sinuosity/ghana-food-security

Ghana Food Security

Part of the Ghana Agricultural Data Commons v0.2.0 public dataset release by Sinuosity Physical Intelligence Lab. Repository payloads include redistributed source data only where licensing and publication gates allow; other entries are registry, provenance, governance or builder metadata.

This dataset is part of the Ghana Agricultural Data Commons from Sinuosity Physical Intelligence Lab.

GhanaAgData is a systematic, publicly documented catalogue and machine-learning-ready collection of Ghanaian agricultural datasets discoverable through the sources and search procedures described in the project methodology as of 2026-07-25. It does not claim literal completeness across all Ghanaian agricultural data.

Dataset Summary

Catalogue and available redistributable crop-yield inputs relevant to food-security analysis.

This repository includes actual redistributable data payloads that passed the release builder gates.

Sinuosity Physical Intelligence Lab did not create the source data produced by Ghanaian institutions, international institutions, or original publishers. This release adds an interoperability, provenance, validation, and machine-learning layer around discoverable and legally reusable sources.

Supported Tasks

• food-security source discovery
• yield context analysis
• builder planning

Contents

• data/dataset_registry.csv, .jsonl, .parquet: filtered registry rows for this repository.
• metadata/SOURCE_MANIFEST.json and .parquet: source URL, license, retrieval, checksum, and access classification per row.
• metadata/FILE_MANIFEST.csv and .json: checksums for files in this repository.
• reports/: validation, license, quality, and publication evidence.
• docs/: search methodology, limitations, user guide, provenance notes, and reproduction guidance.
• data/source/<dataset_id>/: original redistributable source files copied without modification.
• data/processed/crop_observations.parquet: normalized long-format crop observations with row-level provenance.
• data/source_tables/: source-shaped tabular extracts where generated from source files.

Release Status

• Release version: 0.2.0
• Build date UTC: 2026-07-25
• Registry rows in this repository: 3
• Open/API-downloadable registry rows: 3
• Controlled, restricted, unclear, or metadata-only rows: 0
• Redistributable source files included: 1
• Normalized processed observation rows: 110
• Benchmark rows: 0

Dataset Structure

The registry files describe every source. Where actual data payloads are included, the release preserves both source provenance and normalized observations.

Key normalized observation fields include record_id, dataset_id, year, admin1_name, admin2_name, commodity_name, indicator_id, value_original, unit_original, value_standardized, unit_standardized, source_file_sha256, and provenance_uri.

Loading Example

from huggingface_hub import hf_hub_download
import pandas as pd

path = hf_hub_download(
    repo_id="sinuosity/ghana-food-security",
    repo_type="dataset",
    filename="data/dataset_registry.parquet",
)
registry = pd.read_parquet(path)

For repositories with processed crop observations, replace the filename with data/processed/crop_observations.parquet.

Licensing And Use

No single blanket license is assigned to the underlying source material. License, access, redistribution, attribution, sensitivity, and source fields are recorded per row in the registry and source manifest. Users must follow the original publisher terms for every source.

Redistributable files in data/source/ are included only when the registry marks redistribution as allowed, the file exists locally, the checksum is recorded, and the file is an actual tabular payload rather than an HTML landing page or metadata snapshot.

Personal And Sensitive Information

This release does not publish restricted survey microdata, personal farmer or household records, precise private farm coordinates, images, EXIF data, or controlled-access source files.

Known Limitations

• Discovery coverage is systematic but not exhaustive.
• Some public pages do not provide redistribution rights, so they remain metadata-only.
• Some API records are published as builder guidance until source-specific extraction and redistribution terms are verified.
• Administrative-boundary crosswalks and unit ontologies are represented as release scaffolding, not final authoritative mappings.
• The average-yield source file does not embed a unit label in the downloaded CSV, so normalized rows preserve original values without inventing a standardized unit.

Recommended Uses

• Source discovery and legal triage for Ghana agricultural data projects.
• Reproducible crop-statistics analysis where payloads are included.
• Baseline benchmarking for district crop-yield forecasting in ghana-ag-bench.
• Planning controlled data-access workflows with clear exclusion of restricted microdata.

Out-of-Scope Uses

• Treating this release as a complete inventory of all Ghanaian agricultural data.
• Using metadata-only rows as if source payloads were included.
• Republishing restricted, unclear-license, or application-controlled source data without permission.
• Making operational farm, credit, insurance, pesticide, or safety decisions without domain review and current source validation.

Citation

Please cite the original data producers for each source and cite this interoperability release as described in CITATION.cff.

Maintenance

Sinuosity Physical Intelligence Lab welcomes corrections from original publishers and users. Report source, license, schema, or provenance issues through the linked project issue process or maintainer channel when available.