| 引用本文: | 李江,王杰,崔玉环,张耀波,晏实江.基于特征优化的随机森林模型探究河岸带景观对入湖河流总氮浓度的影响.湖泊科学,2025,37(4):1290-1301. DOI:10.18307/2025.0423 |
| Li Jiang,Wang Jie,Cui Yuhuan,Zhang Yaobo,Yan Shijiang.Effects of riparian zone landscape on riverine total nitrogen concentrations using a feature-optimized random forest model. J. Lake Sci.2025,37(4):1290-1301. DOI:10.18307/2025.0423 |
|
| 摘要: |
| 河岸带在拦截地表污染物进入河流方面起着重要作用，探究河岸带景观特征对河流总氮（TN）浓度的影响机制对河流水质调控尤为重要。以受面源污染控制的巢湖入湖河流为研究对象，根据实测水质数据与同期Sentinel-2 MSI影像构建机器学习模型反演河流TN浓度，引入递归特征消除算法优化景观特征指标，利用随机森林回归模型探究不同宽度河岸带景观对河流TN浓度的影响，确定影响河流TN浓度的最有效河岸带宽度与关键景观指标。研究表明：（1）适用于巢湖入湖河流的TN浓度反演模型为梯度提升回归树，其反演精度 R 2 、均方误差和平均绝对百分误差分别达到0.93、0.35 mg/L和28.86%；（2）与传统的冗余分析等方法相比，本文将递归特征消除算法与随机森林回归模型相结合能更为有效捕捉河岸带景观与河流水质间的复杂非线性关系，其拟合优度R 2 超过了0.87；（3）在干、湿季影响巢湖入湖河流TN浓度的最有效河岸带宽度分别为1500、1000 m，关键景观指标为农田破碎度、城镇面积比例、景观破碎度以及植被覆盖度。建议在上述有效河岸带宽度范围减少农田破碎度、城镇面积比例、景观破碎度并提高植被覆盖度，以降低入湖河流及湖体TN浓度。本研究可为探究地表景观对河流水质的影响机制与受面源污染控制的河流水污染防治提供有效方法与科学依据。 |
| 关键词: 河岸带 随机森林模型 特征优化 景观指标 总氮 巢湖 |
| DOI：10.18307/2025.0423 |
| 分类号: |
| 文献标识码:A |
| 基金项目:国家自然科学基金项目（32171573）、安徽省自然科学基金项目（2308085MD114,2308085Y29）和安徽省重点研究与开发项目（2022l07020027）联合资助 |
|
| Effects of riparian zone landscape on riverine total nitrogen concentrations using a feature-optimized random forest model |
|
Li Jiang1，Wang Jie1，2，Cui Yuhuan3，Zhang Yaobo4，Yan Shijiang1
|
|
1:College of Resources and Environmental Engineering, Anhui University, Hefei 230601 , P.R.China,2:Anhui Province Key Laboratory of Wetland Ecosystem Protection and Restoration, Anhui University, Hefei 230601 , P.R.China,3:College of Resources and Environment, Anhui Agricultural University, Hefei 230036 , P.R.China,4:Anhui Provincial Bureau of Surveying and Mapping, Hefei 230031 , P.R.China
|
| Abstract: |
| Riparian zone plays an important role in intercepting surface pollutants from entering rivers, so it is particularly important to explore the influence of riparian landscape on river total nitrogen(TN) concentration for the regulation of river water quality. However, it is difficult to quantitatively analysis the complex relationship between riparian landscape and river water quality to identify the key landscape metrics and optimal riparian strip. Taking the inlet rivers of Lake Chaohu controlled by non-point source pollution as the study area, a remote sensing inversion model was firstly constructed to retrieval TN concentration in rivers using the machine learning regression algorithms according to the measured water quality data and the synchronous Sentinel-2 MSI images, then the recursive feature elimination algorithm was introduced to optimize the landscape indices, and a new random forest regression model was lastly constructed to explore the influence of different width riparian landscape on river TN concentration, to determine the most effective riparian zone width and key landscape indices affecting river TN concentration. Results showed that (1) The retrieval model suitable to TN concentration in the inlet rivers of Lake Chaohu was gradient boosted regression model, and its inversion accuracy of R 2, mean squared error and mean absolute percentage error reached 0.93, 0.35 mg/L and 28.86%, respectively. (2) Compared with the traditional methods such as redundancy analysis (RDA), combining the recursive feature elimination with random forest regression algorithms was more effective method to capture the complex nonlinear relationships between landscape and water quality, with the goodness of fit R 2 ＞0.87. (3) The most effective widths of riparian zone for influencing river TN concentrations in the dry and wet seasons were 1500 m and 1000 m, respectively, and the key landscape metrics contained the farmland fragmentation, proportion of urban and town, landscape fragmentation and vegetation coverage. It is suggested to reduce the farmland fragmentation, urban construction proportion, landscape fragmentation and improve the vegetation coverage in the effective width of riparian zone, so as to reduce the TN concentration in the river and the lake. Our study can provide an effective method and scientific basis for investigating the influence of surface landscape on river water quality and the prevention of river water pollution controlled by non-point source pollution in the agricultural basin. |
| Key words: Riparian zone random forest model feature optimization landscape metrics total nitrogen Lake Chaohu |
