Accuracy of a forecast delivery is being considered during all stages:
Before the forecasts delivery starts
- The models are calibrated with measurements from the area of interest
- Assimilation of online measurements or satellite data into the models up until the point when the forecast starts provides a big improvement of the forecast accuracy.
During the daily forecast delivery
- The forecast models are monitored by the duty forecaster
- For areas with online measurements these are compared to the forecasts/nowcasts from the last couple of days
- Ensemble models will provide error estimates for the daily forecasts when operational. These models require large computational resources (like those available at ECMWF - European Centre for Medium Weather Forecasting) but have a high priority among the new developments within the Water Forecast.
At regular intervals during and/or after forecast delivery
- When measurements are available within the area of interest a detailed statistical analysis can be carried out on a regular basis (for example monthly or seasonally). Typical error statistics include mean, bias, absolute mean error, root mean square, scatter index, correlation coefficient and peak ratio computed typically for the nowcast, the 12-hour forecast, the 24-hour forecast, the 48-hour forecast and the 72-hour forecast.
In all cases the accuracy of the input from the meteorological models is also an important factor - especially for the surge forecasts and the wave forecast.