Atmospheric Modeling, Data Assimilation and Predictability

If the numerical model forecasts are skillful, the forecast variables should be strongly related to the weather parameters of interest to the “person in the street” and for other important applications. These include precipitation (amount and type), surface wind, and surface temperature, visibility, cloud amount and type, etc. However, the model output variables are not optimal direct estimates of local weather forecasts. This is because models have biases, the bottom surface of the models is not a good representation of the actual orography, and models may not represent well the effect of local forcings important for local weather forecasts. In addition, models do not forecast some required parameters, such as visibility and probability of thunderstorms.

In order to optimize the use of numerical weather forecasts as guidance for human forecasters, it has been customary to use statistical methods to “post-process” the model forecasts and adapt them to produce local forecasts. In this appendix we discuss three of the methods that have been used for this purpose.

Model Output Statistics (MOS)

This method, when applied under ideal circumstances, is the gold standard of NWP model output post-processing (Glahn and Lowry, 1972, Carter et al., 1989). MOS is essentially multiple linear regression, where the predictors hnj are model forecast variables (e.g., temperature, humidity, or wind at any grid point, either near the surface or in the upper levels), and may also include other astronomical or geographical parameters (such as latitude, longitude and time of the year) valid at time tn.