Scholarlink

<p>Hydroclimatic forecasting is vital for sustainable water resource management amid climate variability. This study evaluates the performance of deep learning models for predicting key hydroclimatic variables, rainfall, temperature, evapotranspiration, and runoff using TerraClimate data from Morocco’s Middle Atlas Causse and Meknes Plateau regions. Algorithms analyzed include CNNs, LSTMs (Vanilla, Stacked, Bidirectional), and hybrid models like CNN-LSTM and ConvLSTM. Performance was assessed using metrics such as Mean Absolute Error, Root Mean Squared Error, Mean Forecast Error, Pearson Correlation Coefficient, and runtime. CNNs demonstrated high accuracy across most variables, with ConvLSTM also excelling due to its ability to capture temporal features. Among LSTM variants, Bidirectional LSTMs balanced strong predictive performance with computational efficiency. The results highlight CNNs as the most reliable for hydroclimatic forecasting. This research underscores the potential of deep learning models in improving hydroclimatic predictions, offering valuable insights for regions facing water scarcity and the impacts of climate change. By optimizing accuracy and efficiency, these findings support better decision-making and resource management in dynamic environmental conditions.</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 of time series forecasting with SANN models for TerraClimate’s hydroclimatic data and trend monitoring

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月27日

查看全文
  • Yassine Hammoud,
  • Youssef Allali,
  • Abderrahim Saadane

摘要

Hydroclimatic forecasting is vital for sustainable water resource management amid climate variability. This study evaluates the performance of deep learning models for predicting key hydroclimatic variables, rainfall, temperature, evapotranspiration, and runoff using TerraClimate data from Morocco’s Middle Atlas Causse and Meknes Plateau regions. Algorithms analyzed include CNNs, LSTMs (Vanilla, Stacked, Bidirectional), and hybrid models like CNN-LSTM and ConvLSTM. Performance was assessed using metrics such as Mean Absolute Error, Root Mean Squared Error, Mean Forecast Error, Pearson Correlation Coefficient, and runtime. CNNs demonstrated high accuracy across most variables, with ConvLSTM also excelling due to its ability to capture temporal features. Among LSTM variants, Bidirectional LSTMs balanced strong predictive performance with computational efficiency. The results highlight CNNs as the most reliable for hydroclimatic forecasting. This research underscores the potential of deep learning models in improving hydroclimatic predictions, offering valuable insights for regions facing water scarcity and the impacts of climate change. By optimizing accuracy and efficiency, these findings support better decision-making and resource management in dynamic environmental conditions.