Scholarlink

<p>Improved rainfall prediction models contribute to more effective disaster preparedness and response strategies. Early warnings of potential floods or droughts allow authorities to implement necessary measures, reducing the risk of property damage and loss of life. By leveraging historical data and sophisticated machine learning techniques, this research establishes a framework for predicting monthly precipitation to enhance urban services. The study examines 24 years (2000–2023) of meteorological records from three stations in Victoria, Australia. To optimize model efficiency and minimize complexity, a univariate approach utilizing readily available data was adopted. The Average Mutual Information (AMI) method was applied to determine the most relevant lag values for each station. Subsequently, 80% of the dataset was designated for training, while the remaining 20% was reserved for testing. The models employed include a deep neural network (DNN) and two hybrid architectures: a convolutional neural network combined with long short-term memory (CNN-LSTM) and a recurrent neural network-integrated with CNN-LSTM (RNN-CNN-LSTM). To evaluate and compare model performance, visual representations and four key metrics were utilized: coefficient of determination (R²), root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE), and mean absolute percentage error (MAPE). The findings highlight the superiority of the RNN-CNN-LSTM model, which achieved an average error of 5.45&#xa0;mm and a coefficient of determination of 0.968 in forecasting monthly precipitation. These results indicate that the model excels in univariate prediction and holds significant potential for broader applications across different meteorological stations.</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

Leveraging hybrid deep learning architectures for predicting monthly precipitation in Victoria, Australia

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Erfan Abdi,
  • Tina Vahid,
  • Osama Ragab Ibrahim,
  • Mousa Alizadeh

摘要

Improved rainfall prediction models contribute to more effective disaster preparedness and response strategies. Early warnings of potential floods or droughts allow authorities to implement necessary measures, reducing the risk of property damage and loss of life. By leveraging historical data and sophisticated machine learning techniques, this research establishes a framework for predicting monthly precipitation to enhance urban services. The study examines 24 years (2000–2023) of meteorological records from three stations in Victoria, Australia. To optimize model efficiency and minimize complexity, a univariate approach utilizing readily available data was adopted. The Average Mutual Information (AMI) method was applied to determine the most relevant lag values for each station. Subsequently, 80% of the dataset was designated for training, while the remaining 20% was reserved for testing. The models employed include a deep neural network (DNN) and two hybrid architectures: a convolutional neural network combined with long short-term memory (CNN-LSTM) and a recurrent neural network-integrated with CNN-LSTM (RNN-CNN-LSTM). To evaluate and compare model performance, visual representations and four key metrics were utilized: coefficient of determination (R²), root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE), and mean absolute percentage error (MAPE). The findings highlight the superiority of the RNN-CNN-LSTM model, which achieved an average error of 5.45 mm and a coefficient of determination of 0.968 in forecasting monthly precipitation. These results indicate that the model excels in univariate prediction and holds significant potential for broader applications across different meteorological stations.