Abstract
This paper presents a method to estimate shape parameters and track the motion velocity of pack ice using FY-3/MERSI satellite imagery. Focusing on the Fram Strait, the study employs sea ice recognition, pack ice extraction, and parameter estimation to generate high-resolution motion vectors. Results indicate that pack ice concentration decreases faster than overall sea ice concentration, while geometric properties like roundness and convexity remain stable across sizes. The method successfully tracked over 80% of pack ice floes, matching well with NSIDC grid data while providing superior detail.
Introduction: The Challenge of Arctic Ice Tracking
Monitoring Arctic sea ice is critical for climate modeling and maritime navigation. Traditionally, ice motion is tracked using microwave radiometers, which offer all-weather capability but low spatial resolution (several kilometers per pixel). This typically results in "grid-to-grid" motion vectors that average velocity over large areas.
However, regions like the Fram Strait—the primary exit for Arctic ice—are characterized by rapid fragmentation and deformation. Traditional methods like Maximum Cross-Correlation (MCC) struggle to capture the rotation and individual movement of smaller ice floes. This research addresses the need for an object-based approach using high-resolution (250m) optical data to estimate specific shape parameters and individual velocities of pack ice floes.
Methodology
The proposed pipeline consists of four primary stages:
- Sea Ice Recognition: Using the Medium Resolution Spectral Imager (MERSI) on Fengyun-3 (FY-3) satellites, the Normalized Difference Snow Index (NDSI) and specific spectral bands filter out clouds and open seawater.
- Pack Ice Extraction: To separate distinct "pack ice" from "trash ice" (small fragments), gradient operators are applied with OTSU adaptive thresholding to highlight solid floe edges.
- Parameter Estimation: Six shape parameters are calculated for each floe: area, perimeter, mean clamp diameter, roundness, convexity, and width-height ratio.
- Motion Tracking: A two-step matching process identifies the same floe in subsequent images. Candidates are first filtered by area, then the best match is determined using "Euclid closeness," a measure of shape parameter similarity.
Data and Study Area
The study utilizes data from the Medium Resolution Spectral Imager (MERSI) onboard the Chinese FY-3A/B/C satellites from June to August 2014.
- Spatial Resolution: 250m per pixel.
- Study Area: The Fram Strait (77°N to 81°N, 10°W to 10°E).
- Validation: Results were compared against National Snow and Ice Data Centre (NSIDC) polar daily grid sea ice motion vectors.
Results and Discussion
Geometric Self-Similarity and Melting Patterns
A key finding is that "pack ice concentration" decreases significantly faster than "sea ice concentration." This suggests that summer melting causes ice to fragment into untrackable "trash ice" rather than just disappearing. Interestingly, while the perimeter and mean clamp diameter are highly sensitive to edge melting, parameters like roundness, convexity, and width-height ratio remain stable. This geometric self-similarity allows the object-based tracking to remain effective even as floes shrink.
Tracking Performance
The method successfully tracked more than 80% of the pack ice in the study region. The resulting motion vectors matched NSIDC daily grid products but provided significantly higher detail. While NSIDC provides a 25km grid, this method captures the actual motion of individual floes. Observed patterns showed ice flowing from north to southwest, driven by the Transpolar Drift Stream, with occasional cyclonic motion due to atmospheric turbulence.
Conclusion
The object-based tracking method using FY-3/MERSI imagery provides a detailed view of pack ice dynamics in the Arctic. While limited by cloud cover inherent to optical sensors, the approach offers superior resolution for understanding floe-level movement. Future work could involve fusing this high-resolution optical tracking with microwave data to create a robust, all-weather system that maintains individual floe dynamics.
Summary updated