Geometric performance evaluation of the Geostationary Ocean Color Imager

The Geostationary Ocean Color Imager (GOCI) instrument acquires eight channels of multispectral images, which consist of 16 slots positioned in four lines and columns. GOCI Level 1B data, therefore, consist of a mosaic of 16 images, geometrically corrected with the Image Navigation and Registration Software Module (INRSM) system based on automatic point landmark matching for each slot and band. A study of the geometric performance characteristics of the Level 1B data was conducted over a period from August 2010 to September 2011 using residual data from Bands 7 and 8. To evaluate the geometric performance in detail, this paper examines the following four types of image navigation and registration errors: navigation performance, within-frame registration, frame-to-frame registration, and band-toband registration. In addition to the performance statistics based on mosaic images, we used a slot-based analysis method for the rainy season (here, June 2011) to understand the local distribution of the geometric performance. From the image-based results, the navigation and frame-to-frame accuracies were better than 1 pixel and the band-to-band registration accuracy was better than 0.4 pixels, while the within-frame registration accuracy was less than 1 pixel. However, for the band-to-band performance, the percentage of observations that fell within the specifications was slightly less than 99.7% for all 240 frames from June 2011. The within-frame performance was much lower than the other performance categories and the residual error for the east-west direction was higher than that for the north-south direction. The results from the slotbased performance evaluation suggest that abnormal errors (e.g. above 53 ??rad for navigation) occur in some slots, although the performance during an estimation period of 7 continuous days was within the desired criteria.