agrimanager/wofost-weather-regime-pool

WOFOST Weather-Regime Pool

This repository contains a WOFOST-Gym weather-scenario pool for the AgriManager weather-regime out-of-distribution benchmark. Each parquet row is a scenario key, not a full rollout trajectory: the crop is encoded by the file name, and the year, latitude, and longitude columns identify the weather scenario consumed by WOFOST-Gym at runtime.

Intended Use

Use this dataset to evaluate policies trained on non-extreme WOFOST-Gym weather scenarios and validated on interpretable weather-regime shifts. The intended benchmark compares in-distribution validation against drought, wet, hot, and cold validation regimes.

This dataset is not intended to be a standalone climate extreme detector, a global climate product, or a source of real crop yield labels.

Provenance

• Weather source: NASA POWER meteorological data, retrieved through PCSE's NASAPowerWeatherDataProvider.
• Simulator/runtime source: WOFOST-Gym, PCSE, and WOFOST.
• Derived artifact: AgriManager generated, filtered, labeled by weather-regime split, and packaged scenario rows and the accompanying PCSE weather cache archive.
• Ownership statement: this dataset does not claim ownership of NASA POWER, WOFOST, PCSE, or WOFOST-Gym.

Synthetic Data Status

croissant_rai.json sets rai:hasSyntheticData to true. The meteorological inputs are derived from NASA POWER weather data, but the released benchmark rows are generated and simulator-derived WOFOST-Gym scenario keys curated into fixed weather-regime evaluation splits by a deterministic AgriManager/WOFOST pipeline with documented preprocessing. This disclosure applies to the benchmark scenarios, splits, and regime labels, not to the underlying NASA POWER source data.

Crops and Regimes

The pool contains two crop shards:

chickpea and potato.

Weather-regime labels are encoded by split directory:

• train: non-extreme in-distribution training scenarios.
• val_id: held-out non-extreme in-distribution validation scenarios.
• val_drought: drought validation scenarios.
• val_wet: wet validation scenarios.
• val_hot: hot validation scenarios.
• val_cold: cold validation scenarios.
• val: convenience aggregate of val_id, val_drought, val_wet, val_hot, and val_cold.

Use the named validation split directories when preserving regime labels. The aggregate val/{crop}.parquet files do not include a weather_regime column.

Files

Path pattern	Count	Rows per file	Total rows	Description
train/{crop}.parquet	2	1,600	3,200	Non-extreme training scenarios.
val_id/{crop}.parquet	2	128	256	Held-out non-extreme validation scenarios.
val_drought/{crop}.parquet	2	128	256	Drought validation scenarios.
val_wet/{crop}.parquet	2	128	256	Wet validation scenarios.
val_hot/{crop}.parquet	2	128	256	Hot validation scenarios.
val_cold/{crop}.parquet	2	128	256	Cold validation scenarios.
val/{crop}.parquet	2	640	1,280	Aggregate validation files equal to the union of the five named validation splits.
meteo_cache.tar.gz	1	n/a	n/a	Bundled PCSE/NASA POWER cache files required by the scenarios.
croissant_rai.json	1	n/a	n/a	Croissant JSON-LD metadata with core fields and Responsible AI metadata.

Do not double-count val/ and the named validation directories. The aggregate val/ files are a convenience view over the named validation splits.

Schema

All crop parquet files use the same schema:

Column	Type	Units	Description
year	int64	calendar year	Year passed into the WOFOST agromanagement/weather scenario.
latitude	float64	decimal degrees	Scenario latitude, rounded to 2 decimal places.
longitude	float64	decimal degrees	Scenario longitude, rounded to 2 decimal places.

The crop label is encoded by the parquet file name. The weather-regime label is encoded by the parent split directory. The parquet files do not contain raw daily weather variables or rollout rewards.

Croissant Metadata

croissant_rai.json is the completed Croissant JSON-LD metadata file for this dataset. It contains the core Croissant metadata fields and the Responsible AI metadata fields used for NeurIPS 2026 Evaluations & Datasets review, including dataset limitations, data biases, personal or sensitive information, intended use cases, social impact, synthetic-data status, source datasets, and provenance activities.

Weather Cache

meteo_cache.tar.gz contains the PCSE NASAPowerWeatherDataProvider cache files needed by these rows. The AgriManager runtime helper agrimanager.env.wofost_gym.weather_pool.ensure_pool() downloads or locates the dataset and extracts this archive before WOFOST-Gym execution.

The cache is included so reviewers and users can:

• run the exact same weather scenarios without re-downloading from NASA POWER,
• avoid NASA POWER API failures, rate limits, or future data/version drift,
• reproduce WOFOST-Gym runs from the fixed scenario rows, and
• load the pool through AgriManager's runtime helper without manually preparing PCSE weather files.

The packaging script includes cache files for each scenario year plus a one-year padding window on both sides (year_padding=1) so WOFOST-Gym has the weather context required around each season.

Generation, Filtering, and Labels

The experiment configs consume this materialized Hugging Face weather pool; they do not recompute the source-level filtering logic at training time.

The underlying weather-scenario construction follows the same AgriManager WOFOST weather-pool process:

• scenarios are keyed by (crop, year, latitude, longitude),
• years are drawn from the WOFOST weather-pool year window,
• latitude and longitude are sampled from crop-specific windows and rounded to 2 decimals,
• candidate scenarios are executed in a no-op WOFOST-Gym episode,
• scenarios are kept only when PCSE/NASA POWER weather loads completely, the crop reaches at least min_dvs_threshold: 1.5, and storage organ biomass is non-zero (max WSO > 0).

The named validation regimes are equal-size, crop-specific 20% buckets. For a crop-specific candidate weather set S_c, define:

• season_rain(s): cumulative RAIN over the WOFOST growing season.
• season_temp(s): mean daily TEMP over the WOFOST growing season.

With p = 0.20, the regime predicates are:

drought(s) = season_rain(s) is in the lowest  p fraction of S_c
wet(s)     = season_rain(s) is in the highest p fraction of S_c
hot(s)     = season_temp(s) is in the highest p fraction of S_c
cold(s)    = season_temp(s) is in the lowest  p fraction of S_c
normal(s)  = not (drought(s) or wet(s) or hot(s) or cold(s))

AgriManager experiment configs use sampling.generation_seed: 42 and attach labels.weather_regime values for id, drought, wet, hot, and cold. The local T1.1 dataset builder reads the already materialized val_* splits and propagates their weather_regime labels into generated parquet artifacts.

Split Construction and Leakage Prevention

The hosted validation aggregate was verified to match the union of the named validation split files for each crop:

• val/{crop}.parquet contains 640 unique rows,
• the five named validation splits contain 640 unique rows in total,
• all 640 aggregate rows match the named validation split rows.

Benchmark code should sample from the physical split directories and deduplicate by (crop, year, latitude, longitude) when constructing downstream training and evaluation artifacts.

Limitations

• The regime labels are benchmark labels for WOFOST-Gym scenario evaluation, not observed climate-disaster annotations.
• The pool is simulator-oriented: it contains scenario keys and cache files for WOFOST-Gym, not observed management decisions or measured crop yields.
• The WOFOST viability filter removes scenarios that fail simulator completeness or maturity/yield checks, so the pool is not an unbiased sample of all NASA POWER grid cells.
• Reproduction requires compatible versions of AgriManager, WOFOST-Gym, PCSE, and their crop configuration files.

Size and Inspection Sample

The current downloadable meteo_cache.tar.gz runtime artifact for this repository resolves to 289,718,662 bytes, which is under 4 GB. The tabular scenario files are small parquet files and are previewable through the Hugging Face dataset viewer.

Because the current hosted dataset files and runtime cache artifact are under the NeurIPS 4 GB inspection-sample threshold, this repository does not require a separate size-triggered small inspection sample. Users can inspect the parquet scenario rows directly and download the full cache artifact for reproduction.

License

This dataset is released under the CC-BY-4.0 license.

Users may share and adapt the dataset with appropriate attribution.

Citation

If you use this dataset, cite the accompanying AgriManager benchmark paper or code release, NASA POWER, and the upstream WOFOST-Gym/PCSE/WOFOST software used to execute the scenarios.