Lagrangian errors: Feature relative verification
A method of forecast verification is introduced that examines the errors in forecasts relative to a specific weather feature. The method is demonstrated on cyclones in and around North America during the winter season of 2013-2014. A simple method for identifying cyclone centers is employed based on the location of relative maximums in relative vorticity and relative minimums in height at the 850 hPa level. A grid is created, centered on the cyclone, onto which the forecast and observed values are interpolated. Errors are created from the difference between the forecast and observation. The errors from each cyclone are collected to make an error series and the mean error for each grid point is found. In this way, errors are analyzed "following" the feature(s), in a Lagrangian sense, as opposed to a more traditional method which allows weather features to be created in or pass through a fixed geographical region, in an Eulerian sense. Significance testing is performed in order to determine if the mean error field is field significant. The intent of this new method is to allow forecasters to know and understand the errors when a particular weather event is forecasted and to ultimately make better forecasts with that knowledge. Additionally, model developers will be able to learn how the model handles a weather feature to allow for better development in future iterations of the model.
Baldwin, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our