Inferring the Relative Quality of Reanalysis Data from a Historical Reforecast Using a Vorticity Equation Model

The improvement of numerical weather prediction models in recent decades has led to a measurable decrease in forecast error, for instance as quantified by the 500-hPa geopotential height anomaly correlation. This improvement can be attributed in part to the refinements in the numerical architecture and the representation of physical processes in the model. It is more difficult to attribute the reduction in forecast error to improvements in the accuracy of the models' initial conditions, not least because of the complexity of the data assimilation process. As a contribution to this problem that will also be pedagogically useful, a reforecast experiment was conducted using a simple regional vorticity equation model for the 500-hPa flow over North America. This model is comparable in complexity to the Charney-Fjörtoft-von Neumann model used for the first numerical weather prediction experiment in 1949. Six-hourly predictions are made out to 36 hours for the period from 1948-2010, using the NCEP Reanalysis dataset for the initial conditions. Because initialization is trivial for this "filtered" model, one can clearly attribute the change in forecast skill to the changing quality of the initial state. The choice of the 500 hPa level is relevant since it will reflect the effect of the significant change in the density of upper-air observations from the mid- to late 20th Century. As an important byproduct of this exercise, a trend of improved forecast over time will imply an improved relative quality of the reanalysis data for the mid-troposphere. Therefore, the experiment serves two purposes of studying the forecast error and checking the quality of the reanalysis. An analysis of selected cases indicates that the predictions made for the 21st century have smaller errors compared to those made for the 1950s. A comprehensive analysis for the whole set of forecast is underway and the outcome will be discussed.