GloFAS Historical

This dataset is a processed runoff-field forecasting dataset for China constructed from the GloFAS Historical dataset obtained through the Copernicus Climate Data Store. The original variable is river_discharge_in_the_last_24_hours from the historical reanalysis product cems-glofas-historical, with system_version set to version_4_0, hydrological_model set to lisflood, and product_type set to consolidated. The original data were downloaded in NetCDF format by temporal chunks and then organized into a continuous daily gridded time series covering the period from 2020-01-01 to 2025-11-01.

The dataset was cropped to the geographic range of China with longitude [73°, 135°] and latitude [18°, 54°]. A valid-region mask was further constructed based on the China outline derived from Natural Earth administrative boundaries to remove grid cells outside the national boundary or otherwise invalid areas. The released dataset includes the cropped daily runoff fields, the valid-region mask, and the corresponding chronological split information for training, validation, and testing.

For preprocessing, the raw runoff values were transformed using x_log = log10(x + 1) to alleviate the heavy-tailed distribution and large dynamic range of runoff data. After logarithmic transformation, all samples were normalized to the [0,1] range using statistics computed from the training set only, and the same normalization parameters were applied to the validation and testing sets. No additional temporal smoothing or manual filtering was introduced beyond cropping, masking, logarithmic transformation, and normalization.

Following the forecasting setting of the associated study, the historical input length was set to 30 days and the forecasting horizon was set to 7 days. The daily samples were divided chronologically into training, validation, and testing subsets with a ratio of 7:1:2 to preserve temporal order and avoid information leakage. This dataset is intended for reproducible research on gridded runoff forecasting, hydrological spatiotemporal modeling, and basin-scale environmental prediction over China.