Abstract
The Moon, due to its extremely stable surface, has been a great interest to the satellite instrument calibration community for instrument in-orbit calibration, sensor-to-sensor inter-calibration, and historical data re-analyses. Yet using the lunar surface as a solar diffuser is always challenged with its non-uniform and non-Lambertian reflectance. To minimize the reflectance variation caused by the phase angle and libration variations, NOAA, in collaboration with Japan Meteorological Agency (JMA), are developing a lunar radiance calibration model over a set of selected region of interest (ROI) which are identified with both lunar hyperspectral and broad-band measurements, and lunar laser altimeter data as well. Yet accurate image-to-image registration to extract accurate ROI radiance is critical for the model development, especially for the images of relatively low spatial resolution bands. Two automatic image registration methods have been developed at NOAA over the past few years – a theoretic method to predict the ROI positions and an empirical algorithm based on the matched local features. Although the theoretic algorithm can well predict the central Moon position, the accuracy of ROI locations is often affected by fluctuations of instrument performance. The current empirical algorithm can successfully achieve the registration accuracy at sub-pixel level for the lunar images within certain phase angle difference, but the registration error increases when the images have large phase angle differences, due to the misclassification of potentially matched features over the global images. In this study, we are proposing to combine the theoretical and empirical methods to increase the number of matched features by restricting the searching area. The resulted lunar ROI radiance will be used to analyze the GOES-16 ABI and Himawari-8 AHI instrument degradations and compared with the solar calibrated data.
Improvement in ABI/AHI Lunar Image Registration Algorithm for the Extraction of Lunar ROI Radiance
The Moon, due to its extremely stable surface, has been a great interest to the satellite instrument calibration community for instrument in-orbit calibration, sensor-to-sensor inter-calibration, and historical data re-analyses. Yet using the lunar surface as a solar diffuser is always challenged with its non-uniform and non-Lambertian reflectance. To minimize the reflectance variation caused by the phase angle and libration variations, NOAA, in collaboration with Japan Meteorological Agency (JMA), are developing a lunar radiance calibration model over a set of selected region of interest (ROI) which are identified with both lunar hyperspectral and broad-band measurements, and lunar laser altimeter data as well. Yet accurate image-to-image registration to extract accurate ROI radiance is critical for the model development, especially for the images of relatively low spatial resolution bands. Two automatic image registration methods have been developed at NOAA over the past few years – a theoretic method to predict the ROI positions and an empirical algorithm based on the matched local features. Although the theoretic algorithm can well predict the central Moon position, the accuracy of ROI locations is often affected by fluctuations of instrument performance. The current empirical algorithm can successfully achieve the registration accuracy at sub-pixel level for the lunar images within certain phase angle difference, but the registration error increases when the images have large phase angle differences, due to the misclassification of potentially matched features over the global images. In this study, we are proposing to combine the theoretical and empirical methods to increase the number of matched features by restricting the searching area. The resulted lunar ROI radiance will be used to analyze the GOES-16 ABI and Himawari-8 AHI instrument degradations and compared with the solar calibrated data.