Objective   Spaceborne infrared hyperspectral sensors and multi-channel spectral sensors can continuously observe the earth for a long period of time, and have important applications in the fields of climate prediction, weather change, environmental monitoring, etc. The high-precision spectral calibration and radiation calibration of their observation data are crucial to the quantitative application of remote sensing. With the increase of operational time of satellite after being launched, the performance of the spaceborne sensors will change, which will lead to the deviation of observation data accuracy. Therefore, it is necessary to effectively improve the calibration accuracy and the data quality of the instrument through on-orbit inter-calibration. The samples of inter-calibration are generally collocated and filtered through the method of the on-orbit alternative calibration of the Global Space-based Inter-Calibration Sytem (GSICS), including spatial, temporal, observation geometry and spectral collocation through simultaneous nadir overpass (SNO) observations, and consequently achieve the goal of inter-calibration with the target sensor. The SNO observations can make two satellite sensors observe the earth from different heights at the similar time and place, which fully reduces the comparison uncertainty caused by different observation time and angle of satellites. This is a necessary prerequisite for the feasibility of inter-calibration, but these factors are also the main source of calibration uncertainty, and the uncertainty of collocating bias will have effects on the inter-calibration accuracy finally. Therefore, we analyze the uncertainty of the samples collocating processing in this paper, including spatial collocation, observation angle collocation and spectral response function collocation between sensors.

  Methods  We establish the sifting process of inter-observation sample pairs above uniform clear-sky background scenes (Fig.1) of the infrared hyperspectral atmospheric sounder HIRAS-II and the low-light medium-resolution spectral imager MERSI-LL onboard the same platform of the FY-3E of China Fengyun-3 series sun-synchronous orbit meteorological satellite. Collocating MERSI-LL pixels within HIRAS-II nadir instantaneous field of view (IFOV) based on line-of-sight (LOS) vectors, HIRAS-II projects the FOV footprint from the satellite to the earth's surface at a fixed solid angle, and all coordinates are converted into Earth Centered Earth Fixed (ECEF) coordinate system after calculation. All MERSI-LL pixels in the coverage area of HIRAS-II FOV footprint can be determined by calculating the line-of-sight vector (Fig.3). The uncertainty of the samples collocation introduced by spatial, observation geometry and spectral collocating bias is separately analyzed by simulating IFOV shift, observation zenith angle deviation and spectral response function change, respectively.

  Results and Discussions   The results of uncertainty analysis above each section of collocating process through cross observation of sensors on the same platform, radiation transmission model simulation and statistical analysis show that, in terms of spatial collocation, we evaluated the percentage deviation and standard deviation of radiance brightness temperature between the disturbed value and the standard value (Fig.5) by comparing the standard value of radiance brightness temperature in the target area with the disturbed value of radiance brightness temperature after simulating pixel offset, the spatial mis-collocation causes the changes of radiance brightness temperature above observed background scenes, the relative uncertainty is approximately 10% when the IFOV is shifted by half a pixel. In terms of geometric collocation, we evaluated the deviation and relative accuracy of the brightness temperature of the observed and simulated spectrum by comparing the brightness temperature sample of spectrum observed by HIRAS-II with the simulated spectral brightness temperature after changing the satellite zenith angle, it is found that the misalignment of observation geometry causes deviation of spectrum radiance brightness temperature, the uncertainty is less than 0.2% when the observed zenith angle is shifted by 20 degree (Fig.7). In terms of spectral collocation, the hyperspectral equivalent radiance can be obtained by simulating and calculating the HIRAS-II infrared hyperspectral radiance and channel spectral response function of MERSI-LL. The difference of the spectral response function causes bias of spectral equivalent radiance brightness temperature, the uncertainty of the absorption channel and window channle is approximately 2.5% and 0.4% respectively for expanding the response function, and the uncertainty is better than 0.3% overall for shrinking the response function, the uncertainty is relatively small for shifting response function, and it is better than 0.1% when shifting five times the wavelength interval (Fig.9).

  Conclusions   In this study, we analyzed the uncertainty and its influence introduced by observation collocation in terms of spatial, observation geometry and spectral collocation, which are aimed at the spaceborne infrared hyperspectral sensors and multi-channel spectral sensors before inter-calibration. We used the pixel matching method above observation field based on the line-of-sight vector to separately analyze the uncertainty introduced by spatial, observation geometry and spectral collocating bias. The spatial mis-collocation caused by IFOV shift leads to the change of observation background radiance, the relative uncertainty is approximately 25%-30% when the IFOV is shifted by a pixel. In order to reduce the uncertainty introduced by pixel offset, the offset distance should be limited to half of the spatial resolution of the nadir instantaneous field of view. The misalignment of observation geometry caused by observation zenith angle difference leads to the bias of observation background radiance, and the bias is more obvious in vapor channel, the deviation of observation zenith angle should be constrained within 10 degree or more less. The deviation of hyperspectral equivalent radiance caused by the difference of spectral response function has an impact on the calibration accuracy, the effective bandwidth change of spectral response function will cause greater uncertainty relative to the central wavelength shift of spectral response function. This study provides a reference for setting reasonable threshold in the condition of sifting collocated samples before inter-calibration, and also provides support for improving accuracy of inter-comparison and calibration.