Annual, 30 m maps of irrigated agriculture across the western United States, 1985–present, produced with Random Forest classification of Landsat imagery in Google Earth Engine.
Coverage: Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, Wyoming.
The current product version is 1.2, published in the Earth Engine public data catalog:
import ee
ee.Initialize()
# public catalog collection: irrigated pixels = 1, others masked
irr = ee.ImageCollection('UMT/Climate/IrrMapper_RF/v1_2')The working collections carry one image per state-year (e.g. MT_2024):
| Collection | Classes |
|---|---|
UMT/Climate/IrrMapper_RF/v1_2 (public catalog) |
irrigated = 1, others masked |
projects/ee-dgketchum/assets/IrrMapper/IrrMapperComp |
0 irrigated, 1 rainfed, 2 uncultivated, 3 wetland |
projects/ee-dgketchum/assets/IrrMapper/version1_2 |
irrigated only (class 0, masked) |
Version 1.1 (deprecated, still served as UMT/Climate/IrrMapper_RF/v1_1;
1986–2018) is the original 2020 Remote Sensing training data re-run on
Landsat Collection 2. Version 1.2 expanded the training data, moved to
per-state Random Forest models, and added the validation and uncertainty
analysis published with the 2023 Communications Earth & Environment paper.
Training data (labeled polygons: irrigated with year annotation, dryland, fallow, uncultivated, wetland) are published at Zenodo, DOI 10.5281/zenodo.17980068.
- Points are sampled from labeled polygons (
irrmapper/sampling/points.py). - A multi-season Landsat feature stack — surface reflectance composites,
vegetation indices, GRIDMET climate, terrain, and ancillary layers — is
sampled at the points in Earth Engine (
irrmapper/features/stack.py,irrmapper/sampling/extracts.py). - Per-state training tables are assembled and the most important ~50 features
selected (
irrmapper/sampling/tables.py,irrmapper/models/rf_sklearn.py); the feature lists used by every production run are archived inprovenance/variable_importance/. - A Random Forest (150 trees) is trained per state and applied per year
(
irrmapper/models/rf_ee.py::export_classification). - State-specific rules clean the raw classifications (frequency, NDVI, slope,
center-pivot evidence) before export to the published collections
(
irrmapper/postproc/,irrmapper/assets/).
Metadata manifests for every published asset are in provenance/. The
product's full history and reproduction chain are documented in
docs/provenance.md and
docs/reproducibility.md.
irrmapper/— production pipeline package (Earth Engine + scikit-learn):ingest/(Landsat, CDL, WorldClim),features/(the classifier's feature stack),sampling/(points, extracts, training tables),models/(EE and scikit-learn Random Forests),postproc/(cleanup rules, raster exports),validation/,assets/, and thecli.pyrunnerconfigs/— canonical run configurations (TOML);irrmapper_v1_2.tomlreproduces the current production flowlegacy/— quarantined v1.2-era and dissertation scripts, kept runnable against the package but outside itdocs/— product provenance and reproducibility documentationtraining_redevelopment/— experimental ML (MLP/U-Net, embedding-based classification) toward IrrMapper v2provenance/— asset metadata manifests and archived per-run feature lists
Managed with uv:
uv sync --all-extras
# dry run: print the planned Earth Engine tasks for a stage
uv run irrmapper configs/irrmapper_v1_2.toml classify
# start the tasks (writes a resolved-run manifest to provenance/runs/)
uv run irrmapper configs/irrmapper_v1_2.toml classify --executeIrrMapper is under active development with NASA ROSES support (OSIRIS and
FM4Irr projects, 2026–2029): the pipeline is being refactored into a
config-driven, pip-installable irrmapper package with multi-sensor inputs
(Sentinel-1/2, VIIRS) and expanded coverage. The v1.2-2025 tag preserves the
as-run December 2025 production state.
Ketchum, D., Jencso, K., Maneta, M.P., Melton, F., Jones, M.O., Huntington, J. (2020). IrrMapper: A Machine Learning Approach for High Resolution Mapping of Irrigated Agriculture Across the Western U.S. Remote Sensing 12(14), 2328. https://doi.org/10.3390/rs12142328
Ketchum, D., Hoylman, Z.H., Huntington, J., Brinkerhoff, D., Jencso, K.G. (2023). Irrigation intensification impacts sustainability of streamflow in the Western United States. Communications Earth & Environment. https://doi.org/10.1038/s43247-023-01152-2
Apache-2.0