A New Potential of the Deep Convective Clouds as the Calibration Target for a Geostationary UV/VIS Hyperspectral Spectrometer

As one of GEO-constellation for environmental monitoring in the next decade, Geostationary Environment Monitoring Spectrometer (GEMS) is designed to observe the Asia Pacific region to provide the information on the atmospheric chemicals, aerosol and cloud properties. For the continuous monitoring of the sensor performance after its launch in early 2020, here we suggest deep convective clouds (DCCs) as a possible target for the vicarious calibration of GEMS, the first UV/VIS hyperspectral sensor onboard a geostationary satellite. Tropospheric Monitoring Instrument (TROPOMI) and Ozone Monitoring Instrument (OMI) are used as a proxy of GEMS, and a conventional DCC detection approach applying the thermal threshold test is used for the DCC detection based on the collocations with Advance Himawari-8 Imager (AHI) onboard Himawari-8 geostationary satellite. DCCs are frequently detected over the GEMS observation area on average over 200 pixels in a single observation scene. Considering the spatial resolution of GEMS, 3.5 km×7 km which is similar to TROPOMI, and its temporal resolution (8 times a day), availability of DCCs for vicarious calibration of GEMS is expected to be sufficient. Inspection of the DCC reflectivity spectra estimated from the OMI and TROPOMI data also shows a promising result. Even though, their observation geometry and sensor characteristics are quite a different, the estimated DCC spectra agree quite a well within a known uncertainty range with comparable spectral features. When the DCC detection is further improved by applying both visible and infrared tests, the variability of DCC reflectivity from the TROPOMI data is reduced by half, from 10% to 5%. This is mainly due to the efficient screening of cold thin cirrus with the visible test and of bright warm clouds with the infrared test. The precise DCC detection is also expected to contribute to the accurate characterization of the cloud reflectivity, which will be further investigated.


Introduction
With the global transport of anthropogenic chemicals in the atmosphere becoming a controversial issue over recent years, satellite has been considered as one of key measures to keep track of the chemicals with its large spatial coverage. In the Asia-Pacific region, Geostationary Korea Multi-Purpose Satellite-2B (GEO-KOMPSAT-2B, GK-2B) is expected to perform such a role after its launch planned in February 2020 with an ultraviolet (UV) and visible (VIS) hyperspectral sensor, Geostationary Environment Monitoring Spectrometer (GEMS). GEMS is designed to observe the Asia Pacific region including the Korea Peninsula and its surrounding area to continuously monitor the atmospheric conditions by retrieving concentration of atmospheric chemicals and aerosol properties [1,2]. For the consistency of the measurements, on-board calibration with solar diffusers and lightemitting diode (LED) are deployed in the GEMS calibration system which converts the light from the

GEMS
GEMS covers the Asia-Pacific region (5°S-45°N and 75°E-145°E) observing the Earth in the eastwest direction with the fixed north-south field of view (FOV) of 7.73° [2]. For the retrieval of the concentrations of atmospheric gases (O3, NO2, SO2 and HCHO) and aerosol properties, GEMS is designed to provide continuous spectrum from 300 to 500 nm with spectral resolution of better than 0.6 nm at every 0.2 nm. As the first hyperspectral UV-VIS sensor onboard GEO satellite, GEMS is expected to provide critical information to monitor the regional transport of atmospheric chemicals at an hourly temporal resolution during the daytime as a part of GEO-constellation [35].
Prior to the launch of the satellite, on-ground sensor characterization and calibration have been conducted in the preparatory phase for GEMS. While in orbit, GEMS relies on the on-board calibration system consisting of solar diffuser and LED to evaluate and maintain calibration quality. As a part of the on-board calibration system, LED serves as a stable light source to check the nonlinearity of electronic response and the aliveness of each pixels at the detector-level. Solar measurements are also designed to monitor and calibrate sensor response changes with two transmissive diffusers, a working and reference diffuser. The working diffuser is planned to observe the Sun on a daily basis, yet it is known to be gradually degraded during the entire mission because of the daily exposure to the Sun. Thus, a reference diffuser identical with the former but observing the Sun once every six months is included together in the calibration system. However, since most parts of the sensor are expected to be degraded as time passes, it is important to decompose the degradation from each part of the sensor and properly calibrate the changes. Since the on-board calibration is included in the calibration system, an independent way to evaluate the overall performance of the calibration system could be useful to maintain the data quality for GEMS in the long-term operation as a back-up calibration plan.

OMI and TROPOMI
OMI and TROPOMI are both hyperspectral sensors encompassing the spectral range as well as the observation area of GEMS. Operating in a sun-synchronous polar orbit, both sensors take radiance measurement in the ascending node of the satellites at around the local solar time (LST) of 13:30. The top-level specification of GEMS, OMI and TROPOMI is summarized in Table 1. As newly launched in October 2017, TROPOMI especially has stricter data quality requirements compared to other sensors. Since the spatial and spectral resolution of GEMS is also quite similar to the resolution of TROPOMI, these sensors are highly expected to be reciprocal candidates for the inter-calibration after the operation of GEMS.

DCC Climatology
To check whether there are enough DCCs available within the GEMS field of regards (FOR) especially over the TWP region, we apply the conventional DCC detection approach, threshold tests for TB and uniformity of the clouds [26]. Threshold values used for each test along with the constraints of observation angles and spatial coverage are summarized in Table 2. For the TB test, we use the 11 µm window channel with the threshold of 205 K which is set considering the trade-off between the precision of the DCC detection and the sufficiency of the DCC data as presented by previous studies [27,28]. Also for the uniformity test, a relaxed threshold value (from 1 K to 2 K) is used to account for the reduced spatial resolution of GEMS. The relaxation of the threshold could make the available data much larger as relatively less affecting to the effectiveness of the DCC detection [28]. The maximum solar and viewing zenith angle are also limited to 40° as the DCC reflectivity significantly changes when the solar and viewing angles are too large [21]. As an input for the DCC detection, AHI measurements are used since the imager on-board GEO satellite provides VIS (R0.47) and IR (TB10.4) channels with higher temporal resolution while fully covering the TWP region with its full-disk observation (See Table. 3). Since R0.47 has higher spatial resolution compared to TB10.4, spatially averaged R0.47 is prepared. To test the availability of DCCs under the GEMS observation condition, the spatial resolution of GEMS is simulated with each 4 × 4 pixels of the VIS and IR channels and the mean of TB10.4 and standard deviation of R0.47 and TB10.4 are used for the DCC detection.

Frequency Distribution
A-year long DCC climatology data of AHI from July 2016 to June 2017 are constructed with the spatial grid and sampling frequency matched to the GEMS. In the Asia-Pacific region, most of DCCs are observed at near the tropics and distributed quite evenly over the GEMS observation area, similar to the previous analysis [36][37][38]. In Figure 1, the spatial distribution also shows a unique arc-shaped boundary which is attributed to the limitation imposed for current study (the viewing zenith angle (θ) should be smaller than 40°). Considering that the viewing zenith angle is a fixed value over time for each pixel of a GEO sensor, the spatial condition might be limited to the pixels satisfying the angle condition. Figure 2 shows the temporal variation of the number of detected DCCs. In a specific day, most of DCCs are detected at noon time (02:00-04:00 UTC) when the Sun directly passes over the target area. This might be attributed to the constraint of solar zenith angle for the DCC detection as the solar zenith angle is also limited in the same way as the viewing zenith angle. The constraint might also affect to the seasonal variation along with the seasonal deep convection in the Northern Hemisphere. As shown in Figure 2-(b), DCCs mostly occur from late summer to early autumn over the TWP region since the atmospheric convection is strongly dependent on the abundant moisture and the latent heat accumulated during summertime [37,[39][40][41]. Even with the limitation imposed by the viewing angular geometry and the seasonality, the average number of DCCs in a single observation scene is still larger than 200 pixels even in the month presenting the minimum frequency. Since GEMS observes the Earth eight times a day, at least over 50,000 DCC pixels could be detected in a month from GEMS when using the conventional DCC detection approach with the collocated AHI data. The number could increase if we use Advanced Meteorological Imager (AMI) onboard GEO-KOMPSAT-2A which is stationed over 128.2°E with its coverage expanding further the western part of AHI.
In section 3, DCC reflectivity spectra obtained from OMI and TROPOMI are compared to confirm whether the DCC measurements show similar spectral features and stable signal enough to be compared for the different sensor characteristics and optical paths. In the UV-VIS spectral region, the reflected radiation from ice clouds is largely affected by the angle condition [42] and this makes it important to finely detect DCCs for the accurate characterization of the cloud reflectivity.

Collocation Process
Since OMI and TROPOMI only cover the UV-VIS and UV-SWIR spectral region, respectively, the DCC detection for both sensors could be performed with the collocated VIS and IR channels of AHI. For the collocation between GEO and LEO sensors, we apply the collocation criteria suggested by the Global Space-based Inter-Calibration System (GSICS) community [43]. Since the collocated data between LEO and GEO are not directly compared in this study, viewing angle is not matched between sensors. As shown in Figure 3, the hyperspectral data satisfying the spatial and angle condition (see Table 2) are collected first and then the AHI VIS and IR channels matching temporal collocation criteria are called. With the collected data, the AHI pixels observing nearly at the same time (Δt < 5 minutes) with each pixel of OMI and TROPOMI are collocated when the pixels simultaneously satisfy the spatial threshold (located within a half of shorter FOV of a LEO sensor). With the collocated AHI pixels, averaged TB10.4 and standard deviation of TB10.4 and R0.47 are calculated for the DCC detection.

Apparent Reflectivity of DCCs
GEMS, OMI and TROPOMI provide the spectral radiance data to be used as an input for the retrieval of the geophysical information of the atmosphere. Since the uncertainty in the measured radiance due to the optical path of the instrument could be mitigated by using irradiance which has the same optical depth [44], here we use reflectivity for the spectral analysis. As OMI and TROPOMI provide solar observation once a day, the timely matched irradiance with the radiance is used for the calculation of the reflectivity. The radiance measured over the DCCs could be written as (by neglecting backscattered radiation by the atmosphere above DCCs); Where ( 0 , , ) is the measured upwelling radiance at wavelength λ with the solar zenith angle 0 , satellite zenith angle , and relative azimuth angle . The measured radiance is strongly affected by the cloud reflectivity ( 0 , , ) and incoming solar irradiance at the top of atmosphere (TOA) which is attenuated by atmospheric extinction. The atmospheric optical depth ( ) from the cloud top altitude to TOA is determined by both absorption and scattering. Here we consider only Rayleigh scattering to make the problem simpler. The angle component is cosine of the zenith angle. Thus, the cloud reflectivity using the measured radiance and irradiance can be given as; Here, the optical depth ( ) is estimated using an approximation as suggested by Bodhaine et al. [45] considering the altitude and Rayleigh scattering atmosphere. Cloud altitude is approximately set to 16 km since the cloud top of the DCCs nearly reaches to the tropopause at the equatorial region [31,46]. This makes the optical depth of the atmosphere above the clouds in a range of 0.0005-0.0025 from 300 to 500 nm. Since Mie scattering and atmospheric absorption in the upper troposphere are not included in the calculation, the reflectivity is the apparent reflectivity of DCCs, even though it is simply called as the DCC reflectivity in this study. Figure 4 shows the DCC pixels identified by OMI and TROPOMI data for a particular cloudy scene (July 3, 2018 06:10 UTC and July 3, 2018 06:40 UTC, respectively). TROPOMI observes the Earth at about 30 minutes earlier than OMI and thus the cloud distributions are slightly different. As shown in the figure, the number of DCC pixels obtained from OMI-AHI collocations is quite smaller than the TROPOMI-AHI collocations. One of the main reasons for the difference is attributed to the lower spatial resolution of OMI. As the size of one OMI pixel is about 15 times larger than that of the TROPOMI, many of the small scale DCCs which are detected as DCCs by TROPOMI data are missed by the OMI because of both threshold test and uniformity test. Because the lower spatial resolution of OMI with the small scale DCC increases not only brightness temperature but also the spatial variability in the IR and VIS making such a pixel as a non-DCC cloud.  The data quality issues arisen during the long-term operation of OMI also affect to the availability of the OMI observations. For instance, the row anomaly (RA) effect [32] makes nearly a quarter of OMI pixels (especially close to nadir observations) unavailable for the analysis. Figure 5 shows the measured reflectivity of OMI as a function of the position (Row) of the detector (Charge Coupled Device, CCD) and the reflectivity spectrum affected by the RA effect. The measurements in the row number 24-41 contaminated by the RA effect are primarily eliminated during the data processing. However, as shown in the figure 5 (a), the reflectivity at the row numbers close to the nadir port also shows significantly lower reflectivity even though the rows are not flagged as the RA affected pixels. When these observations are detected together as the DCCs, the reflectivity spectrum shows significantly lower value compared to the DCC reflectivity of TROPOMI. Thus, these pixels near the nadir port are also eliminated in this study and it makes the number of DCC pixels of OMI largely reduced. Since the rows close to the nadir port generally have low viewing zenith angle which satisfies the angle condition for the DCC detection, the RA effect significantly affects to the availability of DCC observations of OMI.

DCC Reflectivity Spectrum
The DCC spectra of OMI and TROPOMI observed in a year with ten days interval are presented in Figure 6 showing the mean and standard deviation of radiance and reflectivity at each wavelength. Solar measurements observed on the same day and the scan angle position (the position on the detector) of each DCC measurement are also shown together. Because of the data quality issues as previously mentioned, the number of DCCs for OMI observations in a year is only 3% of the number of TROPOMI. However, even with the distinct difference in the number of measurements, the mean reflectivity of OMI and TROPOMI are almost identical with the higher reflectivity of about 0.9 over the whole wavelength. The DCC spectra also clearly show nearly invariant spectral features except for the both ends of the wavelength range which are attributed to the ozone absorption (300-345 nm) and the pixel saturation of TROPOMI (450-500 nm) [47]. The results indicate that the DCCs observed by the satellite sensor reflect almost stable signal even with the different sensor characteristics, the number of the measurements as well as the observation angle geometry. However, some of the differences are also observed between the spectra of OMI and TROPOMI. At shorter wavelength, TROPOMI reflectivity is slightly higher than OMI and as the wavelength increases, the difference becomes much smaller. This might be caused by the diffuser degradation in the solar measurements of TROPOMI since the degradation more significantly occurs at shorter wavelengths. The degradation is to be addressed in the future updates of TROPOMI L1B data in early 2020 as notified by the S5P validation report [47]. TROPOMI DCCs also have smaller spectral noise than OMI since the solar measurements of OMI show higher spectral noise over the whole wavelength as shown in Figure 6 (a). There are also sudden peaks at 393 and 397 nm each corresponding to the Ca II K and Ca II H of Fraunhofer lines, which are caused by the beam-filling effect from the atmosphere above the clouds. As the rotational Raman scattering occurs in the atmosphere, the scattered radiation is also added to the upwelling radiation from the clouds [48]. However, OMI reflectivity shows negative peaks which are seemingly unrealistic considering that the beam filling effect only prominently occurs at the radiance. The unrealistic features may be caused by the unavailable data at particular wavelengths in the OMI irradiance. For the calculation of the reflectivity, the sparse data are approximated by the interpolation which might not properly reproduce the spectral features especially at the deeper peaks.
These results show that TROPOMI still needs further minor updates but the DCC can be a promising target considering the well-matched spectral features with the theoretical background and smaller spectral noises. The abundance of the data also can be an advantage using the measurements for the further study. However, even with the well-explained spectral features, the DCC measurements still have larger systematic difference among the measurements of nearly up to 10% and 12% for OMI and TROPOMI, respectively over the whole wavelength. Since the systematic difference shows increasing trend as the number of the measurements increases, TROPOMI has higher systematic difference compared the OMI. This indicates that as the observation period becomes longer, the dispersion among the DCCs also could significantly increase. The difference might be too large to regard the DCC detection as properly done and this also can make the characterization of the cloud reflectivity quite complicated without knowing the reason of the dispersion. Thus, in Section 4, the thresholds of the conventional DCC detection are tested to reduce the systematic difference among the DCC measurements and improve to finely detect DCC pixels. Since OMI has some of data quality issues, we only use the TROPOMI and AHI observations for the analysis.  Figure 7 shows horizontal distribution of DCCs found over the typhoon Chaba in October 2016. For a one to one comparison, AHI R0.47 image is binned to match with the spatial resolution of the TB10.4. As shown in Figure 7 (a), DCCs identified as blue dots are mainly found over the typhoon center having cold TB10.4 with quite a symmetric distribution around the center. However, Figure 7 (b) presenting detected DCDs over the R0.47 image shows an interesting result, as the blue dots over the right side of the typhoon center having lower R0.47 of about 0.7. These are the thin cirrus clouds which are spread out from the typhoon center following strong upper air outflows. As these cirrus clouds have colder cloud tops composed of the ice particles, the clouds are detected as DCCs by the conventional detection method, even though the reflectivity is much lower than the actual DCC clouds.  These results can be interpreted that the TB10.4 threshold might be less effective as a DCC detection threshold especially for the UV-VIS measurements. With only the conventional DCC detection thresholds, it is highly possible that the darker cirrus clouds might be also detected as DCCs. One of few attempts using the DCCs for the monitoring of the UV-VIS hyperspectral sensor also uses only the UV reflectivity threshold for the DCC detection [32]. In the study, the OMI pixels having higher reflectivity at 354 nm (R0.354 > 0.9) are regarded as DCCs and then used for the monitoring of the temporal stability of radiometric calibration of OMI. Even with such a simple detection, the averaged cloud reflectivity shows a fairly constant value regardless of wavelength (which is the characteristics of DCC reflectivity). Thus, they use the DCC reflectivity for a long term monitoring of the spectral dependence in the sensor performance. However, it was also clear that the DCC reflectivity still showed a seasonal and inter-annual variations which were attributed to the different cloudiness, angle dependence and residual atmospheric effects (refer Figure 32 of [32]). Although it is not easy to quantify, it is highly possible to increase such attributions when the detected DCCs are bright but low lying warm clouds. For example, the optical path for the warm clouds is much longer than the DCCs, causing increased variability of the measured reflectivity due to the increased contribution from the tropospheric air. With the same token, angular variation of the measured reflectivity also increases with increasing optical depth.

Comparison of the VIS and IR Radiation
For a further clarification of the issues with the warm clouds, Figure 9 and 10 show the spatial distribution of TB10.4 and R0.47 of the warm clouds and the spectral reflectivity of the clouds having different TB10.4 values, respectively. This demonstrates the importance of the IR threshold method for a proper detection of the DCCs especially to select the high altitude clouds which provide homogeneous spectral features with the minimum influence from the troposphere. The blue dots on Figure 9 show the DCC pixels having high reflectivity (TROPOMI R0.354 > 0.9) with warm IR temperature (AHI TB10.4 > 260 K). In this case, most of the blue dots are located over the cloud edges having bright reflectivity, although their temperatures are much warmer than the nearby convection core. Thus, if we use the UV-VIS radiation threshold only, it would be difficult to screen out the bright but warm clouds which are close to the cloud cores.  Figure 10 shows the reflectivity spectrum of the bright DCCs having different brightness temperatures. Blue line in the Figure 10 (a) shows the averaged reflectivity spectrum of the blue dots shown in Figure 9, while the black line is that of the clouds satisfying the conventional DCC detection thresholds. The reflectivity spectrum (blue line) including the bright but warm clouds clearly shows smaller reflectivity compared to the bright and cold clouds. The different spectral features and increased atmospheric scattering caused by the radiative interactions with the tropospheric atmosphere and the clouds could introduce a significant reduction in the measured reflectivity. The tropospheric effects in the measured reflectivity is further depicted in Figure 10 (b) as the beam filling effects increases the cloud reflectivity with the increased rotational Raman scattering from the tropospheric atmosphere [48]. Thus, Figure 10 shows that the clouds having warmer TB10.4 give a lower reflectivity as well as the stronger beam filling effects especially at the Fraunhofer lines compared to the cold clouds. This makes it clear that using both VIS and IR information together could effectively screen out the cirrus and cloud edges as well as detect only colder cloud tops for a better utilization of the DCC reflectivity.

DCC Detection with the Additional VIS Reflectivity
Based on the previous analysis, we develop an updated DCC detection approach utilizing both reflectivity and brightness temperature. For the inclusion of the reflectivity test, it is important to set a proper threshold value; a tighter threshold for the reflectivity such as 0.9 could produce stable statistics while with the reduced availability of the data, while a relaxed threshold such as 0.6 could increase the data points but increasing variability. This indicates that the optimal threshold of reflectivity for the DCC detection needs to be set weighing both sides, the data availability as well as the stability of the reflectivity distribution. Here, we choose an optimal value by analyzing the variation of statistical parameters as a function of the different threshold values. For the test, AHI R0.47 data are used as the reflectivity value since it is independent to the TROPOMI observations.
First of all, Figure 11 (a) and (b) show the frequency distribution of TROPOMI R0.354 to the AHI R0.47 threshold and the uniformity threshold of AHI R0.47 for the DCC detection, respectively. Utilization of R0.354 is based on the previous use for OMI study [32] along with the fact that it is less affected by the ozone absorption as well as the directly transmitted light which increases the angle dependence [33]. As shown in Figure 11, application of the AHI R0.47 test reduces the speediness of the TROPOMI R0.354 distribution and generates a distribution closely following a normal distribution. This is achieved mainly by reducing data located in left tails, the lower reflectivity data. However, some of the left tails still remains since there are residual variations (due to the atmospheric effect above the cloud tops) along with the collocation uncertainty between the AHI and TROPOMI measurements. Since the DCC detection is based on the collocated data between sensors, the possibility for the mis-match between the sensors might increase when the spatial uniformity of a certain cloud is quite lower. Figure 11 (b) also indicates that the cloud pixels having relatively higher spatial inhomogeneity take a large portion in the center of the distribution. This means that the cloud cores vertically shooting up nearly up to the tropopause might have the relatively lower spatial uniformity, which could not be eliminated by the reflectivity threshold. In other words, the TROPOMI R0.354 distribution might be closer to a normal as thoroughly detecting only DCCs with a stricter spatial uniformity test as considering the trade-off between the number of data and the stability of the distribution.  Table 4 shows the statistics of the TROPOMI DCC R0.354 by applying different AHI R0.47 thresholds to find the effective value which makes the distribution fairly normal without eliminating too many observations. TROPOMI R0.354 is also applied together as the detection threshold to reduce the collocation uncertainty by limiting the tail of the distribution to a minimum (TROPOMI R0.354 > 0.7). Results show that as the AHI R0.47 threshold increases, the distribution becomes very close to a normal even though the number of DCCs detected by the thresholds decreases exponentially. Even with the dramatic change of the number of data, the mode of distribution is relatively stable compared to other indicators since the indicator is not significantly affected by the values far away from the center, such as the left tails in this case. Standard deviation of the reflectivity decreases linearly along with the increasing VIS threshold when the kurtosis increases exponentially. Interestingly, only skewness converges at a particular AHI R0.47 threshold of 0.64 as having lowest skewness. Since the TROPOMI R0.354 reflectivity is skewed to the left because of the darker cirrus having lower reflectivity, skewness of the distribution has a negative value regardless of the AHI R0.47 thresholds.   In summary, since the AHI R0.47 threshold for the DCC detection largely affects to the left side of the reflectivity distribution of R0.354, the skewness might increase with the brighter AHI R0.47 threshold until the number of DCCs is decreased by a significant portion. When it comes to the spatial inhomogeneity, some of DCC pixels having relatively low uniformity are eliminated with the stricter uniformity test over the whole range of the TROPOMI R0.354, but mostly at the center of the distribution. Since the AHI R0.47 threshold as well as the uniformity test might simultaneously affect to the statistics of the reflectivity distribution of TROPOMI R0.354, the optimal threshold value for the DCC detection needs to be set considering both effects. Figure 12

Verification of the Updated Method
The effects of the updated DCC detection method to the DCC reflectivity spectra would be different for different test. In case of the thermal threshold test, it would be most effective for the screening of the low altitude clouds. Thus, it would decrease the variation of reflectivity by reducing the variations caused by the atmospheric and cloud effects, in this case the warm clouds which have longer optical path lengths. VIS reflectivity also can be a useful indicator to detect only the optically thick clouds which are bright enough to reflect most of the incoming radiation. By using both radiative properties, DCC detection can be further improved to detect only optically thick and very bright cloud targets which show homogeneous spectral features as well as the higher reflectivity with the smaller variation. Figure 13 verifies the advantages applying the updated method for the DCC detection with the threshold values as suggested in the previous section. To clearly show the spectral features of the DCCs depending on the different DCC detection methods, reflectivity spectra at the Fraunhofer line (393 nm, the Ca II K line) are compared together. Figure 13 (a) shows the anomaly spectra of TROPOMI DCCs for the comparison of the beam filling effects. As clear shown, the DCCs detected by the updated DCC detection method present relatively less pointing peak at the Fraunhofer line compared to the DCCs detected with only the reflectivity condition only (TROPOMI R0.354). This indicates that the atmosphere above the clouds might be much thinner when the DCCs are detected by the thermal radiation threshold. Figure 13 (b) also presents the comparisons of DCCs detected by the conventional DCC detection method as well as the updated DCC detection. The clouds detected by the updated DCC detection method show relatively higher reflectivity as having lower variability with the standard deviation of about 6%.

Conclusions
As the first UV-VIS hyperspectral sensor onboard GEO satellite, the field of regards of GEMS covers Asia Pacific region including the tropical western pacific (TWP) region. To devise a vicarious calibration approach with the given availability of the calibration targets, in this study, DCCs are tested whether the optically thick clouds provides a stable and bright signal enough to characterize the radiometric calibration of sensors having a different hardware characteristics as well as the observation conditions such as viewing angles, especially in the UV-VIS spectral region. For a feasibility test, VIS and IR channels of advanced himawari imager (AHI) onboard Himawari-8 satellite are used along with the UV-VIS hyperspectral data of OMI and TROPOMI as a surrogate of GEMS. To mitigate the calibration uncertainty caused by degradation and low frequency perturbation of the instrument optical path, the reflectivity (ratio between radiance and irradiance) is used. For a further reduction of uncertainty caused by the observation geometry determined by the satellite and solar zenith angle with the atmospheric effects, the cloud reflectivity derived by considering solar zenith angle, satellite zenith angle, and Rayleigh scattering above the clouds are used for the analysis.
To make sure that the number of DCCs over the GEMS observation area, a-year long AHI data matching the spatial and temporal resolution of GEMS has been analyzed. The detected DCCs using the conventional approach relying on the conventional method, i.e. thermal temperature test and uniformity tests show that DCCs occur with clear seasonality, showing a maximum in September and a minimum in April. It also shows an effect of solar zenith angle during a specific day due to the application of limiting condition in the solar zenith angle to reduce the longer optical path length. Spatially, the viewing zenith angle also limits the availability of DCCs since AHI observes the target area with higher viewing zenith angle compared to GEMS. This limitation with the satellite zenith angle is expected to be improved with the AMI (Advanced Meteorological Imager) onboard GEO-KOMPSAT-2A (GK-2A) which is collocated with GEMS at 128.2° E. Even with such limitations, occurrence of DCCs in a single observation scene shows more than 200 pixels in average which appears to be enough for the statistical approach considering the observation frequency and spatial resolution of GEMS.
Although the number of DCCs detected for OMI and TROPOMI shows a significant difference, mainly due to the poor spatial resolution and degraded data quality of OMI data, the comparison between the estimated spectral reflectivity of DCCs shows comparable results with high reflectivity and weak spectral dependence of reflectivity, even with the disparate sensor characteristics, viewing geometry, and number of data. However, a closer inspection of the reflectivity spectra shows that there is a fairly large variability with the standard deviation of nearly up to 10% which is mainly due to the false DCC detection of thin cirrus which has cold cloud temperature with small optical depth. Furthermore, inspection of an alternative approach, using only reflectivity test for the DCC detection, shows a false detection of warm clouds which have a large reflectivity and lower cloud top altitude. Thus, we devise an updated DCC detection approach using both the thermal and reflectivity tests to screen out cold thin cirrus and bright warm clouds. Based on the variation of statistical parameters of the DCC reflectivity with the different reflectivity threshold value, the threshold value for the reflectivity test is determined to be 0.7 which makes the distribution close to normal and the location values of the distribution converged each other retaining as many observations as possible. However, there are still remaining issues causing the speediness in the reflectivity such as the variations in the cloud properties as well as the angle dependence such as the bidirectional reflectivity distribution of DCCs. The long term variability of DCC reflectivity with the updated detection method are going to be analyzed which could be used to minimize the such variation and to demonstrate applicability of new approach for hyperspectral UV-VIS sensors.