--- license: mit tags: - wildfire - geospatial - weather - earth-observation - foundation-models - evaluation - pytorch pipeline_tag: image-segmentation library_name: pytorch pretty_name: WildFIRE-FM ---

WildFIRE-FM

A wildfire-specialized reference backbone for fixed-contract Earth-FM transfer evaluation

Five seeded California regional PyTorch checkpoints, paper-aligned evaluation artifacts, final figure previews, source-data access notes, and a CONUS retraining recipe for 12-hour wildfire occupancy prediction.

Paper PDF Checkpoints PyTorch Data notes Results Figures Grid Input Tasks License

Read the paper PDF · Download checkpoints · Training/data loader · Custom spatial serving · Load the model · Inspect task-contract results

5 California Seeded Checkpoints · 16-Channel Gridded Input · 12-Hour Occupancy Lead · CONUS Retraining Recipe

Why WildFIRE-FM · Release Navigation · Visual Tour · Quick Start · Training Pipeline · Spatial Serving · CONUS Retraining · Integration Handoff · Task Snapshot · Data Sources · How to Cite

---

WildFIRE-FM overview

## Release Navigation

Paper PDF

Read the compiled manuscript that defines the fixed-contract wildfire transfer setting and reports the final results.

Model Checkpoints

Access five seeded WildFIRE-FM weights with manifest metadata and SHA-256 hashes for release auditing.

Model Code

Inspect the compact U-Net implementation used to load the released wildfire occupancy backbone.

Training/Data Loader

Read the original cache builder, tensor assembly, tile sampler, full-map loader, and training loop.

Custom Spatial Serving

Aggregate native grid probabilities to any user-defined polygon granularity.

Quick Start

Clone the repository, instantiate the model, load a seeded checkpoint, and run the artifact check.

Task Snapshot

View the final-paper task-contract summary for occupancy, spread, retrieval, burned area, smoke, and heat.

Numeric Artifacts

Use sanitized CSV and JSON summaries that back the public model-card tables and figure previews.

Visual Tour

Browse final-paper previews for matching-rule sensitivity, selection regret, and task-form ranking changes.

Data Sources

See the public provider links and roles for HRRR, FIRMS, LANDFIRE, WRC, LandScan, WFIGS, and MTBS.
--- ## Why WildFIRE-FM Wildfire Earth-FM transfer scores depend strongly on the contract used for comparison: task form, metric, matching rule, spatial scope, and head-selection criterion. The paper studies these choices by holding outputs or features fixed and then changing only the evaluation contract. WildFIRE-FM is the in-region reference model used in those comparisons. It is trained for 12-hour gridded wildfire occupancy on a California 5 km grid, then evaluated under the same task-specific contracts as the transferred Earth-FM backbones. The released weights are California regional checkpoints, not nationwide-trained weights. For nationwide use, the release includes a CONUS cache and training recipe under `training/NATIONWIDE_RETRAINING.md`. This Hub repository is model-first. It provides the released weights, model-loading code, compact result summaries, paper previews, and source-data access notes needed to inspect the release boundary before running larger local reproductions. --- ## Key Features - **Wildfire-specialized reference model:** Compact U-Net for 12-hour occupancy prediction on a projected California grid. - **Five seeded checkpoints:** Seeds `1`, `7`, `42`, `99`, and `123` are released with manifest hashes. - **Fixed-contract artifacts:** Compact summaries cover matching-rule, head-selection, and task-form comparisons from the final paper. - **Source-data aware release:** Provider links and data roles are documented for each public resource used by the study. - **Compiled paper PDF:** The readable manuscript is available directly from the Hub release. - **Paper-aligned previews:** Figure assets summarize selection regret, supporting-task rank changes, and primary-task rank changes.

Reference model for wildfire transfer

WildFIRE-FM combines regional weather, active-fire supervision, and static fuel, canopy, housing, and population context on the same grid used by the occupancy contract.

Contract-first evaluation artifacts

The released summaries keep metric roles separate across decision, overlap, ranking, regression, and supporting-task checks.

Local-data reproducibility path

Full raw-data reruns use provider-hosted sources and local feature caches. The Hub release documents those inputs and provides lightweight artifact checks.

Final-paper result previews

The model card highlights the paper's fixed-output, fixed-feature, and task-form findings without requiring manuscript source files.
--- ## Visual Tour

Release contents

**Release contents.** The Hub repository is organized around weights, model code, source-data notes, compact artifacts, and final-paper visual previews.

Head-selection regret

**Head-selection regret.** Choosing a lightweight head by a ranking metric can lose decision performance under the same frozen features.

Supporting task rank map

**Supporting-task rank map.** Backbone ordering changes across burned area, analog retrieval, smoke PM2.5, and extreme heat task contracts.

Primary rank changes

**Primary-task rank changes.** Primary-task rankings change under fixed matching-rule and task-form choices. --- ## At A Glance | Signal | Release value | |---|---| | Core task | 12-hour gridded wildfire occupancy prediction | | Released checkpoint region/grid | California regional grid, 5 km, EPSG:5070 | | Nationwide path | CONUS cache and training templates in `training/NATIONWIDE_RETRAINING.md` | | Input channels | 16 channels: weather fields, validity masks, static fuel/canopy/exposure layers | | Architecture | Compact U-Net with occupancy and auxiliary spatial-support heads | | Training split | June-August 2024 train, September 2024 validation, October 2024 test | | Released checkpoints | 5 seeded PyTorch checkpoints | | Data access path | Source links and roles documented in `data_sources/` | | Training/data loader path | Original cache builder and PyTorch loader in `training/` | | Spatial serving path | Custom polygon-granularity aggregation in `spatial_serving/` | | Paper artifact path | Compiled PDF in `paper/`; manuscript source managed in the authors' paper workspace | ### Input Channel Contract The pretrained checkpoints expect a fixed 16-channel tensor in `[channel, y, x]` order. Channel order matters. Inputs use native source units with no mean/std normalization, min/max scaling, or unit conversion. Invalid or missing values are zero-filled, with validity masks included as channels 10 and 11. | Channels | Names | Source | |---:|---|---| | 0-9 | `t2m`, `d2m`, `u10`, `v10`, `cape`, `sp`, `blh`, `vis`, `prate`, `tp` | NOAA HRRR | | 10 | `firewx_valid` | Cache-level dynamic input presence mask; 1.0 everywhere in this release | | 11 | `static_valid` | Fraction of static layers valid after reprojection | | 12-15 | `fuel_fbfm40`, `canopy_cover`, `housing_density`, `population` | LANDFIRE, WRC housing density, LandScan | FIRMS provides the occupancy target, not an input channel. WFIGS and MTBS are used for supporting event-level tasks and are not part of the 16-channel pretrained occupancy input. Full channel-level details are in `data_sources/DATA_SOURCES.md`; the same contract is available as `models/wildfire_fm/input_channels.json`. The `cape` channel is the HRRR surface instantaneous CAPE field selected with `typeOfLevel=surface` and `stepType=instant`. For validity channels, `firewx_valid` is 1.0 everywhere in the released California regional cache and is not a per-variable HRRR missing-data mask; `static_valid` is the fraction of the four static layers valid after reprojection at each grid cell. --- ## Hugging Face Release This Hub repository is a model release with paper-aligned reproducibility artifacts. | Folder | Purpose | Example | |---|---|---| | `models/wildfire_fm/` | Model code, manifests, checkpoint metadata | `models/wildfire_fm/modeling_unet.py` | | `models/wildfire_fm/input_channels.json` | Machine-readable 16-channel input contract | `models/wildfire_fm/input_channels.json` | | `models/wildfire_fm/checkpoints/` | Five seeded PyTorch checkpoints | `seed_1/best_firms_prauc.pt` | | `training/` | Original cache builder, data loader, tile sampler, and training loop | `train_cold_tiled_mainline.py` | | `training/NATIONWIDE_RETRAINING.md` | CONUS retraining recipe and data coverage notes | `stage1_cache_conus_hrrr_us_5km_l12_template.json` | | `spatial_serving/` | Grid-to-polygon serving adapter for custom spatial granularities | `grid_to_polygons.py` | | `data_downloader/hrrr_downloader.py` | Direct NOAA HRRR public archive downloader | `hrrr_download_manifest.json` | | `docs/hugh_handoff_status.md` | Integration answers for channel order, masks, normalization, HRRR, and serving boundary | handoff status | | `paper/` | Compiled paper PDF | `wildfire_fm_evaluation_contracts.pdf` | | `assets/` | Hub-page visuals and final-paper previews | `selection_regret_final.png` | | `paper_outputs/figures/` | Selected final-paper figure PDFs | `fig_rank_heatmap1.pdf` | | `artifacts/results/` | Sanitized compact numeric summaries | `selection_regret_scope_sweep_20260505.csv` | | `data_sources/` | Public data-source roles and access notes | `DATA_SOURCES.md` | | `experiments/` | Sanitized raw-rerun references and Slurm template | `slurm/submit_template.sbatch` | Full local reruns require users to prepare the provider-hosted raw data and comparator assets described in `data_sources/`. --- ## Quick Start Clone the Hub repository or download the files you need: ```bash git clone https://huggingface.co/RAI-Lab/Wildfire-FM cd Wildfire-FM ``` Load a seeded checkpoint: ```python import torch from models.wildfire_fm.modeling_unet import UNetSmallFlex model = UNetSmallFlex( in_ch=16, base=32, dropout=0.1, norm_type="group", norm_groups=8, use_aux_spatial_head=True, ) checkpoint = torch.load( "models/wildfire_fm/checkpoints/seed_1/best_firms_prauc.pt", map_location="cpu", ) state = checkpoint.get("model", checkpoint) model.load_state_dict(state) model.eval() ``` Verify the public release artifacts: ```bash python3 scripts/reproduce_paper_outputs.py ``` The checkpoint expects the same fixed-order 16-channel gridded input described above and in `data_sources/DATA_SOURCES.md`. --- ## Task-Contract Snapshot The card below reports the best final-paper mean for each displayed task contract, with the winning backbone named in the final column. | Task contract | Best final-paper mean | Winner | |---|---:|---| | Occupancy union F1 | `60.1506 ± 7.5865` percent | ClimaX | | Fire-spread spatial F1 | `80.9700 ± 2.0200` percent | WildFIRE-FM | | Final burned-area log-RMSE | `1.1657 ± 0.0126`, lower is better | WildFIRE-FM | | Analog retrieval nDCG@10 | `0.5099 ± 0.0336` | WildFIRE-FM | | Smoke PM2.5 RMSE | `4.4403 ± 0.0488`, lower is better | AlphaEarth | | Extreme-heat RMSE-C | `0.2179 ± 0.0043`, lower is better | WildFIRE-FM | The release provides sanitized CSV/JSON summaries used to audit displayed values. --- ## Data Sources The study uses public or provider-hosted resources. Source roles and access paths are documented in `data_sources/DATA_SOURCES.md`. - NOAA HRRR fields for regional weather inputs. - NASA FIRMS active-fire detections for occupancy supervision. - LANDFIRE fuel and canopy layers for static landscape context. - Wildfire Risk to Communities housing density and LandScan population for exposure context. - WFIGS and MTBS event-level resources for burned-area and analog tasks. - External Earth-FM and backbone assets for comparator features. --- ## Repository Layout ```text WildFIRE-FM/ README.md LICENSE requirements.txt models/wildfire_fm/ modeling_unet.py checkpoint_manifest.json checkpoints/seed_*/best_firms_prauc.pt training/ build_phase1_cache_regional_hrrr.py train_cold_tiled_mainline.py configs/ spatial_serving/ grid_to_polygons.py data_downloader/ hrrr_downloader.py paper/ wildfire_fm_evaluation_contracts.pdf assets/ wildfire_fm_model_card.svg release_contents.svg selection_regret_final.png supporting_rank_map_final.png primary_rank_change_final.png paper_outputs/figures/ matching.pdf fig_task_contract_tiles.pdf fig_selection_regret_scatter.pdf fig_rank_heatmap1.pdf artifacts/ manifests/ results/ data_sources/ DATA_SOURCES.md experiments/ raw_reference/ slurm/ scripts/ audit_release.py reproduce_paper_outputs.py ``` --- ## How To Cite If you use WildFIRE-FM, the released checkpoints, the fixed-contract evaluation artifacts, or the paper-aligned scripts, please cite: ```bibtex @misc{wildfire_fm_evaluation_contracts_2026, title = {Does Your Wildfire Prediction Model Actually Work, or Just Score Well?}, author = {Yangshuang Xu and Yuyang Dai and Liling Chang and Qi Wang and Yushun Dong}, year = {2026}, note = {WildFIRE-FM model and fixed-contract wildfire evaluation artifacts} } ``` The citation will be updated with arXiv metadata after the preprint is public. --- ## News - **2026-05-18** - The model card was reorganized with a larger navigation panel and final-paper visual previews. - **2026-05-13** - The Hub release added the compiled paper PDF without checklist pages. - **2026-05-13** - Five seeded WildFIRE-FM checkpoints were uploaded with SHA-256 metadata. --- ## Scope WildFIRE-FM is scoped to the paper's California regional grid and fixed-contract wildfire transfer comparisons. Use in a new region or grid should start with local preprocessing, validation, and contract-specific evaluation. The release is designed for research inspection, artifact-level checking, and model-loading experiments rather than operational alerting. --- ## License This release is provided under the MIT license. External data and comparator assets remain governed by their original providers and terms of use. --- WildFIRE-FM is released to make wildfire Earth-FM transfer comparisons easier to inspect, reproduce at the artifact level, and evaluate under explicit contracts.