AZ-Hydro — Historical and Projected Arizona Annual Water Use: Software, Input Data, Models, Raster and Well Package Predictions, and Validation at 2 km Resolution (1896–2099)
Year: 2026
Type: dataset
Abstract: AZ-Hydro: Historical and Projected Arizona Annual Water Use, 1896–2099 (v1.2) A 2 km-resolution gridded dataset of Arizona groundwater and surface-water withdrawals, irrigation consumptive use, and pumping-induced stream capture, spanning 204 years (1896–2099), with per-pixel uncertainty from a six-component framework and a well-level package for ~170,000 registered consumptive wells. Companion data archive for Majumdar et al. (submitted to Nature Scientific Data; preprint https://doi.org/10.31223/X56N63) and Majumdar et al. (in prep., AGU Earth’s Future). Version 1.2 (code release v1.2.0). See What’s new in v1.2 below. What’s in this deposit File Size Contents Audience az-hydro-headline.7z ~8.7 GB Self-contained subset of every key output: six-band augmented per-pixel, per-year withdrawal, consumptive-use, and stream capture rasters (prediction, σ, CV, SNR, lower and upper 95 % CI); the GeoParquet well package; per-basin and per-sub-basin time series; the CAP shortage-scenario sweep; era-mean and trend figures; validation and intercomparison outputs. Most users want this. Reviewers and downstream users of the published product az-hydro-data.7z ~80 GB Full reproducibility archive (~220 GB uncompressed): everything in the headline archive plus the raw inputs, cross-validation outputs, intermediate input stacks, and per-component uncertainty rasters (σ_MACA, σ_model, σ_irr, σ_LULC, σ_GW, σ_USBR, σ_CU). Anyone re-running the pipeline from scratch az-hydro-1.2.0.zip ~64 MB Tagged source-code release (Python pipeline, GEE export and upload scripts, documentation), identical to the GitHub repository at tag v1.2.0. Anyone running the pipeline README.md, HEADLINE_README.md, ADWR_readme.txt ~50 KB Current copies of the guides inside the archives: Data/README.md (full archive), Data/HEADLINE_README.md (headline archive), and the ADWR folder’s readme.txt. They supersede the copies inside the .7z files, which predate the final author list and citations. Everyone Inside each .7z, Data/HEADLINE_README.md (headline archive) or Data/README.md (full archive) gives the per-directory inventory and the external-source citations. The standalone copies of these files in this deposit are the current versions. How to unpack the .7z files The archives use LZMA2 with solid-block compression, which macOS Archive Utility cannot open. Use one of: macOS: Keka or The Unarchiver Windows: 7-Zip, WinRAR, or Bandizip Linux: p7zip (for example apt install p7zip-full, then 7z x az-hydro-data.7z) What’s new in v1.2 Post-2026 Colorado River shortage. The central projection now uses the 2027–2028 Operating Guidelines (U.S. Department of the Interior, August 2026), which reduce Arizona’s Colorado River apportionment by 760 kAF yr⁻¹ (CAP delivery factor 0.49, groundwater boost 7.5). The guidelines cover 2027–2028; 2029–2099 hold that latest official condition as a stand-in for continued shortage, not as a forecast. v1.0–v1.1 held the Tier 1 cut (512 kAF yr⁻¹) through 2099. Projected 2027–2099 groundwater withdrawal is 5.5 % higher and surface-water withdrawal 5.5 % lower than in v1.1; total withdrawal and all 1896–2026 predictions are unchanged. CAP scenario sweep. A ninth scenario applies the 2027–2028 Operating Guidelines: 652 kAF yr⁻¹ of CAP delivery, the 900 kAF yr⁻¹ post-Tier baseline less the 248 kAF yr⁻¹ by which the guidelines’ 760 kAF yr⁻¹ Arizona reduction exceeds the Tier 1 cut that baseline is taken to include. It adds 1.30 MAF of groundwater pumping over 2027–2060 and 3.15 MAF over 2027–2099. The other scenarios are unchanged. Uncertainty ensemble. The ensemble members now apply the same dynamic CAP caps and floors as the central partition (a missing import had disabled them). σ for the groundwater and surface-water categories changes in every year; σ for total withdrawal does not. The scenario sweep keeps its full-delivery counterfactual. ADWR partner CSVs (Outputs/.../ADWR/). In v1.1, 14 of the 16 CSVs listed every basin-year twice, because the export also picked up the all-basin aggregate files. Each row now appears once, and Total_Predicted is built from the same per-basin series as the other categories, so Total_GW + Total_SW = Total_Predicted holds row by row. Basin totals count boundary pixels in every basin they touch, so summing basins exceeds the statewide total by about 2–3 %; statewide totals are in Annual_Summaries/. Headline archive now also includes the out-of-distribution probability rasters (OOD_Rasters/, 68 MB). GEE assets for the AZ-Hydro Explorer moved to projects/azhydro/assets/az-wu-v1-2/. Methods overview AZ-Hydro is a physics-constrained machine learning pipeline: XGBRF prediction of total annual withdrawal depth per 2 km pixel, trained on ADWR metered pumping records (1984–2024) with 22 predictors (actual and reference ET, precipitation, effective precipitation, temperature, land-cover fractions and densities, irrigated and groundwater-irrigation fractions, soil properties, water-table depth, well, surface-water-rights and canal densities, watershed streamflow, and pixel coordinates). Climate, ET, and land-cover inputs from different eras are joined into continuous 1896–2099 series by overlap-period bias correction. Density-ratio partition of the total into irrigation and non-irrigation withdrawals from groundwater and surface water, with consumptive use derived as CU = irrigation efficiency × irrigation withdrawal. The few free partition parameters were tuned to statewide USGS and ADWR records (USGS OFR 94-476, 1915–1990; USGS Circulars, 1950–2015; ADWR Annual Report totals and the 2019 ADWR shares); the machine learning totals are never adjusted. Six-component uncertainty quantification: σ_MACA (5 GCMs), σ_model (10 XGBRF seeds), σ_irr (irrigated-area input), σ_LULC (4 USGS land-use scenarios), σ_GW (5 recent well-registry snapshots), and σ_USBR (5 Upper Colorado River streamflow members), combined in quadrature and propagated to consumptive use (σ_CU). Stream capture index combining connectivity (exponential decay with water-table depth) and nearby surface-water availability, with pumping and connectivity uncertainty propagated to the capture volumes. Key validations Statewide agency reconciliation (calibration adequacy, not an independent test): 2017 ADWR total withdrawal 6.81 vs 7.00 MAF (−2.7 %); 2015 USGS groundwater withdrawal 2.96 vs 3.09 MAF (−4.1 %); 2019 ADWR irrigation share 74.2 % vs 72 %; 1955 and 1980 pumping peaks under-predicted by 6–11 %, with every reported total inside the model’s 95 % σ_total interval. Independent comparisons against USGS NHM withdrawals and consumptive use, USGS Reitz irrigation rasters, the USGS public-supply reanalysis, and CAP and SRP delivery records (statewide surface water within 0.8 % of the 1985–2024 CAP + SRP mean). WestWater Research (2026) CAP shortage anchor: Basic Coordination cumulative additional groundwater pumping 7.24 MAF over 2027–2060 vs 8.0 MAF (−9.5 %); Extreme Shortage 13.08 MAF vs 8.7 MAF (the gap reflects AZ-Hydro’s no-regulatory-ceiling framing). CAP delivery shortage scenario sweep Nine scenarios (the no-cut Baseline_900kAF; DCP Tiers 0, 1, 2a, 2b, and 3; the 2027–2028 Operating Guidelines; WestWater Basic Coordination; Extreme Shortage) re-partitioned over 2026–2099. Cumulative additional groundwater pumping over 2027–2060: Operating Guidelines 1.30 MAF, DCP Tier 1 4.05 MAF, DCP Tier 3 10.69 MAF, Basic Coordination 7.24 MAF, Extreme Shortage 13.08 MAF. Basin and per-pixel maps, cumulative σ, and signal-to-noise figures are included for the 2027–2060 and 2027–2099 windows. External datasets required to reproduce the pipeline (not redistributed here) The pipeline reads four USGS ScienceBase data products that must be downloaded separately: USGS NHM withdrawals (Haynes et al. 2023): irrigation withdrawals and efficiency by HUC12, 2000–2020 USGS NHM consumptive use reanalysis (Martin et al. 2023; Martin et al. 2025): irrigation consumptive use by HUC12 USGS public-supply reanalysis (Luukkonen et al. 2023; Alzraiee et al. 2024): public-supply withdrawals by HUC12 USGS Reitz historical ET (Reitz et al. 2023; Reitz, Sanford & Saxe 2023): 800 m gridded ET Bundled here directly: ADWR Well Registry and metered pumping records, GW basin / AMA-INA / CAP / SRP / streamflow / USBR vectors, HarDWR water rights (Lisk et al. 2024), GRAIN canal network (Suresh et al. 2026), and the Ma et al. (2026) water-table depth. See Data/README.md inside each .7z for the exact file names and folders. Citation Data archive (this deposit): Majumdar, S., Smith, R.G., ReVelle, P., Hasan, M.F., & Wogenstahl, C. (2026). AZ-Hydro — Historical and Projected Arizona Annual Water Use: Software, Input Data, Models, Raster and Well Package Predictions, and Validation at 2 km Resolution (1896–2099) (Version 1.2). Zenodo. https://doi.org/10.5281/zenodo.19057935 Companion papers: Majumdar, S., Smith, R., ReVelle, P., Hasan, M. F., & Wogenstahl, C. (2026). Historical and projected freshwater withdrawals and irrigation consumptive use for Arizona, 1896–2099. Submitted to Nature Scientific Data. Preprint at https://doi.org/10.31223/X56N63 Majumdar, S., Smith, R.G., ReVelle, P., Hasan, M.F., & Wogenstahl, C. (2026). Where Arizona’s Water Goes: Declining Agricultural Dominance and Rising Urban Demand Drive a Two-Century Shift in Withdrawal Patterns (1896–2099). In prep. for AGU Earth’s Future. License Data archives (az-hydro-data.7z, az-hydro-headline.7z): CC-BY-4.0 Source code (az-hydro-1.2.0.zip): BSD 3-Clause “Revised” (see LICENSE inside the zip) External datasets bundled here (HarDWR, GRAIN, WTD, ADWR products) retain their original upstream licenses Acknowledgments This work was supported by NASA (Grant numbers 80NSSC21K0979, 80NSSC23K1453, and 80NSSC26K0983) and U.S. Army Corps of Engineers (Grant number W912HZ25C0016). We thank the open-source software and data communities, the OpenET consortium,
Institutions: Desert Research Institute, Colorado State University
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7160256695