How Should a Numerical Weather Prediction (NWP) Model Forecast be Used: Full Field or Anomaly?

An NWP forecast can be decomposed into two parts: model climate and anomaly. Both parts contribute to forecast error. Since observed climate can be derived from historical data, climate part does not need to be predicted. If model climate is replaced by observed climate, the forecast errors stemming from model climate could be eliminated, which might dramatically increase forecast accuracy.

This hypothesis has been tested using a 3-variable Lorenz model. Here are the results: (1) Bias mainly stems from climate part. Therefore, replacing model climate with observed climate can effectively eliminate bias. (2) Flow-dependent error mainly stems from anomaly, so it is less effective to reduce it through this approach. (3) A more accurate “anomaly forecast” needs to be constructed with respect to model climate rather than observed climate by avoiding model bias. This approach is effective for both short- and long-range forecasts, with larger benefit for longer range forecasts.