基于BP神经网络的东北寒区河流水生态安全评估以安肇新河流域为例

Aquatic ecological security assessment of rivers in the cold regions of Northeast China based on BP neural networks: a case study of the Anzhaoxinhe River Basin

  • 摘要: 为科学评估东北寒区河流水生态安全状况,以安肇新河流域为研究对象,结合寒区河流生态系统具有多因子耦合、时空变化显著等特点,构建包含社会经济影响、水生态系统健康、生态服务功能、调控管理共4类37项指标的综合评价体系,其中特别纳入流速、昼夜温差、盐碱度等寒区特色指标。引入BP神经网络模型,利用2022—2023年实测监测数据对模型进行训练与验证,系统评估流域水生态安全状况及其关键影响因子。结果表明:安肇新河流域生境质量总体一般(栖息地质量指数均值111.44),水质达标状况不稳定,底质有机污染整体较轻但存在局部与季节性高值;水生生物多样性偏低,群落结构趋于退化,耐污种类占优势。所构建的BP神经网络模型拟合与预测效果良好(训练集R2>0.90,预测集R2>0.85),能够有效反映水生态安全的时空变化特征,适用于寒区河流评估需求。流域生态健康状况季节性差异显著,春季最佳(88.89%点位属于“一般”等级),冬季最差(评价结果为“一般”或“差”),关键影响因子为水体CODCr及沉积物有机质含量。本研究可为寒区河流水生态安全评估提供基于神经网络评价的新途径,也可为安肇新河流域的生态治理与管控提供依据。

     

    Abstract: To scientifically assess the aquatic ecological security status of rivers in the cold regions of Northeast China, this study took the Anzhaoxinhe River Basin as a case study. A comprehensive evaluation system was constructed, taking into account the characteristics of cold-region river ecosystems, such as multi-factor coupling and significant spatiotemporal variations. This system included 37 indicators across four categories (socio-economic impact, aquatic ecosystem health, ecological service functions, and regulatory management), incorporating specific cold-region characteristics such as flow velocity, diurnal temperature variation, and salinity-alkalinity. On this basis, the BP neural network model was introduced, and the model was trained and validated by using monitoring data from 2022 to 2023, to systematically assess the aquatic ecological security status of the basin and its key influencing factors. The research results showed that the overall habitat quality of the basin was fair (with a mean Habitat Quality Index of 111.44), water quality compliance was unstable, and the organic pollution of the bottom sediment was generally light with local and seasonal peaks. Aquatic biodiversity was low, and the community structure was deteriorating, with pollution-tolerant species being dominant. The constructed BP neural network model performed well in training and prediction (R2 of the training set > 0.90, R2 of the prediction set > 0.85), accurately capturing the spatiotemporal dynamics of aquatic ecological security, and demonstrating its applicability in the assessment of cold-region rivers. The river basin's ecological health status exhibited significant seasonal differences, with the best conditions in spring (88.89% of sampling sites rated as "fair") and the worst in winter (all sites rated as "fair" or "poor"). The key influencing factors were identified as the CODCr concentration in water and the organic matter content in sediments. This study provides a new neural network-based method for aquatic ecological security assessment in cold-region rivers and offers data and a scientific basis for the ecological management and governance of the Anzhaoxinhe River Basin.

     

/

返回文章
返回