Error Growth in a Large Numerical Forecast System

Abstract

Internal and external z500 global total rms errors followed quadratic growth laws quite well in NMC Medium-Range Forecast (MRF) Model 0-10-day forecasts for 1988-93. Growth parameters and model and analysis errors for many winters were estimated using the quadratic rms error growth assumption. Both the MRF model error and analysis error have nearly halved during 1988-93. But at the same time the growth parameters have nearly doubled: smaller errors grow faster. Thus while the limit of deterministic predictability (rms error 71% of saturation) has been going up, the limit of dynamic predictability (rms error 97.5% of saturation) seems to be set at around 20 days in large horizontal scales, dropping to 6-7 days in small scales.