TL;DR
This paper presents a novel, four-step modified approach for accurately extracting landfast ice edges in the Gulf of Bothnia using Sentinel-1A InSAR coherence imagery. By combining multiscale segmentation and morphological operations, the method significantly improves detail preservation and accuracy compared to existing techniques, offering a more robust tool for monitoring crucial coastal ice dynamics.
What Problem Does This Paper Address?
Landfast ice, which is sea ice attached to the coast, plays a critical role in coastal ecosystems and human activities like transportation and resource exploration. Accurately mapping its extent and boundaries is essential for understanding climate change impacts and for safe navigation. However, distinguishing landfast ice from drifting ice, especially in single-image analysis, is challenging due to subtle visual differences and the dynamic nature of sea ice. Traditional methods often struggle with preserving fine details of the ice edge or are limited by the availability and quality of SAR data. This paper aims to address these limitations by developing a more refined and reliable method for landfast ice edge extraction.
Methods
The core of this research is a modified four-step approach for extracting landfast ice edges from InSAR coherence images. The process begins with obtaining the coherence image from Sentinel-1A SAR data, which is sensitive to ice stability. The key innovations lie in the subsequent steps:
- Image Segmentation: An object-oriented multiscale segmentation algorithm is employed to divide the coherence image into meaningful units. This is crucial because simple pixel-based thresholding can be unreliable due to noise and the gradual transition between ice types.
- Median Filtering: A median filter is applied to reduce noise and remove small, isolated patches of high coherence that might be misidentified as landfast ice.
- Morphological Erosion: This step is used to further refine the ice edge and shrink potential ice areas, helping to define a clearer boundary.
- Rejection of Small Patches: Remaining small, disconnected high-coherence areas are removed to ensure the final output represents the main landfast ice extent.
Finally, a floodfill algorithm is used to fill internal holes within the landfast ice area, ensuring a complete representation of its extent. The land-water boundary is also extracted using a similar coherence-based approach during ice-free periods.
Datasets
- Sentinel-1A SLC (Single Look Complex) data: Four InSAR image pairs acquired between July 2018 and February 2019 over the Gulf of Bothnia were used. These pairs were chosen to have a temporal baseline of 12 days, balancing coherence and accuracy.
- ACE2 DEM (Altimeter Corrected Elevations, Version 2): Provided by NASA, this Digital Elevation Model was used for InSAR processing.
- SMHI (Swedish Meteorological and Hydrological Institute) daily ice charts: Used as validation data for the seaward landfast ice edge.
- NOAA (National Oceanic and Atmospheric Administration) global coastline data: Used for validating the extracted land-water boundary.
Key Insights and Ideas
- Coherence is King: The study strongly emphasizes that InSAR coherence, rather than just backscatter intensity, is a more reliable indicator for distinguishing stable landfast ice from dynamic drifting ice. Coherence values are significantly higher for landfast ice due to its immobility.
- Multiscale Segmentation is Crucial: Recognizing that landfast ice edges aren't always sharp and can be affected by noise, the use of multiscale segmentation allows for a more nuanced initial division of the image, preserving more local edge details.
- VV Polarization Superiority: For C-band SAR data like Sentinel-1A, the VV polarization mode provides a clearer distinction between landfast ice and drifting ice compared to VH polarization, likely due to its sensitivity to surface roughness and dielectric properties.
- Temporal Baseline Matters: A temporal baseline of around 12 days was found to be optimal for the Gulf of Bothnia, offering a good balance between maintaining high coherence and capturing sufficient temporal information for accurate landfast ice detection. Longer baselines led to significant coherence loss.
- Preserving Local Details: A key advantage of the modified approach is its ability to preserve finer details along the landfast ice edge, which is often lost in simpler edge detection methods.
Key Results
- The modified approach successfully extracted seaward landfast ice edges and land-water boundaries in the Gulf of Bothnia.
- Visual comparison with SMHI ice charts showed good agreement in the overall shape and distribution of landfast ice.
- The extracted land-water boundary showed consistency with NOAA global coastline data, with a mean distance of 109.1 m and a standard deviation of 99.8 m over a 345 km coastline.
- Quantitative comparison with a reference method revealed a small area difference of only 1.36 km² over the 345 km coastline, highlighting the stability and accuracy of the proposed method.
- The modified approach preserved more local edge features compared to the reference method.
- The study successfully mapped landfast ice extents for three different periods, demonstrating the method's potential for monitoring changes.
Limitations and Future Work
- Temporal Baseline Dependency: The method's performance is tied to the availability of suitable image pairs with adequate coherence, which is influenced by the satellite's revisit cycle and environmental conditions.
- Coherence Degradation: Temporal decorrelation remains a challenge, especially during periods of significant sea ice change.
- Future Work: The authors suggest that future advancements in radar systems, such as multi-band, multi-sensor, and shorter revisit cycles, will further enhance the potential for landfast ice research. Integrating these advanced datasets could lead to even more precise and comprehensive monitoring of landfast ice dynamics.
Summary updated