| 基于HY-2B波形特征的北极海冰分类算法 |
| Arctic sea ice classification algorithm based on HY-2B waveform features |
| 投稿时间：2020-07-31 修订日期：2020-11-04 |
| DOI： |
| 中文关键词: HY-2B，高度计，海冰分类，波形特征，北极 |
| 英文关键词:HY-2B, altimeter, sea ice classification, waveform feature, Arctic |
| 基金项目:国家重点研发计划项目(2018YFC1407203)；国家自然科学基金(41976173)；中欧国际合作龙计划项目(57889) |
|
| 摘要点击次数: 83 |
| 全文下载次数: 0 |
| 中文摘要: |
| 海冰类型识别对进行全球气候研究至关重要，利用高度计进行北极海冰监测是目前研究的热点。为探索国产HY-2卫星高度计在海冰类型识别中的可用性，本文利用2019年12月和2020年3月的HY-2B高度计数据，通过提取HY-2B的波形功率最大值（Maximum Power，MAX）、脉冲峰值（Pulse Peakiness，PP）、前缘宽度（Leading Edge Width，LEW）和后向散射系数（Sigma0）四个波形特征，分析了HY-2B卫星高度计精确识别薄一年冰（Thin First-year ice，TFYI）、一年冰（First-year ice，FYI）、多年冰（Multi-year ice，MYI）、冰间水道（LEAD）和开阔水域（Open Water，OW）的能力，这也是国内外范围内对此项研究的首次尝试。通过与俄罗斯北极和南极研究所（Arctic and Antarctic Research Institute，AARI）冰况图产品和MODIS冰间水道产品对比发现，在单特征识别方面，通过柯尔莫哥洛夫—斯米尔诺夫检验（Kolmogorov-Smirnov test，K-S test）可知，MAX对海冰和OW之间有着一定的区分能力，PP对OW的区分度最好，LEW对FYI有着较强的识别能力，Sigma0可用于MYI与LEAD的区分。在多特征识别方面，综合PP、LEW及Sigma0和K最近邻法（K-Nearest Neighbor，KNN），平均最高海冰分类精度可达到91.96%。另外，本文还分析了上述四种特征在北极海冰分类中的重要性，并给出了KNN分类器的推荐使用设置方案。四类特征的重要性排序从高到底分别是PP、LEW、Sigma0和MAX。在KNN分类器设置方面，推荐使用欧氏距离作为KNN分类器的度量，k值设置为3。 |
| 英文摘要: |
| The type of sea ice monitoring is an important parameter for global climate research. The use of altimeters to monitor Arctic sea ice is a current research hotspot. In order to explore the usability of domestic satellite altimeters for sea-ice type identification, HY-2B satellite altimeter data in December 2019 and March 2020 were used for research. Five types of objects can be identified, including thin first-year ice(TFYI), first-year ice(FYI), multi-year ice(MYI), open water(OW) and LEAD, by extracting the four waveform features of HY-2B, including MAX, PP, LEW, and Sigma0, which is also the first attempt at such research at home and abroad. Compared with the AARI ice chart products and MODIS lead products, it is found that in terms of single feature recognition, the K-S test shows that, MAX has a good distinction between sea ice and LEAD; PP and Sigma0 have a better distinction between OW and sea ice; LEW has the highest ability to identify the LEAD among the four features. In the multi-feature recognition, integrating PP, LEW, Sigma0 and KNN classifiers, the average maximum sea ice classification accuracy can reach 91.96%. In addition, this article also analyzes the importance of the above four features in the classification of Arctic sea ice, and gives the recommended settings for the KNN classifier. The order of importance of the four types of features from highest to lowest is PP, LEW, Sigma0 and MAX. In the KNN classifier setting, the Euclidean distance is used as the metric of the KNN, and the k is set to 3. |
| 查看/发表评论 下载PDF阅读器 |
|
|
|
