SRGAN Outputs for Test Period (2012-2021) and GraphCast/HRES Forecasts (2022)

Overview This repository provides sample NetCDF files from applying different SRGAN models to precipitation datasets (both test period and forecast). Because the full dataset is large (tens of GB), we only include a small subset of files here on Zenodo. To access the entire dataset, please see our Google Drive link: [Google Drive URL here](The directory structure and file naming convention on Google Drive are identical to what is described below.) The folder structure is as follows (note that only a few sample files are included here on Zenodo): Data/ ├─ Test_Data/ │ ├─ All-SR/ │ ├─ Prec-SR/ │ ├─ PrecIVT-SR/ │ ├─ PrecMSLP-SR/ │ └─ PrecTemp-SR/ └─ Forecast_Data/ ├─ GraphCast/ │ ├─ All-SR/ │ ├─ Prec-SR/ │ ├─ PrecIVT-SR/ │ ├─ PrecMSLP-SR/ │ └─ PrecTemp-SR/ └─ HRES/ ├─ All-SR/ ├─ Prec-SR/ ├─ PrecIVT-SR/ ├─ PrecMSLP-SR/ └─ PrecTemp-SR/ Test_Data These folders contain SRGAN-applied test period precipitation data. Specifically, annual NetCDF files for different years (2012–2021) are included (6-hourly resolution), used to evaluate the model performance after training. Examples: All-SR/Uses precipitation, temperature, MSLP (Mean Sea Level Pressure), and IVT (Integrated Water Vapor Transport) as inputs. Prec-SR/, PrecTemp-SR/, PrecMSLP-SR/, PrecIVT-SR/Each uses a different subset of input variables (see the paper for details). Each subfolder (e.g., Data/Test_Data/All-SR) contains files named: predicted_precipitation_YYYY.nc (e.g., predicted_precipitation_2012.nc), which store SRGAN-enhanced precipitation over 30–46N, 128–146E for each year. Forecast_Data These folders contain SRGAN-applied forecast precipitation data for 2022-01-01 to 2022-12-21 (daily resolution), with lead times up to 10 days (so the last forecast covers data up to December 31, 2022). Two forecast sources were used: GraphCast: Forecast outputs from GraphCast (original 0.25°) enhanced by different SRGAN models. HRES: ECMWF HRES (regridded to 0.25°) forecast outputs, also enhanced by the same SRGAN variants. Subfolders All-SR, Prec-SR, etc. follow the same naming convention and input-variable differences as in Test_Data.For example, in Data/Forecast_Data/GraphCast/All-SR/, you will find: predicted_precipitation_YYYY-MM-DD.nc representing daily SR-enhanced precipitation over 30–46N, 128–146E on each date of 2022. File Details NetCDF variables: total_precipitation_24hr – 24-hour accumulated precipitation (mm/24h) time – forecast initialization or valid time lat, lon – geographical coordinates Spatial resolution: SR model output is 0.05° × 0.05° over Japan (land area). Temporal coverage: Test_Data: 2012–2021 (each file spans one calendar year, with data typically at 6-hourly resolution). Forecast_Data: 2022-01-01 to 2022-12-21 (daily resolution), with lead times up to 10 days (so the last forecast covers data up to December 31, 2022). Notes / License These data were generated by applying our SRGAN models to precipitation forecasts or reanalysis. Refer to our paper [Reference] for details on the SR training, model architecture, and validation. License: CC-BY 4.0 (Attribution). Users are free to reuse or redistribute, provided appropriate credit is given. If you use these data or SR outputs in a publication, please cite our paper: [Paper citation / DOI here]