FPA-FOD Weekly

Weekly FPA-FOD aggregates packaged for next-week wildfire count forecasting by size group.

Overview

FPA-FOD Weekly builds rolling lookback windows from weekly wildfire incident counts and predicts next-week counts for grouped size classes.

PyHazards exposes it as a loadable forecasting dataset with a micro mode so sequence models can be validated without the full source archive.

At a Glance

Provider

Fire Program Analysis Fire-Occurrence Database (FPA-FOD) adaptation in PyHazards

Hazard Family

Wildfire

Source Role

Weekly Forecasting

Coverage

User-provided FPA-FOD coverage

Geometry

Temporal tabular sequences

Spatial Resolution

Weekly aggregate windows

Temporal Resolution

Weekly

Update Cadence

User-managed local inputs or deterministic micro mode

Period of Record

Depends on the supplied FPA-FOD source files

Formats

SQLite, DB, CSV, and Parquet inputs

Inspection CLI

python -m pyhazards.datasets.fpa_fod_weekly.inspection --micro --lookback-weeks 12

Registry Entry

fpa_fod_weekly

Data Characteristics

  • Predicts next-week counts for grouped size classes A/B/C/D/EFG.

  • Supports feature modes with counts only or counts plus seasonal time features.

  • Uses chronological splits to preserve the forecasting setting.

  • Returned splits follow the DataBundle contract with sequence inputs and floating-point targets.

Typical Use Cases

  • Weekly wildfire forecasting experiments.

  • Sequence-model smoke tests for wildfire activity prediction.

  • Lightweight benchmarking of tabular temporal wildfire baselines.

Access

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

PyHazards Usage

Use this dataset through the public inspection or registry surface documented below.

Registry Workflow

Primary dataset name: fpa_fod_weekly

from pyhazards.datasets import load_dataset

data = load_dataset(
    "fpa_fod_weekly",
    micro=True,
    features="counts+time",
    lookback_weeks=12,
).load()

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

  • features=’counts’ uses only the five weekly count channels.

  • features=’counts+time’ adds sinusoidal week-of-year features for seasonality.

Inspection Workflow

Use the documented inspection path below to validate local files before training or analysis.

python -m pyhazards.datasets.fpa_fod_weekly.inspection --micro --lookback-weeks 12

Reference