The anomaly correlation coefficient (ACC) is the correlation between anomalies of forecasts and anomalies of verifying values (for example, actuals).
When applied to forecasts of energy dispatch targets the verifying values may be the actual dispatch targets ‘a’ at runtime, the forecasts may be predispatch forecasts ‘f’. This allows anomalies relative to a climatology, ‘c’, to be calculated.
It is possible to calculate ‘c’ in many different ways, depending on the data and historical changes. One way is to calculate the 14-day rolling mean of the actual dispatch targets (separated by each by half hour in the day). The ACC can be calculated as:
Anomaly Correlation Coefficient Formula (uncentered version)
ACC values of 1 indicate perfect association of forecast anomalies with the actual anomalies. Values less than 1 indicate weaker association.
The ACC helps highlight where forecast variability (about the climatological expectation) is more or less aligned with the actual variability and can be considered an indicator of forecast skill.
The ACC is not sensitive to forecast bias. The ACC should be evaluated in the context of an understanding of the magnitude of the anomalies and forecast differences as tiny anomalies can generate spurious correlations. For these reasons the ACC is best used in conjunction with other measures of forecast accuracy such as the root mean square error.