改进DINEOF算法在中国渤海叶绿素a遥感缺失数据重构中的应用研究
张志恒(1995— ),男,山东泰安人,硕士生,主要从事遥感影像云覆盖区域数据恢复的研究。E-mail: nyzhangzhiheng@163.com |
收稿日期: 2020-11-05
要求修回日期: 2021-01-08
网络出版日期: 2021-06-25
基金资助
国家自然科学基金项目(41971339)
国家重点研发计划项目(2018YFC1407605)
山东科技大学科研创新团队支持计划项目(2019TDJH103)
版权
Application of Improved DINEOF Algorithm in the Reconstruction of Missing Remote Sensing Data of Chlorophyll a in the Bohai Sea, China
Received date: 2020-11-05
Request revised date: 2021-01-08
Online published: 2021-06-25
Supported by
National Natural Science Foundation of China(41971339)
National Key Research and Development Program of China(2018YFC1407605)
Shandong University of Science and Technology Research Innovation Team Support Program(2019TDJH103)
Copyright
叶绿素a(chl-a)是重要的海洋环境水色参数,但是受云雾覆盖的影响,卫星遥感chl-a产品中普遍存在数据缺失的现象,大大降低了数据的应用效果。经验正交函数插值方法(DINEOF)是目前在长时间序列缺失数据重构方面应用最广泛的数据插值重构方法,本研究针对DINEOF方法在空间小尺度上过度平滑的缺陷,设计了一种面向渤海chl-a的分层重构方法(SDS-DINEOF),该方法重点考虑了渤海chl-a分布呈现近岸高、中部低的规律,将渤海海域等距离分为32个区域,对位于每个区域的子数据集分别进行重构;利用该方法对2019年全年每日10时13分的渤海GOCI卫星chl-a产品进行了重构,并将其重构结果与DINEOF重构结果进行了对比分析。分析结果表明:应用SDS-DINEOF方法相比DINEOF方法,chl-a重构精度和时间效率上均得到了提升,其中整体精度提高了3.52%,重构时间节约了125%,尤其是在距离陆地最远的渤海中部区域,应用该方法重构精度提升最为显著。本文取得的研究结果对于海洋遥感数据产品的质量提高和应用效率的提升,具有较为重要的理论意义和实际应用价值。
张志恒 , 章超 , 孟麟 , 唐凯 , 朱红春 . 改进DINEOF算法在中国渤海叶绿素a遥感缺失数据重构中的应用研究[J]. 地球信息科学学报, 2021 , 23(4) : 737 -748 . DOI: 10.12082/dqxxkx.2021.200669
Chlorophyll a (chl-a) is an important parameter of water color in marine environment. Quantitative observation from satellite remote sensing is the main way to obtain large-scale oceanic chl-a. However, due to the influence of cloud and fog coverage, data missing is a common phenomenon in satellite remote sensing chl-a products, which greatly reduces the application effect of data. The data-interpolating empirical orthogonal function (DINEOF) method is currently the most widely used data interpolation and reconstruction method for time series data. In this study, a layered reconstruction method for Bohai Sea chl-a (SDS-DINEOF) is designed to address the shortcoming of excessive smoothing at fine scales using the DINEOF method. This method utilizes the distributional characteristics of Bohai chl-a, that is high values in the nearshore and low values in the central water, and divides the Bohai Sea into 32 equidistant regions. We reconstruct the missing data of the daily Bohai GOCI satellite chl-a products at 10:13 a.m in each region separately using this method. The reconstructed results were compared with the DINEOF results. The results show that: (1) Both the SDS-DINEOF and the DINEOF methods can reconstruct missing data completely. The reconstruction results of the former have more detailed information than the latter. The seasonal average results of SDS-DINEOF show that the value of chl-a in the Bohai coastal waters is generally high, while the value of central water is low in summer and autumn, and high in winter and spring; (2) The SDS-DINEOF has a higher overall reconstruction accuracy and reconstruction efficiency compared with the DINEOF. The average accuracy is increased by 3.52%, and the reconstruction time is saved by 125%. The reconstruction accuracy of each layered region has been improved, with the most significant improvement in the 31st and 32nd floors located at the central water in Bohai Sea. The reconstruction accuracy of most single images in the time series has been improved. The reconstruction results of chl-a images in July and August 2019 are significantly improved; (3) As the data missing rate increases, the reconstruction accuracy and efficiency of DINEOF and SDS-DINEOF will both decrease, though the reconstruction accuracy and efficiency of the SDS-DINEOF method are always higher than that of the DINEOF method; (4) During the construction process, the interpretation rate of the data and the reconstruction time by the best model of each layer are restricted and affected by the reconstructed sub-data set itself. The results obtained in this paper have important theoretical and practical significance for improving the quality of marine remote sensing data products.
Key words: satellite remote sensing; Bohai Sea; GOCI; chl-a products; missing ratel; reconstruction; DINEOF; SDS-DINEOF; layer
表1 chl-a重构结果Tab. 1 Reconstruction results of chl-a |
重构方法 | 平均相对精度/% | RMSE | 时间/s |
---|---|---|---|
DINEOF | 88.82 | 0.2489 | 1564 |
SDS-DINEOF | 92.34 | 0.1848 | 694 |
表2 DINEOF方法和SDS-DINEOF方法的5组chl-a重构实验的结果Tab. 2 The results of 5 sets of chl-a reconstruction experiments using DINEOF and SDS-DINEOF |
组别 | 缺失率/% | 方法 | 精度/% | RMSE | 时间/s |
---|---|---|---|---|---|
1 | 9.45 | DINEOF | 88.29 | 0.2497 | 120 |
SDS-DIENOF | 89.04 | 0.2463 | 100 | ||
2 | 25.62 | DINEOF | 85.26 | 0.3534 | 349 |
SDS-DIENOF | 87.96 | 0.3281 | 260 | ||
3 | 45.83 | DINEOF | 83.61 | 0.3796 | 613 |
SDS-DIENOF | 85.24 | 0.3629 | 451 | ||
4 | 64.91 | DINEOF | 80.68 | 0.4242 | 917 |
SDS-DIENOF | 81.91 | 0.4107 | 745 | ||
5 | 86.33 | DINEOF | 73.54 | 0.4744 | 973 |
SDS-DIENOF | 76.77 | 0.4633 | 651 |
表3 DINEOF重构收敛过程数据Tab. 3 DINEOF reconstructing convergence process data |
模态 | 交叉验证误差 | 迭代数 | 均方根误差 | 模态 | 交叉验证误差 | 迭代数 | 均方根误差 |
---|---|---|---|---|---|---|---|
1 | 0.1151 | 35 | 0.0009555 | 11 | 0.0800 | 86 | 0.0009883 |
2 | 0.0968 | 44 | 0.0009600 | 12 | 0.0796 | 129 | 0.0009615 |
3 | 0.0903 | 52 | 0.0009885 | 13 | 0.0793 | 55 | 0.0009709 |
4 | 0.0876 | 64 | 0.0009855 | 14 | 0.0793 | 46 | 0.0009946 |
5 | 0.0856 | 63 | 0.0009879 | 15 | 0.0788 | 46 | 0.0009892 |
6 | 0.0837 | 64 | 0.0009792 | 16 | 0.0793 | 488 | 0.0009842 |
7 | 0.0825 | 51 | 0.0009792 | 17 | 0.0795 | 86 | 0.0009922 |
8 | 0.0818 | 36 | 0.0009645 | 18 | 0.0794 | 188 | 0.0009348 |
9 | 0.0807 | 56 | 0.0009948 | 19 | 0.0798 | 358 | 0.0009396 |
10 | 0.0804 | 73 | 0.0009987 | 20 | 0.0799 | 88 | 0.0009991 |
表4 SDS-DINEOF重构实验各层最佳模态Tab. 4 The optimal mode of each layer in SDS-DINEof reconstruction experiment |
分层 | 最佳模态 | 时间/s | 方差贡献率/% | 分层 | 最佳模态 | 时间/s | 方差贡献率/% |
---|---|---|---|---|---|---|---|
1 | 7 | 7.4 | 88.32 | 17 | 15 | 33.7 | 75.16 |
2 | 8 | 10.1 | 81.66 | 18 | 12 | 22.0 | 86.33 |
3 | 9 | 8.6 | 87.65 | 19 | 15 | 34.1 | 93.11 |
4 | 15 | 29.2 | 87.21 | 20 | 11 | 12.0 | 96.48 |
5 | 13 | 11.2 | 93.36 | 21 | 9 | 13.9 | 89.55 |
6 | 17 | 37.7 | 88.72 | 22 | 11 | 20.1 | 84.56 |
7 | 16 | 31.2 | 85.45 | 23 | 9 | 8.8 | 87.88 |
8 | 16 | 36.3 | 86.78 | 24 | 4 | 3.9 | 98.65 |
9 | 11 | 9.8 | 89.00 | 25 | 6 | 6.4 | 91.02 |
10 | 15 | 34.4 | 96.45 | 26 | 17 | 32.2 | 86.33 |
11 | 16 | 36.5 | 94.12 | 27 | 9 | 11.4 | 89.32 |
12 | 11 | 12.4 | 88.75 | 28 | 11 | 9.2 | 84.56 |
13 | 19 | 75.7 | 78.66 | 29 | 5 | 3.9 | 83.77 |
14 | 17 | 39.6 | 89.84 | 30 | 16 | 40.0 | 87.56 |
15 | 10 | 15.1 | 89.47 | 31 | 11 | 7.6 | 93.45 |
16 | 13 | 25.7 | 84.15 | 32 | 7 | 12.4 | 91.05 |
注:最佳模态的数值大小表示表3重构收敛过程中交叉误差极小值时对应的重构模态数。 |
[1] |
马万栋, 王桥, 吴传庆, 等. 基于反射峰面积的水体叶绿素遥感反演模拟研究[J]. 地球信息科学学报, 2014,16(6):965-970.
[
|
[2] |
梁超, 刘利, 刘建强, 等. 基于HY-1C CZI影像光谱指数重构数据MNF变换的红树林提取[J]. 海洋学报(中文版), 2020,42(4):104-112.
[
|
[3] |
|
[4] |
林明森, 何贤强, 贾永君, 等. 中国海洋卫星遥感技术进展[J]. 海洋学报(中文版), 2019,41(10):99-112.
[
|
[5] |
|
[6] |
姜德娟, 张华. 渤海叶绿素浓度时空特征分析及其对赤潮的监测[J]. 海洋科学, 2018,42(5):23-31.
[
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
蒋雪冰, 胡月明, 刘振华, 等. 基于线性内插的扩展卡尔曼滤波法的NDVI时间序列重构研究[J]. 科技通报, 2017,33(2):137-142.
[
|
[12] |
|
[13] |
胡娅敏, 丁一汇, 沈桐立. 基于遗传算法的四维变分资料同化技术的研究[J]. 大气科学, 2006,30(2):248-256.
[
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
蒋锦刚, 徐曜, 聂晨晖, 等. 海表面温度时空变异特征及对验证误差影响[J]. 遥感学报, 2019,23(2):336-348.
[
|
[19] |
|
[20] |
王跃启, 刘东艳 .基于 DINEOF 方法的水色遥感数据的重构研究——以黄、渤海区域为例[J]. 遥感信息, 2014,29(5):51-57.
[
|
[21] |
李铜基, 林明森, 何贤强 等. 中国近海海洋:海洋光学特性与遥感[M]. 北京: 海洋出版社, 2012:324-348.
[
|
[22] |
陈奕君, 张丰, 杜震洪, 等. 基于DINEOF的静止海洋水色卫星数据重构方法研究[J]. 海洋学研究, 2019,37(4):14-23.
[
|
[23] |
|
[24] |
王建乐. 基于遥感数据的海洋水色水温重构方法研究[D]. 天津:天津大学, 2011.
[
|
[25] |
钱莉, 刘文岭, 郑小慎. 基于MODIS数据反演的渤海叶绿素浓度时空变化[J]. 海洋通报, 2011,30(6):683-687.
[
|
[26] |
|
[27] |
许士国, 富砚昭, 康萍萍. 渤海表层叶绿素a时空分布及演变特征[J]. 海洋环境科学, 2015,34(6):898-903,924.
[
|
[28] |
韩国海洋卫星中心[DB/OL]. http://kosc.kiost.ac.kr/eng/.
[ Korea Ocean Satellite Center[DB/OL]. http://kosc.kiost.ac.kr/eng/. ]
|
[29] |
|
[30] |
丁又专. 卫星遥感海表温度与悬浮泥沙浓度的资料重构及数据同化试验[D]. 南京:南京理工大学, 2009.
[
|
[31] |
孙从容, 张正, 张为良, 等. 利用卫星遥感进行海上透明云薄云的检测[J]. 海洋预报, 2005,22(z1):87-93.
[
|
/
〈 | 〉 |