Sam Lillo-March 28

Upscale Error Growth and Propagation in Numerical Weather Prediction

Despite improvements in the overall performance of numerical weather prediction (NWP), there are still times when forecast models experience errors that are drastically greater than the mean. These periods of poor performance are referred to as dropouts, or busts, and are characterized by a verifying atmosphere that is significantly different from the forecast. The skill of a synoptic-scale forecast is directly related to how accurately Rossby waves are captured around the globe. Mishandling of these features result in errors that often take on the appearance of wave structures themselves. NWP errors have been observed to behave like wave packets, propagating at a group velocity. These observations provide motivation to examine error in a wave-like framework.

A phase-independent quantity is derived based on the error streamfunction to describe its total amplitude, and wave activity flux is manipulated to project error propagation. Packets of error can then be analyzed and tracked through a forecast period. These tools are applied to ECMWF operational forecasts. Previous research from the ECMWF have noted linkage between organized convection over the US and decreased forecast skill over Europe 6 days later. Using ECMWF forecasts between 2007 and 2014, medium-range regional forecast performance over Europe is objectively evaluated using RMSE of 500hPa heights. Composites of all Spring forecasts reveal preferred areas of initial error growth over the Midwest US, East Asia, and the tropical eastern Pacific. However composite differences between the worst and best forecasts indicate significant bifurcation does not develop until after two days, over the North Atlantic, downstream of the features associated with the initial error growth. Two cases are presented as examples of this downstream error development