Abstract
The reliability of reductions of forecasting error derived from changes in the quality of the initial data or model formulation is considered using a signal-to-noise analysis. Defining the initial data error as the data error source and the model error as the modelling source, we propose the use of the modeling error as a baseline against which potential reductions in data error may be calibrated. In the reverse sense, the data error can also be used to calibrate the reduction in the modeling error. A simple nonlinear model is used to illustrate examples of the above reliability test. Further applications of this test to actual numerical forecast experiments using analyses from both the augmented FWE database and the operational NMC data base are shown. Forecast comparisons using various suites of physical parameterizations are also presented.
