In this paper we review various techniques of initialization. We also consider Lewis Fry Richardson's remarkable numerical forecast, and review the implications of initialization for his results.
The story of Richardson's forecast, made about seventy years ago, is well known and is recounted in a recent biography by Ashford. Richardson forecast the change in surface pressure and other parameters at a point in central Europe, using the mathematical equations and observations valid at 0700 GMT on 20 May, 1910. He described his methods and results in his book "Weather Prediction by Numerical Process". His results implied a change in surface pressure of 145 millibars in 6 hours. As Sir Napier Shaw remarked, "the wildest guess ... would not have been wider of the mark ..." Yet, Richardson claimed that his forecast was "... a fairly correct deduction from a somewhat unnatural initial distribution"; he ascribed the unrealistic value of pressure tendency to errors in the observed winds, leading to a spuriously large value of the calculated divergence. This large tendency reflects the fact that the atmosphere can support motions with a great range of timescales.