基于机器学习的长江口表层水体溶解有机碳遥感反演研究

doi:10.3969/j.issn.1000-5641.2024.04.012

摘要/Abstract

摘要：

溶解有机碳 (dissolved organic carbon, DOC) 是海洋中最大的活跃有机碳库. 精确刻画大河河口及其近海水体表层DOC浓度的时空分布, 有助于更好地理解河流输送的有机碳在河口近海经历的生物地化过程及在该区域的归宿. 本研究采用机器学习方法, 通过反演水体溶解态有机碳库中的有色溶解有机物 (colored dissolved organic matter, CDOM) 的吸收光谱信息, 并基于其与水体DOC浓度的相关关系, 发展了基于地球静止轨道水色成像仪 (geostationary ocean color imager, GOCI) 的DOC遥感反演模型. 结果表明, Nu支持向量回归 (nu-supporting vector regression, NuSVR) 方法可准确反演CDOM光谱吸收特性 (如验证集CDOM在300 nm处的吸收系数a_CDOM(300)和275 ~ 295 nm处的光谱斜率S_275–295的平均绝对误差 (mean absolute percent differences, MAPD) 分别为32%和8.6%). 分别基于该区域表层水体CDOM光谱吸收特性与DOC浓度之间表现的3种不同的相关关系进行DOC算法构建, 结果表明, 基于a_CDOM(300)与DOC浓度之间的线性相关, 并考虑这一相关关系的季节性差异所构建的DOC反演算法可较为准确地反演水体DOC浓度, DOC反演现场数据验证集和卫星验证集的MAPD分别为11%和14%. 将构建的DOC算法模型应用到GOCI卫星图像上, 结果显示, 受长江径流影响, 季节尺度上, 长江口夏季水体表层DOC浓度显著高于冬季; 而受潮汐、风场等因素的影响, 小时尺度上河口近岸海域DOC分布呈现逐时高动态变化. 本研究利用卫星遥感反演河口近海水体DOC浓度, 为进一步在不同时间尺度上研究该区域水体DOC动态变化及驱动因素提供了有效手段.

关键词: 地球静止轨道水色成像仪, 有色溶解有机物, 机器学习, 长江口, 溶解有机碳

Abstract:

Dissolved organic carbon (DOC) is the largest reservoir of active organic matter in the ocean. Accurate characterization of the spatial and temporal patterns of DOC in large-river estuaries and neighboring coastal margins will help improve our understanding of biogeochemical processes and the fate of fluvial DOC across the estuary−coastal ocean continuum. By retrieving the absorption properties of colored dissolved organic matter (CDOM) in the dissolved organic matter (DOM) pool using machine learning models, and based on the correlation between CDOM absorption and DOC concentrations, we developed an ocean DOC algorithm for the GOCI satellite. The results indicated that the Nu-Supporting Vector Regression model performed best in retrieving CDOM absorption properties, with mean absolute percent differences (MAPD) of 32% and 8.6% for the CDOM absorption coefficient at 300 nm (a_CDOM(300)) and CDOM spectral slope over the wavelength range of 275 ~ 295 nm (S_275–295). Estimates of DOC concentrations based on the seasonal linear relationship between a_CDOM(300) and DOC were achieved with high retrieval accuracy, with MAPD of 11% and 14% for the training dataset using field measurements and validation datasets on satellite platforms, respectively. Application of the DOC algorithm to GOCI satellite imagery revealed that DOC levels varied dramatically at both seasonal and hourly scales. Elevated surface DOC concentrations were largely associated with summer and lower DOC concentrations in winter as a result of seasonal cycles of Yangtze River discharges. The DOC also changed rapidly on an hourly scale due to the influence of the tide and local wind regimes. This study provides a useful method to improve our understanding of DOC dynamics and their environmental controls across the estuarine −coastal ocean continuum.

Key words: GOCI (geostationary ocean color imager), colored dissolved organic matter, machine learning, the Yangtze River Estuary, dissolved organic carbon

中图分类号:

P733.3

陈灏, 何贤强, 李润, 曹芳. 基于机器学习的长江口表层水体溶解有机碳遥感反演研究[J]. 华东师范大学学报（自然科学版）, 2024, 2024(4): 123-136.

Hao CHEN, Xianqiang HE, Run LI, Fang CAO. Machine learning-based remote sensing retrievals of dissolved organic carbon in the Yangtze River Estuary[J]. Journal of East China Normal University(Natural Science), 2024, 2024(4): 123-136.

图/表 12

图1

表1

图2

图3

图4

表2

图5

表3

图6

图7

图8

图9

参考文献 38

1	HANSELL D A, CARLSON C A.. Deep ocean gradients in dissolved organic carbon concentrations. Nature, 1998, 395 (6699): 263- 266.
2	戴民汉, 翟惟东, 鲁中明, 等.. 中国区域碳循环研究进展与展望. 地球科学进展, 2004, 19 (1): 120- 130.
3	潘德炉, 刘琼, 白雁.. DOC遥感研究进展——基于全球大河DOC与CDOM保守性特征. 海洋学报(中文版), 2012, 34 (4): 1- 11.
4	FICHOT C G, BENNER R.. The spectral slope coefficient of chromophoric dissolved organic matter (S275–295) as a tracer of terrigenous dissolved organic carbon in river-influenced ocean margins. Limnology and Oceanography, 2012, 57 (5): 1453- 1466.
5	MASSICOTTE P, ASMALA E, STEDMON C, et al.. Global distribution of dissolved organic matter along the aquatic continuum: Across rivers, lakes and oceans. Science of the Total Environment, 2017, 609, 180- 191.
6	GRIFFIN C G, MCCLELLAND J W, FREY K E, et al.. Quantifying CDOM and DOC in major Arctic rivers during ice-free conditions using Landsat TM and ETM + data. Remote Sensing of Environment, 2018, 209, 395- 409.
7	CHEN J, ZHU W, TIAN Y Q, et al.. Monitoring dissolved organic carbon by combining Landsat-8 and Sentinel-2 satellites: Case study in Saginaw River estuary, Lake Huron. The Science of the Total Environment, 2020, 718, 137374.
8	MANNINO A, RUSS M E, HOOKER S B.. Algorithm development and validation for satellite-derived distributions of DOC and CDOM in the U. S. Middle Atlantic Bight. Journal of Geophysical Research, 2008, 113 (C7): C07051.
9	HELMS J R, STUBBINS A, RITCHIE J D, et al.. Absorption spectral slopes and slope ratios as indicators of molecular weight, source, and photobleaching of chromophoric dissolved organic matter. Limnology and Oceanography, 2008, 53 (3): 955- 969.
10	FICHOT C G, BENNER R.. A novel method to estimate DOC concentrations from CDOM absorption coefficients in coastal waters. Geophysical Research Letters, 2011, 38 (3): L03610.
11	VANTREPOTTE V, DANHIEZ F, LOISEL H, et al.. CDOM-DOC relationship in contrasted coastal waters: Implication for DOC retrieval from ocean color remote sensing observation. Optics Express, 2015, 23 (1): 33- 54.
12	YU X, SHEN F, LIU Y.. Light absorption properties of CDOM in the Changjiang (Yangtze) estuarine and coastal waters: An alternative approach for DOC estimation. Estuarine, Coastal and Shelf Science, 2016, 181, 302- 311.
13	CAO F, TZORTZIOU M, HU C, et al.. Remote sensing retrievals of colored dissolved organic matter and dissolved organic carbon dynamics in North American estuaries and their margins. Remote Sensing of Environment, 2018, 205, 151- 165.
14	BAI Y, HE X, PAN D, et al.. Summertime Changjiang River plume variation during 1998–2010. Journal of Geophysical Research: Oceans, 2014, 119 (9): 6238- 6257.
15	WU H, SHEN J, ZHU J, et al.. Characteristics of the Changjiang plume and its extension along the Jiangsu Coast. Continental Shelf Research, 2014, 76, 108- 123.
16	YANG B, CAO L, LIU S, et al.. Biogeochemistry of bulk organic matter and biogenic elements in surface sediments of the Yangtze River Estuary and adjacent sea. Marine Pollution Bulletin, 2015, 96 (1): 471- 484.
17	GUO L, SANTSCHI P H, WARNKEN K W.. Dynamics of dissolved organic carbon (DOC) in oceanic environments. Limnology and oceanography, 1995, 40 (8): 1392- 1403.
18	RUDDICK K G, DE CAUWER V, PARK Y, et al.. Seaborne measurements of near infrared water-leaving reflectance: The similarity spectrum for turbid waters. Limnology and Oceanography, 2006, 51 (2): 1167- 1179.
19	RYU J, HAN H, CHO S, et al.. Overview of geostationary ocean color imager (GOCI) and GOCI data processing system (GDPS). Ocean Science Journal, 2012, 47 (3): 223- 233.
20	HE X, BAI Y, PAN D, et al.. Atmospheric correction of satellite ocean color imagery using the ultraviolet wavelength for highly turbid waters. Optics Express, 2012, 20 (18): 20754- 20770.
21	HE X, BAI Y, PAN D, et al.. Using geostationary satellite ocean color data to map the diurnal dynamics of suspended particulate matter in coastal waters. Remote Sensing of Environment, 2013, 133, 225- 239.
22	LIU D, BAI Y, HE X, et al.. Satellite estimation of particulate organic carbon flux from Changjiang River to the estuary. Remote Sensing of Environment, 2019, 223, 307- 319.
23	ZHAO J, CAO W, XU Z, et al.. Estimating CDOM concentration in highly turbid estuarine coastal waters. Journal of Geophysical Research: Oceans, 2018, 123 (8): 5856- 5873.
24	吴志明, 李建超, 王睿, 等.. 基于随机森林的内陆湖泊水体有色可溶性有机物(CDOM)浓度遥感估算. 湖泊科学, 2018, 30 (4): 979- 991.
25	孙璐, 蒋锦刚, 朱渭宁.. 基于GOCI影像的长江口及其邻近海域CDOM遥感反演及其日内变化研究. 海洋学报, 2017, 39 (9): 133- 145.
26	LIU H, LI Q, BAI Y, et al.. Improving satellite retrieval of oceanic particulate organic carbon concentrations using machine learning methods. Remote Sensing of Environment, 2021, 256, 112316.
27	SHEN M, LUO J, CAO Z, et al.. Random forest: An optimal chlorophyll-a algorithm for optically complex inland water suffering atmospheric correction uncertainties. Journal of Hydrology, 2022, 615, 128685.
28	SUN X, ZHANG Y, ZHANG Y, et al.. Machine learning algorithms for chromophoric dissolved organic matter (CDOM) estimation based on Landsat 8 images. Remote Sensing, 2021, 13 (18): 3560.
29	MOREL A, GENTILI B.. A simple band ratio technique to quantify the colored dissolved and detrital organic material from ocean color remotely sensed data. Remote Sensing of Environment, 2009, 113 (5): 998- 1011.
30	SISWANTO E, TANG J, YAMAGUCHI H, et al.. Empirical ocean-color algorithms to retrieve chlorophyll-a, total suspended matter, and colored dissolved organic matter absorption coefficient in the Yellow and East China Seas. Journal of Oceanography, 2011, 67 (5): 627- 650.
31	MANNINO A, NOVAK M G, HOOKER S B, et al.. Algorithm development and validation of CDOM properties for estuarine and continental shelf waters along the northeastern US coast. Remote Sensing of Environment, 2014, 152, 576- 602.
32	MÉLIN F, ZIBORDI G, BERTHON J, et al.. Assessment of MERIS reflectance data as processed with SeaDAS over the European seas. Optics Express, 2011, 19 (25): 25657- 25671.
33	MOON J, AHN Y, RYU J, et al.. Development of ocean environmental algorithms for Geostationary Ocean Color Imager (GOCI). Korean Journal of Remote Sensing, 2010, 26 (2): 189- 207.
34	高源, 明玥, 高磊.. 2019年长江口及其邻近海域溶解有机物的分布和季节变化特征. 海洋环境科学, 2022, 41 (1): 40- 47.
35	张淑坤, 明玥, 高磊.. 2020年夏季长江流域特大洪水期间长江口POC和DOC的分布特征. 海洋环境科学, 2022, 41 (5): 653- 659.
36	李倩, 张珊珊, 线薇微.. 长江口有机碳的时空分异及耦合行为. 海洋环境科学, 2022, 41 (1): 24- 31.
37	叶君, 姚鹏, 徐亚宏, 等.. 长江口盐度梯度下不同形态碳的分布、来源与混合行为. 海洋学报, 2019, 41 (4): 15- 26.
38	SONG S, GAO L, GE J, et al.. Tidal effects on variations in organic and inorganic biogeochemical components in Changjiang (Yangtze River) Estuary and the adjacent East China Sea. Journal of Marine Systems, 2022, 227, 103692.

模型	训练集(N = 49)			验证集(N = 22)
模型	RMSE/m^?1	MAPD/%	Bias/%	RMSE/m^?1	MAPD/%	Bias/%
SVR	0.67	27	2.68	1.00	38	5.78
NuSVR	0.75	26	?2.97	1.05	32	?1.89
RF	0.45	16	0.76	1.42	45	?5.14
BPNN	0.54	19	?0.01	0.92	34	4.55
XGB	0.24	6	?4.15	1.23	39	?7.35
Moon	??	??	??	3.56	95	?96.15

模型	训练集 (N = 49)			验证集 (N = 22)
模型	RMSE /nm^-1	MAPD /%	Bias /%	RMSE /nm^-1	MAPD /%	Bias /%
SVR	0.0020	7.2	?0.342	0.0029	8.8	?2.712
NuSVR	0.0022	7.3	?0.934	0.0031	8.6	?3.260
RF	0.0012	4.4	?0.350	0.0030	9.5	?2.277
BPNN	0.0015	5.8	?0.003	0.0034	10.5	?2.224
XGB	0.0008	3.2	1.314	0.0030	10.0	?1.266

[1]	王余意, 黄晔, 杨静, 丁方方, 何天豪, 李雨珊, 黄琳, 李晔, 刘敏. 长江口及其近岸海域抗生素的时空分异特征与生态风险评估[J]. 华东师范大学学报（自然科学版）, 2024, 2024(6): 136-150.
[2]	张勇, 王慧, 朱传华, 周浩, 詹宇, 李灿, 肖逸凡, 杨丽丽, 刘佳奇. 基于机器学习的卫星遥感水质富营养化评价——以合肥市环城河为例[J]. 华东师范大学学报（自然科学版）, 2024, 2024(1): 1-8, 112.
[3]	高祥云, 孟丹, 罗明凯, 王俊, 张丽平, 孔超. 支持隐私保护的端云协同训练[J]. 华东师范大学学报（自然科学版）, 2023, 2023(5): 77-89.
[4]	游智湧, 刘博林, 刘程, 高灯州. 长江口沉积物固氮过程的温度敏感性及影响因素[J]. 华东师范大学学报（自然科学版）, 2022, 2022(3): 101-108.
[5]	周童晖, 柳银萍. 利用机器学习技术确定同伦分析解中的收敛控制参数[J]. 华东师范大学学报（自然科学版）, 2022, 2022(2): 34-44.
[6]	杨梦晨, 陈旭栋, 蔡鹏, 倪葎. 早期时间序列分类方法研究综述[J]. 华东师范大学学报（自然科学版）, 2021, 2021(5): 115-133.
[7]	余阳, 胡卉芪, 周煊. 基于非易失性内存的LSM-tree存储系统优化[J]. 华东师范大学学报（自然科学版）, 2021, 2021(5): 37-47.
[8]	宋云平, 朱建荣. 长江口余水位时空变化的数值模拟和分析[J]. 华东师范大学学报（自然科学版）, 2021, 2021(4): 121-133.
[9]	朱宜平. 长江口青草沙水域外海正面盐水入侵特点分析[J]. 华东师范大学学报（自然科学版）, 2021, 2021(2): 21-29.
[10]	葛灿, 张卫国. 基于粒级分离的长江口及邻近陆架沉积物磁性特征及其环境意义[J]. 华东师范大学学报（自然科学版）, 2021, 2021(2): 30-41.
[11]	杨正东, 朱建荣, 宋云平, 顾靖华. 长江口余水位时空变化及其成因[J]. 华东师范大学学报（自然科学版）, 2021, 2021(2): 12-20.
[12]	张梦霞, 郑艳玲, 尹国宇, 董宏坡, 韩平, 高娟, 刘程, 常永凯, 刘敏, 侯立军. 纳米银对河口潮滩硝酸盐异化还原成铵过程的影响[J]. 华东师范大学学报（自然科学版）, 2020, 2020(3): 68-77.
[13]	朱坤, 吴莹, 齐丽君. 上海城市内河中有机碳含量的时空变化及影响因素分析[J]. 华东师范大学学报（自然科学版）, 2020, 2020(1): 150-158.
[14]	王浩斌, 杨世伦, 杨海飞. 台风对长江口表层悬沙浓度的影响[J]. 华东师范大学学报(自然科学版), 2019, 2019(2): 195-208.
[15]	朱平, 吴辉. 夏季长江口与苏北海域之间的水体运动及其对动力因子的响应[J]. 华东师范大学学报(自然科学版), 2018, 2018(4): 171-183.