|
We propose a Gaussian framework for coupling the error and uncertainty associated with multiple sources of data in the context of numerical weather prediction (NWP) model assessment and weather forecast evaluation. The aim is to characterize an unobserved physical process.
This framework is specified in a weather forecast evaluation context. In particular, we model the error associated with the verification dataset and investigate the effect of this error on a decision-making process. Namely, evaluating and comparing forecasts through proper scoring rules with respect to a verification dataset that contains errors is likely to produce mis-leading decisions. Using an estimate of the hidden underlying process as a verification referential helps to avoid mis-selection of forecasts.
An extension of this framework can be used to investigate the impacts of the spatial resolution of an NWP model on the prediction of wind speed made from a statistical combination of the NWP outputs and some measurements.
|
