Archived forecast - its window has ended. This is not current weather. Issued ; window ends Nov 14, 2026 at 3:00 AM EST at the next scheduled update. Latest published outlook.
At issue time, 2,468 of 2,469 catalogued Flex pickup sites in United States had an available until-next-update lightning-related closure estimate. 1 were not rated. An unavailable estimate is not 0%, an all-clear, or evidence that a station is operating.
Possible closures until the next scheduled update
0 rated sites had a model estimate of at least 20% until the next scheduled update; 0 were at least 50%. These are descriptive reporting bands, not validated alert thresholds. 0 of the available estimates were still learning and have not passed the model’s reliability checks. These estimates concern a new lightning-related closure indication, not every possible weather closure, an official Amazon decision, a cancelled block, or payment eligibility. Weather systems can affect many nearby stations together; do not add percentages or read the count as an expected number of closures.
Forecast snapshot: Nov 13, 2026 at 9:01 PM EST. Weather-status snapshot: Nov 13, 2026 at 9:01 PM EST. Input coverage varies by location. Japan has no local model percentage in this report. Stations already marked closed at issue time are not assigned a new-closure probability.
This forecast extension starts experimental and is marked learning. It adapts the nearest trained forecast to the time left before the next update, scaling the estimate for windows shorter than six hours. The exact-window calibration still needs testing against later outcomes. Treat it as an estimate, not a confirmed closure.
Your stations at issue time
Sign in to view issue-time forecasts for the stations your plan already covers. Open your account or check your stations on the map. This article does not unlock other stations or hidden forecast horizons.
Browse areas
Alphabetical by country and area. Select an area or expand a country. Counts include learning estimates, not confirmed closures; groups need at least 10 rated sites to protect individual forecasts.
United States · 41 areas
| Area | Rated / catalogued | At least 20% | At least 50% | Still learning |
|---|---|---|---|---|
| Alabama, United States | 15 / 15 | 0 | 0 | 0 |
| Arizona, United States | 65 / 65 | 0 | 0 | 0 |
| Arkansas, United States | 13 / 13 | 0 | 0 | 0 |
| California, United States | 373 / 373 | 0 | 0 | 0 |
| Colorado, United States | 77 / 77 | 0 | 0 | 0 |
| Connecticut, United States | 36 / 36 | 0 | 0 | 0 |
| District of Columbia, United States | 12 / 12 | 0 | 0 | 0 |
| Florida, United States | 183 / 183 | 0 | 0 | 0 |
| Georgia, United States | 79 / 79 | 0 | 0 | 0 |
| Idaho, United States | 11 / 11 | 0 | 0 | 0 |
| Illinois, United States | 110 / 110 | 0 | 0 | 0 |
| Indiana, United States | 37 / 37 | 0 | 0 | 0 |
| Iowa, United States | 20 / 20 | 0 | 0 | 0 |
| Kansas, United States | 22 / 22 | 0 | 0 | 0 |
| Kentucky, United States | 18 / 18 | 0 | 0 | 0 |
| Louisiana, United States | 24 / 24 | 0 | 0 | 0 |
| Maryland, United States | 60 / 60 | 0 | 0 | 0 |
| Massachusetts, United States | 78 / 78 | 0 | 0 | 0 |
| Michigan, United States | 44 / 44 | 0 | 0 | 0 |
| Minnesota, United States | 31 / 31 | 0 | 0 | 0 |
| Mississippi, United States | 10 / 10 | 0 | 0 | 0 |
| Missouri, United States | 38 / 38 | 0 | 0 | 0 |
| Montana, United States | 12 / 12 | 0 | 0 | 0 |
| Nebraska, United States | 10 / 10 | 0 | 0 | 0 |
| Nevada, United States | 29 / 29 | 0 | 0 | 0 |
| New Hampshire, United States | 12 / 12 | 0 | 0 | 0 |
| New Jersey, United States | 86 / 86 | 0 | 0 | 0 |
| New Mexico, United States | 17 / 17 | 0 | 0 | 0 |
| New York, United States | 127 / 127 | 0 | 0 | 0 |
| North Carolina, United States | 68 / 68 | 0 | 0 | 0 |
| Ohio, United States | 70 / 70 | 0 | 0 | 0 |
| Oklahoma, United States | 25 / 25 | 0 | 0 | 0 |
| Oregon, United States | 31 / 31 | 0 | 0 | 0 |
| Pennsylvania, United States | 103 / 103 | 0 | 0 | 0 |
| South Carolina, United States | 27 / 27 | 0 | 0 | 0 |
| Tennessee, United States | 37 / 37 | 0 | 0 | 0 |
| Texas, United States | 202 / 202 | 0 | 0 | 0 |
| Utah, United States | 25 / 25 | 0 | 0 | 0 |
| Virginia, United States | 66 / 66 | 0 | 0 | 0 |
| Washington, United States | 85 / 85 | 0 | 0 | 0 |
| Wisconsin, United States | 22 / 22 | 0 | 0 | 0 |
10 smaller or insufficiently rated areas are included in the coverage total but not listed individually.
Next-day forecast check & learning
Reviewed Nov 14, 2026 at 3:10 AM EST. Across 2467 scored until-next-update estimates: Brier score 0.0024 and log loss 0.0178 (lower is better). 1 estimates were unscored because observation coverage or timing was uncertain.
Review details and limitations
At a descriptive 50% reporting cutoff: 0 hits, 0 false alarms and 6 missed events. This cutoff is not a validated alert rule. A single day, shared weather events and many non-closures cannot establish 97% accuracy. Probabilities are assessed with proper probability scores, not labelled right or wrong solely because a low-probability event occurred.
How the model learns safely
Finished forecasts are matched to timestamped tracker-recorded lightning closure indications. The same storm learner uses recent weather guidance, station and nearby closure history, available lightning features and time-of-day patterns. It fits in the background as settled outcomes arrive, including a check after daily review. Candidate weights and calibration must improve probability scores on later, time-separated cases before adoption; overlapping outcome windows are purged at the training boundary. Too little evidence leaves the existing model unchanged and probabilities marked learning. Until-next-update estimates use actual-window guidance with nearest trained horizon weights (with duration scaling below six hours); this extension starts experimental. A small, bounded exact-window calibrator learns from saved issued estimates and their observed outcomes, separately by region and issue time. It must improve later-case probability scores before adoption, and nonstandard daylight-saving windows remain labelled learning. No full feature history or weather downloads are duplicated, and no weights are forced after one day. Tracker indications are not independent verification of an actual Amazon closure, and no accuracy target is guaranteed.
Other weather that can disrupt pickup
Wind, heat, cold, air quality, rain and winter weather can affect pickup even when the lightning estimate is low. Where an authorized station had a weather-specific advisory at issue time, its context appears above. Those indications describe the tracker’s reading, not a calibrated probability for those hazards. This report does not turn a rain forecast, temperature or general warning into a made-up closure percentage.
Before your block
- Check the live map and your station’s latest reports shortly before travel; this bulletin is one scheduled snapshot.
- Check official local warnings and the Flex app for pickup or block instructions. A closure indication does not guarantee cancellation or payment.
- Plan a safe shelter and route. For US lightning guidance, see the National Weather Service lightning safety page. Follow your local weather service and emergency authorities elsewhere.
How this bulletin is made
Automatically generated from the same station catalogue, weather observations, storm guidance and learned closure models already used by this tracker. No additional weather API calls or language-model service are required. Counts and sentences are assembled from the saved issue-time data, not a human-written forecast. The model can be wrong, especially in bands marked learning; healthy publication data does not mean every station or weather input is covered. This article is not meteorological advice or an official closure notice. Read how to interpret accuracy, our editorial and corrections policy, and report an error to support.