A Rapid and Automatic Orthorectification of MethaneAIR/MethaneSAT Imagery thorough the A-KAZE Algorithm
Amir Souri, Harvard-Smithsonian Center for Astrophysics, ahsouri@cfa.harvard.edu (Presenter)
Joshua Benmergui, Harvard University, benmergui@g.harvard.edu
Eamon Conway, Harvard-Smithsonian Center for Astrophysics, eamon.conway@cfa.harvard.edu
Jenna Samra, Harvard-Smithsonian Center for Astrophysics, jsamra@cfa.harvard.edu
Jonathan Franklin, Harvard University, jfranklin@g.harvard.edu
Xiong Liu, Harvard-Smithsonian Center for Astrophysics, xliu@cfa.harvard.edu
Kelly Chance, Harvard-Smithsonian Center for Astrophysics, kchance@cfa.harvard.edu
Steven Wofsy, Harvard University, swofsy@seas.harvard.edu
Accurate and precise geolocation is required for XCH4 retrievals from MethaneAIR data. MethaneAIR is an airborne instrument precursor to MethaneSAT. MethaneSAT is a satellite under development by MethaneSAT, LLC, an affiliate of the Environmental Defense Fund. XCH4 retrievals are sensitive to geolocation errors for 3 reasons: 1) because the instrument is composed of 2 sensors (an O2 and a CH4 camera), whose data must be properly aligned, 2) because they integrate information about the viewing geometry of the instrument, and 3) because they integrate information about surface properties that vary widely over small spatial scales. Geolocation is achieved through processes called “orthorectification”, “registration”, and “bundle adjustment”. Orthorectification uses a model of the instrument camera properties (the intrinsic model), and its position and attitude (the extrinsic model) to model pixel boresights and calculate their intersection with a digital elevation map. The extrinsic model is derived from aircraft avionics data. Registration is the process of aligning the initial orthorectification to an accurate reference dataset. We used a multi-scale feature extraction method called Accelerated KAZE (A-KAZE) to extract distinctive features from MethaneAIR O2 and CH4 greyscale images and Landsat Band 6 imagery. We matched features using the Hamming distance of the properties of the images at and surrounding each keypoint, and removed outliers. We then applied a linear transformation to register the MethaneAIR images. We applied bundle adjustment to optimize the extrinsic model to the registered imagery, holding the intrinsic model and digital elevation map fixed. Results show significant improvements in the geolocation accuracy for the majority of scenes (>90%). We also demonstrate the practicality of the proposed method for geolocation of data from the upcoming MethaneSAT satellite using MSI band 11 with unprecedented spatial (20 m) and temporal resolutions (5 days). The algorithm performs reasonably well under complex situations including cloudy areas (up to 30% cloud cover) and low signals (signal-to-noise > 12), which we attribute to the wide swath of MethaneSAT and the proficiency of the A-KAZE algorithm.
Poster: Poster_Souri__105_25.pdf
Presentation Type: Poster
Session: 1.2a Results from current missions
Session Date: Monday (6/14) 9:45 AM