The fixThe proposalLaw of the RiverAfter 2026Making waterObjections answered
STEPS VENTURES · BASIN 2.0 · UPDATED 4 AUGUST 2026

What the Post-2026 rules change

Reclamation released the Final Environmental Impact Statement on 31 July 2026. The seven states missed both consensus deadlines, so Interior advanced its own preferred alternative. Operations under it begin 1 January 2027. This page re-runs the analysis already on this site against the new numbers.

The short version. The new framework is adaptive rather than fixed: rules re-set every two years inside sideboards, instead of a twenty-year schedule. Lake Powell releases are bounded between 5.0 and 12.0 million acre-feet. Lower Basin shortages of up to 3.0 maf a year are permitted. Conservation storage accounts are enlarged to 8.0 maf in Powell and 3.0 maf in Mead. Three consequences follow that were not obvious before: a 3.0 maf shortage cannot be bought out of Lower Basin agriculture at any sane price. The release sideboards alone swing Glen Canyon's output by more than a third. And the storage accounts are now far larger than the water anyone can plausibly free up to fill them.
Sourcing note. Final EIS headline figures verified against the DOI release on 2026-08-09. Release tiers keyed to 3,540 ft are NOT verified as part of the new framework. They should still be checked against the Record of Decision when it issues. Everything else here is measured or modelled from public data, and the code is linked.

1. What actually changed

What it means, and what it does not settle

Sideboards are not a schedule. The 2007 Guidelines set operations by elevation tier, so a reservoir level implied a release. This framework sets outer bounds and re-decides inside them every two years. That is more responsive to hydrology and materially less predictable for anyone financing a twenty-year asset against a delivery assumption. Any number in this analysis keyed to a fixed release tier should be read as illustrative rather than as an operating rule.

Interior acted because the states did not. Seven states missed both consensus deadlines, so the federal government advanced its own preferred alternative. The framework explicitly leaves room for a consensus agreement to supersede it, which means the negotiation has not ended, it has been given a default outcome to negotiate against.

Where the exposure lands. Shortage of up to 3.0 maf is roughly 40% of the Lower Basin's apportionment. Because reductions beyond the negotiated tranche revert toward prior appropriation, which protects the most senior rights, the burden falls on junior users first. California's Imperial Valley rights are among the oldest on the river and the Central Arizona Project, perfected in 1968, is among the most junior large rights on the system. Arizona therefore carries more of the downside than its share of the apportionment would suggest.

Reactions have ranged from encouraged to draconian. Interior has presented the framework as balancing flexibility against predictability. Basin coverage since publication records sharply divided state and stakeholder reaction, with some parties treating the adaptive structure as workable and others as an imposition. We have not attempted to characterise any individual state's position here, because the record is still moving and misattributing a position is worse than omitting it.

What this does not change. Nothing in the framework charges Upper Basin allocations for reservoir evaporation, establishes basin-wide groundwater accounting, closes the shepherding gap, or quantifies outstanding tribal rights. Those are the structural failures set out on the methodology page, and a rules package that redistributes shortage does not address any of them.

A Record of Decision is expected around October 2026.

2. A 3.0 maf shortage cannot be bought

This site already published a measured demand curve for Lower Basin agricultural water: satellite-measured consumptive use across four counties, allocated over the crop map and priced by gross output per acre-foot. Walking up that curve, buying the cheapest water first:

Shortage to coverWater availableAverage $/AF Marginal $/AFAnnual cost
0.5 maf0.50 maf$274$320$0.14B
1.0 maf1.00 maf$310$362$0.31B
1.5 maf1.50 maf$327$362$0.49B
2.0 maf2.00 maf$336$362$0.67B
3.0 maf2.92 maf$565$4,231$1.65B

The first 1.5 maf is genuinely cheap, because it is wheat, hay and the low-value end of alfalfa. Past roughly 2.7 maf the curve runs out of sub-$400 water entirely, and the next acre-foot costs thousands, because what remains is lettuce, grapes and carrots.

The finding. The measured four-county pool is 2.92 maf in total. A 3.0 maf shortage is larger than all the agricultural water in those counties. It cannot be met by buying out farming there at any price, which means the new rules contemplate shortages that conservation markets alone cannot absorb.

3. The release sideboards swing the power

Glen Canyon's generation scales with release volume. Running the measured 2015 hourly release shape, scaled to each volume with the LTEMP minimum flows preserved:

Annual releaseGenerationEnergy value at $30-60/MWh
5.0 maf2.58 TWh$77M - $155M
6.0 maf3.09 TWh$93M - $185M
7.0 maf3.61 TWh$108M - $216M
8.0 maf4.12 TWh$124M - $247M
8.23 maf4.24 TWh$127M - $254M
10.0 maf5.15 TWh$155M - $309M
12.0 maf6.18 TWh$185M - $371M

The sideboards alone swing output by about 3.6 TWh a year. That is the Colorado River Storage Project Basin Fund's exposure, and under the new framework it is set biennially by a discretionary federal decision, not by hydrology alone. Any revenue model that shares that interconnection inherits the same exposure.

4. The storage accounts are bigger than the water

Eleven million acre-feet of conservation storage is now authorised across the two reservoirs. The entire measured agricultural pool in the four counties studied is 2.92 maf, and acquiring all of it would cost roughly $1.6B at measured gross values.

The finding. Storage is no longer the binding constraint. Acquisition is. That is an argument for supply-side measures that add water without taking it from someone, and against assuming demand-side conservation can fill the accounts.

5. Floating solar, re-tested against the water question

Our earlier floating-solar page found the idea is energy-led and bounded by the transmission tie at about 1 GW per dam. The new rules make the water side worth stating precisely, using measured reservoir areas and Reclamation evaporation rates.

ReservoirTotal evaporationWater saved per GWp Surface covered per GWp
Lake Powell343,741 AF/yr11,007 AF/yr4.3%
Lake Mead456,402 AF/yr11,934 AF/yr3.5%
Water and energy scale in opposite directions. At the deployment the grid can absorb, about 1 GW per dam, the arrays save roughly 23,344 acre-feet a year. That is 0.78% of a 3.0 maf shortage. Reaching 300,000 AF/yr would take roughly 26 GWp and about 50% coverage of both lakes, 13 times what the ties carry. The water benefit only becomes material at a scale the grid cannot absorb.

There is a harder problem underneath. Mainstem reservoir evaporation is a system loss: it is not charged against Arizona's, California's or Nevada's apportionment. Intentionally Created Surplus and system conservation both require a verifiable reduction in otherwise beneficial consumptive use, meaning a foregone diversion at a headgate. No non-diversion measure has ever been credited. So suppressed evaporation raises the pool for everyone and earns whoever paid for it nothing. Valued at the $400/AF Reclamation and IID actually paid in 2025-26, the water from 2 GWp is worth about $7.6M a year against roughly $2.5B of capital, a return of about 0.3%.

That is not an argument against doing it. It is an argument that it is public infrastructure rather than a private project, and that the accounting rule, not the physics, is what currently makes it unfinanceable.

6. Poking a hole in our own work with the satellite

We set out to test a suspicion: that a moored array has to stay afloat through the whole operating range, so the shallow margins that go dry on drawdown are not buildable, and every coverage fraction quoted against current surface area is therefore too generous. We measured the persistently inundated area , pixels wet in every summer from 2019 to 2026.

ReservoirMax extent (any year)Current (2026) Persistent (every year)Persistent share of current
Lake Mead82,31266,59065,41098%
Lake Powell91,36244,76641,72293%

The suspicion was wrong, and that is worth saying plainly. Persistent water is 93-98% of today's surface at both reservoirs. Both lakes have already drawn down so far that what remains is deep and stable, so the buildable footprint is not the binding constraint. The transmission tie still is. At 10-20% of persistent water the deployable array would be 4.3-8.7 GWp, against ties that carry about 2 GWp and a 300,000 AF/yr target that needs roughly 26 GWp.

One real caveat did survive. Powell's persistent water is only 46% of its maximum extent since 2019. The constraint binds on the upside: an array sited on today's shoreline is safe, but half of Powell's historic footprint is now intermittent, so a recovery scenario is where mooring design gets expensive.

The cross-check that did find something

The same run pulled gridMET reference evapotranspiration over each lake and converted it to an open-water estimate. It disagreed sharply with a published evaporation rate we had been using for Powell, and the disagreement went against us:

We had reconciled the gap by assuming a large bank-storage residual seeping into the Navajo sandstone. With the corrected rate no residual is needed, so that explanation was probably wrong too. Powell's evaporation and the water an array there could save are both about 50% higher than we first published. The headline conclusion did not move: at the interconnection limit the arrays still save under 1% of a 3.0 maf shortage, which is the useful thing to know about a finding that survives a 50% error in one of its main inputs.

Sentinel-2 summer median MNDWI at 30 m. gridMET reference ET times an arid open-water coefficient of 1.05-1.25. Persistent water is necessary but not sufficient for siting: it says nothing about depth, anchoring, navigation, intake exclusion zones or Park Service restrictions, all of which subtract further.

7. Whose wires are they, anyway

The floating-solar analysis on this site, and the white paper it responds to, both assumed that a dam generating well below its nameplate leaves usable room on its transmission tie. We checked the actual position at Glen Canyon and that assumption does not hold.

There is a named mechanism for exactly this situation. Surplus Interconnection Service under FERC Order 845 lets a new generator use the unneeded portion of an existing generator's approved interconnection service without a full new request. WAPA is non-jurisdictional as a federal power marketing administration but has adopted Order 845 and processes such requests out-of-queue. The obstacle is vintage: Glen Canyon was built in the 1960s and its interconnection predates modern large-generator interconnection agreements, so WAPA's business practice requires a bespoke multi-party agreement among Reclamation, WAPA's merchant function, WAPA transmission and the new customer.

Line thermal ratings are Critical Energy Infrastructure Information and are not public. WAPA's OASIS nodes (CRCM and WALC) require a registered digital certificate, so we could not pull actual ATC figures or queue positions. Note also that as of April 2026 WAPA transferred functional control of some CRSP transmission to SPP while retaining the Glen Canyon to Pinnacle Peak lines under its own tariff in the WALC balancing authority.

8. The full candidate portfolio, measured

The white paper names Mead, Powell, Flaming Gorge, the reach below Hoover and Lake Las Vegas. We measured all five: persistently inundated area (wet every summer 2019-2026, the honest denominator for a coverage fraction) and gridMET-derived open-water evaporation.

ReservoirMax extentCurrentPersistent Persistent share of maxGWp at 10-20% coverEvaporation ft/yr
Lake Mead82,31266,59065,41079%2.65-5.296.6-7.9
Lake Powell91,36244,76641,72246%1.69-3.385.4-6.4
Flaming Gorge33,93628,13427,44681%1.11-2.224.0-4.7
Lake Mohave26,32624,99023,91791%0.97-1.947.1-8.4
Lake Las Vegas25522415159%0.01-0.017.2-8.6

Two of four tests came back cleanly. A thermal test of the warm-epilimnion claim at Glen Canyon returned a plausible forebay warming trend but a corrupted mid-lake control series, so we are not reporting a number from it. A chlorophyll baseline succeeded at four of five reservoirs and exceeded Earth Engine's memory limit at Powell.

9. The load nobody is talking about, twenty miles away

Everything above says the same thing from different directions: the constraint is not the water surface, it is what you can do with the electricity. Behind-the-meter load solves that in a way grid export cannot, because it never touches WAPA's wires and has no exposure to negative prices. It is also the only route to a scale at which the water benefit becomes material.

That is not hypothetical near Page, Arizona:

The structural catch, and the way through it. WAPA cannot sell federal hydropower to a private data centre. Reclamation law limits it to preference customers: municipalities, cooperatives, federal and state agencies, and tribes. Two routes exist. A load can sit inside a preference customer's territory, and the City of Page and the Navajo Tribal Utility Authority both qualify. Or a load can co-locate with a new non-federal generator and pay only for transmission service. A floating array is exactly that generator, which is what makes a private load at a federal dam workable without touching federal power at all.

And the trap. Cooling consumes water. At the commonly cited 20.6 acre-feet per MW per year for evaporative cooling, a 1 GW load would consume about 20,600 acre-feet a year, roughly twice what a matched 1 GWp array saves. Closed-loop or dry cooling drops that to a few hundred acre-feet and keeps almost the whole benefit. The choice of cooling decides whether co-location helps or hurts, and it is a design decision rather than a constraint.

One loop runs in the right direction. Closed-loop cooling raises power usage effectiveness from about 1.15 to between 1.4 and 1.8, so the same computing needs 20 to 50% more electricity. That additional demand is more solar that can be self-consumed, which is more covered surface, which is more water saved. Going water-free increases the water benefit rather than trading against it.

10. The thermal claim, tested properly

The white paper argues that falling Powell elevations brought the warm epilimnion close to Glen Canyon's intakes, which let smallmouth bass establish below the dam, and that shading the surface addresses this at source. A first attempt to test it failed: the band-ratio water mask let cloud through and returned physically impossible values for the control site. Redone with the Landsat quality-band flags, discarding cloud, cloud shadow, cirrus and snow, keeping only flagged water, and reporting the valid pixel count for every year so thin years can be thrown out rather than quietly reported.

YearForebay CpxMid-lake Cpx
201431.2262532.142945
201529.565731.153096
201631.266832.93087
201731.7268832.953247
201832.0560332.742926
201931.8464232.722887
202031.2760431.692846
202132.6552733.012482
202234.1347134.182190
202331.7956732.832806
202431.9859232.852880
202532.2753732.542561
202634.1840833.111970

All thirteen years survive the pixel threshold in both zones.

This is the first test here that supports the white paper rather than complicating it. It is a necessary condition for their argument, not proof of it. A satellite sees skin temperature and cannot see the temperature at penstock depth, which is what the claim is actually about. One honest confound: the forebay pixel count falls from 625 to 408 across the record as the pool drops, so part of the trend could be the sampled patch of water moving rather than the water warming. Resolving that needs in-situ profiles, which is exactly what a field campaign would provide.

11. Does the tie limit hold across hydrology, and what about Hoover?

Every interconnection number here previously came from a single year at one dam. Both gaps are now closed. Solar is held at one irradiance year on purpose so that any change is attributable to the release record alone.

Glen Canyon, measured sub-daily gage:

YearMean release cfsSolar-hydro corr0.5 GW1 GW2 GW3 GW
201512,320-0.020.0%1.8%31.4%50.0%
201712,241-0.070.0%1.5%30.6%49.2%
201912,701+0.010.0%2.4%32.7%50.9%
202110,817-0.010.0%0.7%27.6%46.7%
202312,115+0.071.0%5.9%33.1%51.1%

At 1 GW curtailment runs 0.7 to 5.9 percent across five real hydrology years, so the original single-year result was not an artifact of 2015. The correlation between solar output and the dam's own generation stays near zero or slightly negative in every year, which is the physical basis for saying the two share the wire rather than compete for it.

Hoover, real daily volumes with a synthetic intraday shape:

YearMean release cfsSolar-hydro corr0.5 GW1 GW2 GW3 GW
201513,004+0.110.0%0.0%2.2%19.0%
201712,058+0.120.0%0.0%1.4%17.4%
201911,763+0.130.0%0.0%1.1%17.2%
202112,630+0.120.0%0.0%1.9%18.5%
202310,284+0.120.0%0.0%0.7%14.8%

Hoover looks like the better host, taking about 2 GW where Glen Canyon takes 1, which is what the operations literature predicts since nearly all of Hoover's sub-daily swing is energy-driven. But this is the weaker of the two models. Lake Mohave re-regulates Hoover's releases and there is no public sub-daily gage below the dam, so the intraday shape is constructed rather than measured. Treat it as indicative and worth a proper look, not as evidence on a par with Glen Canyon.

A first version of this run aligned the solar and release series by array index. PVGIS returns UTC and USGS returns local standard time, so that put solar noon at 07:00 local and produced a spurious positive correlation. Corrected with a 7 hour shift, after which the 2015 correlation reproduces the original model's value to within 0.01. Curtailment figures remain PHYSICAL headroom and therefore upper bounds, per section 7.

What would change our mind

Sources

Grouped by how far we can vouch for each. The first two we used or checked directly. The last two we have not read, and those are the ones to verify before relying on anything here that rests on them.

Data used directly. Landsat 8 and 9 Collection 2 Level 2 surface temperature (LANDSAT/LC08/C02/T1_L2, LC09) with QA_PIXEL cloud and water flags. Sentinel-2 Level-2A (COPERNICUS/S2_SR_HARMONIZED). gridMET reference ET (IDAHO_EPSCOR/GRIDMET), Abatzoglou, J. T. (2013), International Journal of Climatology 33(1) 121 to 131, doi:10.1002/joc.3413. PVGIS v5.2 with the PVGIS-NSRDB radiation database, returned in UTC. USGS gages 09380000 at Lees Ferry (instantaneous) and 09421500 below Hoover Dam (daily only), returned in local standard time. CAISO day-ahead prices at node TH_SP15_GEN-APND. Measured crop consumptive use from OpenET.

Literature, citations checked. Rosenlieb, E., Rivers, M. and Levine, A. (2025), "Floating photovoltaic technical potential: a novel geospatial approach on federally controlled reservoirs in the United States", Solar Energy 287. McKuin, B. et al. (2021), "Energy and water co-benefits from covering canals with solar panels", Nature Sustainability 4, 609 to 617.

Named but not read by us. Reclamation's Post-2026 Operational Guidelines Final EIS, FERC Order 845 and WAPA's surplus interconnection business practice, Reclamation Manual LND 08-01, 43 U.S.C. 485h(c), 54 U.S.C. 100902, the Glen Canyon LTEMP Record of Decision, and the Intentionally Created Surplus rules.

Secondary reporting, not independently verified. Navajo Generating Station retirement, transmission and water allocation. The proposed Page data-centre campus and the nearby pumped-storage preliminary permit. Data-centre load realisation rates. Prices paid for conserved water in 2025 and 2026. Data-centre cooling intensities, where published figures disagree by about a factor of two and both are carried rather than one chosen. One point we did check directly: the public record shows no operating US floating array above 10 MW.

The code

Steps Ventures. Measured where possible, modelled where not, and labelled either way. Figures derived from a Final EIS read second-hand are flagged above.