【摘要】 为证明间接遥感反演总磷的可操作性，明确总磷与光敏水质参数之间的定量关系，依据山东省南四湖、云蒙湖、崂山水库、峡山水库2021—2024年的国控地表水断面监测数据，先使用统计分析方法进行总磷与光敏水质参数的相关分析，明确相关程度，然后采用分类与回归树（CART）和随机森林2种机器学习方法构建定量关系模型。结果表明：（1）总磷与叶绿素、藻密度、浊度、电导率、水温的斯皮尔曼（Spearman）相关程度平均占比分别为62.50%,64.06%,75.00%,73.44%,75.00%,说明总磷与这5种光敏参数具有较强的相关性。（2）使用2种机器学习方法构建季节和区域差异下的定量关系模型，CART模型的均方误差（MSE）均值和决定系数（R²）均值分别为0.3×10^-5～29×10^-5和0.361～0.832,随机森林模型的MSE和R²分别为0.04×10^-5～8.3×10^-5和0.781～0.967,证明了机器学习方法的有效性，且随机森林模型的性能更优。（3）在不同季节、不同区域条件下，不同光敏参数对定量关系建模的贡献程度不同。本研究探明了湖库总磷与光敏水质参数的显著相关性，验证了使用机器学习方法构建二者定量关系模型的可行性，为间接遥感反演总磷提供了方法支撑与科学依据。更多还原

【Abstract】 In order to prove the operability of indirect remote sensing inversion of total phosphorus and clarify the quantitative relationship between total phosphorus and photosensitive water quality parameters, based on the national surface water section monitoring data of Nansi Lake, Yunmeng Lake, Laoshan Reservoir and Xiashan Reservoir from 2021 to 2024, the correlation analysis between total phosphorus and photosensitive water quality parameters was first conducted using statistical analysis methods to clarify the degree of correlation, then the quantitative relationship models were constructed by using Classification and Regression Tree（CART） and Random Forest machine learning methods. Results showed that:（1） The mean proportion of Spearman correlation degree between total phosphorus and chlorophyll, algal density, turbidity, electrical conductivity, and water temperature was 62.50%, 64.06%, 75.00%, 73.44%, and 75.00%, respectively, indicating that total phosphorus had a strong correlation with the five photosensitive parameters.（2） When constructing quantitative relationship models using the two machine learning methods under seasonal and regional differences, the mean distribution intervals of mean-square error（MSE） and R² were 0.3×10^-5～29×10^-5,0.361～0.832 and 0.04×10^-5～8.3×10^-5,0.781～0.967 respectively, which proved that machine learning methods were effective and the performance of the Random Forest model was better.（3） Under different regional and seasonal conditions, the contribution of different photosensitive water quality parameters to the quantitative relationship modeling was different, and the contribution of electrical conductivity was the greatest. This study has clarified the significant correlation between total phosphorus and photosensitive water quality parameters in lakes and reservoirs, verified the feasibility of using machine learning methods to construct a quantitative relationship model between the two, and provided methodological support and scientific basis for the indirect remote sensing inversion of total phosphorus.更多还原