Calibration can be defined as the process by which the forecast is adjusted / modified based on what has been observed in the past.
For example, if we note that for a specific location, a probabilistic forecast of 80% below normal is actually correct (observed) 60% of the time, an ideally calibrated probabilistic forecast would be 60%.
The 30 year database of historical seasonal forecasts is too small to enable a perfect calibration to be achieved at every location. However, the optimization procedure employed in producing the forecasts brings the forecast probabilities much closer to the observed frequencies than the previous procedure, which consisted of simply counting forecast ensemble members in the tercile bins.
Comparisons of historical forecast probabilities to the observed frequencies are shown in the reliability diagrams that accompany each forecast. There is one curve for each forecast category (above, near and below normal), which for perfectly calibrated forecasts would lie along the dashed diagonal.
What we are doing
- Date modified: