In 1939 Rossby demonstrated the usefulness of the linearized perturbation of the equations of motion for weather prediction and thus made possible the first successful numerical forecasts of the weather by Charney et al. In 1951 Charney wrote a paper on the science of numerical weather prediction (NWP), where he predicted with remarkable vision how NWP would evolve until the present. In the 1960's Lorenz discovered that the chaotic nature of the atmosphere imposes a finite limit of about two weeks to weather predictability. At that time this fundamental discovery was “only of academic interest” and not really relevant to operational weather forecasting, since at that time the accuracy of even a 2-day forecast was rather poor. Since then, however, computer-based forecasts have improved so much that Lorenz's limit of predictability is starting to become attainable in practice, especially with ensemble forecasting, and the predictabilty of longer-lasting phenomena such as El Niño is beginning to be successfully exploited.
The skill of operational weather forecasts has at least doubled over the last two decades. This improvement has taken place relatively steadily, driven by a large number of scientific and computational developments, especially in the area of NWP. It has taken place in all the operational NWP centers, as friendly competition and information sharing make scientific improvements take place faster than they would in a single center. Because the improvements have occurred steadily, rather than suddenly, the overall increase in forecast skill due to NWP has not been clearly recognized by the media and the public despite the impact that improved forecasts have on the national economy and on the lives of every American.
In this paper the authors review several measures of operational forecast skill that quantify improvements in NWP at the National Centers for Environmental Prediction (NCEP, formerly the National Meteorological Center) of the National Weather Service, although they are representative of improvements in all major NWP operational centers. The authors point out that there are three major requirements for improved numerical weather prediction: better atmospheric models, better observational data, and better methods for data assimilation. These improvements are generally very computer intensive and can only be made operational with the availability of more powerful supercomputers. Operational forecasts are compared with “reforecasts” from the NCEP–NCAR 40-Year Reanalysis, showing that, if the present-day NWP systems had been available many decades ago, skillful 5-day forecasts would have been possible in the Northern Hemisphere with the upper-air network of the late 1950s. The authors discuss new approaches in the use of observations (variational assimilation of remote observations) and of numerical weather prediction guidance (ensemble forecasting) that have allowed the recent extension of operational predictions into longer ranges and the possibility of adaptive observing systems. The extension of operational forecast skill into seasonal predictions of the El Niño–Southern Oscillation phenomena using coupled ocean-atmosphere models is also discussed. In the last section the authors attempt to “forecast” the future of NWP.
Editor's note: This is the third of a series of papers about Carl-Gustaf Rossby that will be published in the Bulletin. All papers were presented at the 1998 Annual Meeting of the American Meteorological Society as part of the Special Session Honoring the Centennial of the Birth of Carl-Gustaf A. Rossby. Other papers in the series will appear in upcoming issues of the Bulletin.
* Additional affiliation: School of Meteorology, University of Oklahoma, Norman, Oklahoma.