A plain-language explanation of the forecasting model, its data sources, and how well it performs.
The model answers a simple question: where will the water level be next month?
It uses a mass-balance approach. Think of Lake Powell as a giant bathtub. Water flows in (snowmelt, rain, tributaries), water flows out (releases to the Grand Canyon, evaporation), and the level changes accordingly. In equation form:
"Storage" is measured in million acre-feet (MAF) — the volume of water in the reservoir. But most people think in terms of the water surface elevation (feet above sea level), so the model converts between the two using a lookup table published by the Bureau of Reclamation (BOR).
All data is pulled from public federal sources. No proprietary data or API keys are required.
The BOR publishes a table mapping water elevation to storage volume. The model stores 20 points from this curve and interpolates linearly between them. Round-trip accuracy (elevation → storage → elevation) is better than 0.5 ft.
Key reference elevations: 3,700 ft is full pool (24.3 MAF), 3,490 ft is minimum power pool (5.0 MAF), and 3,370 ft is dead pool (1.7 MAF).
Each spring, snowmelt in the Rocky Mountains feeds the Colorado River and refills Lake Powell. The amount of snow on the ground — specifically the Snow Water Equivalent (SWE) on April 1 — is a strong predictor of how much water will flow in during spring and summer.
The model uses a linear regression of basin-average April 1 SWE (averaged across 10 stations) against the observed net storage gain from April through the summer peak (typically July). This is calibrated on 2020–2025 data, with R² = 0.899.
A key design choice: the regression predicts the net effect — inflow minus releases minus evaporation — not just raw inflow. This is more useful because it directly predicts what happens to the lake level, not an intermediate quantity you'd need to further adjust.
Outside the spring refill window (August through April), the model applies average monthly net storage changes derived from 2020–2025 historical data. During these months, releases exceed inflow, so the lake steadily declines:
| Month | Net Change (MAF) | Notes |
|---|---|---|
| August | −0.355 | Peak release month |
| September | −0.182 | |
| October | −0.073 | |
| November | −0.147 | |
| December | −0.225 | |
| January | −0.273 | |
| February | −0.221 | |
| March | −0.163 | |
| April | −0.041 | Transition to refill |
During spring refill (May–July), the SWE-derived net gain is distributed across three months: 35% in May, 45% in June, and 20% in July. This distribution reflects the typical timing of snowmelt runoff in the Upper Colorado Basin.
The interactive forecast allows users to adjust two parameters:
These controls let users explore "what-if" scenarios without modifying the underlying model.
The model was validated by hindcasting five known water years (2020–2024). For each year, the model starts from the actual January elevation and April 1 SWE reading, then forecasts forward. Results are compared against observed peak (summer high) and trough (winter low) elevations.
| Year | Start Elev | SWE | Pred. Peak | Actual Peak | Peak Err | Pred. Low | Actual Low | Low Err |
|---|---|---|---|---|---|---|---|---|
| 2020 | 3600.2 | 4412 | 3619.6 | 3610.6 | +9.0 | 3598.6 | 3559.4 | +39.2* |
| 2021 | 3566.2 | 3356 | 3566.2 | 3561.8 | +4.4 | 3531.3 | 3522.1 | +9.2 |
| 2022 | 3523.1 | 3662 | 3528.0 | 3539.5 | −11.5 | 3494.1 | 3519.5 | −25.4* |
| 2023 | 3521.6 | 5527 | 3580.6 | 3584.3 | −3.7 | 3553.9 | 3557.6 | −3.7 |
| 2024 | 3558.4 | 4445 | 3583.5 | 3586.8 | −3.3 | 3557.4 | 3557.3 | +0.1 |
* 2020 and 2022 had atypical release rates. WY2021 released 8.28 MAF vs. the current 7.48 MAF tier; WY2022 was drought-reduced to 7.07 MAF. The model is calibrated for the current Mid-Elevation Release Tier (7.48 MAF/year).
Bottom line: Under current operating conditions (2023–2024), forecast errors are under 4 feet for both peak and low predictions. When release rates deviate significantly from the calibration period, errors can reach 25–40 ft — which is expected, since the model assumes current policy will continue.
As of April 19, 2026, Lake Powell sits at 3,527 ft — less than 25% full. Basin snowpack collapsed in late winter: April 1 SWE averaged about 1,168 tenths-mm across our 10 stations (roughly 32% of the 2020–2025 median), and a record-breaking March heat wave accelerated melt. The April 2026 Colorado Basin River Forecast Center projects Apr–Jul unregulated inflow of only 1.40 MAF (22% of average) and water-year total of 3.87 MAF (40% of average) under the Most Probable scenario.
The Bureau of Reclamation's April 2026 Most Probable 24-Month Study projects Lake Powell to decline from 3,527 ft today to a low of approximately 3,456 ft in March 2027, with recovery beginning in May 2027 once spring runoff arrives. That trajectory crosses minimum power pool (3,490 ft) in August 2026 without intervention.
On April 17, 2026, Secretary of the Interior Doug Burgum and Reclamation announced unprecedented emergency actions to protect Glen Canyon Dam (see BOR news release and Circle of Blue's reporting):
The combined intervention adds approximately 2.48 MAF to Lake Powell and is expected to lift the elevation about 54 ft above the Most Probable trajectory, targeting at least 3,500 ft by April 2027. The press release acknowledges Powell is expected to stay above 3,500 ft "but just barely" — a narrow margin that depends on average conditions continuing.
The app's Policy Scenario selector models these regimes:
The Bureau of Reclamation publishes monthly 24-Month Studies using the CRSS model, which is the authoritative federal projection for Lake Powell and Lake Mead operations. The April 2026 Most Probable scenario projects end-of-month elevations that are overlaid on our chart as a dashed gray reference line.
Our mass-balance model and BOR's CRSS differ in important ways:
When driven by BOR-comparable inputs (SWE equivalent to 40% of average, Auto-tier policy), the model projects Lake Powell at approximately 3,480 ft in December 2026 — within 10 ft of BOR's Most Probable (3,471 ft). Large divergences appear at the tails (very dry or very wet years) where the SWE regression is extrapolating.