Feature-based predictability of 500 hpa height in the Australia-New Zealand region

Summary Feature-based predictability stratifies forecast model errors on the basis of individual weather systems. We examine only one level and chose very simple categories: high, cut-off low, trough and block. European Centre for Medium-range Weather Forecasts data are used. We emphasize systems found near Australia and New Zealand during winter 1987. Calculations for the preceding summer and fall and for other midlatitude regions of the southern hemisphere yield similar results.The approach herein is fully automated and simple to implement. Features are identified in the verification field. Then an error calculation is made on moving grids that are each centered upon and contain one system.The total error is the root mean square difference (RMS) between forecast and verification. The structural error is the RMS difference when the forecast and verification small grids are independently centered upon the corresponding feature in each field. The difference between the structural and total errors, called the locational error, is typically a quarter of the total. Even when normalized by presistence, highs are forecast better than troughs; cut-off lows and troughs have similar errors. The distance between the forecast and observed positions is typically 3° longitude west and 0.5° latitude north of where features should be at 72 hours. The model has a systematic bias of too small amplitude of vorticity. No relation is found between skill and jet stream splitting.With 4 Figures