Scholarlink

<p>Machine learning methods provide new alternative methods and ideas for runoff prediction. To improve the application of machine learning methods in runoff prediction, five Japanese rivers spanning a latitudinal gradient across northern to southern regions, exhibiting heterogeneous hydrological regimes, were selected as study sites. A systematic comparison of six distinct machine learning architectures was conducted to evaluate the accuracy and applicability of these methods for daily runoff prediction in different watersheds and improve the commonality problem found in the prediction process. The results showed significant improvements across all assessment indicators. Before the improvement, the model performed well in only three basins (Kushiro, Yodogawa, and Shinano), with an average NSE value greater than 0.65, meeting the minimum threshold for the applicability of the hydrological model. After the improvement, the prediction indexes of the six methods increased significantly in the five basins (NSE increased by 4.94% to 190.32% on average), among which the NSE score of the enhanced Deep Temporal Convolutional Network (DeepTCN) reached more than 0.94. In general, the improved DeepTCN has the best comprehensive prediction effect, and has the potential to be widely recommended for runoff prediction.</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

Multiple machine learning methods for runoff prediction: contrast and improvement

  • 期刊论文

  • 发表日期：

    2025年6月11日

查看全文
  • Yuechao Chen,
  • Jiajia Gao,
  • Yue Zhang,
  • Xiaolei Fan,
  • Xue Song,
  • Zhaohui Bin,
  • Hao Ma,
  • Qing Zhang,
  • Li zhao

摘要

Machine learning methods provide new alternative methods and ideas for runoff prediction. To improve the application of machine learning methods in runoff prediction, five Japanese rivers spanning a latitudinal gradient across northern to southern regions, exhibiting heterogeneous hydrological regimes, were selected as study sites. A systematic comparison of six distinct machine learning architectures was conducted to evaluate the accuracy and applicability of these methods for daily runoff prediction in different watersheds and improve the commonality problem found in the prediction process. The results showed significant improvements across all assessment indicators. Before the improvement, the model performed well in only three basins (Kushiro, Yodogawa, and Shinano), with an average NSE value greater than 0.65, meeting the minimum threshold for the applicability of the hydrological model. After the improvement, the prediction indexes of the six methods increased significantly in the five basins (NSE increased by 4.94% to 190.32% on average), among which the NSE score of the enhanced Deep Temporal Convolutional Network (DeepTCN) reached more than 0.94. In general, the improved DeepTCN has the best comprehensive prediction effect, and has the potential to be widely recommended for runoff prediction.