Public weather forecasting heralded the beginning of modern meteorology less than 150 years ago. Since then, meteorology has been largely a forecasting discipline. Thus, forecasting could have easily been used to test and develop hypotheses, consequently enhancing the potential of the scientific method to increase knowledge of meteorology. The outcome has proved to be rather different.
In the day-to-day operations of meteorological services, the constant stream of predictions (some of which are wrong) should quickly demolish or modify doubtful hypotheses. In fact, feedback between prediction and hypothesis is rare. In research, forecasting is value neutral—a wrong forecast may contribute as much to understanding as a right forecast. Routine public weather forecasting is an end in itself and is not value neutral—a right forecast is much more valuable than a wrong forecast.
Despite the flood of data from satellites and radar, forecasts barely improved. This suggested shortcomings in our understanding of the atmosphere as expressed by conceptual and numerical models. Statistics, which had supported weather forecasting in only a minor way, has been increasingly called on to compensate for what appears to be an ingrained inability to understand atmospheric processes and their interactions.