Scholarlink

<p>Accurate and timely drought forecasting is crucial for sustainable water resource management and effective agricultural planning, especially in regions prone to water stress. This study introduces a novel Hybrid Convolutional Bi-Kernel Ensemble (HCBKE) model, which synergistically integrates convolutional neural networks (CNN), bidirectional long- and short-term memory (BiLSTM), and kernel-based k-Nearest Neighbors (KNN) for multiscale drought prediction using the standardized precipitation index (SPI) at timescales of 3, 6, and 12 months. The model has been evaluated using data from six meteorological stations in Ankara province, Turkey (1971-2022), incorporating local climate indices and global teleconnection patterns (NAO, ENSO, MOI). Comparative results demonstrate that HCBKE consistently outperforms traditional and deep learning models. It achieved a maximum <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> of 0.788, an NSE of 0.788, and a minimum RMSE of 0.487 at Polatlı for SPI-12 predictions. At SPI-3, despite higher variability, the model maintained moderate accuracy with <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> values ranging from 0.363 to 0.448. The proposed model also showed statistically significant improvements (p &lt; 0.01) in Mean Absolute Error (MAE) across all timescales compared to baseline models, as confirmed by the Wilcoxon signed-rank test. By effectively capturing complex nonlinear and spatiotemporal patterns while maintaining numerical stability through multicollinearity assessment, the HCBKE model presents a robust, generalizable framework for drought monitoring. Its superior predictive performance highlights its potential for operational implementation in climate-sensitive regions.</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

Modeling meteorological drought across scales with regional and global climate indicators

  • 期刊论文

  • 发表日期：

    2025年11月12日

查看全文
  • Muhammad Ilyas,
  • Rizwan Niaz,
  • Hefa Cheng,
  • Luca Di Persio,
  • Maysaa Elmahi Abd Elwahab,
  • Ali Danandeh Mehr,
  • Ijaz Hussain

摘要

Accurate and timely drought forecasting is crucial for sustainable water resource management and effective agricultural planning, especially in regions prone to water stress. This study introduces a novel Hybrid Convolutional Bi-Kernel Ensemble (HCBKE) model, which synergistically integrates convolutional neural networks (CNN), bidirectional long- and short-term memory (BiLSTM), and kernel-based k-Nearest Neighbors (KNN) for multiscale drought prediction using the standardized precipitation index (SPI) at timescales of 3, 6, and 12 months. The model has been evaluated using data from six meteorological stations in Ankara province, Turkey (1971-2022), incorporating local climate indices and global teleconnection patterns (NAO, ENSO, MOI). Comparative results demonstrate that HCBKE consistently outperforms traditional and deep learning models. It achieved a maximum \(R^2\) of 0.788, an NSE of 0.788, and a minimum RMSE of 0.487 at Polatlı for SPI-12 predictions. At SPI-3, despite higher variability, the model maintained moderate accuracy with \(R^2\) values ranging from 0.363 to 0.448. The proposed model also showed statistically significant improvements (p < 0.01) in Mean Absolute Error (MAE) across all timescales compared to baseline models, as confirmed by the Wilcoxon signed-rank test. By effectively capturing complex nonlinear and spatiotemporal patterns while maintaining numerical stability through multicollinearity assessment, the HCBKE model presents a robust, generalizable framework for drought monitoring. Its superior predictive performance highlights its potential for operational implementation in climate-sensitive regions.