What an ensemble-mean anomaly shows
The ensemble mean anomaly is the average departure of all forecast members from the model climate. What it can and cannot tell you about the coming season.
Members, mean and anomaly
A seasonal forecast is not one run of a model but an ensemble: a few dozen runs sampling uncertainty in initial states and, in some systems, model physics or different start dates. The ensemble mean is the average of all of them. The anomaly is that mean minus the model’s own climatological average for the same start month and lead, taken from its hindcasts. The ensemble-mean anomaly map on this site is therefore “how far, on average, this model expects the valid period to depart from its normal”.
What the mean is good at
Averaging reduces variations that differ between members and can reveal a shared large-scale response. It does not remove shared model errors. Large-scale patterns such as a warm anomaly across a whole continent or a tropical Pacific SST anomaly show up clearly. Because every model on this site is expressed relative to its own hindcast climate, the maps of different centres can be compared in the same units even though their absolute climatologies differ. Their biases, variability and skill can still differ.
What the mean hides
The mean says nothing about spread. A +0.5 °C anomaly can come from fifty members that all sit near +0.5, or from half the members at +2 and half at −1. Those distributions imply very different uncertainty. Anomalies need not shrink with lead: a persistent or developing climate signal can remain strong. A weak mean alone cannot distinguish a narrow ensemble near its climate mean from a wide ensemble whose departures cancel.
Units and conversions
Temperature and SST anomalies are published in kelvin and shown in °C without any offset, because a difference of one kelvin is a difference of one degree Celsius. Precipitation is an anomalous accumulation rate, converted to mm/day. Pressure is converted from Pa to hPa. Monthly means come from the C3S anomaly product without additional statistical calibration. The map display interpolates the stored grid; source missing values are retained.
For a view of the distribution rather than its centre, switch the forecast type to a tercile or quintile probability.
Sources and method
These references explain the underlying science and provider conventions. Descriptions of this site’s calculations, colours and coverage were checked against its processing code and imported data.
- C3S seasonal product definitions
- Scaife and Smith (2018): a signal-to-noise paradox in climate science