极地研究 ›› 2019, Vol. 31 ›› Issue (4): 431-440.DOI: 10.13679/j.jdyj.20180070

• 研究论文 • 上一篇    下一篇

北极海冰薄冰厚度遥感反演模型的船测验证

马雪沂1 赵羲1,2 屈猛1 程子桉1   

  1. 1.中国南极测绘研究中心, 湖北 武汉 430079
    2.极地测绘科学国家测绘地理信息局重点实验室, 湖北 武汉430079
  • 收稿日期:2018-12-11 修回日期:2019-01-25 出版日期:2019-12-30 发布日期:2019-12-30
  • 通讯作者: 赵羲
  • 基金资助:
    国家自然科学基金面上基金(41576188, 4157060613)和国家重点研发计划课题(2016YFC1402704, 2018YFC1407100)资助

Assessing the accuracy of thin ice thickness retrieval by Arctic ship observations

Ma Xueyi1, Zhao Xi1,2, Qu Meng1, Cheng Zian1   

  1. 1.Chinese Antarctic Center of Surveying and Mapping, Wuhan University, Wuhan 430079, China;
    2.Key Laboratory of Polar Surveying and Mapping, NASG, Wuhan 430079, China
  • Received:2018-12-11 Revised:2019-01-25 Online:2019-12-30 Published:2019-12-30

摘要: 利用美国科考船Sikuliaq在2015年10月至11月初在北极波弗特海区域获得的512处船测海冰数据, 分别对三种不同雪厚参数的MODIS (Moderate-resolution Imaging Spectroradiometer) 热力学反演冰厚模型、德国汉堡大学和不莱梅大学发布的SMOS (Soil Moisture and Ocean Salinity mission) 海冰厚度产品进行精度验证。结果表明, 将船测雪厚作为反演模型中的雪厚参数, 得到薄冰厚度与船测薄冰厚度的平均偏差为0.02 m, 均方根误差为0.12 m, 两者均为三种模型中最低而被认为是最优模型。不考虑雪厚的裸冰冰厚模型得到的薄冰厚度与船测薄冰厚度的相关系数为0.72, 相关性最高但只适用于冰上无雪盖的情况。以经验雪厚关系推算出的雪厚参数加入反演模型中, 得到的冰厚结果在研究验证中精度最不理想。另外, 两种SMOS产品与船测冰厚的相关性与均方根误差结果优于MODIS反演结果, 且不莱梅大学的SMOS冰厚产品精度更优。因此研究认为现有的MODIS薄冰厚度反演算法的反演精度有待提高, 暂不适合作为评定其他薄冰厚度产品精度的标准, 只能作为比较数据。

关键词: 海冰厚度, 走航观测, MODIS, SMOS

Abstract: In this study, MODIS SIT products derived from three retrieval models under different snow thickness parameterization schemes and two SMOS (Soil Moisture and Ocean Salinity mission) SIT maps released from the University of Hamburg (SSIT-UH) and the University of Bremen (SSIT-UB) are validated against ship-based SIT observations (SIT-OBS) acquired at 512 sites from October 4th to November 5th in 2015, in Beaufort Sea, Arctic. Results showed that the SIT retrievals derived from Model III with input from ship-based snow data observations offer the best performance with the lowest bias of 0.02 m and root-mean-square error (RMSE) of 0.12 m. Model I with the snow-free retrieval model matched quite well with SIT-OBS (R=0.72), and is more accurate under less snowy weather conditions. Model II takes the empirical snow-ice thickness relationship as snow thickness parameterization, and exhibits overall poor accuracy. Results also illustrate that the SSIT-UB retrievals agree better with the observations than SSIT-UH retrievals. Both of these methods are more accurate than Model III SIT retrievals, with a higher correlation and smaller RMSE. These results indicate that MODIS SIT products are not always ideal as the reference in evaluating other SIT results, and there are still some problems in the current MODIS SIT retrieval models that require further improvement.

Key words: thin ice thickness, ship observations, MODIS, SMOS