March 2011. Thereafter, the error of the simulated root-mean-square departureis about 0. 2 K, which is comparable to that of theMetOp-A/AMSUA. The corrected data are releasedto the public domain by the NSMC of the CMA.

The module for correcting frequency measurement errors and nonlinear radiative effects, similar tothat of the FY3A MWTS, has also been implementedin the FY3B's MWTS operational preprocessing system. Therefore, in terms of the FY3B MWTS, thesetwo errors are not dominant. In fact, scanning anglebiases related to the scanning position are the dominant biases, which can be corrected through a biascorrection algorithm in the assimilation system.

FY3C was launched successfully in September2013. Currently, it is in the on-orbit testing phase. SIPOn-Opt is being used to detect and correct thedominant observing system biases to further improvedata quality. 5. Summary

The innovations to the algorithms discussed aboveare summarized as follows:

(1)The traditional image navigation techniquedepends on l and mark matching, while the FY2 image navigation technique uses vectors pointing at theearth disk center. The latter technique not only drawsthe outline of the navigation equation more clearly butalso more easily achieves an accurate input parameter. In the traditional image navigation technique, the fi angle is expressed in an empirical formula, whilethe FY2 image navigation technique uses an analyticexpression. Advanced coordinate systems ensure thequality of solutions to the image navigation equations.

(2)Before 2012, FY2 geostationary meteorological satellites mainly used inter-calibration with othersatellites for operational calibration processing. However, the calibration timeline(7-15 days) and accuracy(2-3 K)needed to be improved due to the temporal, spectral, and spatial differences between differentsensors. By using CIBLE, two main difficulties havebeen conquered. First, by correcting the radiometric contributions of the FCCs, hourly high frequencycalibration in the IR bands has been realized withthe IBB observations, which had previously only beenused for on-orbit monitoring of radiometric stability. Second, the on-orbit observed IR radiance from thelunar surface is used to correct the systematic errorof the IBBC, which results in the realization of highlyfrequent(hourly), high accuracy(less than 1 K)IRradiometric calibration.

(3)It has been assumed that the instruments ina meteorological satellite on-orbit are consistent withtheir design specifications, that the instrumental performance parameters on-orbit are consistent with thevalues measured before the satellite is launched, and that even if certain differences exist, there is no effective way to calculate such values. However, practicalapplications show that these parameters do vary and that these variations have an impact on data quality and on the effectiveness of their application to numerical weather prediction. For the first time, an SIPOn-Opt has been developed and used to variationally retrieve the performance parameters of instruments on-orbit. More exact instrumental parameters can beretrieved through such observation constraints as thechannel central frequency, nonlinear parameters, theantenna main-beam efficiency, and the spectral response function. These parameters may be used tocorrect satellite observation system biases, making thedata of FY3 consistent with its counterparts from theEuropean and the U. S. space agencies. When thedata of FY3 are assimilated with the ECMWF model, the prediction accuracy of the model is also improved. When this model is used in meteorological satelliteinstruments(MSU and AMSU-A)from Europe and the U. S., the quality and effectiveness of the data fromthese instruments, applied over the past 40 years, arealso improved.