{"success":true,"data":{"author":"oovanger (AORC Texas downscaling benchmark authors)","bucketAccessType":"public","bucketName":"oovanger/aorc-texas-downscaling","bucketPrefix":"","category":"Climate & Weather","cloudProvider":"huggingface","createdAt":"2026-11-16T15:39:55.971Z","description":"A benchmark for statistical climate downscaling (super-resolution) built from NOAA's Analysis of Record for Calibration (AORC) v1.1 hourly 1 km reanalysis over two Texas sub-regions (south-west and north-east), 2021 to 2024. Targets are 512 x 512 pixel patches at 1 km of three variables: 2 m air temperature, near-surface specific humidity and precipitation rate. Low-resolution inputs are made with a fixed degradation operator (Gaussian blur, strided subsampling, bicubic upsampling) at five scaling factors, 2x, 4x, 8x, 16x and 32x, so models from bicubic and U-Net baselines to diffusion models can be compared on identical inputs. Values are stored normalised: z-scores for temperature and humidity and log1p then z-score for precipitation, with training-split statistics in stats.json and code in the card to recover physical units. Layout: processed/<split>/<variable>.zarr as Zarr v2 float32 arrays chunked (32, 512, 512), about 4.7 GB across 979 files, plus a Croissant metadata file. Read with zarr or xarray. Licensed CC BY 4.0 by the authors; AORC itself is a NOAA product.","displayTitle":"AORC Texas Climate Downscaling Benchmark (1 km, 2021-2024)","huggingFaceUrl":"https://huggingface.co/datasets/oovanger/aorc-texas-downscaling","id":"6a3f7820-5ce4-5d0a-b1b8-cacb52a9c481","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/oovanger/aorc-texas-downscaling","objectCountEstimate":6890,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"oovanger/aorc-texas-downscaling\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"oovanger/aorc-texas-downscaling\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=oovanger, license:cc-by-4.0, size_categories:10B<n<100B, downloads=46"],"format":"mixed","generatedAt":"2026-11-16T15:40:04.484Z","layout":{"example":"processed/test_ood/humidity.zarr/.zarray","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"{\n    \"chunks\": [\n        32,\n        512,\n        512\n    ],\n    \"compressor\": {\n        \"blocksize\": 0,\n        \"clevel\": 5,\n        \"cname\": \"lz4\",\n        \"id\": \"blosc\",\n        \"shuffle\": 1\n    },\n    \"dimension_separator\": \".\",\n    \"dtype\": \"<f4\",\n    \"fill_value\": 0.0,\n    \"filters\": null,\n    \"order\": \"C\",\n    \"shape\": [\n        8784,\n        512,\n        512\n    ],\n    \"zarr_format\": 2\n}","encoding":"utf8","key":"processed/test_ood/humidity.zarr/.zarray","truncatedAt":512}],"sampleObjects":["croissant_metadata.json","processed/stats.json","processed/test_ood/humidity.zarr/.zarray","processed/test_ood/humidity.zarr/0.0.0","processed/test_ood/humidity.zarr/1.0.0"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\ntask_categories:\n  - other\ntask_ids: []\ntags:\n  - climate\n  - downscaling\n  - super-resolution\n  - weather\n  - geospatial\n  - zarr\npretty_name: AORC Texas Climate Downscaling Benchmark\nsize_categories:\n  - 10B<n<100B\nlanguage: []\nmultilinguality: monolingual\n---\n\n# AORC Texas Climate Downscaling Benchmark\n\nA benchmark dataset for statistical climate downscaling (super-resolution) based on NOAA AORC 1 km reanalysis data over Texas (2021–2024). The dataset supports training and evaluating models that reconstruct high-resolution (1 km) meteorological fields from coarsened low-resolution inputs at five scaling factors (2×, 4×, 8×, 16×, 32×).\n\n## Dataset Description\n\n| Field | Value |\n|---|---|\n| **Source** | NOAA Analysis of Record for Calibration (AORC) v1.1 |\n| **Spatial domain** | Texas (two sub-regions: SW and NE), 1 km grid |\n| **Temporal range** | 2021–2024, hourly |\n| **Patch size** | 512 × 512 pixels at 1 km resolution |\n| **Variables** | 2 m air temperature, near-surface specific humidity, precipitation rate |\n| **Format** | Zarr v2, float32, chunks (32, 512, 512) |\n| **Total size** | ~43 GB |\n\n## Variables\n\n| Short name | Long name | Physical unit | Normalization |\n|---|---|---|---|\n| `temp` | 2 m air temperature | K | z-score: `(T − 290.58) / 10.01` |\n| `humidity` | Near-surface specific humidity | kg kg⁻¹ | z-score: `(q − 6.58×10⁻³) / 4.16×10⁻³` |\n| `precip` | Precipitation rate | mm h⁻¹ | log1p then z-score: `(log1p(p) − 0.01637) / 0.1322` |\n\nNormalization statistics were computed from the **training split** only and are stored in `stats.json`.  \nValues stored in the zarr arrays are **already normalized** (approximately zero-mean, unit-variance).\n\nTo recover physical values:\n```python\nimport numpy as np, json, zarr\n\nwith open(\"stats.json\") as f:\n    stats = json.load(f)\n\nz = zarr.open(\"processed/train/temp.zarr\", \"r\")\ny_norm = z[0]  # (512, 512), normalized\n\n# Temperature (K)\ntemp_K = y_norm * stats[\"temp\"][\"std\"] + stats[\"temp\"][\"mean\"]\n\n# Humidity (kg/kg)\nhum = y_norm * stats[\"humidity\"][\"std\"] + stats[\"humidity\"][\"mean\"]\n\n# Precipitation (mm/h) — undo z-score then undo log1p\nprecip_log = y_norm * stats[\"precip\"][\"std\"] + stats[\"precip\"][\"mean\"]\nprecip_mmh = np.expm1(precip_log)\n```\n\n## Degradation Operator\n\nLow-resolution inputs are generated from the high-resolution targets using a **Gaussian blur → strided subsampling → bicubic upsample** operator:\n\n```python\nfrom scipy.ndimage import gaussian_filter, zoom\n\ndef degrade(y_hr, scale):\n    \"\"\"y_hr: (512,512) normalized HR field → (512,512) degraded LR-upsampled field.\"\"\"\n    sigma  = 0.5 * scale\n    y_blur = gaussian_filter(y_hr, sigma=sigma)\n    x_lr   = y_blur[::scale, ::scale]          # (512//scale, 512//scale)\n    x_up   = zoom(x_lr, scale, order=3)[:512, :512]  # bicubic upsample back to 512×512\n    return x_up\n```\n\nSupported scaling factors: **2×, 4×, 8×, 16×, 32×**.\n\n## Splits\n\n| Split | Region | Years | Frames (per variable) | Purpose |\n|---|---|---|---|---|\n| `train` | SW Texas | 2021–2022 | ~29 334 | Training |\n| `val` | SW Texas | 2023 | 8 760 | Validation / model selection |\n| `test_temporal` | SW Texas | 2024 | 8 784 | Temporal generalization |\n| `test_spatial` | NE Texas | 2021–2022 | 17 520 | Spatial generalization |\n| `test_ood` | NE Texas | 2024 | 8 784 | Spatio-temporal OOD |\n\nThe train/val/test_temporal splits share the SW Texas domain (in-distribution spatially) while test_spatial and test_ood use an unseen NE Texas domain. test_ood is the hardest split: unseen region **and** unseen year.\n\n## Directory Structure\n\n```\nprocessed/\n├── stats.json                  ← normalization statistics (training split)\n├── train/\n│   ├── temp.zarr/              ← (29334, 512, 512) float32 z-score normalized\n│   ├── precip.zarr/            ← (29451, 512, 512) float32 log1p + z-score normalized\n│   └── humidity.zarr/          ← (29316, 512, 512) float32 z-score normalized\n├── val/\n│   └── ...                     ← same structure,","encoding":"utf8","key":"README.md","name":"README.md","size":6163,"truncated":true}],"sourceObjectKey":"processed/test_ood/humidity.zarr/.zarray","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"oovanger/aorc-texas-downscaling\", filename=\"processed/test_ood/humidity.zarr/.zarray\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"processed/test_ood/humidity.zarr","totalObjects":6890},"summary":"A benchmark for statistical climate downscaling (super-resolution) built from NOAA's Analysis of Record for Calibration (AORC) v1.1 hourly 1 km reanalysis over two Texas…","title":"AORC Texas Climate Downscaling Benchmark (1 km, 2021-2024)","updatedAt":"2026-11-17T17:16:43.762Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/oovanger/aorc-texas-downscaling"}}