Scholarlink

<p>Precise runoff forecasting is important for regional water resources planning. Reliable prediction remains difficult in the middle and lower Yangtze River because traditional hydrological models often underperform, while many machine learning methods act as black boxes. We propose a two-stage long short-term memory (LSTM) approach for multi-step runoff forecasting. The framework integrates the Xinanjiang (XAJ) model, LSTM, ensemble empirical mode decomposition (EEMD), and error correction (EC). This design addresses compatibility issues between hydrological processes and signal decomposition, improving predictive accuracy while retaining physical interpretability. For a 7-day lead time, EC-X-LSTM improves correlation coefficient (r, + 4.63%), Nash–Sutcliffe Efficiency (NSE, + 9.80%), Kling–Gupta Efficiency (KGE, + 2.93%), and Willmott Index (WI, + 2.27%). It also reduces Root Relative Mean Square Error (RRMSE, –54.26%) and Mean Absolute Percentage Error (MAPE, –52.41%). SHAP analysis shows that hybrid model output, dew point temperature, maximum temperature, mean temperature, and reference evapotranspiration are the top five predictors at Datong station. Overall, the method provides a reliable and interpretable tool for short- and medium-term runoff forecasting, offering practical insights for water management in the Yangtze River basin.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of a Two-Stage LSTM for Multi-Step Runoff Forecasting Using a XAJ Model and EEMD

  • 期刊论文

  • 发表日期：

    2026年1月21日

查看全文
  • Zihao Yang,
  • Qianjin Dong,
  • Xu Zhang,
  • Hongyu Zhu,
  • Zhetao Cheng

摘要

Precise runoff forecasting is important for regional water resources planning. Reliable prediction remains difficult in the middle and lower Yangtze River because traditional hydrological models often underperform, while many machine learning methods act as black boxes. We propose a two-stage long short-term memory (LSTM) approach for multi-step runoff forecasting. The framework integrates the Xinanjiang (XAJ) model, LSTM, ensemble empirical mode decomposition (EEMD), and error correction (EC). This design addresses compatibility issues between hydrological processes and signal decomposition, improving predictive accuracy while retaining physical interpretability. For a 7-day lead time, EC-X-LSTM improves correlation coefficient (r, + 4.63%), Nash–Sutcliffe Efficiency (NSE, + 9.80%), Kling–Gupta Efficiency (KGE, + 2.93%), and Willmott Index (WI, + 2.27%). It also reduces Root Relative Mean Square Error (RRMSE, –54.26%) and Mean Absolute Percentage Error (MAPE, –52.41%). SHAP analysis shows that hybrid model output, dew point temperature, maximum temperature, mean temperature, and reference evapotranspiration are the top five predictors at Datong station. Overall, the method provides a reliable and interpretable tool for short- and medium-term runoff forecasting, offering practical insights for water management in the Yangtze River basin.