Construction of a continuous spatiotemporal sea ice concentration dataset for the Bohai Sea based on sub-pixel convolutional neural network super-resolution technology
-
摘要: 构建连续的海冰时空序列是实现渤海海域更准确、及时且高时空分辨率海冰预测的前提。本文针对可见光与被动微波数据在渤海海冰监测中的固有局限，提出通过多源协同与超分辨率融合的技术路径予以互补。首先，优化DT-ASI算法并设定渤海局地系点值，获取6.25 km分辨率AMSR时序数据；进而采用亚像素卷积神经网络（Pixel Shuffle）进行超分辨率重建，确定多Pixel Shuffle阶段超分辨率为最优策略，使平均绝对误差降低8.79 %，相关系数提升0.19。通过融合AMSR超分结果与MODIS数据，构建出2002−2025年1 km分辨率高时空连续序列，冰期内覆盖率从不足28.31 %提升至95.86 %以上。该数据结果初步揭示近10年海冰面积缩减、盛冰期缩短及年际波动增强的趋势。2024−2025年渤海冰期内呈现多次“发展−消融殆尽−再发展”的循环特征，具有一定的特殊性。本研究以协同融合思路克服了单一数据源缺陷，可为渤海海冰精细化监测与预测提供时空连续的关键数据基础。Abstract: Constructing continuous spatiotemporal sequences of sea ice is a prerequisite for achieving more accurate, timely, and high-resolution sea ice predictions in the Bohai Sea. To address the inherent limitations of visible light and passive microwave data in sea ice monitoring, this paper proposes a technical approach based on multi-source synergy and super-resolution fusion. First, the DT-ASI algorithm is optimized and local tie points for the Bohai Sea are established to obtain time-series AMSR data at 6.25 km resolution. Subsequently, a sub-pixel convolutional neural network (Pixel Shuffle) is employed for super-resolution reconstruction, identifying the multi-stage Pixel Shuffle strategy as optimal. This approach reduces the mean absolute error by 8.79% and increases the correlation coefficient by 0.19. By integrating the super-resolution AMSR results with MODIS data, a highly continuous spatiotemporal sequence at 1 km resolution from 2002 to 2025 is constructed, with ice coverage during the ice season rising from less than 28.31% to over 95.86%. The dataset reveals preliminary trends of sea ice area reduction, a shortened ice season, and enhanced interannual variability over the past decade. During the 2024–2025 ice season, the Bohai Sea exhibited a distinctive cyclic pattern of “developing-almost completely melting-redeveloping”. This study, through a synergistic and fusion-based approach, overcomes the limitations of single data sources and provides a spatiotemporally continuous data foundation for refined sea ice monitoring and prediction in the Bohai Sea.1)
1 1数据由中国海洋大学极地海洋过程与全球海洋变化重点实验室提供，可通过其公开数据平台获取：https://coas.ouc.edu.cn/pogoc/2021/0110/c20830a312061/page.htm 。进入网站后，通过“数据下载”栏目下的下载地址即可下载数据。 -
图 1 渤海地理位置水深、高程及可见光真彩图（子图）
Fig. 1 Bathymetry and elevation of the Bohai Sea and true-color visible imagery (inset)
图 2 渤海参数分布与北极对比
ICE为冰点，OW为开阔水点
Fig. 2 Parameter Distributions in the Bohai Sea Compared with the Arctic Ocean
ICE represents ice tie points, and OW represents open water tie points
图 3 AMSR 89.0 GHz水平极化、垂直极化亮温及天气滤波前、后海冰密集度反演结果
Fig. 3 Brightness temperatures at 89.0 GHz horizontal and vertical polarizations from AMSR and retrieved sea ice concentration before and after weather filtering in the Bohai Sea
图 4 基于亚像素重排的Pixel Shuffle（像素重组）超分辨率模型结构
Fig. 4 Structure of the Pixel Shuffle super-resolution model based on sub-pixel rearrangement
图 5 2025年2月7日MODIS真彩图、海冰密集度反演结果及AMSR反演结果（双线性插值、超分辨率）
Fig. 5 Comparison of MODIS true-color imagery, retrieved sea ice concentration, and AMSR-derived sea ice concentration using Bilinear interpolation and super-resolution on 7 February 2025 in the Bohai Sea
图 6 超分辨率模型验证评价指标对比
Fig. 6 Comparison of evaluation metrics for super-resolution model validation
图 7 AMSR超分辨率结果与MODIS反演结果对比
Fig. 7 Comparison of super-resolved AMSR sea ice concentration with MODIS-derived results in the Bohai Sea
图 8 测试集（2024–2025年度）超分辨率结果时间序列及误差分析
Fig. 8 Time series and error analysis of super-resolved sea ice concentration for the 2024–2025 test set in the Bohai Sea
图 9 海冰密集度超分辨率融合结果（2025年1月25日至2025年2月23日）
Fig. 9 Super-resolved and fused sea ice concentration from 25 January to 23 February 2025 in the Bohai Sea
图 10 各个数据源及融合产品的渤海海冰数据逐月覆盖率
颜色为覆盖率，数字为缺失天数
Fig. 10 Monthly Coverage of Sea Ice Data from Individual Data Sources and the Fused Product in the Bohai Sea
Shaded areas represent coverage percentage, and numbers indicate the number of missing days
图 11 多年气候态及近3年渤海海冰面积时间序列
Fig. 11 Time series of climatological and recent three-year sea ice extent in the Bohai Sea
图 12 近3年渤海海冰面积和ERA5 2 m气温对比
Fig. 12 Comparison of the Bohai Sea ice area and ERA5 2-meter air temperature over the past three years
表 1 超分辨率技术的关键变量信息表
Tab. 1 Key variables used in the super-resolution technique
变量符号 变量名称 数学定义 / 维度 物理意义 / 说明 r 上采样倍数 标量（正整数） 目标分辨率相对于原始分辨率在单维度上的放大倍数。 Fin 输入特征图 $ {\mathbb{R}}^{{{\mathrm{C}}_{\text{in}}}\times {{\mathrm{H}}_{\text{in}}}\times {{\mathrm{W}}_{\text{in}}}} $ 模型的输入数据。Cin为通道数量，对于海冰密集度数据，初始Cin = 1；中间网络层
Cin = 64。Hin，Win为输入空间尺寸，分别代表输入空间的高度和宽度。Fexp 通道扩展后特征图 $ {\mathbb{R}}^{{{\mathrm{C}}_{\exp }}\times {{\mathrm{H}}_{\text{in}}}\times {{\mathrm{W}}_{\text{in}}}} $ 通过卷积层将通道数扩展至Cexp = Cin × r2，为后续空间重组储备信息。 Fout 上采样后特征图 $ {\mathbb{R}}^{{{\mathrm{C}}_{\text{in}}}\times {{\mathrm{H}}_{\text{out}}}\times {{\mathrm{W}}_{\text{out}}}} $ 空间重组后的输出特征图。空间尺寸放大为Hout = Hin × r，Wout = Win × r，通道数恢复为Cin。 YHR 最终超分结果 $ {\mathbb{R}}^{{{\mathrm{H}}_{\text{out}}}\times {{\mathrm{W}}_{\text{out}}}} $ 网络输出的高分辨率海冰密集度图像。本研究最终目标为输出1 km分辨率网格。 
下载: 导出CSV
表 2 数据融合试验设置
Tab. 2 Configuration of the data fusion experiment
试验序号 超分辨率步骤 其余参数设置 Exp0
(6x_Bilinear)①6倍Pixel Shuffle →
②1.25倍双线性插值中间特征层：64
激活函数：LeakyReLU
损失函数：Smooth L1 Loss
优化器：AdamW
初始学习率：0.0001
权重衰减：0.00001
训练轮数：30
批次大小：8Exp1
(2x3x_Bilinear)①2倍Pixel Shuffle →
②3倍Pixel Shuffle →
③1.25倍双线性插值Exp2
(3x2x_Bilinear)①3倍Pixel Shuffle →
②2倍Pixel Shuffle →
③1.25倍双线性插值
下载: 导出CSV
表 3 渤海海冰密集度超分辨率融合结果及双线性插值结果的误差分析
Tab. 3 Error analysis comparing super-resolution fusion and Bilinear interpolation results for sea ice concentration in the Bohai Sea
偏差/% 平均绝对误差/% 均方根误差/% 相关系数 6.25倍双线性插值 −10.05 22.01 32.04 0.63 Exp0
(6x_Bilinear)9.92 20.30 31.03 0.69 Exp1
(2x3x_Bilinear)−7.68 13.55 26.33 0.81 Exp2
(3x2x_Bilinear)−7.97 13.22 25.66 0.82
下载: 导出CSV
表 4 渤海海冰密集度超分辨率结果逐日平均绝对误差
Tab. 4 Daily mean absolute error of super-resolved sea ice concentration in the Bohai Sea
日期 Bilinear
InterpolationExp0
6x_Bilinear_SRExp1
2x3x_
Bilinear_SRExp2
3x2x_
Bilinear_SR2025/1/4 17.69 % 3.17 % 2.63 % 3.87 % 2025/1/6 24.98 % 32.34 % 17.38 % 16.07 % 2025/1/7 20.36 % 14.53 % 8.83 % 8.48 % 2025/1/8 18.66 % 17.67 % 10.95 % 9.34 % 2025/1/10 24.30 % 27.54 % 20.99 % 19.80 % 2025/1/11 25.72 % 28.04 % 20.46 % 20.51 % 2025/1/12 23.09 % 22.05 % 15.66 % 15.22 % 2025/1/15 25.14 % 39.22 % 25.04 % 23.29 % 2025/1/16 27.15 % 31.46 % 20.71 % 19.68 % 2025/1/17 20.42 % 7.75 % 4.72 % 5.95 % 2025/1/18 17.08 % 7.52 % 4.84 % 5.35 % 2025/1/19 16.57 % 9.10 % 5.35 % 5.43 % 2025/1/20 19.49 % 21.09 % 9.93 % 10.02 % 2025/1/21 21.21 % 23.68 % 10.69 % 10.26 % 2025/1/22 21.73 % 20.56 % 12.04 % 11.92 % 2025/1/23 40.81 % 42.25 % 22.27 % 22.75 % 2025/1/24 14.89 % 8.53 % 4.20 % 4.99 % 2025/1/28 20.93 % 17.57 % 11.92 % 11.13 % 2025/2/1 20.89 % 22.85 % 16.05 % 15.21 % 2025/2/4 17.66 % 14.68 % 10.48 % 10.19 % 2025/2/7 21.06 % 20.67 % 17.10 % 15.96 % 2025/2/8 22.54 % 17.93 % 14.85 % 15.29 % 2025/2/9 22.80 % 24.50 % 21.20 % 21.54 % 2025/2/10 21.83 % 19.50 % 15.46 % 15.40 % 2025/2/11 31.30 % 35.09 % 28.49 % 26.17 % 2025/2/12 23.81 % 24.01 % 19.33 % 19.06 % 2025/2/16 14.46 % 3.56 % 2.59 % 2.54 % 2025/2/17 19.86 % 18.91 % 10.71 % 10.20 % 2025/2/18 21.96 % 12.96 % 8.09 % 7.82 % Mean 22.01 % 20.30 % 13.55 % 13.22 %
下载: 导出CSV
表 5 渤海及其各个海域超分辨率结果测试集误差分析
Tab. 5 Test set error analysis of super-resolved sea ice concentration for the Bohai Sea and its subregions
Mean Bias Error
(MBE)/km2Mean Absolute Error
(MAE)/km2Root Mean Square Error
(RMSE)/km2Pearson Correlation
Coefficient (R)AMSR-SR
mean/km2MODIS
mean/km2Liaodong Bay −414.28 965.53 1 213.18 0.92 2 962.82 3 377.10 Bohai Bay −398.90 403.56 856.76 0.89 63.17 462.06 Laizhou Bay −120.39 132.06 305.96 0.69 37.82 158.21 Bohai Sea −857.26 1 336.08 1 816.11 0.91 3 071.15 3 928.41
下载: 导出CSV
表 6 AMSR原始渤海海冰密集度及超分辨率结果误差
Tab. 6 Errors of original AMSR Bohai Sea ice concentration and super-resolution results
AMSR
平均误差AMSR
平均绝对误差AMSR-SR
平均误差AMSR-SR
平均绝对误差−5.01% 12.11% −5.98% 13.22%
下载: 导出CSV
-
[1] 王相玉, 张惠滋, 严素, 等. 渤、黄海北部海冰年代时空变化特征分析[J]. 海洋预报, 2007, 24(2): 26−32. doi: 10.3969/j.issn.1003-0239.2007.02.004Wang Xiangyu, Zhang Huizi, Yan Su, et al. The characteristic of the spatio-temporal transformation of the sea-ice in northern part of the Bohai sea and the Yellow sea[J]. Marine Forecasts, 2007, 24(2): 26−32. doi: 10.3969/j.issn.1003-0239.2007.02.004 [2] 刘煜, 吴辉碇. 第1讲 渤、黄海的海冰[J]. 海洋预报, 2017, 34(3): 94−101. doi: 10.11737/j.issn.1003-0239.2017.03.012Liu Yu, Wu Huiding. Sea ice in the Bohai Sea and the northern Yellow Sea[J]. Marine Forecasts, 2017, 34(3): 94−101. doi: 10.11737/j.issn.1003-0239.2017.03.012 [3] 王安良, 赵倩, 隋俊鹏, 等. 渤海和黄海北部工程设计冰厚变化趋势分析[J]. 哈尔滨工程大学学报, 2024, 45(4): 659−664. doi: 10.11990/jheu.202208041Wang Anliang, Zhao Qian, Sui Junpeng, et al. The trend of sea ice design thickness for the Bohai Sea and Northern Huanghai Sea[J]. Journal of Harbin Engineering University, 2024, 45(4): 659−664. doi: 10.11990/jheu.202208041 [4] Gong Daoyi, Wang Shaowu, Zhu Jinhong. East Asian winter monsoon and Arctic oscillation[J]. Geophysical Research Letters, 2001, 28(10): 2073−2076. doi: 10.1029/2000GL012311 [5] 白珊, 刘钦政, 吴辉碇, 等. 渤海、北黄海海冰与气候变化的关系[J]. 海洋学报, 2001, 23(5): 33−41. doi: 10.3321/j.issn:0253-4193.2001.05.004Bai Shan, Liu Qinzheng, Wu Huiding, et al. Relation of ice conditions with climate change in the Bohai Sea and the northern Huanghai Sea[J]. Haiyang Xuebao, 2001, 23(5): 33−41. doi: 10.3321/j.issn:0253-4193.2001.05.004 [6] 李彦青, 苏洁, 汪洋, 等. 渤海海冰外缘线候平均离岸距离的变化及其关键影响因子[J]. 中国海洋大学学报, 2013, 43(7): 7−16.Li Yanqing, Su Jie, Wang Yang, et al. Variability of the pentadly average distance between the sea ice edge and the coast in the Bohai sea and its key impact factors[J]. Periodical of Ocean University of China, 2013, 43(7): 7−16. [7] 唐茂宁, 刘煜, 李宝辉, 等. 渤海及黄海北部冰情长期变化趋势分析[J]. 海洋预报, 2012, 29(2): 45−49. doi: 10.3969/j.issn.1003-0239.2012.02.007Tang Maoning, Liu Yu, Li Baohui, et al. Analysis of the sea ice long-term trend in the Bohai Sea and the northern Yellow Sea[J]. Marine Forecasts, 2012, 29(2): 45−49. doi: 10.3969/j.issn.1003-0239.2012.02.007 [8] 药蕾, 苏洁. 渤海海冰与西伯利亚高压之间的关系及与北大西洋涛动之间的可能联系[J]. 中国海洋大学学报, 2018, 48(6): 1−12.Yao Lei, Su Jie. Relationships between Bohai sea ice and Siberian high and possible connections between Bohai sea ice and North Atlantic oscillation[J]. Periodical of Ocean University of China, 2018, 48(6): 1−12. [9] Lin Zhikun, Yan Yu, Xu Yingjun, et al. Recent decline of sea ice area in the Bohai Sea over the period 2001 to 2023[J]. Journal of Marine Systems, 2025, 250: 104089. doi: 10.1016/j.jmarsys.2025.104089 [10] 中华人民共和国自然资源部. 2016年中国海洋灾害公报[EB/OL]. (2017-03-22)[2026-02-25]. https://gc.mnr.gov.cn/201806/t20180619_1798020.html.Ministry of Natural Resources of the People’s Republic of China. China marine disaster bulletin 2016[EB/OL]. (2017−03−22)[2026−02−25]. https://gc.mnr.gov.cn/201806/t20180619_1798020.html. [11] 郑新江, 邱康睦, 陆风. 定量计算渤海海冰参数的遥感方法[J]. 应用气象学报, 1998, 9(3): 359−363.Zheng Xinjiang, Qiu Kangmu, Lu Feng. Quantitative calculation of sea ice over the Bohai Sea using NOAA/AVHRR imagery[J]. Journal of Applied Meteorological Science, 1998, 9(3): 359−363. [12] 罗亚威, 张蕴斐, 孙从容, 等. “海洋1号”卫星在海冰监测和预报中的应用[J]. 海洋学报, 2005, 27(1): 7−18.Luo Yawei, Zhang Yunfei, Sun Congrong, et al. Application of the “HY-1” satellite in sea ice monitoring and forecasting[J]. Haiyang Xuebao, 2005, 27(1): 7−18. [13] 刘志强, 苏洁, 时晓旭, 等. 渤海AVHRR多通道海冰密集度反演算法试验研究[J]. 海洋学报, 2014, 36(11): 74−84.Liu Zhiqiang, Su Jie, Shi Xiaoxu, et al. Study on the multi-band retrieval algorithm for the Bohai Sea ice concentration using AVHRR data[J]. Haiyang Xuebao, 2014, 36(11): 74−84. [14] Drüe C, Heinemann G. High-resolution maps of the sea-ice concentration from MODIS satellite data[J]. Geophysical Research Letters, 2004, 31(20): L20403. doi: 10.1029/2004gl020808 [15] Drüe C, Heinemann G. Accuracy assessment of sea-ice concentrations from MODIS using in-situ measurements[J]. Remote Sensing of Environment, 2005, 95(2): 139−149. doi: 10.1016/j.rse.2004.12.004 [16] Zhang Dong, Ke Changqing, Sun Bo, et al. Extraction of sea ice concentration based on spectral unmixing method[J]. Journal of Applied Remote Sensing, 2011, 5(1): 053552. doi: 10.1117/1.3643703 [17] 史凯琦, 邹斌, 陈树果, 等. 中分辨率成像光谱仪的海冰密集度遥感反演[J]. 遥感学报, 2021, 25(3): 753−764.Shi Kaiqi, Zou Bin, Chen Shuguo, et al. Remote sensing inversion of sea ice concentration by a middle-resolution imaging spectrometer[J]. National Remote Sensing Bulletin, 2021, 25(3): 753−764. [18] Liu Yinghui, Key J, Mahoney R. Sea and freshwater ice concentration from VIIRS on Suomi NPP and the future JPSS satellites[J]. Remote Sensing, 2016, 8(6): 523. doi: 10.3390/rs8060523 [19] Shi Wenqi, Yuan Shuai, Liu Chengyu, et al. Inversion of sea ice concentration in the Liaodong Bay from MODIS data[J]. Remote Sensing, 2022, 14(18): 4439. doi: 10.3390/rs14184439 [20] Shi Wenzhe, Caballero J, Huszár F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 1874−1883. [21] Mei Shaohui, Yuan Xin, Ji Jingyu, et al. Hyperspectral image spatial super-resolution via 3D full convolutional neural network[J]. Remote Sensing, 2017, 9(11): 1139. doi: 10.3390/rs9111139 [22] Lu Xiaochen, Yang Dezheng, Zhang Junping, et al. Hyperspectral image super-resolution based on spatial correlation-regularized unmixing convolutional neural network[J]. Remote Sensing, 2021, 13(20): 4074. doi: 10.3390/rs13204074 [23] Cai Yuanhao, Lin Jing, Lin Zudi, et al. MST++: multi-stage spectral-wise transformer for efficient spectral reconstruction[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New Orleans: IEEE, 2022: 745−755. [24] Xiao Kangliang, Xiao Shaozhang, Zhao Yiyan. Super resolution reconstruction of bidirectional feature flow images based on SRGAN[J]. Innovative Applications of AI, 2025, 2(4): 53−60. doi: 10.70695/IAAI202504A15 [25] Yang Jingyu, Ahmed Y A E, Lv Panjie, et al. UBR-Net: road extraction from high-resolution remote sensing imagery using multi-scale attention and cross-residual encoding[J]. Canadian Journal of Remote Sensing, 2025, 51(1): 2586320. doi: 10.1080/07038992.2025.2586320 [26] Ge Linyao, Chen Ge, Wang Guiyu, et al. Multiscale dynamics-informed deep learning reconstruction of global ocean offshore near-surface current[J]. GIScience & Remote Sensing, 2025, 62(1): 2596455. doi: 10.1080/15481603.2025.2596455 [27] Liu Bingjie, Wu Wei, Wu Hao, et al. KERF: a knowledge-enhanced relearning framework for tailings pond detection from high-resolution remote sensing images[J]. GIScience & Remote Sensing, 2025, 62(1): 2541429. doi: 10.1080/15481603.2025.2541429 [28] Feng Tiantian, Jiang Peng, Liu Xiaomin, et al. Applications of deep learning-based super-resolution networks for AMSR2 Arctic sea ice images[J]. Remote Sensing, 2023, 15(22): 5401. doi: 10.3390/rs15225401 [29] Rusin J, Lavergne T, Doulgeris A P, et al. Resolution enhanced sea ice concentration: a new algorithm applied to AMSR2 microwave radiometry data[J]. Annals of Glaciology, 2024, 65: e15. doi: 10.1017/aog.2024.6 [30] Wang Xue, Chen Zhuoqi, Fang Yan, et al. An improvement in accuracy and spatial resolution of the pattern-matching sea ice drift from SAR imagery[J]. International Journal of Digital Earth, 2023, 16(2): 4073−4094. doi: 10.1080/17538947.2023.2264918 [31] Au C, Tsamados M, Manescu P, et al. ARISGAN: extreme super-resolution of arctic surface imagery using generative adversarial networks[J]. Frontiers in Remote Sensing, 2024, 5: 1417417. doi: 10.3389/frsen.2024.1417417 [32] Hao Guanghua, Su Jie. A study on the dynamic tie points ASI algorithm in the Arctic Ocean[J]. Acta Oceanologica Sinica, 2015, 34(11): 126−135. doi: 10.1007/s13131-015-0659-y [33] Gloersen P, Cavalieri D J. Reduction of weather effects in the calculation of sea ice concentration from microwave radiances[J]. Journal of Geophysical Research: Oceans, 1986, 91(C3): 3913−3919. doi: 10.1029/JC091iC03p03913 [34] Cavalieri D J, St. Germain K M, Swift C T. Reduction of weather effects in the calculation of sea-ice concentration with the DMSP SSM/I[J]. Journal of Glaciology, 1995, 41(139): 455−464. doi: 10.3189/S0022143000034791 -
计量
- 文章访问数: 15
- HTML全文浏览量: 9
- PDF下载量: 0
- 被引次数: 0
