Scholarlink

<p>The gravity recovery and climate experiment (GRACE) satellite offers valuable data for hydrological analysis but its coarse spatial resolution limits its effectiveness for local-scale studies. While statistical downscaling techniques using global hydrological model outputs have shown promise in enhancing the resolution of terrestrial water storage (TWS) estimates from GRACE data, the performance of GRACE TWS downscaling based on regional fully-coupled model outputs remains unexplored. In this study, we analyzed to appraise the feasibility of GRACE TWS downscaling under two scenarios. The first scenario included training a machine learning algorithm with global hydrological (Famine Early Warning Systems Network (FEWS NET) Land Data Assimilation System (FLDAS)) model outputs. In the second scenario, ParFlow and CLM (PFC) models were combined in a fully surface–subsurface coupled modeling, and the outputs were integrated into a Random Forest machine learning downscaling technique. The downscaled TWS values (0.1°) were then evaluated against the GRACE TWS (0.25°) and precipitation observations of the rain gauges over the Rhine basin in Germany. The PFC-based downscaled TWS showed stronger correlation (0.98) than the FLDAS-based downscaled TWS (0.80). Comparison of the downscaled TWS results with precipitation data also emphasized the superiority of the second scenario. The PFC model-based downscaled TWS demonstrated increased correlations with precipitation data over all the sub-basins of the Rhine, suggesting that training downscaling algorithms with the fully-coupled physics-based hydrological model outputs yield better results compared to those of the FLDAS model.</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

Integrating Fully-Coupled Hydrological Modeling and Random Forest to Enhance Spatial Resolution of GRACE-Observed Water Storage Across the Rhine Basin

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月7日

查看全文
  • Fahimeh Youssefi,
  • Samira Sadat Soltani,
  • Shoaib Ali,
  • Behnam Khorrami

摘要

The gravity recovery and climate experiment (GRACE) satellite offers valuable data for hydrological analysis but its coarse spatial resolution limits its effectiveness for local-scale studies. While statistical downscaling techniques using global hydrological model outputs have shown promise in enhancing the resolution of terrestrial water storage (TWS) estimates from GRACE data, the performance of GRACE TWS downscaling based on regional fully-coupled model outputs remains unexplored. In this study, we analyzed to appraise the feasibility of GRACE TWS downscaling under two scenarios. The first scenario included training a machine learning algorithm with global hydrological (Famine Early Warning Systems Network (FEWS NET) Land Data Assimilation System (FLDAS)) model outputs. In the second scenario, ParFlow and CLM (PFC) models were combined in a fully surface–subsurface coupled modeling, and the outputs were integrated into a Random Forest machine learning downscaling technique. The downscaled TWS values (0.1°) were then evaluated against the GRACE TWS (0.25°) and precipitation observations of the rain gauges over the Rhine basin in Germany. The PFC-based downscaled TWS showed stronger correlation (0.98) than the FLDAS-based downscaled TWS (0.80). Comparison of the downscaled TWS results with precipitation data also emphasized the superiority of the second scenario. The PFC model-based downscaled TWS demonstrated increased correlations with precipitation data over all the sub-basins of the Rhine, suggesting that training downscaling algorithms with the fully-coupled physics-based hydrological model outputs yield better results compared to those of the FLDAS model.