The Analog Ensemble Technique Explained

Schematic Diagram

The following schematic diagram shows the four steps to generate a four-member ensemble forecast.

AnEn-scheme

  • Step 1: The process starts with a current deterministic multivariate prediction and a set of historical predictions from a deterministic weather model. The multivariate prediction includes surface temperature, humidity, wind speed, and so on. Corresponding observations to each historical forecasts are also collected.
  • Step 2: A number of historical predictions are identified based on their similarity to the current multivariate prediction. This similarity is also time-dependent, meaning that, instead of point-to-point comparison, it also compares the trend of each weather variable within a short time range.
  • Step 3: The corresponding observations associated with the identified historical predictions are selected.
  • Step 4: These observations become ensemble members in the final forecast.

Simplified Example for Temperature Forecasts

Please navigate through the following slides to see the example.

Animation credited to Laura Clemente-Harding and Guido Cervone

  • Step 1: A deterministic model has been running for a week and a new prediction is generated from the model. Red dots are temperature observations, and black dots are model predictions.
  • Step 2: By comparing current and historical model predictions (black dots), most similar past forecasts are identified.
  • Step 3: The corresponding observations are selected that are associated with the identified past predictions.
  • Step 4: These observations become ensemble members in the final forecast.

Of course, in reality, the similarity metric is a time-dependent and multivariate metric.

References

Comments