Anomaly based numerical weather prediction: preliminary explorations
8 March, 1pm
What will the next generation numerical weather prediction look like? A new idea, anomaly based NWP, will be discussed in this talk. Two aspects of this idea have been explored using simple models. One is an anomaly equation based model, and another is an anomaly based model forecast's post processing.
Any variable can be decomposed into a long term average ("climate") and a departure ("anomaly"). Climate is known and doesn't not need to be predicted since it can be pre-calculated using reanalysis data, only anomalies need to be predicted. Therefore, if a model only predicts anomalies but not the climate part, total forecast error might be dramatically reduced. An example of hurricane track forecasts using Beta Advection Model (BAM) will be demonstrated. The same idea can be applied to model forecast's post processing, which will be demonstrated using a 3 variable Lorenz model.
Is the idea presented in this talk a fantasy world only, or realizable in real world operations?