A practical algorithm for the retrieval of floe size distribution of Arctic sea ice from high-resolution satellite Synthetic Aperture Radar imagery
A practical algorithm for the retrieval of floe size distribution of Arctic sea ice from high-resolution satellite Synthetic Aperture Radar imagery
复制标题
DOI：
10.1525/elementa.154
复制
发表时间：
2017-07-20
影响因子：
3.9
通讯作者：
Aptoula, Erchan
中科院分区：
文献类型：
--
作者：
Hwang, Byongjun;Ren, Jinchang;Aptoula, Erchan
在这项研究中，我们提出了一种算法，夏季海冰条件下，半自动产生的浮冰大小分布的北极海冰从高分辨率卫星合成孔径雷达数据。目前，来自卫星图像的浮冰大小分布数据在文献中非常罕见，主要是由于缺乏可靠的算法来产生这种数据。在这里，我们开发的算法结合各种图像分析方法，包括核图切割，距离变换和分水岭变换，以及基于规则的边界重新验证。所开发的算法已被验证对地面真理，手动提取的帮助下，1米分辨率的可见光卫星数据。全面的验证分析显示了前景和局限性。与地面实况相比，该算法往往无法检测到小的浮冰（大多数平均卡尺直径小于100米），这主要是由于水冰分割的限制。絮凝物粒度分布的幂律指数的一些变化是由于在去噪，核图切割分割，边界重新验证和图像分辨率的阈值的过程中的控制参数的影响。尽管如此，该算法，大于100米的浮冰，已显示出合理的协议与地面实况下，这些控制参数的各种选择。考虑到卫星合成孔径雷达数据的覆盖范围和空间分辨率近年来显着增加，开发的算法打开了一个新的可能性，以产生大量的浮冰大小分布数据，这是必不可少的，以提高我们的理解和预测北极海冰覆盖。
In this study, we present an algorithm for summer sea ice conditions that semi-automatically produces the floe size distribution of Arctic sea ice from high-resolution satellite Synthetic Aperture Radar data. Currently, floe size distribution data from satellite images are very rare in the literature, mainly due to the lack of a reliable algorithm to produce such data. Here, we developed the algorithm by combining various image analysis methods, including Kernel Graph Cuts, distance transformation and watershed transformation, and a rule-based boundary revalidation. The developed algorithm has been validated against the ground truth that was extracted manually with the aid of 1-m resolution visible satellite data. Comprehensive validation analysis has shown both perspectives and limitations. The algorithm tends to fail to detect small floes (mostly less than 100 m in mean caliper diameter) compared to ground truth, which is mainly due to limitations in water-ice segmentation. Some variability in the power law exponent of floe size distribution is observed due to the effects of control parameters in the process of de-noising, Kernel Graph Cuts segmentation, thresholds for boundary revalidation and image resolution. Nonetheless, the algorithm, for floes larger than 100 m, has shown a reasonable agreement with ground truth under various selections of these control parameters. Considering that the coverage and spatial resolution of satellite Synthetic Aperture Radar data have increased significantly in recent years, the developed algorithm opens a new possibility to produce large volumes of floe size distribution data, which is essential for improving our understanding and prediction of the Arctic sea ice cover.