机器学习在水环境典型水质参数遥感反演中的应用

Application of Machine Learning in Remote Sensing Retrieval of Typical Water Quality Parameters in Aquatic Environments

遥感技术是一种可用于大面积水体长时序监测的有效方法,研究综述了机器学习方法在几种典型水质参数遥感反演中的应用。首先,简述了水质反演中几种常用机器学习算法的原理和优缺点。随后,介绍了机器学习模型反演叶绿素a、悬浮物质、溶解性有机质、磷和氮5种参数的研究进展,并进一步分析了面临的问题和挑战。在此基础之上,进行了总结和展望:(1)机器学习模型的反演效果普遍优于传统经验公式和半经验模型;(2)具有量化反演不确定性能力的机器学习模型(如混合密度网络和贝叶斯神经网络等),提供了更为全面和可靠的预测;(3)基于全球性大样本数据集构建的机器学习模型具有较好的泛化能力,存在产品化潜力;(4)未来的工作应主要集中于不确定性估计算法和迁移学习的推广、大气校正算法的评估,以及水环境遥感大数据的发展等。

Remote sensing technology is a highly effective method for long-term monitoring of large-scale water bodies.The application of machine learning techniques in remote sensing retrieval of typical water quality parameters is reviewed in this study.Initially,the principles,advantages,and disadvantages of several commonly used machine learning algorithms for water quality retrieval are briefly described.Subsequently,the research progress of machine learning models in estimating several key parameters,including chlorophylla,suspended matter,dissolved organic matter,phosphorus and nitrogen,is discussed.Additionally,the challenges and issues encountered in this field are analyzed.Based on this review,the following conclusions and prospects are proposed.(1)Machine learning models generally outperform traditional empirical and semi-empirical models in terms of retrieval accuracy.(2)Models capable of quantifying retrieval uncertainty,such as mixture density networks and Bayesian probabilistic neural networks,offer more comprehensive and reliable retrieval.(3)Machine learning models developed from extensive global datasets demonstrate good generalization capabilities and hold potential for productization.(4)Future research should focus on popularizing uncertainty estimation algorithms and transfer learning,evaluating atmospheric correction algorithms,and developing big data applications for remote sensing of aquatic environments.