Streamflow datasets from the high-resolution, multiscale, differentiable HBV hydrologic models

This dataset is from the high-resolution, multiscale, differentiable HBV hydrologic models, provided by the Multi-scale Hydrology, Processes, and Intelligence (MHPI) team from The Pennsylvania State University, led by Dr. Chaopeng Shen's group in Hydrologic Deep Learning and Modeling. (Directly downloading the folder may exceed your browser's file limit, resulting in zero-sized data in the Zarr file. Downloading the zip file is recommended.) Please cite: Song, Y., Bindas, T., Shen, C., Ji, H., Knoben, W. J. M., Lonzarich, L., et al. (2025). High‐resolution national‐scale water modeling is enhanced by multiscale differentiable physics‐informed machine learning. Water Resources Research, 61, e2024WR038928. https://doi.org/10.1029/2024WR038928 The dataset is generated by High-resolution, multiscale, differentiable HBV hydrologic models, dHBV2.0UH and dHBV2.0dMC. dHBV2.0UH is a high-resolution, multiscale model that uses unit hydrograph routing. dHBV2.0dMC is a high-resolution, multiscale model that uses external Muskingum-Cunge routing. Please use the link in the Readme.docx file to access the dataset. A link to the code for loading the dataset is also provided in Readme.docx. The dHBV_streamflow_simulation_gages folder includes 40 years (1980–2020) of streamflow simulations at over 7,000 gage stations from GAGES-II, using both dHBV2.0UH and dHBV2.0dMC models. This data is useful for comparison with observations. The MERIT_flux_states folder includes 40 years (1980–2020) of spatially seamless simulations of hydrologic variables over 180 thousand MERIT unit basins on CONUS from dHBV2.0UH, including baseflow, evapotranspiration (ET), soil moisture, snow water equivalent, and runoff. The dHBV2.0_MERIT_river_network_simulation includes 40 years (1980-2020) of streamflow simulation on seamless MERIT river network by dHBV2.0dMC (New updates!). The dHBV2.0UH code is available at mhpi/generic_deltaModel: High-resolution differentiable model, 𝛿HBV2.0. https://doi.org/10.5281/zenodo.14827983本数据集源自美国宾夕法尼亚州立大学多尺度水文过程与智能（Multi-scale Hydrology, Processes, and Intelligence, MHPI）团队，由沈朝鹏博士领衔的水文深度学习与建模课题组所开发的高分辨率多尺度可微分HBV水文模型。 注意：直接下载文件夹可能超出浏览器允许的文件大小上限，将导致Zarr文件出现零字节数据，建议下载压缩包获取数据。 请引用以下文献： Song, Y., Bindas, T., Shen, C., Ji, H., Knoben, W. J. M., Lonzarich, L., 等(2025). 高分辨率国家尺度水文建模借助多尺度可微分物理知情机器学习实现性能提升. 《水资源研究》, 61, e2024WR038928. https://doi.org/10.1029/2024WR038928 本数据集由两款高分辨率多尺度可微分HBV水文模型dHBV2.0UH与dHBV2.0dMC生成： 1. dHBV2.0UH：采用单位线汇流的高分辨率多尺度水文模型； 2. dHBV2.0dMC：采用外部马斯京根-康吉（Muskingum-Cunge）汇流的高分辨率多尺度水文模型。 请通过Readme.docx文件内的链接获取本数据集，文件中同时提供了数据集加载代码的相关链接。 ### 各数据文件夹说明 1. dHBV_streamflow_simulation_gages 文件夹：包含1980–2020年共40年间，基于dHBV2.0UH与dHBV2.0dMC模型，对GAGES-II数据集内7000余个水文测站的径流模拟结果，该数据可用于与实测径流数据开展对比分析。 2. MERIT_flux_states 文件夹：包含1980–2020年共40年间，由dHBV2.0UH生成的美国本土（Continental United States, CONUS）范围内18万个MERIT单元流域的空间连续水文变量模拟结果，涵盖基流、蒸散发（ET）、土壤含水量、雪水当量与径流。 3. dHBV2.0_MERIT_river_network_simulation 文件夹：包含1980–2020年共40年间，由dHBV2.0dMC模型（本次新增更新！）在连续MERIT河网上生成的径流模拟结果。 dHBV2.0UH 代码可在 mhpi/generic_deltaModel 仓库获取，该仓库对应高分辨率可微分模型𝛿HBV2.0，相关DOI为：https://doi.org/10.5281/zenodo.14827983