Research library
Weather forecasting and climate applications
View the four-date completion graphs and reconstruction notes.
Promise and evidence
AI could improve forecasts, early warnings, renewable-energy planning, and adaptation decisions. GenCast provides a concrete probabilistic weather-forecasting example. Medium-range forecasting skill does not automatically establish reliable multidecadal climate projections.
Apply the shared measurement protocol. Targets are proposed.
Dated baseline coverage
All linked measurable-outcome files below contain an evidence baseline reviewed November 7, 2026, with source dates, findings, limitations and follow-up measurements. These distinguish measured results from proxies and unknown values; they do not assert that proposed milestones are achieved. See the baseline coverage notes.
Progress estimates
Judgment-based estimates as of November 7, 2026, using the same 0–100% scale and five-part rubric as mathematics. 0% anchors the period before ChatGPT's public release in 2022; 100% requires AI to manage this field's continuing advancement without human participation. Confidence is low. These are evidence-based judgments, not empirical completion measurements, forecasts or percentages of jobs replaced.
| Measurable outcome | Estimated progress | Execution / breadth / agenda / validation / improvement (each out of 20) |
|---|---|---|
| Climate application validity | 35% | 10 / 5 / 5 / 10 / 5 |
| Computation | 45% | 15 / 10 / 5 / 10 / 5 |
| Decision benefit | 15% | 5 / 5 / 0 / 5 / 0 |
| Extreme event performance | 40% | 10 / 10 / 5 / 10 / 5 |
| Forecast skill | 45% | 15 / 10 / 5 / 10 / 5 |
| Robustness | 20% | 5 / 5 / 5 / 5 / 0 |
Each linked outcome explains its score, cites its dated evidence baseline and identifies the next measurement. Zero indicates no qualifying post-anchor progress established by the reviewed evidence for that outcome, not proof of no relevant capability. Do not average these overlapping indicators into a field-wide percentage or interpret their differences as a precise ranking of fields. All outcomes reaching 100% would still require a separate field-wide demonstration of sustained autonomous advancement.
Measurable features
| Feature | Measurement and acceptance rule |
|---|---|
| Forecast skill | Continuous ranked probability score for forecast distributions, Brier score for specified events, and appropriate deterministic error metrics, by variable, region, and lead time. |
| Extreme-event performance | Event recall, false-alarm rate, probability calibration, cyclone-track error, and useful warning lead time on prospectively observed events. |
| Computation | End-to-end forecast latency, energy, and cost, including data ingestion and required upstream initialization. |
| Decision benefit | Avoided losses or improved resource allocation from actual forecast-informed decisions against a prespecified operational baseline. |
| Robustness | Skill in poorly observed regions, unusual seasons, and rare events, with uncertainty and missing-data handling reported. |
| Climate application validity | Performance on withheld periods and regions, physical-consistency checks, and uncertainty under distribution shift; report these separately from weather scores. |
Evidence requirements
Archive forecasts before outcomes occur. Use the same information availability, issue times, spatial resolution, and observation reference for comparators. Prespecify which events and variables will be scored to prevent selective reporting. Historical hindcasts support evaluation but cannot directly verify future century-scale projections. Distinguish better warnings from realized reductions in harm, which also depend on response systems.
Proposed milestones
Prospectively improve a registered primary probabilistic score by at least 10% against the contemporary operational baseline over a full year, without unacceptable regional regressions.
Improve useful warning lead time by at least six hours at a fixed false-alarm rate for one defined hazard; report event count and uncertainty, extending evaluation if rare events are insufficient.
Independently reproduce skill gains in another region or operational service.
Demonstrate a measured decision benefit in a comparative deployment, including operating cost. These are concrete societal gains; feedback toward more capable AI requires a separate causal demonstration.