Thermal and near-infrared sensor for carbon observation Fourier transform spectrometer-2 (TANSO-FTS-2) on the Greenhouse gases Observing SATellite-2 (GOSAT-2) during its first year in orbit

The Japanese Greenhouse gases Observing SATellite-2 (GOSAT-2), in orbit since 29 October 2018, follows up the GOSAT mission, itself in orbit since 23 January 2009. GOSAT-2 monitors carbon dioxide and methane in order to increase our understanding of the global carbon cycle. It simultaneously measures carbon monoxide emitted from fossil fuel combustion and biomass burning and permits identification of the amount of combustion-related carbon. To do this, the satellite utilizes the Thermal and Near Infrared Sensor for Carbon Observation Fourier-Transform Spectrometer-2 (TANSO-FTS-2). This spectrometer detects gas absorption spectra of solar radiation reflected from the Earth’s surface in the shortwave-infrared (SWIR) region as well as the emitted thermal infrared radiation (TIR) from the ground and the atmosphere. TANSO-FTS-2 can measure the oxygen A band (0.76 μm), weak and strong CO2 bands (1.6 and 2.0 μm), weak and strong CH4 bands (1.6 and 2.3 μm), a weak CO band (2.3 μm), a mid-wave TIR band (5.5–8.4 μm), and a long-wave TIR band (8.4–14.3 μm) with 0.2 cm−1 spectral sampling intervals. TANSO-FTS-2 is equipped with a solar diffuser target, a monochromatic light source, and a blackbody for spectral radiance calibration. These calibration sources permit characterization of time-dependent instrument changes in orbit. The onboardrecalibrated instrumental parameters are considered in operational level-1 processing and released as TANSO-FTS-2 level-1 version 102102 products, which were officially released on 25 May 2020. This paper provides an overview of the TANSO-FTS-2 instrument, the level-1 processing, and the first-year in-orbit performance. To validate the spectral radiance calibration during the first year of operation, the spectral radiance of the version 102102 product is compared at temporally coincident and spatially collocated points from February 2019 to March 2020 with TANSO-FTS on GOSAT for SWIR and with AIRS on Aqua and IASI on METOP-B for TIR. The spectral radiances measured by TANSO-FTS and TANSO-FTS-2 agree within 2 % of the averaged bias and 0.5 % standard deviation for SWIR bands. The agreement of brightness temperature between TANSO-FTS-2 and AIRS–IASI is better than 1 K in the range from 220 to 320 K. GOSAT-2 not only provides seamless global CO2 and CH4 observation but also observes local emissions and uptake with an additional CO channel, fully customized sampling patterns, higher signal-to-noise ratios, and wider pointing angles than GOSAT.

calibrators. Also, GOSAT-2 spectral radiance products are compared with other satellite data such as TANSO-FTS on 65 GOSAT, AIRS on AQUA (Aumann et al., 2003) and IASI on METOP-B (Clebaux et al., 2009). Finally, we will show that the spectral radiance for GOSAT-2 is consistent with these satellites' inter-calibration data.
This paper first introduces an overview of the GOSAT-2 satellite and the TANSO-FTS-2 and CAI-2 instruments. The processing method for transforming raw on-orbit data to calibrated spectral radiances (level-1 processing) for TANSO-FTS-2 follow. Next is an assessment of the first year of in-orbit performance of TANSO-FTS-2 by its comparison to temporally 70 and spatially coincident data from other satellites. In addition, calibration challenges are identified with current best estimated values.

GOSAT-overview
2.1 GOSAT-2 satellite system 75 GOSAT-2 monitors CO2, CH4, and CO globally from space. GOSAT-2 is placed in a 613 km sun-synchronous orbit at 13:00 local time, with an inclination angle of 98⁰ and a revisit time of six days. Figure 1 shows a prelaunch image of GOSAT-2.
The orbital parameters for GOSAT-2 are listed in Table 1 as well as those for GOSAT.

Operation
Normal scanning, including fixed grid and target observations, and pointing at sun-glint over the ocean, is carried out according to a set pattern and is determined prior to the observation cycle. TANSO-FTS-2 accommodates a fully programmable pointing system to extend observation control capabilities. This system allows operators to point to any location on the earth's surface to observe, and to manage the observation locations day by day. Target (MMO). Due to safety concerns with the pointing mirror scanner motor, the difference of the optical angles between two 95 consecutive observations must be less than 25.5⁰. If the motor receives an angle command greater than 25.5⁰, TANSO-FTS-2 immediately transits from normal operation to safe mode, suspending observations. Nominally, the observation plan is uploaded to the satellite once a day. The observed data is recorded by the onboard data recorder and is transferred periodically from the satellite to ground typically once every two orbits. 100 3 The TANSO-FTS-2 instrument

Instrument overview
The greenhouse gases sensor takes spectroscopic measurements of shortwave infrared sunlight reflected from the earth's surface to the satellite and TIR radiation emitted from the ground and the earth's atmosphere. The gaseous column amounts (column abundances) of CO2 and CH4 in the 1.6 um band, CO2 in the 2.0 um band, and CH4 and CO in the 2.3 um band are 105 estimated, and the gaseous concentration profiles (vertical distribution) are assessed using the TIR region. A Fourier transform spectrometer is used because it allows simultaneous observations with a high spectral resolution (0.2 cm -1 ) over a wide bandwidth. An interferogram measurement created in 4.024 seconds is taken as a baseline. TANSO-FTS-2 uses almost the same type of Fourier-Transform Spectrometer (FTS) mechanism as the TANSO-FTS, and its main characteristics are summarized in Table 2. The optical layout is shown in Fig. 2. 110 A two-axis pointing control mechanism allows the sensor to aim at planned locations for a complete interferogram measurement interval after which it moves to the next observation location. It can also observe sun glint over oceans. The two-axis pointing system for TANSO-FTS-2 has wider coverage than that for TANSO-FTS and provides fully programmable aiming to up to 1246 observation points per orbit. It allows between -40⁰ and +40⁰ in the along-track direction and between -35⁰ and +35⁰ in the cross-track direction. In addition, to keep the lubricant in each bearing uniform, wide travel 115 motion in both along-track and cross-track directions are scheduled once per day.
The input optics, interferometer, band separation optics, and the detector optics are housed in a single, temperaturecontrolled optical box. The scene flux is reflected into the input optics by a bare gold-coated mirror on the two-axis pointing system, and it is divided into two parts by a pickoff mirror. One beam is directed to the CMOS video camera (608 x 1024 pixels) for identifying the scene image, the other to the interferometer (FTS). The camera image is also used to identify cloud 120 positions in the field of view. To reduce the number of cloud-contaminated observations, the TANSO-FTS-2 uses the cloud images to actively avoid clouds by adjusting the line of sight during the FTS turnaround motion just before the measurement scan.
TANSO-FTS-2 employs a double pendulum and double cube corner type of FTS mechanism with an uncoated ZnSe beam splitter (as for TANSO-FTS). The effective aperture size is 73 mm, which is larger than that of TANSO-FTS, and the 125 maximum optical path difference is +/-2.5 cm. A long-life diode laser acts as a metrology light source, with a stable single-https://doi.org/10.5194/amt-2020-360 Preprint. Discussion started: 28 September 2020 c Author(s) 2020. CC BY 4.0 License. mode 1.31 um emission, in order to last through the five-year mission. The interferometer temperature is controlled to 23 +/-3 deg C. The temperature of the blackbody ranges between 293 K and 295 K during the first year's operation.
The scene flux signal is sampled by the FTS at fixed time intervals and measures both the science signal and laser fringe signal simultaneously; this is called uniform time sampling. Many conventional FTS mechanisms, including TANSO-130 FTS, use a laser fringe triggered measurement technique called uniform optical-path-difference sampling. In level-1 processing, the uniform time-based interferogram is converted to a uniform optical-path-difference sampled interferogram, followed by the inverse FFT (Fast Fourier Transform). The detailed processing method is described in the following section.
The modulated scene flux is divided into the various SWIR and TIR bands by dichroic beam splitters. The separated beams are focused on three kinds of detectors, a silicon(Si) detector for the O2 band 1, PhotoVoltaic -Mercury Cadmium 135 Telluride (PV-MCT) detectors for SWIR bands 2 and 3, and for the shortwave TIR band 4, and a Photo Conductive -Mercury Cadmium Telluride (PC-MCT) detector for the longwave TIR band 5. A multistage passive space cooler provides detector cooling. The detector temperatures are set at 215 K for band 1, 130 K for bands 2 and 3 and 100 K for bands 4 and 5, respectively. The field of view for all the bands is limited to 15.8 mrad. The instantaneous ground field of view (GIFOV) becomes a 9.6 km in diameter circle at the sub-satellite location from an altitude of 613 km. 140 Simultaneous observations of two linear polarizations for the 0.76 um, 1.6 um, and 2.0-2.3 um bands are facilitated by two identical detectors coupled with a polarizing beam splitter. TANSO-FTS-2 has a somewhat extended spectral coverage compared to TANSO-FTS. The TANSO-FTS-2 spectral regions, which are defined by bandpass filters, are as follows: band 1 (12950-13250 cm -1 ), band 2 (5900-6400 cm -1 ), band 3 (4200-5200 cm -1 ), band 4 (1188-1800 cm -1 ), and band 5 (700-1188 cm -1 ). The signal voltages from the detectors are converted into numerical representations by 14-bit analog-to-digital 145 convertors (ADCs). Just before an ADC, the DC component of the offset science signal is removed, but its value is included in the data so that the ADC handles only the full dynamic range of the AC component. Also, high frequency pulses are counted for each laser fringe interval as reference times for the optical path difference.

Calibration operation 150
Solar irradiance calibration (Sol. Cal.), deep space calibration (DS Cal.), and blackbody calibration (BB Cal.) are all collected over each orbit using a set timing pattern. Nadir observations are not made during the DS Cal. or BB Cal. period. If a DS Cal. and BB Cal. are to be managed at a fixed latitude, observations of certain latitudes would lead to a loss of nadir observations at certain locations. For this reason, the DS Cal. and BB Cal. are dispersed over the orbit to facilitate a uniform data acquisition coverage that is the same as for TANSO-FTS. TANSO-FTS is also operated six calibrations per orbit -155 twice during the day side and four times at the night side.
The Sol. Cal. uses a solar diffuser activated with a shutter: 17 seconds before the start of calibration measurements, the shutter opens and the primary solar diffuser is exposed to sunlight. In addition, a secondary solar diffuser is used for https://doi.org/10.5194/amt-2020-360 Preprint. Discussion started: 28 September 2020 c Author(s) 2020. CC BY 4.0 License. reference operations, which permits monitoring of the degradation of the reflectance of the primary diffuser. Reference diffuser operations are scheduled once every three months. 160 A lunar calibration that complements the solar diffuser calibration is also done, once a month. The Instrumental Line Shape function (ILS) calibration is performed using two types of diode laser; 0.77 um and 1.54 um diffused via an integrating sphere to illuminate the full field of view of the interferometer. The schematic diagram of a typical calibration operation for one orbit is shown in Fig. 3.

Instrument models
The initial characterization for TANSO-FTS-2 was done during prelaunch testing and the calibration phase. Generally, retrievals of the column abundances of CO2, CH4, and CO require instrument models that are compatible with the retrieval software. Auxiliary information such as radiance conversion factors, signal-to-noise-ratio (SNR) models, and ILS models is available via the GOSAT-2 product archive site (https://prdct.gosat-2.nies.go.jp/en/document.html). 170

Radiance conversion model
For prelaunch radiance calibration, the output signal level of TANSO-FTS-2 was compared to the radiance levels of a 65inch integrating sphere whose inner surface was coated with barium sulfate. The radiance levels emitted by the integrating sphere were characterized based on an NIST standard lamp. The calibration was conducted in a temperature-and humidity-175 controlled room, but the instrument was not installed in a vacuum chamber during the radiance calibration period.
TANSO-FTS-2 is equipped with a passive space cooler and with heaters to control the temperatures of all detectors as well as the optical components. Under laboratory conditions, the passive cooler could not provide power enough sufficient to cool the detectors to their set point temperatures. To assist in cooling, an additional external cooler system was coupled with the TANSO-FTS-2 passive cooler during prelaunch calibration. 180 The instrument was illuminated by the integrating sphere with both along-track and cross-track pointing angles at zero deg. Due to the presence of oxygen and water vapor lines in bands 1 and 3, the envelope of the spectrum is assumed to determine the radiance conversion coefficients. The conversion coefficients from raw spectra to radiance are available on the GOSAT-2 data distribution site (https://prdct.gosat-2.nies.go.jp/en/document.html). In level-1 products, the radiance spectra are processed by applying these conversion coefficients. However, the two-axis orientation dependency of the scanner mirror 185 reflection is not taken into account in the current level-1 products. In orbit, the radiometric response for the SWIR bands has been changed and were recalibrated based on an on-orbit calibration dataset. The details are discussed in section 5.2. In addition, the conversion coefficients are revised by the first-year coefficients. https://doi.org/10.5194/amt-2020-360 Preprint. Discussion started: 28 September 2020 c Author(s) 2020. CC BY 4.0 License.

Polarization sensitivity model
The elements of the 2 x 4 matrix are identified as the polarization model, the Mueller matrix, for TANSO-FTS-2. To characterize the polarization model for TANSO-FTS-2, the instrumental response for each polarization band was characterized by using linearly polarized light during the prelaunch test and calibration periods.
A linear polarizer between the instrument aperture and the integrating sphere was rotated in 10 deg steps while the 200 TANSO-FTS-2 acquired interferograms. These interferograms were analyzed and processed with a polarization model.
Generally, the V component in equation (1) is negligible for atmospheric composition measurement. Then, a 2 x 3 matrix dependent on wavenumber is processed against the , , and components. Figure 4 presents the estimated polarization characteristics based on the prelaunch measurements. To implement the matrix, the retrieval teams have to validate these sensitivities by evaluating their forward calculations (Butz et al., 2011, 205 Kikuchi et al., 2016, O'Dell et al., 2012, Yoshida et al., 2011, Parker et al., 2011, Heymann et al., 2015, Oshcheplov et al., 2013 for particularly polarization-sensitive scenes such as sun-glint data (O'Brien et al., 2013). The implementation and fine-tuning of the 2 x 3 matrix needs to be carried out by a separate level-2 algorithm. The prelaunch Mueller matrix is available from the GOSAT-2 data distribution site as a reference.

Signal-to-noise ratio (SNR) characterization
The signal-to-noise ratio (SNR) for TANSO-FTS-2 is determined by the ratio of in-band signal intensity divided by out-ofband signal intensity. The maximum resolving power achievable with the field of view of the FTS configuration is 32000:1, and the achievable resolution in band 1 is < 0.4 cm -1 . As supplemental information for level-2 processing, the SNR for each observation is stored in the operational level-1 products. Figure 5 presents a typical SNR for TANSO-FTS-2. 215 The temperature of the interferometer for TANSO-FTS-2 is controlled to around 295 K. As a result, the thermal background radiation is not negligible and needs to be calibrated with BB and DS calibration. However, the actual signal, which represents the radiation from the earth and the thermal emission of the instrument itself, has to be balanced. In the TANSO-FTS-2 case, the balanced temperature that creates an almost zero-amplitude interferogram condition coupled with https://doi.org/10.5194/amt-2020-360 Preprint. Discussion started: 28 September 2020 c Author(s) 2020. CC BY 4.0 License. the earth radiation and instrument internal thermal emission, for TIR is around 220K. For this reason, the calculated SNR for 220 TIR with this procedure presents a quick reference of incoming signal intensity.
The simplified SNR is included in the operational L1B products as auxiliary information for level-1 users. It is based on 225 the ratio between the maximum (max) of the in-band signal and the standard deviation (std) of the out-of-band signal (at the lower and the upper ends of the spectral range). The formulation of the simplified SNR is expressed by equation (2). 230 In addition to the simplified SNR estimate, the full SNR model is given by equation (3).
where is the monochromatic radiance and the empirical parameters , , and are listed in Table 3. 235

ILS model
The ILS function is also characterized during prelaunch testing and the calibration period. The monochromatic light sources for bands 1 and 2 are used for this test. In addition, gas cells filled with CO2, and CO are used for bands 3, 4, and 5. The modelled ILS functions are processed according to these measurements and issued as the initial versions of ILS functions on 240 the GOSAT-2 web site.
The in-orbit ILS functions are expected to be different than the test set because of small optical alignment changes that may have happened during and after launch. Updated ILS functions were determined through comparison between on-orbit ILS calibration and solar calibration datasets. The most recent estimated ILS function models are plotted in Fig. 6. A later part of this paper will discuss the challenges related to acquiring ILS knowledge. 245

Level-1 processing algorithm for TANSO-FTS-2
TANSO-FTS-2 acquires the uniform-time-sampled interferogram for each observation and each detector channel using a high rate of sampling to minimize noise. The uniform-time-sampled interferogram is numerically filtered and decimated, which have a factor 5 for bands 1 and 2, a factor 6 for band 3, a factor 12 for bands 4 and 5, in real time to reduce the data 250 volume needing to be stored in the onboard mission data record processer (MDP), and transmitted to the ground once every two orbits.
The first step of processing in the ground facility is generating the level-0 product. Level-0 processing consists of data sorting and decompressing raw data. The second step is to generate level-1A/UTS files, which means that Uniform Timesampled (UTS) interferograms (1A) are created combining a uniform-time-sampled interferogram and ancillary data such as 255 satellite position, line of sight, and housekeeping data for the instrument.

Processing for uniform-time-sampled to uniform OPD-sampled interferogram.
All the processing from interferogram to atmospheric radiance spectra is performed on the ground. In the first step, the decimated uniform-time-sampled interferogram is constructed with a DC offset and gain correction. The equation for the 265 uniform-time-sampled interferogram is described by equation (4).
where, 270 : Bands ( Offset signal. If the interferogram value at zero path difference (ZPD) is equal to or greater than the full bit range for a band, then a quality warning flag is set for that band, since interferograms with data saturation are not suitable for data processing.
However, since the saturation is identifiable by setting a flag, the data will be inverse Fourier transformed and stored as part 285 of the L1B data.
When passing through the South Atlantic Anomaly, spikes may occur in the interferograms due to energetic particle radiation. If a spike is included in the interferogram, then a quality (warning) flag is set in the L1B file. Theoretically, a onepoint spike signal creates a single, generally high-frequency noise component over the complete spectral domain after the inverse Fourier transform. To reduce this noise, a spike signal correction technique is applied. In the case of a spike being at 290 the -th point, the -th sampled data is replaced by the average value of samples − 1 and + 1. In the case of the edge of the interferogram, = 2 = − 1 sample is used.
The acquisition duration for time-sampled interferograms, called the sampling window, is synchronized with the scan motion of the interferometer. When the metrology pulse is active, the counter of the sampling clock is incremented by one for each metrology pulse, then the counted values are transmitted to the MDP as the time interval between metrology pulses. 295 The nominal count is 3458 sampling clock pulses. In parallel, the nominal sampling frequency of the SWIR and TIR signal channels is 117 kHz. The observed interferometric signal is converted from analog to digital by the ADC, which is triggered by the master clock. The original ADC samples are called undecimated samples. The delay between metrology pulses and the science signal is not zero, therefore the time delays have to be included by the processing algorithm. In the nominal observation mode, the signal is decimated with specific values. The decimated science signal is transmitted to the ground and 300 retrieved for atmospheric spectra. In parallel, the metrology time data is also transmitted as counts to the ground. The number of sampling pulses of the metrology signal is fixed at 76789: Two sample pulses are generated for each OPD change of 1 wavelength of the metrology laser so, at a wavelength of 1.31um, the number of sampling pulses correspond to a total OPD change of 5 cm.
To process OPD-sampled science interferograms, time-sampled science interferograms are coupled with the time-305 sampled metrology signals, because the metrology signal contains both time and OPD domain information. OPD-sampled interferograms are obtained by applying a digital filter to the time-sampled interferogram in ground processing. The coefficients of digital filters are optimized during prelaunch tests.
Bands 2, 3, and 4 use PV-MCT detectors, and band 5 uses a PC-MCT detector. MCT detectors generally have a nonlinear response that is normally more pronounced with PC-MCT detectors. If the nonlinearity is not negligible, a wider 310 range of interferogram frequencies should be considered. Theoretically, the parent signal due to nonlinearity shows the harmonic signal features in out-of-band regions. Then, the nonlinear correction terms are characterized with verifying out-ofband signal intensity. For bands 2, 3, and 4, there are no observed harmonic signal features in the out-of-band regions. Only band 5 for the TANSO-FTS-2 has been implemented with a second-order nonlinear term in operational processing.

Processing for bands 1, 2, and 3 spectra
This section provides an overview of how OPD-sampled interferograms (referred to as interferograms here) are transformed into spectra. The basic processing steps are the same as described for the TANSO-FTS per Kuze et al., 2012Kuze et al., , 2016 where : Operator for Fourier transform ℎ : Operator for shifting the zero-frequency component to the center of the array. 330 Then the phas-corrected spectrum *,, is given by equation (6) *,, = F8//,*,, 335 and the associated wavenumber is given by equation (7) = H , Where 340 G56,*,, : Low resolution spectrum filtered with a gauss function for phase correction * : Sampling interval in inverse wavenumber units https://doi.org/10.5194/amt-2020-360 Preprint. Discussion started: 28 September 2020 c Author(s) 2020. CC BY 4.0 License.
Next the phase-corrected spectra, *,, , in units of V/cm -1 are converted to radiance units through multiplication by the radiance conversion coefficients, * as well as the time-dependent degradation factor: ( , ). The radiance spectra at 345 time and wavenumber are then given by equation (8).

Processing for bands 4 and 5 spectra 350
In contrast to the SWIR (band 1-3 spectra) processing, TIR processing requires not only nadir (earth scene) observation spectra but also calibration spectra both of deep space and black body calibrations. For each nadir spectrum, the most recent calibration spectra are selected. The data-trimming method is the same as for SWIR. Equation (9)  year of operation, the radiometric, geometric, and spectroscopic parameters were characterized via calibration operations.
The following section describes the characterization method and results. In addition, some calibration parameters (such as 380 the degradation correction factor for the radiances) are updated. The updated products are compared with temporally and spatially coincident observations of the TANSO-FTS on GOSAT, AIRS on AQUA, and IASI on METOP-B.

Operation overview
GOSAT-2 operations have been nominal and continuous during the first year with three short suspensions of observation. On 385 8 and 24 April 2019, unplanned satellite maneuvers were carried out to avoid collision with space debris. On 11 December 2019, TANSO-FTS-2 was set to safe mode and suspended observation due to an illogical observation plan that was detected on board. The details of suspended periods are given on the GOSAT-2 website (https://prdct.gosat-2.nies.go.jp/en/index.html).
In the initial on-orbit period, it is also important to establish the stability of operations over periodic cycles, with an 390 emphasis on the operational temperature set points. Figure 9 presents the time series of the cooler stage temperatures and the beam splitter temperature in the interferometer as well as the beta angle (sun-satellite angle) and sun-satellite distance. In the early operation period, the temperature of stage 4 was at the set point of 97 K. However, at this temperature setting, the actual temperature shows a time dependency, probably due to unexpected thermal radiance incoming to the passive radiation cooler. As a result on 12 April 2019, the temperature control setpoint for stage 4 was changed from 97 K to 100 K. 395 After this set point change, the temperature of stage 4 was controlled well at 100 K without time dependency.
In nominal operation, the radiator of the cooler views deep space and maintains the set point temperatures. During lunar calibration, the flight geometry of the satellite is changed and the view of the radiator is contaminated by Earth's thermal emission. As a result, temperatures of stages 2, 3, and 4 increase rapidly. These perturbed temperatures settle down to the operational values only six hours after the lunar calibration. 400 As of 12 July 2019, the beam splitter temperature shifted from 293.5 K to 292.5 K. Thermal analysis with on-orbit data suggested that the temperature gradient around the interferometer is larger than expected. To reduce the temperature gradient, especially at the beam splitter of the interferometer, the controlled set point of the interferometer was changed as https://doi.org/10.5194/amt-2020-360 Preprint. Discussion started: 28 September 2020 c Author(s) 2020. CC BY 4.0 License. well as the metrology laser control temperature. As a result, the metrology laser wavelength made a small shift at this time.
In summary, in the first year, two of the key instrument temperatures were adjusted to account for on-orbit conditions. 405

Radiometric characterization
TANSO-FTS-2 has four types of radiometric calibration sources on board: (1) a monochromatic source which accommodates 768 nm and 1543 nm laser diodes with an integrating sphere; (2) a solar diffuser plate made from Spectralon® equipped with the flip and shutter mechanism; (3) a four-panel black body equipped with three temperature 410 sensors; and (4) a viewing port directed toward deep space. These are called as ILS cal., Sol cal., BB cal., and DS cal., respectively. The ILS cal. has been performed every six days since May 2019 during night and over sea conditions. Solar diffuser calibration is conducted every orbit in the satellite position over Antarctica. Once every three months, the solar diffuser of the primary side is flipped to the back side (secondary side) and exposed to solar light to identify the degradation of the solar diffuser on the primary side. BB and DS calibrations are periodically scheduled. Nominally, two pairs of BB and 415 DS calibrations are executed on the day side of the orbit and four pairs on the night side. In addition, TANSO-FTS-2 points to the moon once a month during the night side of the orbit and while the pitch maneuver is suspended.

SWIR (bands 1, 2, and 3)
Sol cal. and ILS cal. are used to characterize the sensitivity change of the SWIR bands (bands 1, 2, and 3). As mentioned, 420 Sol cal. is conducted in every orbit with the primary solar diffuser. The distance between the sun and satellite positions is taken into account in the normalized signal intensity. In this analysis, the Bidirectional Reflectance Distribution Function (BRDF) of the solar diffuser is not considered. Figure 10 shows the present-time (left panels) and the beta-angle (right panels) dependencies of the signal from the solar diffuser for bands 1p, 1s, 2p, 2s, 3p, and 3s (top to bottom). The wavenumbers considered for the analysis are listed in Table 5. There were substantial sensitivity changes with time and input 425 angle to the solar diffuser, and these plots suggest that bands 2 and 3 are less time dependent than band 1.
Sol cal. measurements with the "secondary" diffuser plate have been conducted on 15 April, 2 July, 6 October 2019, and on 4 January 2020. Figure 10 (circles) shows that there is no difference between reference and routine plate for the signal level. In other words, the primary diffuser plate shows no degradation.
To make an independent assessment of the time-dependent degradation, the signal intensities of the ILS cal. The time-dependent degradation factors for SWIR were determined by assessing Sol cal. data from February 2019 to March 2020. Equation (10) provides the empirical degradation correction ( , ) to be used in equation (8) and table 6 lists  435 the respective parameters , , , and : where U is the time on 5 February 2019. The corrected signal levels for Fig. 10 are plotted in Fig.11. Compared to Fig.10, 440 the lines on Fig.11. are indistinguishable, appearing on top of each other. Since the BRDF of the solar diffuser has not been taken into account so far, the signal level still shows a spurious correlation with the input angle. The effect of BRDF is ignored in the earth scene observation.
The calibrated spectral radiances for TANSO-FTS-2 are compared with temporally and spatially coincident TANSO-FTS spectral radiances. The screening conditions are as follows: 445 -Less than 2 km between the GOSAT and GOSAT-2 pointing locations -Less than 80 degrees of the solar zenith angle -Less than 5 deg difference of viewing angle for GOSAT and GOSAT-2 against pointing location -Brightness temperature for TIR region is greater than 250 K -Quality flags: 0 450 -Less than 10 % of cloud probability assessed by the TANSO-FTS-2 onboard camera Figure 12 compares the TANSO-FTS and TANSO-FTS-2 radiances for six-day averages. Figure 12 shows that the calibrated TANSO-FTS-2 spectral radiances generally agree with TANSO-FTS spectral radiances within 2 % of the averaged bias and 0.5 % standard deviation. The evaluated numbers are listed in table 7.

TIR (bands 4 and 5)
Spectral radiances for the TIR region (bands 4 and 5) are periodically calibrated by applying BB and DS cal. spectra.
Nominally, six pairs of these calibrations are done every orbit. To characterize the noise performance of the TIR bands in orbit, the typical Noise-Equivalent differential radiaNce (NEdN) and Noise-Equivalent differential Temperature (NEdT) are plotted in Fig.13 with the black body temperature trend within a one-revisit time period. Figure 13 upper panel presents the 460 temperature of the black body source which varied by 0.5 K peak-to-peak. To estimate the NEdN and NEdT in orbit, the black body spectra and deep space spectra were processed according to equations (11) and (12) (Chen et al., 2015), where is the -th measurement and the other symbols are the same as defined above for equation (9). Typically, there are up to 24 calibration measurements per orbit. Figure 13 shows that NEdNs for both bands 4 and 5 have almost the same noise level. NEdT is less than 0.3 K against the typical black body temperature condition (around 294.2 K). -Less than +/-100 km and +/-5 min between GOSAT-2 and AIRS/IASI orbit -Smaller than +/-3 degree of cross-track and +/-3 degree of along-track angle for GOSAT-2 -Less than 17 km between GOSAT-2 and AIRS/IASI observing locations 480 Since the spectral resolution of AIRS is lower than that of TANSO-FTS-2, we convolve the TANSO-FTS-2 spectra with the AIRS spectral response function. The same scheme is applied to IASI data. The inter-comparison method is the same as the one used by Kataoka (2019). Figure 15 shows the comparison results. The mean temperature differences between TANSO-FTS-2 and AIRS, and TANSO-FTS-2 and IASI are less than 0.5 % from February 2019 to March 2020. In the case of TANSO-FTS-2 and AIRS 485 comparison, the coincident scenes have moderate brightness temperatures such as 280 K. In contrast, for the TANSO-FTS-2 and IASI comparison, the brightness temperature is colder than that for the AIRS case. Figure 15 (b) shows the differences at four focused channels against the window temperature. This figure suggests that the differences are almost zero over the 240K window temperature. However, it has 1 to 2 K bias in the cold scene with some wavenumber dependencies.

Figures 15 (c) and (d) present the time series of the brightness temperatures difference between TANSO-FTS-2 and 490
IASI and between TANSO-FTS-2 and AIRS, showing that cold scene targets such as the CO2 and CH4 channels exhibit a 1 to 2 K bias. However, the bias is less than 0.5 K in the warm target case.

Spectral radiance characterization challenges
In the first year of operation, the spectral radiances for both the SWIR and TIR were recalibrated and the radiance conversion coefficients were updated. In parallel, several challenges were identified, for example, the SWIR polarization sensitivity has 495 changed, and the dependence of TIR calibration on scene temperature and wavenumber changed.
As described in the previous section, TANSO-FTS-2 shows considerable sensitivity to the U Stokes vector component.
The polarization sensitivity has been characterized on the ground by positioning a linear polarizer in front of the instrument.
In contrast, on-orbit data suggests that these polarization sensitives changed after launch. A future update of the level-1 product will include the best estimated radiances and the related polarization model. For updating the polarization model, we 500 require feedback from retrieval studies that examine retrieval performance for polarizing scenes.
The correction term in equation (9) for calibrating the TIR radiance is ideally equal to 1, but empirically differs from 1 to compensate for the bias. As a result, the spectral radiance differences between TANSO-FTS-2 and AIRS/IASI are minimized, and the brightness temperature difference is less than 0.5 % during the first year when the sensitivity correction factor is set to 1.0198. However, is estimated by an empirical method. Currently, we assume that deviating from 1 505 originates from unaccounted polarization sensitivity in the TIR optics since the input polarized light geometry is completely different between nadir observations and black body observations. In addition, might have wavenumber dependence. For future versions of level-1 products, a theoretical model will be constructed and implemented for level-1 processing.

Geometric characterization 510
To identify a line-of-sight offset, the processed pointing locations based on satellite position and the along-track (AT) and cross-track (CT) pointing angles are compared with the validated ground control position based on co-registered camera images. The time series of the root-mean-square differences of latitude and longitude between the processed pointing locations and the validated locations are plotted in Fig.16 for the period 5 February 2019 to 31 March 2020. Figure 16 (a) shows that there is no time dependence in the differences. Likewise, the latitude and longitude differences are plotted in Fig.  515 16 (b). The graph shows that TANSO-FTS-2 points have almost no offset. The averaged differences are less than 0.02 km in latitude and 0.06 km in longitude. The standard deviation of differences is 0.17 km in latitude and 0.18 km in longitude.

Spectroscopic characterization
Through the match-up analysis between TANSO-FTS and TANSO-FTS-2, we found that the absorption spectra of TANSO-520 FTS-2 show a marginally coarser spectral resolution than TANSO-FTS in bands 1 to 3. Theoretically, the spectral resolution between TANSO-FTS and TANSO-FTS-2 should be the same. However, optical aberration and alignment is slightly different between the two instruments. During prelaunch characterization, the ILS function was derived from monochromatic light source measurements. Theoretically, the response to monochromatic light can provide the proper ILS given that the light beam is uniform and covers the full FOV. Non-uniformity of the light beam or partial illumination of the detector can 525 lead to a narrower line shape. Therefore, the ILS function is reassessed with on-orbit data.

In-orbit ILS calibration
The designed spectral resolution is 0.2cm -1 for all bands. Due to the finite field of view, optical aberration, and misalignment, the theoretical sinc-function is distorted and spectral resolution is worse. 530 TANSO-FTS-2 accommodates monochromatic light sources. These light sources allow us to monitor the changes in the ILS in orbit. Typically, ILS at shorter wavelengths are more susceptible to alignment or illumination changes. The 0.77 um laser diode is preferred for identifying the changes. Figure 17 presents the ILS function for bands 1 and 2 based on the laser diode signal, and the trend of the FWHM (fullwidth-at-half-maximum) for both wavelengths. To create densely sampled spectra, we applied zero-padding to the original 535 OPD-sampled interferograms and retrieved the ILS with higher sampling. The shapes of bands 1 and 2 are significantly different. Band 1 is the most sensitive to optical alignment and has an asymmetric line shape. In contrast, band 2 has an almost symmetric line shape.

Spectral response characterization challenges 540
As described in the previous section, ILSs show a time dependence. This might be due to optical alignment change with time. Not only the ILS function, but also polarization sensitivities were found to change over time because the ratio between p and s polarization signal was changing. The first year of operation data suggest that the rate of change is mild and was likely to become progressively smaller. Since July 2019, the instrumental line shape is almost constant.
For level-1 version102102 auxiliary data, a best-estimate ILS function is provided on the ILS cal. and Sol cal. datasets. 545 The best-estimate ILS function is slightly wider than that of the prelaunch testing. However, a time-dependent term is not implemented. The typical residual between observed spectra and theoretical spectra is reduced with the current best-estimate ILS function and plotted in Fig.18. In a future update, the time-dependent ILS function will have to be improved; it is especially needed for Band 1.

Intelligent pointing functionality
TANSO-FTS-2 carries a camera which can identify cloudy areas in the TANSO-FTS-2 field of view before observation.
Based on onboard processing of the camera images, TANSO-FTS-2 relocates the observation point to a cloud-free area.
Since the processing resources in orbit are limited, a simple and fast cloud identification algorithm was implemented. 555 https://doi.org/10.5194/amt-2020-360 Preprint. Discussion started: 28 September 2020 c Author(s) 2020. CC BY 4.0 License.

Cloud identification algorithm
A camera image pixel is decomposed into three-pixel information (red, green, and blue). A simple cloud-detection algorithm uses these raw three-pixel measures to directly identify the cloud-contaminated pixels in the image. To identify the most probable cloud locations, the following filters are applied with thresholds of , , , , , and equal to 73, 16, 83, 2, 4, and 60, respectively.
If the above conditions for filters 1 or 2 or 3 conditions are confirmed, and those for filter 4 are not, the scene is identified as probably cloud covered.
The result of cloud detection for each scene on orbit is not recorded. However, we did add auxiliary information about 580 onboard cloud identification by applying the same algorithms to the camera images during on-ground processing. Using this information allows on-orbit performance to be evaluated. To this end, for each TANSO-FTS-2 measurement, we take the camera images taken before and after each FTS observation and run the above cloud-detection algorithm on the images. We calculate a cloud cover index for each FTS observation by finding the ratio of cloudy camera image pixels to the total number of camera pixels within the FTS footprint. Thus, the cloud cover index is 0 for cloudless scenes, and the index is 585 100 % for fully cloudy scenes. Figure 19 displays a typical global map for cloud cover index lower than 1 % for September 2019, comparing the cloud cover index before (upper panel) and after (lower panel) intelligent pointing. Clear differences are observed over central America, the Amazon, central Africa, and southeast Asia. In these areas, with intelligent pointing, the number of measurements with a low cloud cover index is increased by 1.7, 1.6, 1.9 and 2.1, respectively. Globally, for the study period 590 from March to December 2019, the number of clear-sky retrievals (cloud cover index up to 1%) was increased by a factor of 1.8 over land for intelligent pointing over standard pointing.

Intelligent pointing challenges
The cloud detection algorithm for intelligent pointing is based on simple brightness and chroma thresholds. For darker 595 scenes, such as ponds and lakes, the method tends to fail to detect cloudy areas. For brighter surfaces, such as concrete buildings in a city, the method has too many cloud detections. Both false positive and false negative detections imply the risk for unsuitable re-pointing operations for the FTS, so it might be better to examine the filter thresholds and the observation plan region by region. If we identify areas on the globe that are unsuitable for re-pointing, intelligent pointing can be switched off and on depending on the location. In the current operation, intelligent pointing is switched off over Cal/Val. 600 sites and user-specified sites.

Conclusions
The Japanese Greenhouse gases Observing SATellite-2 (GOSAT-2), in orbit since October 2018, is the follow-up mission of GOSAT, which has been operating since January 2009. Both satellites are dedicated to the monitoring of global carbon 605 dioxide and methane to further knowledge of the global carbon cycle. This paper has reported on the function and performance of the TANSO-FTS-2 instrument, level-1 data processing, and calibrations for the first year of GOSAT-2 observation. To evaluate its performance, the spectral radiances (level-1 processor version v102102) collected by TANSO-FTS-2 between February 2019 and March 2020 are compared with the spatiotemporally coincident measurements of the TANSO-FTS on GOSAT for the SWIR band, and with AIRS on AQUA, IASI on METOP-B for the TIR bands. We 610 conclude that the spectral radiances measured by TANSO-FTS and TANSO-FTS-2 agree to within 2 % for the SWIR bands.
In the TIR, the agreement between TANSO-FTS-2 and AIRS, IASI is better than 0.5 % (1 K) for scenes brighter than 220 K.
We further evaluated GOSAT-2's intelligent pointing mechanism based on active cloud avoidance. The preliminary analysis indicates that the number of scenes useful for spectral analysis increased by factor 1.8 over a stiff pointing schedule.