Scholarlink

<p>Given the importance of drought and its wide-ranging impacts, this study investigates drought conditions across Iran using climatic data from 34 stations over the period 1970–2024 based on the Reconnaissance Drought Index (RDI) in both temporal and spatio-temporal dimensions. Future drought conditions for 2025–2035 were predicted using two advanced deep learning models: Bidirectional Long Short-Term Memory (BiDr-LSTM) for temporal forecasting and Convolutional LSTM (ConvLSTM) for spatio-temporal analysis. In addition, drought trends were evaluated using Spearman’s Rho test. The core technical novelty of this study lies in formulating a comprehensive dual-dimensional early warning framework that integrates BiDr-LSTM and ConvLSTM, rather than relying on isolated conventional approaches. Analysis of the historical period (1970–2024) shows that the Normal (Nor) drought class is the most dominant condition across Iran, whereas Extremely Wet (Ext-W) and Very Wet (Ver-W) classes have negligible occurrence. Validation results demonstrate the robustness and reliability of both BiDr-LSTM and ConvLSTM models in capturing drought dynamics. Temporal projections indicate that 41.18% of stations will experience exclusively Nor conditions, while only 2.94% will remain under Moderately Dry (Mod-D) conditions; the remaining stations will exhibit a combination of these two classes, with varying dominance. Spatio-temporal projections further reveal that at least 82% of Iran’s area will remain under Nor conditions throughout the 2025–2035 period, while smaller portions will be affected by Mod-D, Ver-D, and Ext-D classes. Trend analysis highlights a decreasing trend in Mod-W and Nor classes and a concurrent increase in Mod-D, Ver-D, and Ext-D classes. Moreover, a significant drying trend is observed, affecting over 88% of stations during 1970–2024 and projected to intensify to more than 91% by 2035.</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

Evaluation and Prediction of Drought Dynamic Using the Convolutional and Bidirectional Long Short-Term Memory Frameworks

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Abdol Rassoul Zarei

摘要

Given the importance of drought and its wide-ranging impacts, this study investigates drought conditions across Iran using climatic data from 34 stations over the period 1970–2024 based on the Reconnaissance Drought Index (RDI) in both temporal and spatio-temporal dimensions. Future drought conditions for 2025–2035 were predicted using two advanced deep learning models: Bidirectional Long Short-Term Memory (BiDr-LSTM) for temporal forecasting and Convolutional LSTM (ConvLSTM) for spatio-temporal analysis. In addition, drought trends were evaluated using Spearman’s Rho test. The core technical novelty of this study lies in formulating a comprehensive dual-dimensional early warning framework that integrates BiDr-LSTM and ConvLSTM, rather than relying on isolated conventional approaches. Analysis of the historical period (1970–2024) shows that the Normal (Nor) drought class is the most dominant condition across Iran, whereas Extremely Wet (Ext-W) and Very Wet (Ver-W) classes have negligible occurrence. Validation results demonstrate the robustness and reliability of both BiDr-LSTM and ConvLSTM models in capturing drought dynamics. Temporal projections indicate that 41.18% of stations will experience exclusively Nor conditions, while only 2.94% will remain under Moderately Dry (Mod-D) conditions; the remaining stations will exhibit a combination of these two classes, with varying dominance. Spatio-temporal projections further reveal that at least 82% of Iran’s area will remain under Nor conditions throughout the 2025–2035 period, while smaller portions will be affected by Mod-D, Ver-D, and Ext-D classes. Trend analysis highlights a decreasing trend in Mod-W and Nor classes and a concurrent increase in Mod-D, Ver-D, and Ext-D classes. Moreover, a significant drying trend is observed, affecting over 88% of stations during 1970–2024 and projected to intensify to more than 91% by 2035.