Drag the slider. The water saved keeps rising. The energy you can actually sell stops.
Covering a reservoir with floating solar panels sounds like it should solve two problems at once: shade the water so less of it evaporates, and generate clean power on infrastructure that already exists. The question is how much of the surface to cover. Proposals in circulation put it at 15–20%.
The answer is that the two benefits pull in different directions. Water saved rises steadily with coverage and never stops. Power you can actually sell does not: each dam's transmission line fills up, and past that point the extra panels generate electricity with nowhere to go. Where that happens is different at every reservoir. Move the slider and watch them come apart.
The dam's own generation takes the line first. This sets how much of what is left a solar array could really sell. 100% is generous: those lines carry firm hydropower contracts through 2057.
Load next to generation reduces what has to leave on the shared line. Lake Havasu has about 300 MW of Central Arizona Project pumping capacity on its shoreline. Read this as a sensitivity, not a design: co-located load is not a transmission bypass, and CAP pumping runs to the canal's delivery schedule rather than to the solar day.
Below Lake Mead elevation 1,035 ft, 12 of Hoover's 17 turbines cannot run and capacity falls about 70%; new wide-head units would soften that to about 58%. As the dam derates it vacates its own line.
Storage lets the array follow the dam's shape instead of dumping into the midday glut.
The higher cases price deep-water mooring in a reservoir that swings 100 feet, and quagga mussel fouling of floats and anchors.
What a generator pays for its own interconnection facilities and local network upgrades, from Berkeley Lab's project-level data. Heavily right-skewed, so the median and the average are far apart. Transmission access charges are billed to load, not generators, so there is no separate wheeling charge on top.
Real layouts need mooring corridors, navigation and maintenance lanes and intake exclusion zones, so the gross figure is optimistic.
Pumping is seasonal and delivery-driven, so not all of it lines up with sunshine.
The two benefits do not scale together. Water saved is a straight line: cover twice as much surface, save twice as much water, at every reservoir, forever. Exported energy bends over and flattens. If floating solar on these reservoirs is worth doing, it is worth doing as a water project that happens to make some power, or as a power project that happens to save some water — and those two framings imply completely different amounts of coverage.
The ceiling is different at every reservoir, and the spread is more than tenfold. Lake Havasu and Lake Mead sit at the top of the range, Havasu because 300 MW of pumping load sits on its shoreline and Mead because Hoover's 2,080 MW line now runs at a capacity factor near 0.22. Everywhere else crosses far sooner. Spill starts well below these figures at every reservoir: this is where it becomes serious, not where it begins. Taking the coverage at which spill first passes 10%:
| Lake Havasu | 8.50% |
| Lake Mead | 7.25% |
| Lake Powell | 4.25% |
| Blue Mesa | 2.50% |
| Lake Mohave | 1.00% |
| Flaming Gorge | 1.00% |
| Navajo | 0.50% |
Reservoirs with large surfaces behind small powerplants are the worst candidates, which is the opposite of what a surface-area screen selects. The spread is more than tenfold, so a single basin-wide coverage target is the wrong unit of argument, and 15–20% is above the ceiling at every reservoir here. These are bounds under this model's export assumption, not measured project limits: only an interconnection study at each bus could establish those.
Every figure on this page is computed from public data at build time. The sources that decide the answer, and what each one fixed:
| Quantity | Source |
|---|---|
| Reservoir surface area | Reclamation daily elevation and storage, area taken as the derivative of a fitted hypsometry. Replaced our own satellite measurement, which read low at every reservoir. |
| Dam output through the day | Reclamation measured daily release at six reservoirs, plus the USGS 15-minute record below Glen Canyon. Replaced an assumed seasonal shape. |
| Solar output | PVGIS-NSRDB hourly, at each reservoir's own coordinates. |
| Prices | Day-ahead LMP at each reservoir's own balancing authority. The two Upper Basin reservoirs use a shape transfer, labelled as such wherever it appears. |
| Evaporation rate | USGS eddy-covariance flux at Mead and Mohave; screening estimates elsewhere. Driver split from Penman on NASA POWER daily weather. |
| Canals and other water | OpenStreetMap centrelines and the National Hydrography Dataset's surveyed polygons. |
Full method, every caveat and the review history are on the technical paper and methods page.
The largest array that still sells nearly everything it makes:
| Reservoir | All idle line generous | 25% of idle line plausible | 10% of idle line non-firm |
|---|---|---|---|
| Lake Mead | 2,375 MW | 594 MW | 238 MW |
| Lake Powell | 1,379 MW | 345 MW | 138 MW |
| Lake Havasu | 556 MW | 478 MW | 462 MW |
| Lake Mohave | 34 MW | 8 MW | 3 MW |
| Flaming Gorge | 131 MW | 33 MW | 13 MW |
| Blue Mesa | 69 MW | 17 MW | 7 MW |
| Navajo Reservoir | 15 MW | 4 MW | 2 MW |
Those megawatts are a stricter test than the coverage percentages above, and the two answer different questions, so they should not be read against each other directly. The percentages ask where spill first passes ten percent. This table asks for the largest array that still sells nearly everything, which we set at 5% spill. The second is always the smaller number.
Read the Havasu row across. The other six rows fall by exactly the factor the line falls: cut the line to a tenth and the array drops to a tenth, because with no load on the shoreline the headroom scales with the line and the output scales with the array, so only their ratio matters. That is arithmetic, not a finding. Havasu is the row that breaks it, dropping only about a sixth across the same range, and the reason is the one thing in this table that is not on the line at all: the Central Arizona Project's Mark Wilmer plant sits on its shore and lifts water 824 feet out of the lake, and Metropolitan's Whitsett plant draws from the same reservoir. That load is there whether or not the transmission line has room, so it holds Havasu's viable array up as every other site's collapses. Havasu's surface is also BLM and Arizona State Parks rather than a National Park Service unit, which removes the ownership barrier that blocks Mead and Powell. On this model it is the least transmission-dependent site in the set, and the one where the case is strongest.
Operators would qualify that and they are right to. Mark Wilmer's synchronous pumping units run to the canal's delivery schedule and to power prices, so the overlap between that load and midday generation is partial and seasonal rather than a standing sink. The model samples that coincidence between 25 and 100 percent for exactly this reason, and Havasu's ranking holds only while it stays above roughly a quarter. Co-located load also does not bypass interconnection: an array here is still a separate generator needing its own agreement, and capacity on the Parker-Davis and CRSP systems is allocated by contract rather than by physical headroom alone. Shoreline load reduces how much has to leave on the shared line. It does not take the wire out of the problem.
Spill is a cost, not a failure. If what you want is water, the marginal panel is worth adding as long as it costs less per acre-foot than the one before it, even if a growing share of its output is thrown away. Optimising on cost per acre-foot instead gives bigger arrays that accept 25–53% spill, and cost per acre-foot barely moves with sizing. Across every reservoir and every sizing at baseline costs it runs $7,896 to $28,517 per acre-foot. Which reservoir you pick matters far more than how big you build: the four Lower Basin sites sit at the bottom of that range and the three Upper Basin ones at the top. Carrying the full uncertainty widens it further, and the two smallest reservoirs have median costs near $35,000.
Floating solar carries the capital cost of a power plant. If the objective is purely to stop evaporation, cheaper covers exist — they simply have no revenue, which is fine if they save more water per dollar. Every option below is annualised at the same 7% cost of capital over its own service life, so a ten-year shade ball and a twenty-five-year array are comparable.
| Option | Capex per acre | Life | Suppression | Cost per acre-foot |
|---|---|---|---|---|
| Chemical monolayer film | $200 | 1 yr | 20% | $172 |
| Shade balls | $197,143 | 10 yr | 90% | $5,014 |
| Floating solar, baseline cost | $597,317 | 25 yr | 75% | $13,590 |
| Floating cover / geomembrane | $1,428,571 | 20 yr | 90% | $24,446 |
| Floating solar, high cost case | $1,214,058 | 25 yr | 75% | $30,139 |
| Pay a Colorado River farmer to conserve | $325–700 | |||
| Desalination | $2,500–3,500 | |||
These figures assume panels suppress 75% of evaporation over the water they cover. That is already below the directly-shaded figure, because a cover does not simply switch evaporation off: heat that would have left the lake as vapour instead warms the water and mixes into the rest of it, so the open water around an array evaporates somewhat faster. At high coverage the benefit is sub-linear, and how much so has not been measured at this scale anywhere.
Every cover loses to conservation by an order of magnitude. Shade balls are about two and a half times cheaper per acre-foot than floating solar, so within the category the ranking matters — but the cheapest credible cover still costs roughly ten times what it costs to pay a farmer not to irrigate.
Desalination beats every cover, which is the surprise. The reason generalises: a cover's yield is capped by physics. An acre of covered water saves only what would have evaporated off it, about 5.6 acre-feet a year in the Lower Basin, and no engineering improves that. A desalination plant has no such ceiling — Carlsbad produces roughly 56,000 acre-feet a year from a footprint of a few acres, on the order of a thousand times more water per acre. Covers spread industrial capital across enormous areas to harvest a thin layer.
And shade balls do not transfer to Lake Mead. Los Angeles put 96 million of them on a 175-acre reservoir in 2015, and the project was for water quality — sunlight reacting with naturally occurring bromide and chlorine forms a carcinogen — with evaporation as the side benefit that got the press. LA Reservoir is small, sheltered and closed. Lake Mead is four hundred times larger, with open wind fetch, 100-foot level swings, boating across a National Recreation Area, and a quagga mussel infestation. Covering it would take on the order of thirty-eight billion balls. Los Angeles has since moved away from them.
Hoover is losing most of its generating capacity. Below Lake Mead elevation 1,035 feet, 12 of its 17 turbines cannot run and output falls about 70%; three new wide-head units would soften that to about 58%. Mead sat near 1,050 feet in mid-2026 and loses roughly 5.7 MW per foot of decline.
An array sized to replace what the dam is losing is not competing for transmission at all. It substitutes for a shrinking resource on a line built for that resource. Set the derate control and the spill stays at essentially zero at every array size, because the derate freed exactly that much room. This is the one configuration where the transmission objection does not apply.
The catch is that a derate is not permanent. Hydrology is cyclical, and a good snowpack year raises the reservoir and brings the turbines back. The headroom an array was sized into then belongs to the dam again, and the solar becomes the marginal resource that gets curtailed. Sizing to a derate is a bet that the derate persists, and it needs a contract that says what happens when the water comes back.
Delivering the power is not free either. None of these reservoirs sit at the Palo Verde hub whose prices are used here. Moving power from Hoover, Davis or Parker to a liquid trading point incurs wheeling charges and basis risk that this model does not include, so the revenue side is optimistic.
But solar and batteries cannot match a dam's shape. A 58% derate takes about 2.3 TWh a year out of Hoover. Arrays run against that gap fill 40–57% of it hour by hour, and no amount of building raises that. A capacity derate is a loss every hour of every day, including nights and including the winter weeks when a desert array runs at a fraction of its summer output. Four or eight hours of storage covers the evening ramp; it does not cover the overnight trough or a January lull. Solar plus storage offsets roughly half a derate. Something firm covers the rest.
Storage helps less than the headline suggests. Adding 1,000 MW of four-hour storage pulls the combined break-even from $94 to about $65/MWh. But most of that is not the solar getting better. At the capacity values used here the battery earns roughly $105M a year in capacity payments against about $67M a year of its own annualised capital, net of the tax credit it still qualifies for, so it is profitable standing alone, before it stores a single megawatt-hour of solar. Bundling the two simply lets that profit offset the array's losses.
Held apart, the array's own break-even goes from $94 to $84/MWh with 1,000 MW of storage attached. That five-to-nine dollar improvement is the real benefit, and it comes from the battery absorbing output the line could not take or that would have sold at a negative price. The rest of the apparent gain is the battery's own profit, about $52M a year, being credited to the solar.
So the honest reading is not that storage rescues reservoir solar. It is that at these dams the battery is the investable asset and the solar is not. A developer who wanted to build only the battery could do so, would not need a reservoir, and would not need to solve surface leasing, mooring, or evaporation measurement. That conclusion depends heavily on the capacity value: at the bottom of the observed range, around $51/kW-yr, the battery no longer covers its own cost and the effect disappears.
The array gets no tax credit. The battery does. The federal credit for solar ended for anything placed in service after 2027 unless construction had already begun by July 2026, and no project here could have met that: none has a lease on the water, a finished environmental review or an interconnection study, and reservoir arrays take five to ten years to build. Energy storage was deliberately spared the same cut and keeps a 30% credit into the 2030s, so the battery figures above are net of it. That asymmetry matters: it makes the storage half of a hybrid materially cheaper than the generating half, which is the opposite of how these projects are usually pitched.
The remaining gap is a contract question, not an engineering one. Break-even sits around $65–94/MWh; the array captures $24–32/MWh selling into the midday market, depending on which market the dam sells into. That gap does not close through better dispatch. It closes on what the power is worth to the dam's customers — when the dam cannot deliver, its firm contracted power gets bought on the market instead. Anyone underwriting this should be pricing avoided replacement power and capacity, not midday megawatt-hours.
Pumped storage does not rescue it inside the decade. New pumped storage in the United States runs about ten years from development through permitting to operation, and federal licensing alone typically takes five years or more. The proposed scheme at Hoover has no licence application on file. The one project genuinely aimed at this corridor, a 2,200 MW station using Lake Powell water and the retired Navajo Generating Station's transmission, took its preliminary permit in 2021 and targets operation around 2030 — and it would consume that transmission rather than free it. Glen Canyon cannot do on-site pump-back at all, because the Grand Canyon is below it.