Scholarlink

<p>Urban water security is increasingly threatened by climate variability and intensifying droughts, posing critical challenges to food production, ecosystem services, and sustainable city development—particularly in monsoon-influenced transition zones. To improve the accuracy and reliability of meteorological drought forecasting in urban contexts, this study proposes a hybrid VMD–BiLSTM–XGBoost model that synergistically integrates variational mode decomposition (VMD), deep learning, and ensemble machine learning. The model first decomposes the non-stationary Standardized Precipitation Evapotranspiration Index (SPEI) into a set of intrinsic mode functions (IMFs) using VMD, effectively disentangling multi-scale drought signals. Low-frequency components—reflecting persistent climate trends—are predicted by a bidirectional long short-term memory (BiLSTM) network, while high-frequency fluctuations—associated with short-term meteorological variability—are captured by XGBoost. The approach is evaluated using daily multi-scale SPEI datasets from three urban meteorological stations in Henan Province (Anyang, Zhengzhou, and Xinyang), which span a pronounced climatic gradient across China’s urbanizing monsoon transition region. Results demonstrate that the VMD–BiLSTM–XGBoost model consistently outperforms benchmark alternatives (including VMD–ARIMA, VMD–LSTM, and VMD–BiLSTM) across all sites and time scales. Notably, it achieves an R² of 0.997 and an MSE of 0.001 for 12-month SPEI forecasting at the Anyang station, with average R² values exceeding 0.97 across all three cities. These findings highlight the model’s superior capability in capturing extreme drought events and forecasting multi-scale drought dynamics with high fidelity. By delivering robust, high-resolution drought projections, this method offers a valuable decision-support tool for enhancing urban climate resilience, informing adaptive water governance, and advancing drought early-warning systems in rapidly urbanizing regions vulnerable to hydroclimatic extremes.</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

A hybrid VMD–BiLSTM–XGBoost approach for multi-scale drought forecasting in urbanizing monsoon transition zones

  • 期刊论文

  • 发表日期：

    2026年4月6日

查看全文
  • Zhe Zhu,
  • Weihua Lin,
  • Fujiang Liu,
  • Yan Guo,
  • Bo Li

摘要

Urban water security is increasingly threatened by climate variability and intensifying droughts, posing critical challenges to food production, ecosystem services, and sustainable city development—particularly in monsoon-influenced transition zones. To improve the accuracy and reliability of meteorological drought forecasting in urban contexts, this study proposes a hybrid VMD–BiLSTM–XGBoost model that synergistically integrates variational mode decomposition (VMD), deep learning, and ensemble machine learning. The model first decomposes the non-stationary Standardized Precipitation Evapotranspiration Index (SPEI) into a set of intrinsic mode functions (IMFs) using VMD, effectively disentangling multi-scale drought signals. Low-frequency components—reflecting persistent climate trends—are predicted by a bidirectional long short-term memory (BiLSTM) network, while high-frequency fluctuations—associated with short-term meteorological variability—are captured by XGBoost. The approach is evaluated using daily multi-scale SPEI datasets from three urban meteorological stations in Henan Province (Anyang, Zhengzhou, and Xinyang), which span a pronounced climatic gradient across China’s urbanizing monsoon transition region. Results demonstrate that the VMD–BiLSTM–XGBoost model consistently outperforms benchmark alternatives (including VMD–ARIMA, VMD–LSTM, and VMD–BiLSTM) across all sites and time scales. Notably, it achieves an R² of 0.997 and an MSE of 0.001 for 12-month SPEI forecasting at the Anyang station, with average R² values exceeding 0.97 across all three cities. These findings highlight the model’s superior capability in capturing extreme drought events and forecasting multi-scale drought dynamics with high fidelity. By delivering robust, high-resolution drought projections, this method offers a valuable decision-support tool for enhancing urban climate resilience, informing adaptive water governance, and advancing drought early-warning systems in rapidly urbanizing regions vulnerable to hydroclimatic extremes.