Inter-calibration of HY-1B/COCTS thermal infrared channels with MetOp-A/IASI

: The Chinese Ocean Color and Temperature Scanner (COCTS) on board the Haiyang-1B (HY-1B) satellite has two thermal infrared channels (9 and 10) centred near 11 µ m and 12 µ m, respectively, which are intended for sea surface temperature (SST) observations. To improve the accuracy of COCTS SSTs, inter-calibration of COCTS thermal infrared radiance is carried out. The Infrared Atmospheric Sounding Interferometer (IASI) on board MetOp-A satellite is used as inter-calibration reference owing to its hyperspectral nature and high-quality measurements. The inter-calibration of HY-1B COCTS thermal infrared radiances with IASI is undertaken for data from the Period 2009–2011 located in the northwest Paciﬁc. Collocations of COCTS radiance with IASI are identiﬁed within a temporal window of 30 min, a spatial window of 0.12 ◦ and an atmospheric path tolerance of 3%. Matched IASI spectra are convolved with the COCTS spectral response functions, while COCTS pixels within the footprint of each IASI pixel are spatially averaged, thus creating matched IASI-COCTS radiance pairs that should agree well in the absence of satellite biases. The radiances of COCTS 11 and 12 µ m channel are lower than IASI with relatively large biases, and a strong dependence of difference on radiance in the case of 11 µ m channel. We used linear robust regression for four different detectors of COCTS separately to obtain the inter-calibration coefﬁcients to correct the COCTS radiance. After correction, the mean values of COCTS 11 and 12 µ m channel minus IASI radiance are − 0.02 mW m − 2 cm sr − 1 and − 0.01 mW m − 2 cm sr − 1 , respectively, with corresponding standard deviations of 0.51 mW m − 2 cm sr − 1 and 0.57 mW m − 2 cm sr − 1 . Striped noise is present in COCTS original radiance imagery associated with inconsistency among the four detectors, and inter-calibration is shown to reduce, although not eliminate, the striping. The calibration accuracy of COCTS is improved after inter-calibration, which is potentially useful for improving COCTS SST accuracy in the future.


Introduction
Satellite measurements are critical for quantifying Earth's environmental parameters, monitoring climate change and informing numerical weather prediction.Satellite calibration is fundamental to the error characteristics of derived satellite products.Calibration accuracy, stability and consistency over time are important for the retrieval quality of geophysical and atmospheric parameters [1].
The data we used include the COCTS Level 1B (L1B) radiance data provided by NSOAS [19] and IASI Level 1C (L1C) radiance data provided by EUMETSAT [20].The IASI L1C products are the geolocated, calibrated and apodized radiances, consisting of 8461 spectral samples every 0.25 cm −1 [6].This paper is organised as follows.In Section 2, the inter-calibration method is introduced.In Section 3, we compare the COCTS thermal infrared radiances from 11 and 12 µm channel with IASI.In Section 4, we describe the calculation of coefficients for COCTS radiance correction.In Section 5, we validate the COCTS corrected radiance.In Section 6, the post-recalibration improvement in striped noise reduction is analysed.In Section 7, the main results are summarised.

Calculation of IASI-Convolved Radiance
IASI L1C radiance data include 8461 spectral samples in every pixel with a fixed spectral binning of 0.25 cm −1 .The spectral ranges of COCTS thermal infrared spectral response functions (SRFs) for 11 and 12 µm channel are 9.50-12.50µm and 10.50-13.50µm, respectively (shown in Figure 1), both with a bin size of 0.002 µm.Due to the spectral resolution difference between IASI spectral radiance and COCTS SRFs, linear interpolation is used to generate COCTS spectral response value at each IASI radiance samples.Then, the IASI-convolved radiance corresponding to COCTS SRFs could be calculated using Equation (1): where subscript i indicates COCTS thermal infrared channel number, L(λ) is the IASI spectral radiance, S i (λ) is the COCTS spectral response, and λ 1 and λ 2 are COCTS bandpass boundary values.
Remote Sens. 2018, 10, x FOR PEER REVIEW 3 of 16 geolocated, calibrated and apodized radiances, consisting of 8461 spectral samples every 0.25 cm −1 [6].This paper is organised as follows.In Section 2, the inter-calibration method is introduced.In Section 3, we compare the COCTS thermal infrared radiances from 11 and 12 μm channel with IASI.
In Section 4, we describe the calculation of coefficients for COCTS radiance correction.In Section 5, we validate the COCTS corrected radiance.In Section 6, the post-recalibration improvement in striped noise reduction is analysed.In Section 7, the main results are summarised.

Calculation of IASI-Convolved Radiance
IASI L1C radiance data include 8461 spectral samples in every pixel with a fixed spectral binning of 0.25 cm −1 .The spectral ranges of COCTS thermal infrared spectral response functions (SRFs) for 11 and 12 μm channel are 9.50-12.50μm and 10.50-13.50μm, respectively (shown in Figure 1), both with a bin size of 0.002 μm.Due to the spectral resolution difference between IASI spectral radiance and COCTS SRFs, linear interpolation is used to generate COCTS spectral response value at each IASI radiance samples.Then, the IASI-convolved radiance corresponding to COCTS SRFs could be calculated using Equation ( 1): (1) where subscript indicates COCTS thermal infrared channel number, is the IASI spectral radiance, is the COCTS spectral response, and and are COCTS bandpass boundary values.

Generation of Matchups
According to the COCTS and IASI spatial resolution of 1 km and 12 km near nadir, respectively, COCTS and IASI-convolved radiance are projected to equal-angle maps in the research region (0°-48°N and 100°E-148°E) with the grid size of 0.01° and 0.12°, respectively.The collocations of COCTS and IASI radiance data are generated with a spatial window of 0.12° and a temporal window of ±30 min.We need to ensure the matchups of COCTS and IASI are measured under similar conditions.

Generation of Matchups
According to the COCTS and IASI spatial resolution of 1 km and 12 km near nadir, respectively, COCTS and IASI-convolved radiance are projected to equal-angle maps in the research region (0 • -48 • N and 100 • E-148 • E) with the grid size of 0.01 • and 0.12 • , respectively.The collocations of COCTS and IASI radiance data are generated with a spatial window of 0.12 • and a temporal window of ±30 min.
We need to ensure the matchups of COCTS and IASI are measured under similar conditions.Due to the effect of atmospheric path on radiance measurements, the satellite zenith angle of COCTS and IASI should also be matched when generating the matchup datasets.We set the threshold of difference in the secant of the satellite zenith angles as 0.03, constraining the difference of COCTS and IASI atmospheric paths to be less than 3% to ensure the view conditions are close.The variances of COCTS minus IASI radiance difference against temporal window value and atmospheric paths difference have been checked.There is no obvious increasing trend presented, demonstrating that these window thresholds of finding collocations we set is reasonable.The radiances of COCTS pixels across all detectors inside each IASI IFOV are averaged as the collocated COCTS radiance with IASI.Typically, there are about 50-140 pixels go towards average under different satellite zenith angles.All the COCTS pixels are evenly weighted, despite the effect of IASI point spread function [3,6].Considering the inconsistency between COCTS four parallel detectors, COCTS pixels inside each IASI IFOV measured by per detector are averaged separately later, to estimate the difference between COCTS detectors and remove the striped noise.

Filtering of Matchups
Because the spatial resolution of IASI is much larger than COCTS, the homogeneity of each IASI IFOV is very important.Radiance nonuniformity within the IASI IFOV increases the spatial uncertainties of matchups because of the different point spread functions between IASI and COCTS as well as increasing the error due to geolocation errors [15,16].We used the relative standard deviation, which is the standard deviation/mean of the COCTS valid pixels inside each IASI IFOV, to quantify the homogeneity.We also used a perimeter region outside the central collocation area to reduce the likelihood of time variable components such as errors caused by differences in the cloud and/or clear distributions in a similar way to that suggested by GSICS [21].For the COCTS and IASI gridded radiance after equal-angle projection, we calculated the relative standard deviation of COCTS pixels at 0.12 • × 0.12 • region (shown as the yellow region of Figure 2) to estimate the uniformity of the corresponding IASI IFOV, and the relative standard deviation of surrounding pixels around the target region within 0.16 • (shown as the green region) as the estimation of homogeneity in the wider surroundings.Figure 3 shows the variations of COCTS 11 and 12 µm channel radiance minus IASI radiance difference against the relative standard deviation of COCTS pixels in target region (Figure 3a,c) and surrounding environment region (Figure 3b,d), respectively.The plotted colour indicates the number of collocations in every 0.0005 relative standard deviation bin and in 0.15 mW m −2 cm sr −1 radiance difference bin.The radiance differences in both target region and surrounding environment region become larger with the larger relative standard deviations, indicating that the relative standard deviation is an effective factor to control the matchup spatial uncertainties.According to the distribution pattern for different channels and different regions, we chose different thresholds (shown as black vertical lines in Figure 3) to select matchups located in uniform scenes.For 11 µm channel, we chose the matchups with the COCTS relative standard deviation in the target and surrounding region smaller than 0.006 and 0.01, respectively.For 12 µm channel, we chose the matchups with the COCTS relative standard deviations in target and surrounding region smaller than 0.01 and 0.013, respectively.Both criteria on relative standard deviation must be met for both channels for the matchup to be retained.The total number of filtered matchups of COCTS and IASI in uniform scenes from 2009 to 2011 is 11,250 and all of the subsequent analysis and correction are based on these filtered matchups.

Comparison of COCTS Radiance with IASI
Based on these 11,250 filtered matchups presented in Section 2, COCTS radiance from 11 and 12 μm channel are compared with IASI. Figure 4a,c    There is distinct striped noise with a pattern approximately repeating every four scan lines in COCTS radiance image, due to the inconsistency between four parallel detectors [4].Similar striping is also seen on other whiskbroom scanning radiometers, such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS) [22][23][24][25].
To estimate the measurement difference between different detectors, the COCTS pixels inside each IASI IFOV measured by individual detectors are averaged separately.Then, the COCTS radiance from four detectors (called Detector 1, Detector 2, Detector 3, and Detector 4 hereafter) are compared with IASI separately.Figure 5  There is distinct striped noise with a pattern approximately repeating every four scan lines in COCTS radiance image, due to the inconsistency between four parallel detectors [4].Similar striping is also seen on other whiskbroom scanning radiometers, such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS) [22][23][24][25].To estimate the measurement difference between different detectors, the COCTS pixels inside each IASI IFOV measured by individual detectors are averaged separately.Then, the COCTS radiance from four detectors (called Detector 1, Detector 2, Detector 3, and Detector 4 hereafter) are compared with IASI separately.Figure 5       Figure 6a represents a strong negative correlation between 11 µm channel radiance difference and IASI radiance that is consistent with overall three-year comparison results, also leading to strong fluctuations of daily biases.In other words, the fluctuation of biases from 11 µm channel is dominated by a radiance-dependent variation, not by a time-dependent variation.For 12 µm channel, there is no obvious daily radiance difference dependence on IASI radiance shown, and the biases keep relatively stable, except for a difference jump about 1 mW m −2 cm sr −1 decreasing at the junction point between March and April 2011 (shown as the black circle of the lower line in Figure 6b).Thus, the overall calibration of COCTS thermal infrared channels was stable from 2009 to 2011, except for the calibration drifting at the beginning of April 2011.

Calculation of Coefficients for COCTS Radiance Correction
The coefficients of COCTS radiance correction are obtained based on the filtered matchups of COCTS with IASI radiance.We use the robust linear regression to obtain the slope, , and offset, ,

Calculation of Coefficients for COCTS Radiance Correction
The coefficients of COCTS radiance correction are obtained based on the filtered matchups of COCTS with IASI radiance.We use the robust linear regression to obtain the slope, a, and offset, b, of the relationship between COCTS minus IASI radiance difference and IASI radiance, represented as the Equation (2): where the L COCTS and L I ASI represent the COCTS and IASI original radiances, respectively.The COCTS original radiances are fitted to IASI data using Equation ( 2), therefore the COCTS corrected radiance L COCTS (substituting L COCTS for L I ASI ) are given using Equation (3).A randomly selected subset, 2/3 of the whole matchup dataset, was used for regression and the remaining 1/3 were used for validation.

Validation of COCTS Corrected Radiance
The comparison results of COCTS 11 and 12 µm channel corrected radiance with IASI are shown as Figure 7. Figure 7a,c shows scatter plots, while Figure 7b,d presents the variations of radiance difference against IASI radiance after correction.Table 3 is the statistics of comparison results before and after correction based on the validation matchups.In Table 3, the mean differences and standard deviations of COCTS 11 and 12 µm channel radiance with IASI in two periods separately and combined are provided.The large biases between COCTS original radiance and IASI radiance are removed after correction, with the value of −0.02 mW m −2 cm sr −1 and −0.01 mW m −2 cm sr −1 for 11 and 12 µm channel during the whole period from January 2009 to December 2011, respectively.For 11 µm channel, the significant radiance-dependent pattern of radiance difference is corrected, with the reduced overall standard deviation from 0.95 mW m −2 cm sr −1 to 0.51 mW m −2 cm sr −1 .A reduction in the bias and standard deviation between the COCTS and IASI is also seen for 12 µm channel.We have also provided the same statistics in brightness temperature (BT) space for ease of comprehension.As shown in Table 3, the overall mean differences of COCTS 11 and 12 µm channel original BTs with IASI are −4.08K and −4.76 K, with the corresponding standard deviations of 0.50 K and 0.39 K.After correction, both mean values of COCTS 11 and 12 µm channel corrected BTs minus IASI are −0.01K, with the corresponding standard deviations of 0.33 K and 0.35 K respectively.not represented again in Figure 8b, indicating that the different coefficients applied in two periods separately is reasonable to correct COCTS calibration drifting.
The time series plots of COCTS 11 and 12 μm channel corrected radiance minus IASI are shown in Figure 8a,b respectively.Similar to Figure 6, the upper black lines are daily mean values of IASI radiance, and the lower blue lines are the daily variations of COCTS corrected radiance minus IASI difference with the corresponding uncertainties shown as red vertical lines.For 11 μm channel, compared with Figure 6a, the fluctuations of radiance difference become much smoother.In addition, the jump of 12 μm channel radiance difference at the junction point between March and April 2011 is not represented again in Figure 8b, indicating that the different coefficients applied in two periods separately is reasonable to correct COCTS calibration drifting.The radiance images, as well as the significant fluctuations of original radiance variations, display the strong striping in COCTS radiance data.In addition, the larger difference between neighbour detectors of 12 µm channel indicates that the striping of 12 µm channel radiance is stronger than 11 µm channel.After the correction of COCTS original radiance using individual coefficients for four detector measurements, the striped noise shown in corrected COCTS radiance images is reduced, with smoother variations along track, demonstrating that the inter-calibration method in this study is useful in some extent for COCTS striped noise elimination.Moreover, the histogram statistics of local standard deviations over 3 by 3 pixels box (LSD 3×3 ) of the whole scene radiance is applied to evaluate the noise level quantitatively.Figure 11a,b shows the LSD 3×3 histogram curves of original radiance (red curve) and corrected radiance (blue curve) from COCTS 11 and 12 µm channels on 12 May 2011, respectively.The LSD 3×3 value corresponding to the peak number reflects the noise level.For 11 µm channel original radiance and corrected radiance, the LSD 3×3 peak values are 0.18 mW m −2 cm sr −1 and 0.13 mW m −2 cm sr −1 respectively.For 12 µm channel, after correction, the LSD 3×3 peak value decrease from 0.41 mW m −2 cm sr −1 to 0.11 mW m −2 cm sr −1 .The histogram patterns and LSD 3×3 peak values indicate the noise COCTS 12 µm channel in the original data is stronger than 11 µm channel, which is consistent with the striped noise analysis results.After inter-calibration, the noise levels in both channels are decreased, and the improvement of 12 µm channel is greater than 11 µm channel, mainly due to the distinguishable difference between neighbour detectors of 12 µm channel.The radiance inter-calibration does not fully eliminate apparent striping, presumably because of residual error or temporal evolution of the calibration differing between detectors.Nonetheless, the improvements of COCTS radiance accuracy after inter-calibration are essential for improved SST retrieval from COCTS, which will be the topic of future work.

Conclusions
In this study, the COCTS radiances from thermal infrared channels were inter-calibrated with IASI in the northwest Pacific region during the period from 2009 to 2011.The comparison results indicate that the mean values of COCTS 11 and 12 µm channel original radiance minus IASI radiance difference are −6.35mW m −2 cm sr −1 and −7.56 mW m −2 cm sr −1 , with the corresponding standard deviations of 0.96 mW m −2 cm sr −1 and 0.63 mW m −2 cm sr −1 , respectively.The mean values of COCTS 11 and 12 µm channel original BT minus IASI BT difference are −4.08K and −4.75 K, with the corresponding standard deviations of 0.50 K and 0.39 K, respectively.In addition, the radiance difference between COCTS and IASI is radiance-dependent for both 11 and 12 µm channel with a stronger dependence in the 11 µm channel radiance difference with IASI radiance than 12 µm channel.The time series plot of radiance difference demonstrates that there is no obvious time-dependent variation except a slight hopping at the junction point between March and April 2011.
The inter-calibration coefficients are obtained using linear robust regression for COCTS individual detectors separately.After correction, the mean values of COCTS minus IASI radiance difference are −0.02mW m −2 cm sr −1 and −0.01 mW m −2 cm sr −1 , respectively, with the corresponding standard deviations of 0.51 mW m −2 cm sr −1 and 0.57 mW m −2 cm sr −1 .Both mean values of COCTS 11 and 12 µm channel corrected BTs minus IASI are −0.01K, with the corresponding standard deviations of 0.33 K and 0.35 K, respectively.Compared with COCTS original radiance statistic results, the biases are removed and the standard deviations are reduced, and the dependence of radiance difference on IASI radiance is disappeared after correction.The significant striped noise of COCTS original radiance is reduced, which is evident in imagery and distributions of local standard deviation.In conclusion, several strands of evidence indicate the calibration accuracy of COCTS is improved after the correction.In a future study, the COCTS corrected radiance will be applied to SST retrieval in the expectation that this will improve the COCTS SST accuracy.

Figure 2 .Figure 3 .
Figure 2. Illustration of the homogeneity estimation region.The smallest blue square represents one gridded COCTS pixel with the spatial resolution of 0.01°.The yellow square represents the COCTS pixels assumed to correspond to one IASI IFOV with spatial resolution of 0.12°.(Use of a square in the COCTS is an approximation that is only good if the area is adequately homogeneous.)The green region represents the surrounding area within 0.16°.

Figure 2 . 16 Figure 2 .Figure 3 .
Figure 2. Illustration of the homogeneity estimation region.The smallest blue square represents one gridded COCTS pixel with the spatial resolution of 0.01 • .The yellow square represents the COCTS pixels assumed to correspond to one IASI IFOV with spatial resolution of 0.12 • .(Use of a square in the COCTS is an approximation that is only good if the area is adequately homogeneous.)The green region represents the surrounding area within 0.16 • .

Figure 3 .of 15 3 .
Figure 3.The variations of COCTS 11 and 12 µm channel radiance minus IASI radiance difference against relative standard deviations of COCTS radiance in target regions and surrounding regions: (a) 11 µm channel in target region; (b) 11 µm channel in surrounding region; (c) 12 µm channel in target region; and (d) 12 µm channel in surrounding region.
shows the scatter plots of COCTS 11 and 12 μm channel radiance with IASI radiance, respectively.The background colour indicates the collocation numbers in every 0.2 mW m −2 cm sr −1 COCTS and IASI radiance bin.The overall statistics of COCTS with IASI radiance indicate large cold biases of −6.35 mW m −2 cm sr −1 and −7.56 mW m −2 cm sr −1 , with the corresponding standard deviations of 0.96 mW m −2 cm sr −1 and 0.63 mW m −2 cm sr −1 .Figure 4b,d indicates the variations of COCTS 11 and 12 μm channel radiance minus IASI radiance against IASI radiance, respectively.Again, the background colour indicates the number of collocations in every 0.2 mW m −2 cm sr −1 IASI radiance bin and in 0.02 mW m −2 cm sr −1 radiance difference bin.The purple solid line shows mean values of the radiance difference in IASI radiance bins of width 2 mW m −2 cm sr −1 , with the corresponding black vertical bars indicating two times the uncertainty in the mean (i.e., the standard deviation/√ ).For COCTS 11 and 12 μm channel, both radiance differences represent approximately linear dependence on radiance.The difference for COCTS 11 μm channel more obviously depends on scene radiance than for 12 μm channel.For 11 μm channel, the mean value of radiance difference is about −3.3 mW m −2 cm sr −1 when radiance is around 70 mW m −2 cm sr −1 , becoming around −7.4 mW m −2 cm sr −1 when the IASI radiance is about 110 mW m −2 cm sr −1 .The radiance difference of 12 μm channel has a less-strong dependence on IASI radiance, spanning roughly −7 mW m −2 cm sr −1 to −8 mW m −2 cm sr −1 over the matchup range.
represents the comparisons of COCTS radiance measured by four individual detectors with IASI radiance.Similar to the purple solid lines shown in Figure 4b,d, the solid lines in difference colours of Figure 5 are the variations of mean difference in IASI radiance bins of width 2 mW m −2 cm sr −1 , with the corresponding two times the uncertainty in the mean shown as the vertical bars in the same colour.Table 1 indicates the overall statistics of COCTS minus IASI radiance difference, including the comparison results of all detector measurements as well as measurements from four individual detectors.For 11 µm channel, the comparison results show smaller difference between adjacent detectors with the maximum difference of 0.22 mW m −2 cm sr −1 corresponding to Detector 3 and Detector 4. In addition, when the radiances larger than 95 mW m −2 cm sr −1 , the difference among Detector 1, Detector 2 and Detector 4 are very minor.However, for 12 µm channel, both the variation plots and the statistics results indicate the difference between adjacent detectors are larger than 11 µm channel in the whole radiance range, which is consistent with more obvious striped noise shown in COCTS 12 µm channel radiance image.Remote Sens. 2018, 10, x FOR PEER REVIEW 7 of 16

Figure 6
Figure 6 is the time series plot of COCTS minus IASI radiance difference from 2009 to 2011 (Figure 6a for COCTS 11 μm channel and Figure 6b for COCTS 12 μm channel).The lower blue points in each figure indicate the daily mean values of the COCTS minus IASI radiance difference and the

Figure 6
Figure 6 is the time series plot of COCTS minus IASI radiance difference from 2009 to 2011 (Figure 6a for COCTS 11 µm channel and Figure 6b for COCTS 12 µm channel).The lower blue points in each figure indicate the daily mean values of the COCTS minus IASI radiance difference and the

Figures 9
Figures 9 and 10 are regional radiance images of COCTS 11 and 12 μm channel on 12 May 2011, respectively, with Figures 9a and 10a being the original data and Figure 9b and Figure 10b the corrected data.The blue plots indicate radiance variations along track at the location of black lines

Figures 9
Figures 9 and 10 are regional radiance images of COCTS 11 and 12 µm channel on 12 May 2011, respectively, with Figures 9a and 10a being the original data and Figures 9b and 10b the corrected data.The blue plots indicate radiance variations along track at the location of black lines painted in each radiance image.The radiance images, as well as the significant fluctuations of original radiance variations, display the strong striping in COCTS radiance data.In addition, the larger difference between neighbour detectors of 12 µm channel indicates that the striping of 12 µm channel radiance is stronger than 11 µm channel.After the correction of COCTS original radiance using individual coefficients for four detector measurements, the striped noise shown in corrected COCTS radiance images is reduced, with smoother variations along track, demonstrating that the inter-calibration method in this study is useful in some extent for COCTS striped noise elimination.Moreover,

Figure 9 .
Figure 9. COCTS 11 μm channel original and corrected regional radiance of 200 × 200 pixels size on 12 May 2011: (a) original radiance; and (b) corrected radiance.The blue plots in right panels indicate COCTS 11 μm channel radiance variations along track at the location of black lines painted in left panels.

Figure 9 .
Figure 9. COCTS 11 µm channel original and corrected regional radiance of 200 × 200 pixels size on 12 May 2011: (a) original radiance; and (b) corrected radiance.The blue plots in right panels indicate COCTS 11 µm channel radiance variations along track at the location of black lines painted in left panels.

Figure 10 .Figure 10 .Figure 10 .Figure 11 .
Figure 10.COCTS 12 μm channel original and corrected regional radiance of 200 × 200 pixels size on 12 May 2011: (a) original radiance; and (b) corrected radiance.The blue plots in right panels indicate COCTS 12 μm channel radiance variations along track at the location of black lines painted in left panels.
In this study, the COCTS radiances from thermal infrared channels were inter-calibrated with IASI in the northwest Pacific region during the period from 2009 to 2011.The comparison results indicate that the mean values of COCTS 11 and 12 μm channel original radiance minus IASI radiance difference are −6.35mW m −2 cm sr −1 and −7.56 mW m −2 cm sr −1 , with the corresponding standard deviations of 0.96 mW m −2 cm sr −1 and 0.63 mW m −2 cm sr −1 , respectively.The mean values of COCTS

Table 1 .
The statistics of COCTS minus IASI radiance difference from 2009 to 2011.

Table 1 .
The statistics of COCTS minus IASI radiance difference from 2009 to 2011.

.Dev Mean/Std.Dev Mean/Std.Dev Mean/Std.Dev Mean/Std.Dev
Due to the calibration drifting at the junction point between March and April 2011, two periods during 2009 to 2011 are analysed separately.Period 1 is from January 2009 to March 2011 and Period 2 is from April 2011 to December 2011.Coefficients for COCTS radiance corrections were calculated separately in Period 1 and Period 2. In addition, considering the different comparison results for COCTS four detectors shown in Section 3, the regression coefficients for different COCTS detectors were obtained separately.Table2shows the coefficients for COCTS 11 and 12 µm channel radiance correction of different detectors in two periods separately.Then, the COCTS radiances measured by each detector were corrected using individual coefficients to reduce the striped noise due to nonuniformity of individual detectors.

Table 2 .
The coefficients for COCTS radiance correction.