Scholarlink

<p>Accurate parameterization is essential for reliable hydrological model predictions, yet parameter uncertainty remains a significant challenge. This study introduces an innovative framework that integrates a differentiable hydrology model with variational inference (VI), to effectively quantify parameter uncertainty. The differentiable hydrology component handles complex processes through differentiable functions, facilitating gradient-based optimization, while VI provides scalable and computationally efficient uncertainty quantification for probabilistic inference. The framework’s efficacy is exemplified through a case study in the Hieu River Basin, where its outcomes are compared against high-fidelity Markov Chain Monte Carlo analysis results, demonstrating its ability to efficiently estimate parameters and quantify uncertainty. Such proficiency enhances the accuracy and efficiency of hydrological predictions with uncertainty quantification. VI also exhibits an outstanding ability to recognize the equifinality of uncertain parameters by providing parameter sets with wide posterior distributions yet maintaining high streamflow simulation accuracy. Ultimately, this novel approach proves particularly useful in forecasting applications where multiple uncertainties are considered and is especially valuable in real-time forecasting applications due to high computational efficiency.</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

Advancing parameter uncertainty quantification in hydrology models through integration of variational inference with a differentiable hydrology framework

  • 期刊论文

  • 发表日期：

    2025年6月17日

查看全文
  • Giang V. Nguyen,
  • Chien Pham Van,
  • Linh Nguyen Van,
  • Vinh Ngoc Tran,
  • Younghun Kim,
  • Giha Lee

摘要

Accurate parameterization is essential for reliable hydrological model predictions, yet parameter uncertainty remains a significant challenge. This study introduces an innovative framework that integrates a differentiable hydrology model with variational inference (VI), to effectively quantify parameter uncertainty. The differentiable hydrology component handles complex processes through differentiable functions, facilitating gradient-based optimization, while VI provides scalable and computationally efficient uncertainty quantification for probabilistic inference. The framework’s efficacy is exemplified through a case study in the Hieu River Basin, where its outcomes are compared against high-fidelity Markov Chain Monte Carlo analysis results, demonstrating its ability to efficiently estimate parameters and quantify uncertainty. Such proficiency enhances the accuracy and efficiency of hydrological predictions with uncertainty quantification. VI also exhibits an outstanding ability to recognize the equifinality of uncertain parameters by providing parameter sets with wide posterior distributions yet maintaining high streamflow simulation accuracy. Ultimately, this novel approach proves particularly useful in forecasting applications where multiple uncertainties are considered and is especially valuable in real-time forecasting applications due to high computational efficiency.