Scholarlink

<p>Accurate meteorological drought forecasting is critical for sustainable water management and agricultural development in the Yellow River Basin (YRB), where increasing climate variability intensifies drought risk. However, existing models often fail to capture the complex and nonlinear relationships between drought indicators and climatic drivers. This study improves meteorological drought forecasting in the YRB by integrating multiple input strategies into five machine learning models: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Backpropagation Neural Network (BP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). Meteorological drought was quantified using the Standardized Precipitation Evapotranspiration Index (SPEI) at five time scales (SPEI-1, -3, -6, -9, and -12). Three input patterns were designed, including one univariate mode (Input-1) and two multivariate modes (Input-2 and Input-3). Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Nash-Sutcliffe Efficiency (NSE), with Taylor diagrams and violin plots used for visual comparison. Results show that shifting from univariate to multivariate inputs greatly improves forecasting accuracy across all models, with Inputs-2 and -3 performing comparably. Under univariate input, no single model achieved consistent superiority, whereas LSTM and GRU performed best with multivariate inputs. Forecasting accuracy varied among climatic regions, showing lower accuracy for SPEI-6, -9, and -12 in semi-arid zones (RMSE = 0.6) than in arid (RMSE = 0.8) and semi-humid regions (RMSE = 0.82). Overall, multivariate machine learning approaches markedly enhance drought predictability in the YRB and provide guidance for regional drought early warning and sustainable water management.</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

Improving the accuracy of meteorological drought forecasting in the Yellow River Basin of China based on multivariate machine learning

  • 期刊论文

  • 发表日期：

    2026年2月4日

查看全文
  • Dehe Xu,
  • Yaoning Wang,
  • Hengbin Zhang,
  • Nan Jia,
  • Li Guo,
  • Dongbao Zhao,
  • Ke Zhang

摘要

Accurate meteorological drought forecasting is critical for sustainable water management and agricultural development in the Yellow River Basin (YRB), where increasing climate variability intensifies drought risk. However, existing models often fail to capture the complex and nonlinear relationships between drought indicators and climatic drivers. This study improves meteorological drought forecasting in the YRB by integrating multiple input strategies into five machine learning models: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Backpropagation Neural Network (BP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). Meteorological drought was quantified using the Standardized Precipitation Evapotranspiration Index (SPEI) at five time scales (SPEI-1, -3, -6, -9, and -12). Three input patterns were designed, including one univariate mode (Input-1) and two multivariate modes (Input-2 and Input-3). Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Nash-Sutcliffe Efficiency (NSE), with Taylor diagrams and violin plots used for visual comparison. Results show that shifting from univariate to multivariate inputs greatly improves forecasting accuracy across all models, with Inputs-2 and -3 performing comparably. Under univariate input, no single model achieved consistent superiority, whereas LSTM and GRU performed best with multivariate inputs. Forecasting accuracy varied among climatic regions, showing lower accuracy for SPEI-6, -9, and -12 in semi-arid zones (RMSE = 0.6) than in arid (RMSE = 0.8) and semi-humid regions (RMSE = 0.82). Overall, multivariate machine learning approaches markedly enhance drought predictability in the YRB and provide guidance for regional drought early warning and sustainable water management.