One number is rarely the answer
"Next summer will be hot" is not a forecast you can check. "There is a 70% chance that next July is hotter than the 1991–2020 average" is. The second version tells you two things: what is expected, and how confident the forecaster is. It can also be scored later, which is how forecasters, and the models behind them, get better.
How the IPCC turns words into probabilities
Words like "likely" mean different things to different people. To avoid this, the Intergovernmental Panel on Climate Change uses a fixed scale in its reports, including the AR6 Synthesis Report. Each term has a probability range:
| Term | Probability |
|---|---|
| Virtually certain | 99–100% |
| Extremely likely | 95–100% |
| Very likely | 90–100% |
| Likely | 66–100% |
| More likely than not | >50–100% |
| About as likely as not | 33–66% |
| Unlikely | 0–33% |
| Very unlikely | 0–10% |
| Extremely unlikely | 0–5% |
| Exceptionally unlikely | 0–1% |
So when an IPCC report says something is "very likely", it means the authors assess at least a 90% probability. That is a claim you can hold them to.
Forecasting can be learned
Is good forecasting a talent or a skill? Research from a large geopolitical forecasting tournament suggests it can be improved. In a 2014 study in Psychological Science, Mellers and colleagues reported that training in probabilistic reasoning, working in teams and tracking forecasters' accuracy over time all improved results.
The lesson for digital twins is direct: keep score. A twin that never compares its forecasts with what happened cannot know whether it is any good.
How we will present forecasts
- Ranges, not points: we show the spread of outcomes, like the shaded band in the chart on our home page.
- Plain-language likelihoods tied to numbers, following a fixed scale.
- Sources and assumptions next to every result.
- A public track record once we have enough forecasts that can be checked. Until then, we will not claim any accuracy figure.