| 摘要: |
| 叶绿素a浓度是表征水体富营养化状态的关键指标。为构建适用于青海湖的叶绿素a浓度最优遥感反演模型，本研究基于Landsat-8 OLI遥感影像数据，结合同期叶绿素a浓度实测数据，构建并对比了多种青海湖叶绿素a浓度反演模型，分析了2022-2024年春、夏、秋季叶绿素a浓度的时空分布特征。结果表明：Landsat-8 OLI影像的B6、B7波段反射率与青海湖叶绿素a浓度相关性较高，波段组合（B3+B6）-（B4*B5）与叶绿素a浓度的相关性进一步提升。通过对比分析，随机森林（RF）模型相较于传统经验统计模型以及其他机器学习模型（支持向量机SVM、最小二乘支持向量机LSSVM），表现出最佳的反演精度（测试集：R2=0.710，RMSE=0.052，MAE=0.037）。时空反演结果显示，2022-2024年，青海湖叶绿素a浓度在空间分布上，布哈河入湖口始终维持低浓度特征，而湖区其他区域则呈现显著的季节异质性；时序变化上，2022年夏季、2023年秋季和2024年春、夏两季的叶绿素a高值区覆盖范围显著大于其他时期。综上可知，基于机器学习算法的RF模型在高原湖泊叶绿素a浓度的反演方面具有良好的应用效果，为高原湖泊水体富营养化监测与管理提供了可靠的技术支持。 |
| 关键词: 叶绿素a浓度 青海湖 遥感反演 机器学习算法 |
| DOI： |
| 投稿时间：2025-05-19修订日期：2025-06-26 |
| 基金项目:青海省跨高校中试研发能力建设项目(2022ZY042)资助；青海湖水环境健康安全诊断与评估技术研究项目(k991933)资助。 |
|
| Inversion of chlorophyll a concentration in Qinghai Lake based on Landsat 8 remote sensing images |
|
Shi Jing, Niu Hailin, Xie Chaoyang, MA Xujie, CHEN Shengcun
|
| (School of Eco-Environmental Engineering,Qinghai University,Xining,Qinghai) |
| Abstract: |
| Chlorophyll-a concentration serves as a critical indicator for assessing the eutrophication status of water bodies. This study aimed to develop an optimal remote sensing inversion model for chlorophyll-a concentration applicable to Qinghai Lake. By utilizing Landsat-8 OLI remote sensing image data and integrating it with corresponding measured chlorophyll-a concentration data, various inversion models were constructed and compared for estimating chlorophyll-a concentration in Qinghai Lake. The spatiotemporal distribution characteristics of chlorophyll-a concentration during spring, summer, and autumn from 2022 to 2024 were analyzed. Results demonstrated that the reflectance of the B6 and B7 bands in Landsat-8 OLI images exhibited a strong correlation with chlorophyll-a concentration in Qinghai Lake. Furthermore, the band combination formula (B3+B6) - (B4*B5) significantly enhanced the correlation with chlorophyll-a concentration. Through comparative analysis, the Random Forest (RF) model demonstrated superior performance compared to traditional empirical statistical models and other machine learning models (Support Vector Machine SVM, Least Squares Support Vector Machine LSSVM) in terms of inversion accuracy (test set: R2 = 0.710, RMSE = 0.052, MAE = 0.037). The spatiotemporal inversion results revealed that during 2022–2024, chlorophyll-a concentration at the Buha River inflow area remained consistently low, whereas other regions within the lake exhibited significant seasonal heterogeneity. In terms of temporal variations, high-value areas of chlorophyll-a concentration in summer 2022, autumn 2023, and spring and summer 2024 were notably larger than during other periods. In conclusion, the RF model based on machine learning algorithms proved effective for chlorophyll-a concentration inversion in plateau lakes, providing robust technical support for monitoring and managing water body eutrophication in such environments. |
| Key words: chlorophyll a concentration Qinghai Lake remote sensing inversion Machine learning algorithms |
