Green acreage in non-Pueblo MRGCD irrigable parcels, summer 2026
Area of green land in the non-Pueblo Middle Rio Grande Conservancy District, summers 2017–2026. Bars give the area of 10 m Sentinel-2 pixels with median June–August NDVI ? 0.45, summed over the 10,551 parcels of the district’s 2021 irrigation survey (58,145 acres; right axis expresses area as a percentage of that total). Error bars show the same count at NDVI cuts of 0.50 (lower) and 0.40 (upper); they express sensitivity to the cut, not a statistical interval. The parcel boundaries are fixed at their 2021 extent in every year, so each bar is the green area within those parcels, not a survey of that summer. *2021 is the year the cut was calibrated against the survey’s recorded irrigation labels. NDVI cannot distinguish irrigation by canal, well or municipal supply from shallow groundwater or rainfall. 2026 is highlighted (17.3 thousand acres; 13.9–21.2 across the cut range, against 36.2 thousand acres in 2017). Data: Copernicus Sentinel-2 surface reflectance (ESA), scene-classification cloud mask, summer median composite; parcels from the MRGCD 2021 ISOlog. Methodology caption written by Claude.
2026 provided a remarkable convergence of a natural experiment on New Mexico’s Middle Rio Grande (a river and therefore irrigation system with little or no water) and the rapid advancement of what I’ve come to think of as “Curiosity Exploder Tools” (CETs, also more commonly known as LLMs or AI).
We’ve got European Space Agency satellites flying overhead every five-ish days hoovering up data, Google Earth Engine processing power in “the cloud,” and now tools like Codex and Claude Code to handle the fiddly bits of code writing, driving down the transaction costs of asking a question by orders of magnitude.
So I wondered: With little or no Middle Rio Grande Conservancy District ditch water for non-Prior and Paramount Pueblo land this year, what happened? As Robin Sloan said, “‘Ask for what you want’ is a key skill for the 21st century.”
We don’t have meters on irrigated land in the Middle Rio Grande, so remote sensing is the best we can do.
This uses MRGCD’s spatial mapping of irrigated parcels, with Normalized Difference Vegetation Index (NDVI), a measure of plant greenness, using observations from the European Space Agency’s Sentinel-2 satellites. 10 meter resolution roughly every five days. “Green” here is not an either/or, it’s a more-or-less thing. I chose an NDVI cutoff based on a reading of 0.45, benchmarked against an MRGCD dataset from 2021 of parcels either irrigated or fallowed. The cutoff is arbitrary (but sane based on my 2021 control set), the relative scale is what matters.
The satellite can’t tell us the difference among ditch water, wells (both irrigation and domestic), municipal water, and trees with their roots in the shallow aquifer. But as I wrote with a group of colleagues earlier this year, “Agencies don’t use water, people do.” All four can be present on these parcels. I checked precipitation, it’s not throwing off the numbers.
I’ve now got a stunning set of maps of this on my hard drive – MRGCD non-Pueblo land by division and sub-section, year by year – a record of how people are greening the valley ditch by ditch and parcel by parcel. As Robin Sloan said, ask for what you want. Curiosity Exploder Tools.
However you interpret it, the satellite data suggests a significant reduction in 2026 consumptive use of water on these parcels, a remarkable water conservation success story, if you will.
n.b. “Study” design by John Fleck in conversation with Claude Code, data processing pipeline by Claude Code with Google Earth Engine analysis.