[Evaluation and Prediction of Water Quality in Beijing Section of Yongding River Based on CEWI Index and BP Neural...

Aquatic communities serve as critical indicators for evaluating the health of aquatic ecosystems. However, challenges such as technical complexity and high costs in the collection and identification of aquatic organisms hinder assessment efficiency. This study proposes a deep learning-based predictive model for water quality indices to enhance evaluation timeliness and universality. From autumn 2020 to summer 2021, four aquatic ecological surveys were conducted at 16 monitoring points in the Beijing section of the Yongding River, identifying 118 macroinvertebrate species, 159 zooplankton species, and 107 phytoplankton species. By integrating multi-group biodiversity, the comprehensive ecological water quality index (CEWI) was constructed, revealing an overall water quality status of β-moderate pollution in the study area. Canonical Correspondence Analysis (CCA) identified water temperature (WT), pH, flow velocity (CV), water depth (WD), and dissolved oxygen (DO) as key environmental drivers of community structure. A BP neural network model was developed to predict the CEWI index, achieving an overall R² of 0.978, a Mean Square Error (MSE) of 0.106, and a Mean Absolute Error (MAE) of 0.262, thereby validating the model's effectiveness. Through the "environmental factor-biological response-model prediction" framework, this study provides data-driven insights and methodological innovations for ecological restoration in the Yongding River Basin, demonstrating significant practical value and regional guidance significance.