Envelope’s hurricane forecasts, run as a hindcast and scored against the 2014–2025 archive.
Each forecast generates 1,000 storm scenarios. At each lead time, the observed storm center is ranked within the scenario spread, along the direction of motion and across it; a centered spread ranks the observed center at 0.50 on average. Measured ranks are 0.47–0.50 at every lead, on both axes, illustrating that Envelope’s ensemble is spatially well-calibrated.
| Lead time | Along track | Cross track | Forecasts |
|---|---|---|---|
| 24 h | 0.50 | 0.50 | 1,548 |
| 48 h | 0.47 | 0.48 | 1,379 |
| 72 h | 0.47 | 0.48 | 1,196 |
| 96 h | 0.47 | 0.50 | 1,008 |
| 120 h | 0.47 | 0.50 | 835 |
The same rank test, applied to intensity. In a calibrated spread, the observed intensity falls in the weakest 10% of scenarios 10% of the time, and in the strongest 10% of scenarios 10% of the time. Envelope’s stated 10th percentile verifies at 9th–10th percentile; Envelope’s stated 90th percentile verifies at 88th–90th percentile.
| Lead time | In weakest 10% | In strongest 10% |
|---|---|---|
| 24 h | 9.7% | 12.2% |
| 48 h | 9.0% | 12.4% |
| 72 h | 10.2% | 11.0% |
| 96 h | 9.5% | 10.2% |
| 120 h | 9.5% | 10.8% |
A major part of modeling tropical cyclones is modeling the probability that the event dissipates. Envelope probabilities of dissipation are calibrated within 2–4 percentage points.
| Lead time | Scenarios dissipated | Observed |
|---|---|---|
| 24 h | 12% | 8% |
| 48 h | 22% | 18% |
| 72 h | 33% | 29% |
| 96 h | 43% | 41% |
| 120 h | 53% | 51% |
For actual storms, we compare what Envelope’s wind field model would have said conditional on the post-event best track to our best estimate of the actual wind field. The model is well-calibrated, with some conservative over-forecasting of winds at the high-end where our error bars around the true wind speed are highest.
| Model said | Model average | Analysis average | Ratio | Location-hours |
|---|---|---|---|---|
| 40–50 mph | 44 mph | 40 mph | 0.90 | 2,348,671 |
| 50–60 mph | 54 mph | 53 mph | 0.97 | 688,276 |
| 60–70 mph | 64 mph | 63 mph | 0.98 | 213,835 |
| 70–80 mph | 74 mph | 73 mph | 0.99 | 93,909 |
| 80–90 mph | 85 mph | 82 mph | 0.97 | 50,390 |
| 90–100 mph | 95 mph | 89 mph | 0.94 | 37,277 |
| 100–110 mph | 104 mph | 94 mph | 0.90 | 19,888 |
| 110+ mph | 130 mph | 110 mph | 0.84 | 22,137 |
We compiled every probabilistic forecast Envelope would have made in the historical record, binned them by probability band, and compared the forecast probabilities to the realized probabilities, demonstrating forecast reliability.
Our working source of wind field truth is NOAA’s URMA 10-m gust analysis, enhanced with tropical cyclone core winds estimated from National Hurricane Center best track position, intensity, and size (an enhancement that gap-fills wind underestimation in hurricane inner cores due to instrument sparsity). Track, intensity, and lifetime are verified against the National Hurricane Center’s post-season best track. All wind verification is over land, 2014–2025.