基于Sentinel-2A影像的洪泽湖叶绿素a浓度反演分析
DOI:
作者:
作者单位:

江苏省水文水资源勘测局淮安分局

作者简介:

通讯作者:

中图分类号:

x143

基金项目:


Analysis of chlorophyll-a concentration retrieval in Hongze Lake based on Sentinel-2A imagery
Author:
Affiliation:

Huaian Branch of Jiangsu Province Hydrology and Water Resources Investigation Bureau,

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对洪泽湖叶绿素a浓度反演缺少较高分辨率遥感数据支撑、其他湖泊建立的模型因洪泽湖自身水体悬浮物高而导致的泛化能力差的难题,基于洪泽湖2018—2023年间长序列实测叶绿素a浓度以及同期Sentinel-2A遥感影像等数据,采用波谱特征分析、波段相关性分析等方法获得反演分析的最佳波段组合,构建一元线性、二次多项式、指数、幂函数和对数等统计回归模型以及LS-SVM、RF模型、BP神经网络等机器学习模型;通过对比分析不同模型的反演预测精度,探明适用于洪泽湖的最佳反演模型;结合反演结果揭示洪泽湖分区域叶绿素a浓度时空变化规律及影响因子。结果表明:(1)洪泽湖因悬浮物含量高,光谱曲线在整个可见-近红外波段均呈现较高的反射率,基于窄近红外NDVI指数在洪泽湖水体中与叶绿素a浓度的相关性最好。(2)统计回归模型中指数函数模型预测精度最优,RMSE为0.0075 mg·L-1,MAE为0.0058 mg·L-1;机器学习模型中LS-SVM模型预测精度最优,RMSE和MAE分别为0.0061 mg·L-1和0.0049 mg·L-1,预测精度比RF模型、BP神经网络模型均有显著提升;LS-SVM模型优于统计回归模型。(3)LS-SVM模型反演结果显示洪泽湖叶绿素a浓度空间分布上呈北部成子湖区和西部溧河洼区浓度较高、东部过水区浓度较低特征,丰水期空间差异较为显著;气温以及降雨对成子湖区和溧河洼区叶绿素a浓度影响显著,水位变化对其影响较小,出入湖水力交换是东部过水区叶绿素a浓度的抑制因子。

    Abstract:

    To address the challenges of chlorophyll-a(Chl-a) concentration inversion in Lake Hongze—specifically, the lack of high-resolution remote sensing data support and the poor generalization ability of models developed for other lakes due to Lake Hongze’s high suspended particulate matter (SPM) content.This study utilized long-term in-situ Chl-a concentration data (2018–2023) and concurrent Sentinel-2A remote sensing images. Spectral feature analysis and band correlation analysis were employed to identify the optimal band combination for inversion. Statistical regression models (including linear, quadratic polynomial, exponential, power, and logarithmic functions) and machine learning models (LS-SVM, RF, and BP neural network) were constructed. By comparing the inversion accuracy of different models, the optimal inversion model for Lake Hongze was determined. Additionally, the spatiotemporal variation patterns of Chl-a concentration and influencing factors in different lake regions were analyzed based on inversion results.The findings revealed:(1) Due to its high SPM content, Lake Hongze exhibits elevated reflectance across the visible to near-infrared (VNIR) spectrum. Among spectral indices, the narrow near-infrared NDVI showed the strongest correlation with Chl-a concentration.(2) Among statistical regression models,the exponential function model achieved the highest prediction accuracy (RMSE: 0.0075 mg·L-1, MAE: 0.0058 mg·L-1). Among machine learning models, the LS-SVM model performed best (RMSE: 0.0061 mg·L-1, MAE: 0.0049 mg·L-1), significantly outperforming RF and BP neural network models.The LS-SVM model also surpassed all statistical regression models in accuracy.(3) The LS-SVM inversion results indicated that Chl-a concentrations were higher in the northern Chengzi Lake area and western Lihewa area but lower in the eastern water-transfer zone, with pronounced spatial heterogeneity during the wet season. Temperature and rainfall significantly influenced Chl-a concentrations in Chengzi Lake and Lihewa, whereas water level had minimal impact. Hydrological exchange with inflow/outflow rivers acted as an inhibitory factor for Chl-a in the eastern water-transfer zone.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2025-03-11
  • 最后修改日期:2025-07-07
  • 录用日期:2025-07-07
  • 在线发布日期:
  • 出版日期: