欢迎访问现代地质!

现代地质 ›› 2025, Vol. 39 ›› Issue (02): 420-428.DOI: 10.19657/j.geoscience.1000-8527.2024.128

• 表生资源观测模拟与综合评价 • 上一篇    下一篇

基于BP神经网络模型的黄海水体叶绿素a质量浓度反演

刘子华1(), 尤中义2, 刘玖芬3,4,5,6,7(), 刘晓煌5,6,7, 赵晓峰5,6,7, 李子奇4, 张文博5,6, 李洪宇5,6,7, 尹永会8, 石连武5,6()   

  1. 1.山东科技大学计算机科学与工程学院,山东 青岛 266000
    2.山东科技大学测绘与空间信息学院,山东 青岛 266000
    3.自然资源部生态地球化学重点实验室,北京 100037
    4.中国地质大学(北京)地球科学与资源学院, 北京 100083
    5.中国地质调查局自然资源综合调查指挥中心,北京 100055
    6.元素形态与新污染物检验检测技术创新中心,云南 昆明 650000
    7.自然资源要素耦合过程与效应重点实验室,北京 100055
    8.中国地质调查局烟台海岸带调查中心,北京 100055
  • 出版日期:2025-04-10 发布日期:2025-05-08
  • 通信作者: 石连武,男,高级工程师,1967出生,主要从事自然资源管理研究工作。Email:guofangban@163.com
    刘玖芬,女,正高级工程师,1971年出生,实验测试及地球化学专业,主要从事自然资源检测与观测技术研究研究工作。Email: 13863858360@163.com
  • 作者简介:刘子华,男,本科生,2002年出生,软件工程专业,主要从事神经网络机器学习研究工作。Email:1260759413@qq.com
  • 基金资助:
    自然资源部生态地球化学重点实验室开放基金项目(ZSDHJJ202303);中国地质调查局项目“金矿等战略性矿产实验测试技术支撑与服务”(DD20242769);“自然资源观测监测一体化技术体系研究”(DD20230514)

Retrieval of Chlorophyll-a Concentration in the Yellow Sea Based on a BP Neural Network Model

LIU Zihua1(), YOU Zhongyi2, LIU Jiufen3,4,5,6,7(), LIU Xiaohuang5,6,7, ZHAO Xiaofeng5,6,7, LI Ziqi4, ZHANG Wenbo5,6, LI Hongyu5,6,7, YIN Yonghui8, SHI Lianwu5,6()   

  1. 1. Shandong University of Science and Technology, College of Computer Science and Engineering,Qingdao,Shandong 266000, China
    2. Shandong University of Science and Technology,College of Geodesy and Geomatics, Qindao, Shandong 266000, China
    3. Key Laboratory of Eco-geochemistry, Ministry of Natural Resources, Beijing 100037, China
    4. China University of Geosciences (Beijing), School of Earth Sciences and Resources,Beijing 100083, China
    5. Comprehensive Survey Command Center of the China Geological Survey, Beijing 100055, China
    6. Center for Innovation in Elemental Speciation and New Pollutant Detection Technologies, Kunming,Yunnan 650000, China
    7. Key Laboratory of Coupled Processes and Effects of Natural Resources, Beijing 100055, China
    8. Yantai Coastal Zone Investigation Center, China Geological Survey, Beijing 100055, China
  • Published:2025-04-10 Online:2025-05-08

摘要:

海洋水体叶绿素a质量浓度是评估海洋生态环境与水质状况的重要指标之一,对于海洋资源管理和海洋保护具有重要意义。以黄海海域为研究对象,基于2022年VIIRS遥感数据,利用BP神经网络机器学习方法,构建了3隐藏层、4-6-4节点架构的神经网络模型,采用同时段多波段遥感反射率和海洋表面温度作为输入层数据,实现了黄海水体叶绿素a浓度反演模型;并与常用OC2模型、OC3模型、linear模型及cubic模型结果进行了比对研究,表明5种模型均可用于黄海海域叶绿素a浓度的分布反演,但本研究设计的模型反演结果误差更小(相对误差基本小于12%,平均相对误差8.2%),优于其它模型;为进一步评估模型性能,本研究运用综合评价指标对神经网络模型进行了全面评估,证明模型反演效果(MAE=0.122,RMSE=0.153,R2=0.937)满足预期。另外,使用研究区域2021年2月的VIIRS遥感数据与海洋表面温度数据进行了模型泛化能力实验,经过验证,本模型的泛化能力较好,可以完成同区域其他时段水体叶绿素a浓度反演任务。本研究证明基于VIIRS数据的BP神经网络模型监测海洋叶绿素a技术的可能性,并为相关实践和研究提供了基础。

关键词: 黄海, 机器学习, BP神经网络, 叶绿素a, 海表温度, 海洋遥感

Abstract:

The concentration of chlorophyll-a in marine waters is one of the key indicators for assessing the ecological environment and water quality of oceans, which plays a crucial role in marine resource management and protection.Focusing on the Yellow Sea, this study constructs a chlorophyll-a concentration inversion model based on 2022 VIIRS remote sensing data using a BP neural network machine learning method.The model features a three-hidden-layer architecture with 4-6-4 nodes and takes multi-band remote sensing reflectance and sea surface temperature during the same period as input data.The results are compared with commonly used models, such as OC2, OC3, linear, and cubic models.The comparison shows that all five models can be used for chlorophyll-a concentration inversion in the Yellow Sea, but the model designed in this study exhibits smaller inversion errors (with relative errors generally less than 12%, and an average relative error of 8.2%), outperforming the other models.To further assess the model’s performance, a comprehensive evaluation index was applied, demonstrating that the inversion results (MAE=0.122, RMSE=0.153, R2=0.937) meet the expected accuracy.Additionally, the model’s generalization ability was tested using VIIRS remote sensing data and sea surface temperature data from February 2021.The validation shows that the model has good generalization ability and can be used for inversion tasks in other periods within the same region.This study demonstrates the feasibility of using a BP neural network model based on VIIRS data for monitoring ocean chlorophyll-a, which provides a foundation for related practices and research.

Key words: Yellow Sea, machine learning, backpropagation neural network, chlorophyll-a, sea surface temperature, marine remote sensing

中图分类号: