Canadian ensemble forecasts

The daily ensemble forecasts have been available operationnally since January 24, 1996. They were originally performed with eight members. As of August 24, 1999, eight more members were added creating a 16-member ensemble forecast system. On January 12, 2005, the Optimal Interpolation Technique for the analysis cycle was replaced with the Ensemble Kalman Filter Technique.

Each day, 16 "perturbed" 10-day weather forecasts are performed as well as an unperturbed 10-day control forecast. Of the 16 perturbed forecasts, 8 are performed with the SEF global spectral model and 8 with the GEM global model. The 16 models have different physics parametrizations, data assimilation cycles and sets of perturbed observations. Boundary conditions such as sea surface temperature, albedo and roughness length have been perturbed as well. The control forecast is initiated from the ensemble mean and produced with the SEF spectral model.

10 day mean temperature anomaly

View the forecast of the normalised mean temperature anomaly for the next 10 days. The forecast starts at 00Z of the day indicated in the figure. The anomaly is with respect to the climatological mean temperature. The forecast was obtained by applying a regression equation to the ensemble mean 1000-500 hPa thickness. The contours are 0.43 multiples of the standard deviation.

Spaghetti plots

Control Line dam 500 hPa

h forecast

produced

[view]

In the spaghetti plots one can see both the position of a contour line and its uncertainty. We also show the standard deviation for the 500 hPa height (with the background colour). It may happen that the contour lines are far apart but that the gradients are not very important. In such a case, the forecasts may still be considered to be reliable.

Calibrated probability of equivalent precipitation

h forecast

produced

[view]

In these figures we show the probabilities that the accumulated amount of (melted or liquid) precipitation in a 24 hour period exceeds thresholds of 2, 5, 10 or 25mm. Raw probability intervals are at less than 22 % (at most 3 members are above the threshold), between 22 and 47 % (4 to 7 members above the threshold), between 47 and 72 % (8 to 11 members above the threshold) or more than 72 % (12 to 16 members above the threshold).

We performed a calibration of our precipitation forecasts. This allowed us to replace the raw probabilities of 22, 47 and 72 % by calibrated values.

Accumulated quantity of precipitation

h forecast

produced

[view]

First we show a contour plot of the ensemble mean forecast. On this plot we put small blue numbers and red letters for the precipitation centres of the corresponding members. The verifying analysis is added on with the dashed red line. The next plots show the high resolution GEM forecast and the 16 individual forecasts.

Sea level pressure centres

h forecast

produced

[view]

First we show a contour plot of the ensemble mean forecast. On this plot we put small red numbers for low centres and small blue numbers for high centres of the corresponding members. The 16 ensemble members are numbered 1-8 and A-H. The verifying analysis is added on with the dashed red line. The next plots show the high resolution GEM forecast and the 16 individual forecasts.

GZ 500 maps

h forecast

produced

[view]

Ensemble spread of trial fields

Every 6 hours, we compute the standard deviation in dam of the GZ 500 hPa trial fields from the 16 representative assimilation cycles. Where the standard deviation of the trial field is large, it is important to have good observations. One might target additional observations to such areas. Another option is to manually introduce bogus observations. View.