| 摘要: |
| 叶绿素a浓度是评估水质状况的重要参数之一, 然而因为近海的二类水体光谱特征复杂, 影响了其中叶绿素a浓度反演的可靠性。本文提出了一种动态K-means聚类与粒子群最优化算法(Particle Swarm Optimization, PSO)优化的径向基神经网络反演模型, 用以提高叶绿素a浓度反演的精度。以Sentinel-2多光谱成像仪作为遥感数据源, 选择与香港近海的叶绿素a浓度监测点时间相同、云覆盖率低于10% 的遥感影像, 并提取对应叶绿素a浓度监测点的遥感反射率, 并与其进行相关性分析。在此基础上, 选择具有较高相关性的波段及波段组合B2、1/B2、1/B3、B2-B4, 构建优化后的RBF模型, 并与传统经验模型和传统RBF模型作为比较。结果表明: 优化的RBF模型决定系数为0.90, 均方根误差为0.23 μg·L-1, 两种算法优化后的RBF模型对近海二类水体中叶绿素a浓度反演的可靠性远远高于传统经验模型和传统RBF模型。同时反演结果显示香港近海海域叶绿素a浓度从西到东呈现低-高-低的空间分布, 局部上呈现近岸高、外海逐渐降低的特征。 |
| 关键词: 叶绿素a浓度 RBF神经网络 动态K-means聚类 粒子群最优化算法 香港近海 |
| DOI：10.11759/hykx20250118001 |
| 分类号:X87 |
| 基金项目:国家重点研发计划项目(2022YFD2401304) |
|
| Inversion of chlorophyll a concentration in Hong Kong waters based on Sentinel-2 satellite and optimized RBF model |
|
ZHANG Lei1, ZHAO Kuan1, WEI Lai1, GUAN Shoude1,2, ZHAO Wei1,2
|
|
1.Key Laboratory of Ocean Observation and Information of Hainan Province, Sanya Oceanographic Institution, Ocean University of China, Sanya 572024, China;2.Physical Oceanography Laboratory, Qingdao 266100, China
|
| Abstract: |
| Chlorophyll-a concentration is a crucial parameter for the aquatic environment. However, because of the complexity of the spectral characteristics of offshore Case-Ⅱ water, which affects the reliability of the inversion of chlorophyll-a concentration in them. To assess the suitability of the radial basis function (RBF) neural network, enhanced by the dynamic K-means clustering and particles swarm optimization, for chlorophyll-a concentration inversion in Case-Ⅱ water, Sentinel-2 multispectral image remote sensing data were employed. Hong Kong offshore served as the study area. Sampling points with matching remote sensing image data were chosen based on consistent chlorophyll-a collection times, ensuring cloud coverage rates below 10%. Remote sensing image data underwent preprocessing to obtain reflectance values aligned with the monitoring dates. On this basis, bands with high correlation and their combinations B2, 1/B2, 1/B3 and B2-B4 were selected to construct the optimized RBF model, and compared with the traditional empirical model and the traditional RBF model. Results indicated an R2 value of 0.90 for the optimized RBF model, surpassing the traditional RBF models and empirical models. Additionally, the optimized RBF model’s suitability for chlorophyll-a concentration inversion in Case-Ⅱ water was confirmed. Using the trained and optimized RBF model, chlorophyll-a concentration inversion in Hong Kong’s offshore waters was executed using Sentinel-2 MSI data. The spatial distribution exhibited a pattern of low-high-low from west to east. Notably, certain areas within Hong Kong offshore waters displayed higher chlorophyll-a concentrations compared to the surrounding external waters. |
| Key words: chlorophyll-a concentration RBF neural network dynamic K-means clustering algorithm particle swarm optimization Hong Kong offshore waters |
