BHRR v1.0 — Data Archive: CMIP6 ACCESS-CM2 inputs and example outputs for Restormer + ViT temperature downscaling and bias correction over Oceania

This archive bundles the precise input datasets required to train and evaluate BHRR v1.0, the trained model weights used to produce the published results, and a representative subset of example BHRR outputs, supporting the manuscript: Song, Y. H., Kim, H. J., and Chung, E.-S. (2026). BHRR v1.0: a two-stage Transformer framework for simultaneous spatial restoration and quantile-function bias correction of climate model temperature fields. Geoscientific Model Development (submitted, egusphere-2026-1958). The companion source-code archive is available at https://doi.org/10.5281/zenodo.19441661 (Song et al., 2026a). Together, this data archive and that code archive contain everything needed to reproduce the manuscript's results. Total archive size: approximately 36 GB across 22 files (5 loose top-level documents, 10 ZIPs, and 7 per-variable weight files). ## Record contents This record is split into multiple files for reliable downloading. The original on-disk folder layout of the data archive is reconstructed by extracting each ZIP and placing the top-level loose files alongside. ### Loose files (top-level documentation) - README.md — full archive description and reproduction instructions- NOTICES.md — upstream attribution and per-folder license terms- CITATION.cff — machine-readable citation- LICENSE.txt — top-level license (CC BY 4.0)- DATASET_DESCRIPTION.md — per-file inventory ### Input data (cropped to the Oceania evaluation domain: lat 49.875°S – 0.125°N, lon 110.125°E – 179.875°E at 0.25°) - inputs_cmip6_historical_0p25_regrid.zip — ACCESS-CM2 historical regridded to 0.25° (3 NetCDF files, BHRR direct historical input)- inputs_cmip6_future_RAW_gn.zip — ACCESS-CM2 ssp245/ssp585 RAW on native gn grid (6 files) plus a historical tasmax sample (7 NetCDF files total)- inputs_nex_gddp_historical.zip / _ssp245.zip / _ssp585.zip — NEX-GDDP-CMIP6 v1.0 BCSD benchmark for ACCESS-CM2, split by experiment (3 NetCDF files each, 9 total)- inputs_pgfv3_historical_reference.zip — Princeton Global Forcing v3 daily near-surface temperature reference, 1980–2014 (3 NetCDF files) ### Trained model weights (CC BY 4.0) Two complementary sets of trained weights are provided: **(A) Unified weights** — single unified weight set per BHRR stage, packaged inside trained_models.zip: - 2_trained_models/stage1_restormer_sr/best_model.pth, best_model_ema.pth- 2_trained_models/stage2_vit_bc/best_model.pth- 2_trained_models/stage3_microcal/microcal_offset_sr_tmax.json **(B) Per-variable weights** — specialised checkpoints used to produce the manuscript's Fig 7–12 results, provided as loose files for convenient download without extracting the data ZIPs: - restormer_sr_tas.pth, restormer_sr_tasmax.pth, restormer_sr_tasmin.pth (Stage 1 — Restormer spatial restoration, one per variable)- vit_bc_tas.pth, vit_bc_tasmax.pth, vit_bc_tasmin.pth (Stage 2 — ViT bias correction, one per variable)- microcal_offset.json (Stage 3 — micro-calibration offsets) For full reproducibility of the manuscript's figures, combine the per-variable weights with the inputs above and run the inference scripts from the companion code archive (https://doi.org/10.5281/zenodo.19441661). ### Example outputs (representative sample only) - example_outputs_stage1_restormer_sr.zip — Stage 1 (Restormer SR) sample output (1 file, illustrates product format)- example_outputs_stage2_vit_bc.zip — Stage 2 (ViT bias-corrected) sample outputs (4 files, final BHRR product format) These outputs are a representative subset only, not the complete output set from the manuscript. The full output set is reproducible from the inputs and weights above. ### Integrity manifest - manifest.zip — SHA-256 checksums (checksums_sha256.txt) and a 26-column per-file structured inventory (file_inventory.csv) ## Key variant / version information - **CMIP6 ACCESS-CM2**: variant r1i1p1f1, data version v20191108, native grid gn (~1.875° × 1.25°).- **NEX-GDDP-CMIP6**: version 1.0 (BCSD method applied to ACCESS-CM2 with the PGFv3 reference).- **PGFv3**: Princeton/Southampton hydroclimatology group v3 dataset (Sheffield et al., 2006). ## Licenses - **Top-level author-generated material** (weights, example outputs, manifest, documentation): CC BY 4.0.- inputs_cmip6_historical_0p25_regrid.zip and inputs_cmip6_future_RAW_gn.zip: **CC BY-SA 4.0** (Share-Alike, inherited from upstream CMIP6 — derivative works must remain CC BY-SA 4.0).- inputs_nex_gddp_*.zip: NASA Open Data — free reuse with attribution.- inputs_pgfv3_historical_reference.zip: CC BY 4.0. See NOTICES.md for the full text of each license and required attribution. ## Upstream DOIs - CMIP6 ACCESS-CM2 historical: https://doi.org/10.22033/ESGF/CMIP6.4271 (Dix et al., 2019a)- CMIP6 ACCESS-CM2 ssp245: https://doi.org/10.22033/ESGF/CMIP6.4321 (Dix et al., 2019b)- CMIP6 ACCESS-CM2 ssp585: https://doi.org/10.22033/ESGF/CMIP6.4332 (Dix et al., 2019c)- NEX-GDDP-CMIP6 v1.0: https://doi.org/10.7917/OFSG3345 (Thrasher et al., 2022)- PGFv3 reference paper: https://doi.org/10.1175/JCLI3790.1 (Sheffield et al., 2006)- BHRR v1.0 code (companion archive): https://doi.org/10.5281/zenodo.19441661 (Song et al., 2026a) ## Citation Song, Y. H., Kim, H. J., and Chung, E.-S. (2026). BHRR v1.0 — Data Archive. Zenodo. https://doi.org/10.5281/zenodo.20152297. All websites and repositories were last accessed on 13 May 2026.