A prediction market called Kalshi has opened wagers on where Lake Mead's surface will sit at the end of June, with the quoted band sitting near 1,044 feet. It is a small, lightly traded market, and the only water-specific bet on the platform. For a hydrologist the framing is more interesting than the stakes: a price on a water level is just a probability forecast, and the same mass balance the Bureau of Reclamation runs every month produces a far better one.
The point of this piece is not the gamble. It is to show what actually moves a reservoir elevation forecast over a few weeks, why the near-term answer is so tightly bounded, and where the genuine uncertainty sits.
A water level is a storage balance, read off a curve
Reservoir elevation is not forecast directly. What you forecast is storage, in acre-feet, through a mass balance: the volume at the end of a period equals the starting volume, plus inflow from the Colorado River, minus releases through Hoover Dam, minus evaporation, with smaller terms for bank storage and local gains. Convert that ending volume to a surface elevation through the reservoir's stage-storage relationship, also called the area-capacity curve, and you have a predicted water level.
That conversion is the part outsiders underrate. Lake Mead is wide and relatively flat at its current stage, so a large change in stored volume maps to a small change in elevation. A swing of hundreds of thousands of acre-feet can move the surface by only a foot or two. The market is, in effect, betting on the last term of an arithmetic problem whose inputs are already mostly known.
The 24-Month Study and CRSS
Reclamation publishes two products that frame these numbers. The monthly 24-Month Study is a deterministic operations model: it runs the most-probable inflow forecast, plus a maximum-probable and minimum-probable scenario, to project Lake Powell and Lake Mead elevations month by month. Over a horizon of weeks, the spread between these scenarios is small, because the dominant inputs, current storage and scheduled releases, are effectively fixed.
For longer horizons Reclamation runs the Colorado River Simulation System, or CRSS, as an ensemble. CRSS drives the same operating rules with a large set of inflow traces drawn from the historical and modeled record, producing probabilistic projections: a 10th, 50th, and 90th percentile elevation rather than a single line. That ensemble is where the real forecasting craft lives, and it is what underpins formal statements about a Tier 1 shortage or the odds of approaching dead pool.
Where the uncertainty actually is
Three terms carry most of the spread, and their weight depends entirely on lead time.
- Inflow. Over a season, runoff is the largest unknown, governed by snowpack and the runoff forecast above Lees Ferry. Over the next three weeks of June, most of that inflow is already in the system or already measured upstream, so it contributes little.
- Evaporation. Off a surface this large, evaporative loss is a real summer term, on the order of several feet of depth annually. It is modeled from temperature, wind, and surface area, and its short-term error is modest but not zero.
- Releases. Hoover Dam outflow is set by operating decisions and delivery obligations, not weather. For a near-term forecast it is close to a known quantity, which is precisely why the elevation band is tight.
This is the structural reason a one-month elevation bet is nearly a sure thing while a two-year projection is genuinely uncertain. The further out you forecast, the more the inflow term dominates and the wider the ensemble spreads. Pull in close and the mass balance is almost bookkeeping.
Why the level still matters
The number being wagered on is not arbitrary. Elevation, not stored volume, sets the operational thresholds on the Colorado: the roughly 950-foot minimum power pool below which Hoover Dam's turbines stop generating, the January 1 elevations that trigger shortage tiers, and the 895-foot dead pool at which water can no longer pass the dam under gravity. Each is a stage value read off the same area-capacity curve, which is the engineering reason hydrologists track feet above sea level rather than acre-feet in public reporting. We covered the power-side of that threshold in our look at Hoover Dam's hydropower cliff, and the broader logic of probabilistic forecasting in the 100-year flood explained.
So the market is not predicting anything Reclamation has not already projected with more rigor. What it does, usefully, is put a public price on a question the agency answers in a technical appendix. For practitioners the lesson is the reverse of the headline: the closer a reservoir forecast sits to the present, the less there is to bet on, because the hydrology has already mostly happened. For more on how we approach water-resource analysis, see the Ayres River homepage and our editorial standards.