FloodCastBench¶

Synthetic-backed inundation benchmark adapter aligned to the FloodCastBench evaluation ecosystem.

Overview¶

FloodCastBench is the public inundation adapter used by PyHazards for raster flood prediction benchmarks.

The current implementation is synthetic-backed, but it preserves the raster task and metric surface used by the shared flood evaluator.

At a Glance¶

Provider	FloodCastBench ecosystem surfaced through a PyHazards adapter
Hazard Family	Flood
Source Role	Inundation Benchmark
Coverage	Benchmark-aligned flood inundation samples
Geometry	Raster inundation sequences
Spatial Resolution	Benchmark-defined raster tiles
Temporal Resolution	Short history windows with next-horizon inundation targets
Update Cadence	Generated locally for smoke and benchmark-alignment runs
Period of Record	Synthetic-backed benchmark adapter
Formats	PyTorch tensors via the dataset registry
Registry Entry	`floodcastbench_inundation`

Data Characteristics¶

Multi-step raster inputs paired with next-horizon inundation targets.
Registry-backed benchmark adapter rather than a raw external dataset ingestion path.
Intended for pixel-level evaluation such as IoU and pixel MAE.

Typical Use Cases¶

Smoke tests for FloodCast and UrbanFloodCast.
Shared flood benchmark runs on inundation tasks.
Regression checks for raster flood prediction outputs.

Access¶

Use the links below to access the upstream source or its public documentation.

FloodCastBench repository

PyHazards Usage¶

Use this adapter when you want the public FloodCastBench-aligned inundation surface exposed by the flood benchmark.

Registry Workflow¶

Primary dataset name: floodcastbench_inundation

from pyhazards.datasets import load_dataset

data = load_dataset(
    "floodcastbench_inundation",
    micro=True,
    history=4,
    channels=3,
).load()

train = data.get_split("train")
print(train.inputs.shape, train.targets.shape)

Inspection Workflow¶

This dataset is currently surfaced as a registry-backed benchmark adapter, so there is no standalone inspection CLI documented for it.

Notes¶

This is a synthetic-backed benchmark adapter rather than a full FloodCastBench ingestion pipeline.

Reference¶

FloodCastBench.