A Combination of Feature Tracking and Pattern Matching with Optimal Parametrization for Sea Ice Drift Retrieval from SAR Data

• Published in: Remote Sensing
• Main Authors: Korosov, Anton, Rampal, Pierre
• Other Authors: Nansen Environmental and Remote Sensing Center Bergen (NERSC)
• Format: Article in Journal/Newspaper
• Language: English
• Published: HAL CCSD 2017
• Subjects: sea ice drift; feature tracking; pattern matching; Sentinel-1; SAR; [SDU]Sciences of the Universe [physics]; Sea ice
• Online Access: https://hal.univ-grenoble-alpes.fr/hal-03405346; https://hal.univ-grenoble-alpes.fr/hal-03405346/document; https://hal.univ-grenoble-alpes.fr/hal-03405346/file/Korosov2017aRemote_Sensing.pdf; https://doi.org/10.3390/rs9030258

International audience Sea ice drift strongly influences sea ice thickness distribution and indirectly controls air-sea ice-ocean interactions. Estimating sea ice drift over a large range of spatial and temporal scales is therefore needed to characterize the properties of sea ice dynamics and to better understand the ongoing changes of the climate in the polar regions. An efficient algorithm is developed for processing SAR data based on the combination of feature tracking (FT) and pattern matching (PM) techniques. The main advantage of the combination is that the FT rapidly provides the first guess estimate of ice drift in a few unevenly distributed keypoints, and PM accurately provides drift vectors on a regular or irregular grid. Thorough sensitivity analysis of the algorithm is performed, and optimal sets of parameters are suggested for retrieval of sea ice drift on various spatial and temporal scales. The algorithm has rather high accuracy (error is below 300 m) and high speed (the time for one image pair is 1 min), which opens new opportunities for studying sea ice kinematic processes. The ice drift can now be efficiently observed in the Lagrangian coordinate system on an irregular grid and, therefore, used for pointwise evaluation of the models running on unstructured meshes or for assimilation into Lagrangian models.