Machine learning-based remote sensing retrieval model for MODIS chlorophyll-a concentration in adjacent waters of the Yellow River Estuary
-
摘要: 海洋叶绿素a（Chl-a）浓度是海洋浮游植物生物量的重要表征，是海洋生态环境变化的直观体现，准确、高效的估算水体Chl-a浓度是海洋相关研究的基础。卫星遥感技术可用于大范围、高频次的Chl-a浓度监测，为我们理解海洋生态环境变化提供了重要支撑。但是，由于复杂的生物光学特征，近岸浑浊水体Chl-a浓度的遥感反演结果具有很大的不确定性，需要我们利用大量现场实测数据进行验证和优化。本文基于2010-2023年间45个航次的表层Chl-a浓度实测数据，利用MODIS同步卫星遥感反射率数据，构建了黄河口邻近海域Chl-a浓度的机器学习遥感反演模型。结果表明：与传统的全球标准算法和以往的区域算法相比，机器学习算法获得了更高的反演精度；其中，高斯过程模型表现最佳（R2 = 0.62, RMSE = 0.21 mg/m3），结果很好的呈现了该海域Chl-a浓度的时空变化特征。总体来看，该海域Chl-a浓度在空间上呈现近岸高于离岸的特征，季节变化呈现较为明显的夏高冬低的单峰结构，2003−2023年间平均Chl-a浓度以每年0.02 mg/m3的速率增加。研究结果丰富了近岸海域Chl-a浓度遥感反演算法，拓展了机器学习算法的遥感应用，为黄河口邻近海域海洋生态环境评估提供了方法和数据支撑。Abstract: : Chlorophyll-a (Chl-a) concentration in the ocean is an important indicator of the marine phytoplankton biomass and serves as a direct reflection of marine ecological and environmental changes. Accurate and efficient estimation of Chl-a is essential for oceanographic research. Satellite remote sensing facilitates large-scale, high-frequency monitoring of Chl-a, offering important support for understanding evolution of marine ecosystem. However, due to the complex bio-optical properties, remote sensing retrievals of Chl-a in coastal, turbid waters are often uncertain, requiring the use of extensive in situ observations for validation and optimization. In this study, in situ Chl-a observations from 45 cruises conducted between 2010 and 2023 were integrated with synchronous satellite remote sensing reflectance data to develop a machine learning (ML) model for estimating Chl-a concentrations in the Yellow River Estuary adjacent sea areas. The results demonstrate that, compared to traditional standard algorithms and previous regional models, ML algorithms achieve higher accuracy. Among ML models, the Gaussian Process Regression (GPR) model yielded the best performance (R2 = 0.62, RMSE = 0.21 mg/m3), effectively capturing the spatial temporal Chl-a patterns of this area. The chlorophyll-a (Chl-a) concentration in this sea area demonstrates a spatial pattern of nearshore areas being higher than offshore areas, with seasonal variation showing a distinct single-peak structure characterized by higher values in summer and lower values in winter. From 2003 to 2023, the average Chl-a concentration increased at an annual rate of 0.02 mg/m3. This research advances remote sensing retrieval algorithms for Chl-a concentration in coastal waters, expands the application of ML models, and provides both methodological and data supports for evaluating the marine ecological environment in the Yellow River Estuary and its adjacent areas.
-
图 1 研究区及现场实测Chl-a站点分布图（其中绿色圆点代表所有站点，红色圆点代表实测数据与遥感数据相匹配站点）
Fig. 1 Study area and the distribution of in situ Chl-a sites (Green dots represent all of the in situ sites and red dots represent match-ups between in situ measured and remotely sensed datasets)
图 3 泰勒图 （绿色圆点为原位实测数据，红色点为各模型的结果）
Fig. 3 Taylor diagram (Green dots are in situ measured data, red dots are satellite-retrieval models)
图 4 9个方法反演的Chl-a浓度值与实测Chl-a浓度值的散点图(其中X 坐标是模型反演Chl-a浓度值，Y 坐标是实测Chl-a浓度值，均以mg/m3 为单位)
Fig. 4 Scatterplots of Chl-a concentration values between the 9 satellite models versus in situ measurements (The X-coordinate is the satellite-derived Chl-a concentration value and the Y-coordinate is the in situ measured Chl-a concentration value, both in mg/m3)
图 5 2021年11月14日9种模型反演Chl-a浓度的比较
Fig. 5 Comparison of Chl-a estimations from 9 models on November 14, 2021
图 6 GP模型反演气候态月平均Chl-a 浓度空间分布图
Fig. 6 Spatial distribution of climatological monthly Chl-a concentrations derived by GP Model
图 7 实测Chl-a浓度的逐月空间分布图
Fig. 7 Spatial distribution of in situ measured monthly Chl-a concentrations
图 8 2002−2024年研究区月均Chl-a浓度折线图
Fig. 8 Monthly variability of Chl-a in the study area (2002−2024)
图 9 2003−2023年研究区年均Chl-a浓度折线图
Fig. 9 Annual mean Chl-a concentration in the study area (2003−2023)
表 1 MODIS不同波段及波段组合
Tab. 1 MODIS different bands and band combinations
波段组合
方式具体波段内容 单个波段 Rrs(412)、Rrs(443)、Rrs(469)、Rrs(488)、Rrs(531)
Rrs(547)、Rrs(555)、Rrs(645)、Rrs(667)、Rrs(678)波段比 Rrs(443)/Rrs(488)、Rrs(488)/Rrs(555)、Rrs(443)/Rrs(555)
Rrs(555)/Rrs(667)、Rrs(443)/Rrs(667)、Rrs(488)/Rrs(667)波段差值 (Rrs(488)-Rrs(443))/(488-443)、(Rrs(555)-Rrs(488))/(555-488)
Rrs(555)-(Rrs(667)-Rrs(443))*(555-443)/(667-443)
Rrs(555)-(Rrs(667)-Rrs(488))*(555-488)/(667-488)
下载: 导出CSV
表 2 机器学习模型的主要参数设置
Tab. 2 Parameterization of machine learning models
机器学习模型 参数设置 MLR 默认参数 SVM 核函数：高斯；核尺度：0.16；盒约束：1.7；容错边界：0.14 RF 方法：LSBoost；学习周期数：127；学习率：0.02
学习器: templateTree（最小叶子大小：3，每次抽样变量：11）ANN 层尺寸：62；激活函数：relu
正则化参数：4.87E-4；迭代限制：1000 GP 基函数：常数；核函数：指数
核参数：[0.033 0.27]；变异性：0.30；XGB 学习速率：0.99；特征采样比例：0.9；树的最大深度：5；最小子节点权重：5；样本采样比例：0.9；最小分裂损失：1E-3 GAM 链接函数：identity link光滑函数：三次样条；初始结数：3
下载: 导出CSV
表 3 模型训练结果
Tab. 3 Model training results
机器学习模型 RMSE Bias R2 Slope Intercept MLR 0.248 −0.003 0.485 1.000 0.000 SVM 0.136 −0.002 0.860 1.148 −0.043 RF 0.178 −0.020 0.802 1.392 −0.081 ANN 0.181 0.006 0.730 1.080 −0.023 GP 0.070 0.000 0.968 1.104 −0.030 XGB 0.068 −0.001 0.973 1.123 −0.037 GAM 0.208 −0.006 0.640 1.037 −0.010
下载: 导出CSV
表 4 模型验证结果 （10倍交叉验证）
Tab. 4 Model testing results (10-fold cross validation)
机器学习模型 RMSE Bias R2 Slope Intercept OC3 0.460 0.280 0.036 0.435 0.030 LMC 0.366 0.064 0.102 0.438 0.123 MLR 0.260 −0.002 0.435 0.924 0.022 SVM 0.234 0.007 0.541 1.036 −0.015 RF 0.248 −0.018 0.521 1.313 −0.058 ANN 0.228 0.005 0.566 0.945 0.015 GP 0.213 0.004 0.622 1.049 −0.015 XGB 0.244 0.022 0.532 1.291 −0.103 GAM 0.239 −0.005 0.527 0.913 0.024
下载: 导出CSV
-
[1] Falkowski P G, Barber R T, Smetacek V. Biogeochemical controls and feedbacks on ocean primary production[J]. Science, 1998, 281(5374): 200−206. doi: 10.1126/science.281.5374.200 [2] Field C B, Behrenfeld M J, Randerson J T, et al. Primary production of the biosphere: integrating terrestrial and oceanic components[J]. Science, 1998, 281(5374): 237−240. doi: 10.1126/science.281.5374.237 [3] Hu Chuanmin, Feng Lian, Guan Qi. A machine learning approach to estimate surface chlorophyll a concentrations in global oceans from satellite measurements[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(6): 4590−4607. doi: 10.1109/TGRS.2020.3016473 [4] Dierssen H M. Perspectives on empirical approaches for ocean color remote sensing of chlorophyll in a changing climate[J]. Proceedings of the National Academy of Sciences of the United States of America, 2010, 107(40): 17073−17078. [5] 潘应阳, 国巧真, 孙金华. 水体叶绿素a浓度遥感反演方法研究进展[J]. 测绘科学, 2017, 42(1): 43−48.Pan Yingyang, Guo Qiaozhen, Sun Jinhua. Advances in remote sensing inversion method of chlorophyll a concentration[J]. Science of Surveying and Mapping, 2017, 42(1): 43−48. [6] Nazeer M, Nichol J E. Development and application of a remote sensing-based Chlorophyll-a concentration prediction model for complex coastal waters of Hong Kong[J]. Journal of Hydrology, 2016, 532: 80−89. doi: 10.1016/j.jhydrol.2015.11.037 [7] D'Alimonte D, Zibordi G. Phytoplankton determination in an optically complex coastal region using a multilayer perceptron neural network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2003, 41(12): 2861−2868. doi: 10.1109/TGRS.2003.817682 [8] Guan Qi, Feng Lian, Hou Xuejiao, et al. Eutrophication changes in fifty large lakes on the Yangtze Plain of China derived from MERIS and OLCI observations[J]. Remote Sensing of Environment, 2020, 246: 111890. doi: 10.1016/j.rse.2020.111890 [9] 罗婕纯一, 秦龙君, 毛鹏, 等. 水质遥感监测的关键要素叶绿素a的反演算法研究进展[J]. 遥感技术与应用, 2021, 36(3): 473−488.Luo Jiechunyi, Qin Longjun, Mao Peng, et al. Research progress in the retrieval algorithms for Chlorophyll-a, a key element of water quality monitoring by remote sensing[J]. Remote Sensing Technology and Application, 2021, 36(3): 473−488. [10] 张明慧, 苏华, 季博文. MODIS时序影像的福建近岸叶绿素a浓度反演[J]. 环境科学学报, 2018, 38(12): 4831−4839.Zhang Minghui, Su Hua, Ji Bowen. Retrieving nearshore chlorophyll-a concentration using MODIS time-series images in the Fujian Province (China)[J]. Acta Scientiae Circumstantiae, 2018, 38(12): 4831−4839. [11] 滕越, 邹斌, 叶小敏. HY-1C卫星海岸带成像仪叶绿素a浓度反演研究[J]. 海洋学报, 2022, 44(5): 25−34.Teng Yue, Zou Bin, Ye Xiaomin. Study on the chlorophyll a concentration retrieved from HY-1C satellite coastal zone imager data[J]. Haiyang Xuebao, 2022, 44(5): 25−34. [12] Chen Jun, Quan Wenting, Wen Zhenhe, et al. An improved three-band semi-analytical algorithm for estimating chlorophyll-a concentration in highly turbid coastal waters: a case study of the Yellow River estuary, China[J]. Environmental Earth Sciences, 2013, 69(8): 2709−2719. doi: 10.1007/s12665-012-2093-1 [13] Chen Jun, Quan Wenting. An improved algorithm for retrieving chlorophyll-a from the Yellow River Estuary using MODIS imagery[J]. Environmental Monitoring and Assessment, 2013, 185(3): 2243−2255. doi: 10.1007/s10661-012-2705-y [14] Hao Qiang, Chai Fei, Xiu Peng, et al. Spatial and temporal variation in chlorophyll a concentration in the Eastern China Seas based on a locally modified satellite dataset[J]. Estuarine, Coastal and Shelf Science, 2019, 220: 220−231. doi: 10.1016/j.ecss.2019.01.004 [15] 王梦雪, 丁晓坤, 侯兴, 等. 渤海叶绿素a的时空分布及其影响因素[J]. 中国海洋大学学报, 2023, 53(9): 123−131.Wang Mengxue, Ding Xiaokun, Hou Xing, et al. Temporal and spatial variations of chlorophyll a and their influencing factors in the Bohai Sea[J]. Periodical of Ocean University of China, 2023, 53(9): 123−131. [16] Wang Jiru, Tang Jiakui, Wang Wuhua, et al. Quantitative retrieval of chlorophyll-a concentrations in the Bohai–Yellow sea using GOCI surface reflectance products[J]. Remote Sensing, 2023, 15(22): 5285. doi: 10.3390/rs15225285 [17] 王晓霞, 汪健平, 王佳莹, 等. 基于CNN-SVR网络的黄渤海近岸海域叶绿素a浓度预测[J]. 海洋预报, 2024, 41(4): 77−87.Wang Xiaoxia, WANG Jianping, WANG Jiaying, et al. Prediction of chlorophyll a concentration in nearshore waters of the Yellow and Bohai Seas based on CNN-SVR networks[J]. Marine Forecasts, 2024, 41(4): 77−87. [18] Yao Lulu, Wang Xiaopeng, Zhang Jiahua, et al. Prediction of sea surface chlorophyll-a concentrations based on deep learning and time-series remote sensing data[J]. Remote Sensing, 2023, 15(18): 4486. doi: 10.3390/rs15184486 [19] 李鹏. 黄河口及其邻近海域悬浮泥沙多尺度时空动态与驱动机制[D]. 上海: 华东师范大学, 2023.Li Peng. Multi-scale spatiotemporal dynamics and driving mechanisms of suspended sediment in the Yellow River Estuary and its adjacent seas[D]. Shanghai: East China Normal University, 2023. [20] 骆永明. 中国海岸带可持续发展中的生态环境问题与海岸科学发展[J]. 中国科学院院刊, 2016, 31(10): 1133−1142.Luo Yongming. Sustainability associated coastal eco-environmental problems and coastal science development in China[J]. Bulletin of Chinese Academy of Sciences, 2016, 31(10): 1133−1142. [21] Lorenzen C J. Determination of chlorophyll and pheo‐pigments: spectrophotometric equations[J]. Limnology and Oceanography, 1967, 12(2): 343−346. doi: 10.4319/lo.1967.12.2.0343 [22] Cai Lina, Bu Juan, Tang Danling, et al. Geosynchronous satellite GF-4 observations of Chlorophyll-a distribution details in the Bohai Sea, China[J]. Sensors, 2020, 20(19): 5471. doi: 10.3390/s20195471 [23] O'Reilly J E, Maritorena S, Mitchell B G, et al. Ocean color chlorophyll algorithms for SeaWiFS[J]. Journal of Geophysical Research: Oceans, 1998, 103(C11): 24937−24953. doi: 10.1029/98JC02160 [24] 王惠文, 孟洁. 多元线性回归的预测建模方法[J]. 北京航空航天大学学报, 2007, 33(4): 500−504.Wang Huiwen, Meng Jie. Predictive modeling on multivariate linear regression[J]. Journal of Beijing University of Aeronautics and Astronautics, 2007, 33(4): 500−504. [25] Cortes C, Vapnik V. Support-vector networks[J]. Machine Learning, 1995, 20(3): 273−297. [26] Breiman L. Random forests[J]. Machine Learning, 2001, 45(1): 5−32. doi: 10.1023/A:1010933404324 [27] 李明轩, 黎雷. 机器学习在水环境典型水质参数遥感反演中的应用[J]. 净水技术, 2024, 43(8): 45−53.Li Mingxuan, Li Lei. Application of machine learning in remote sensing retrieval of typical water quality parameters in aquatic environments[J]. Water Purification Technology, 2024, 43(8): 45−53. [28] 张亭禄, 贺明霞. 基于人工神经网络的一类水域叶绿素-a浓度反演方法[J]. 遥感学报, 2002, 6(1): 40−44.Zhang Tinglu, He Mingxia. A method to retrieve the oceanic chlorophyll-a concentrations in case I water based on artificial neural network[J]. Journal of Remote Sensing, 2002, 6(1): 40−44. [29] 何志昆, 刘光斌, 赵曦晶, 等. 高斯过程回归方法综述[J]. 控制与决策, 2013, 28(8): 1121−1129,1137.He Zhikun, Liu Guangbin, Zhao Xijing, et al. Overview of Gaussian process regression[J]. Control and Decision, 2013, 28(8): 1121−1129,1137. [30] Seeger M. Gaussian processes for machine learning[J]. International Journal of Neural Systems, 2004, 14(2): 69−106. doi: 10.1142/S0129065704001899 [31] Chen Tianqi, Guestrin C. XGBoost: a scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco: ACM, 2016: 785−794. [32] 傅春阳. 基于XGBoost模型的湖泊水体叶绿素a反演[D]. 阜新: 辽宁工程技术大学, 2023.Fu Chunyang. Retrieval of Chlorophyll-a in Lake water based on XGBoost model[D]. Fuxin: Liaoning Technical University, 2023. [33] 袁伟皓, 王华, 夏玉宝, 等. 基于GAM模型的鄱阳湖叶绿素a与水质因子相关性分析[J]. 生态环境学报, 2021, 30(8): 1716−1723.Yuan Weihao, Wang Hua, Xia Yubao, et al. Relationship of Chlorophyll A and water quality factors in Poyang Lake based on GAM model[J]. Ecology and Environmental Sciences, 2021, 30(8): 1716−1723. [34] Kim Y H, Im J, Ha H K, et al. Machine learning approaches to coastal water quality monitoring using GOCI satellite data[J]. GIScience & Remote Sensing, 2014, 51(2): 158−174. [35] 张乃星, 任荣珠, 吴凤丛, 等. 渤海海峡冬季营养盐的分布特征及影响因素[J]. 海洋通报, 2011, 30(6): 607−614.Zhang Naixing, Ren Rongzhu, Wu Fengcong, et al. Characteristics of vertical distribution and influence factors of the nutrients at Bohai Strait in winter[J]. Marine Science Bulletin, 2011, 30(6): 607−614. [36] Wang Yibo, Sun Yanyu, Wang Caixia, et al. Net-phytoplankton community structure and its environmental correlations in central Bohai Sea and the Bohai Strait[J]. Aquatic Ecosystem Health & Management, 2019, 22(4): 481−493. [37] 王英, 张晶晶, 吕其明, 等. 调水调沙对黄河口近海浮游植物群落结构时空分布影响[J]. 海洋环境科学, 2021, 40(3): 369−378.Wang Ying, Zhang Jingjing, Lv Qiming, et al. Effects of water and sediment regulation scheme on phytoplankton community and abundance in the Yellow River Estuary[J]. Marine Environmental Science, 2021, 40(3): 369−378. [38] 孟庆辉, 王林, 陈艳拢, 等. 2002—2021年渤海叶绿素a浓度变化及环境响应分析[J]. 中国环境监测, 2022, 38(6): 228−236.Meng Qinghui, Wang Lin, Chen Yanlong, et al. Change of Chlorophyll a concentration and its environmental response in the Bohai Sea from 2002 to 2021[J]. Environmental Monitoring in China, 2022, 38(6): 228−236. [39] 赵娜, 王霄鹏, 李咏沙, 等. 黄渤海海域叶绿素a浓度时空特征分布及影响因子分析[J]. 科学技术与工程, 2020, 20(17): 7101−7107.Zhao Na, Wang Xiaopeng, Li Yongsha, et al. Temporal-spatial distribution of chlorophyll-a and impacts of environmental factors in the Bohai Sea and Yellow Sea[J]. Science Technology and Engineering, 2020, 20(17): 7101−7107. [40] Shahriari B, Swersky K, Wang Ziyu, et al. Taking the human out of the loop: a review of Bayesian optimization[J]. Proceedings of the IEEE, 2016, 104(1): 148−175. doi: 10.1109/JPROC.2015.2494218 [41] Wang Yueqi, Liu Dongyan, Tang Danling. Application of a generalized additive model (GAM) for estimating chlorophyll-a concentration from MODIS data in the Bohai and Yellow Seas, China[J]. International Journal of Remote Sensing, 2017, 38(3): 639−661. doi: 10.1080/01431161.2016.1268733 [42] Liang Wen, Wang Yun, Liu Sumei, et al. Nutrient dynamics and coupling with biological processes and physical conditions in the Bohai Sea[J]. Frontiers in Marine Science, 2022, 9: 1025502. doi: 10.3389/fmars.2022.1025502 [43] Liang Wen, Wang Yan, Mu Jinglong, et al. Nutrient changes in the Bohai Sea over the past two decades[J]. Science of the Total Environment, 2023, 903: 166696. doi: 10.1016/j.scitotenv.2023.166696 -
图(9) / 表(4)
计量
- 文章访问数: 5
- HTML全文浏览量: 6
- PDF下载量: 0
- 被引次数: 0
