Probabilistic forecasts: P10, P50 and P90
A probabilistic power forecast gives the likely range of output for every hour instead of a single number. P50 is the median: output is equally likely to be above or below it. P10 and P90 bound the range: there is a 10% chance that output falls below P10 and a 10% chance that it rises above P90.
Updated
Reading P10, P50 and P90
One hour from the synthetic example on our homepage, a 120 MW wind farm at noon as a cold front approaches:
| Quantile | Output | What it means |
|---|---|---|
| P10 | 33 MW | 1 chance in 10 that output is lower |
| P25 | 56 MW | 1 chance in 4 that output is lower |
| P50 | 79 MW | The median |
| P75 | 94 MW | 1 chance in 4 that output is higher |
| P90 | 104 MW | 1 chance in 10 that output is higher |
So the P10–P90 range should contain the actual output in 8 hours out of 10, and the P25–P75 range in half of them. A narrow range means the weather models agree. A wide range means they disagree, typically about the timing of a front, the arrival of cloud or a change in wind direction.
Where the range comes from
Global weather centres run their models many times with slightly different starting conditions. Each run, called an ensemble member, is an equally plausible future. Frontwatt turns every one of these scenarios, over 100 per run, into plant output with the physical model of your plant.
Raw ensembles are often too confident. So we check, on data that was not used to build the model, how often actual output falls inside each range, and calibrate the ranges where there is enough history to do so.
Adding up a portfolio
The P10 of a portfolio is not the sum of the plants’ P10s. Adding P10s assumes every plant has its bad hour at the same time, which overstates the risk of a spread-out portfolio.
Frontwatt adds plants up scenario by scenario. In each weather scenario, plants that see the same weather move together and plants far apart do not. The portfolio range is taken from these totals, so it reflects the diversification you actually have.
How to use the range
- Choosing a schedule. The best day-ahead schedule is usually not the median. It depends on what being long or short costs you. How the schedule is chosen.
- Spotting risky hours. Wide ranges show which hours deserve attention before gate closure and in intraday trading.
- Planning work. Hours with low output and a narrow range are good windows for maintenance.
Rule of thumb: look at the width of the range before the median. The width tells you how much you can trust the median.