Perspective Paper on Lorenz (1982): “Atmospheric Predictability Experiments with a Large Numerical Model”

In 1982, Edward Lorenz published a study evaluating the growth of forecast errors in 10-day European Centre for Medium-Range Weather Forecast (ECMWF) forecasts. This was one of the first articles to examine practical predictability in an operational global forecasting system and to estimate the limit of atmospheric predictability. The method of fitting a second-order polynomial to error growth as a function of error magnitude motivated numerous subsequent studies. The present article reviews some of this later work and uses recent ECMWF forecasts to illustrate how the method can be applied today. The error-growth model is also used to discuss recent developments in machine-learning-based forecasting. Although forecast quality has improved substantially since 1982, considerable errors remain for near-surface variables and still need to be addressed.