FireCastNet¶

Overview¶

firecastnet is a raster wildfire spread baseline that uses a shallow encoder-decoder architecture.

At a Glance¶

Hazard Family

Wildfire

Public catalog grouping used for this model.

Maturity

Implemented

Catalog maturity label used on the index page.

Tasks

Spread

Benchmark Family

Wildfire Benchmark

Primary benchmark-family link used for compatible evaluation coverage.

Description¶

firecastnet is a raster wildfire spread baseline that uses a shallow encoder-decoder architecture.

The PyHazards implementation is optimized for the shared smoke benchmark rather than the full upstream training stack.

Benchmark Compatibility¶

Primary benchmark family: Wildfire Benchmark

Mapped benchmark ecosystems: WildfireSpreadTS

External References¶

Paper: FireCastNet: Earth-as-a-Graph for Seasonal Fire Prediction | Repo: Repository

Registry Name¶

Primary entrypoint: firecastnet

Supported Tasks¶

Spread

Programmatic Use¶

import torch
from pyhazards.models import build_model

model = build_model(name="firecastnet", task="segmentation", in_channels=12)
logits = model(torch.randn(2, 12, 16, 16))
print(logits.shape)

Notes¶

The smoke configuration uses the single-frame wildfire spread raster fixture.