We present CAMELSH (Catchment Attributes and Hourly HydroMeteorology for Large-Sample Studies), the first hourly large-sample hydrometeorological dataset for the contiguous United States. CAMELSH provides hourly streamflow and meteorological time series, along with catchment attributes and boundaries from GAGES-II and HydroATLAS, covering 7,211 catchments across diverse climatic, hydrological, and anthropogenic conditions. Among these, 2,569 catchments include observed streamflow, while the remaining contain only meteorological forcings. The dataset spans 45 years (1980–2024) with 11 meteorological variables from the NLDAS-2 forcing dataset, from which we compute nine climate indices related to precipitation, evapotranspiration, seasonality, and snow fraction. Additionally, CAMELSH includes two sets of catchment attributes: 439 from GAGES-II and 195 derived from HydroATLAS. Developed in accordance with FAIR (Findability, Accessibility, Interoperability, and Reusability) principles, CAMELSH is the first large-sample dataset at an hourly timescale, supporting machine learning applications for short-term streamflow (flood) prediction and advancing data-driven hydrological research across multiple timescales.
\nThe dataset is organized into the following subfolders:
• The attributes folder comprises three subfolders, each corresponding to a source dataset, and contains CSV (comma-separated values) files that store basin attributes. Specifically, the ‘Attributes-Climate’ subfolder includes the ‘attribute_climate.csv’ file, which contains nine climate attributes (Table 2) derived from NLDAS-2 data. The ‘Attributes-GAGESII’ subfolder contains a total of 26 CSV files and one Excel (*.xlsx) file. Among these, the 26 CSV files include information related to basins/streamflow gauging stations and 439 basin attributes. The name of each file represents a distinct group of attributes, as described in Table S.1. The remaining file, named ‘Var_description.xlsx’, provides explanatory details regarding the variable names included in the 26 CSV files, with information similar to that presented in Table S.1. The final subfolder, ‘Attributes-HydroAtlas’, contains a single file, ‘attribute_HydroATLAS.csv’, which includes 195 basin attributes derived from the HydroATLAS dataset. The first column in all CSV files, labeled ‘STAID,’ contains the identification (ID) names of the streamgages. These IDs are assigned by the USGS and are sourced from the original GAGES-II dataset.
• The shapefiles folder contains two sets of shapefiles for the catchment boundary. The first set, CAMELSH_shapefile.shp, is derived from the original GAGES-II dataset and is used to obtain the corresponding climate forcing data for each catchment. The second set, CAMELSH_shapefile_hydroATLAS.shp, includes catchment boundaries derived from the HydroATLAS dataset. Each polygon in both shapefiles contains a field named GAGE_ID, which represents the ID of the stream gauges.
• The timeseries (7zip) file contains a compressed archive (7zip) that includes time series data for 7,211 basins. Within this 7zip file, there are a total of 7,211 NetCDF files, each corresponding to a specific basin. The name of each NetCDF file matches the stream gauge ID. Each file contains an hourly time series from 1980-01-01 00:00:00 to 2024-12-31 23:00:00 for streamflow (denoted as “Streamflow” in the NetCDF file) and 11 climate variables (see Table 1). The streamflow data series includes missing values, which are represented as “NaN”. Among these, a total of 2,569 stations contain observed streamflow data, while 4,642 stations do not have observed data. Note that streamflow is measured in cubic meters per second. All meteorological forcing data and streamflow records have been standardized to the +0 UTC time zone.
• The info.csv file, located in the main directory of dataset, contains basic information for 7,211 stream stations. This includes the stream gauge ID, the total number of observed hourly data points over 45 years (from 1980 to 2024), and the number of observed hourly data points for each individual year from 1980 to 2024. Stations with and without observed data are distinguished by the value in the second column, where stations without observed streamflow data have a corresponding value of 0.