Two linked views of irrigated agriculture. The national view maps where the United States irrigates,
county by county. The High Plains Aquifer view opens the results of my 30-year hydro-economic
optimization model for 180 counties: observed water-level trends, pumping costs, the year irrigation
gives way to dryland farming, and the bioenergy potential of the land that leaves irrigated maize.
Hover for values, click a county for its full trajectory, and press play to run the simulation.
Loading county data…
How to read the High Plains view
Each county's trajectory comes from a profit-maximizing optimization solved over a 30-year horizon.
In every year the model chooses the crop (irrigated maize, irrigated sorghum, or switchgrass) and the
well capacity that maximize net returns, subject to the groundwater remaining under the county, the
cost of lifting it, and the yield penalty of pumping at lower capacity.
Dryland transition year is the first year irrigation is no longer economical.
Storage remaining is groundwater in storage as a share of the starting volume.
Bioenergy NPV is the 30-year net present value of switchgrass on land leaving maize.
Data and methods
National irrigated acres: USGS, Estimated Use of Water in the United States, County-Level Data for 2015. County area is computed on the sphere from the US Atlas county boundaries.
Satellite layers: LANID, the Landsat-based Irrigation Dataset for the United States (Xie, Gibbs, and Lark, 2021, Earth System Science Data; CC BY 4.0). County shares are estimated from a systematic sample of the 30 m annual maps, one pixel per 600 m cell, assigned to counties in GIS.
High Plains results: Ramaswamy, Uludere-Aragon, and Lark (2026), Agricultural Water Management 334, 110675, and the supplementary data package on Zenodo (CC BY 4.0).
Aquifer boundary: USGS High Plains Aquifer boundary (Qi, 2010).
D3 and TopoJSON on a static page, with no tracking and no external services. Model inputs were assembled
with Python and ArcGIS from USDA NASS, USGS, PRISM, and LANID sources; the optimization was solved in GAMS.
The map data files are plain JSON, so the same figures can be reproduced from the Zenodo package.