An Experiment in Non-Linear Prediction

View More View Less
  • 1 The Johns Hopkins University, Baltimore, Md.
© Get Permissions
Full access

Abstract

The role of statistics in numerical weather prediction should be to aid in the handling of random errors and disturbances. The actual prediction is a problem in dynamics. When the initial conditions are observed with error, there is information in the past forecasts which could increase the accuracy of the numerical predictions. The techniques of control theory provide an optimal method for combining past forecasts with current observations. This paper demonstrates the method on simulated non-linear time series.

Abstract

The role of statistics in numerical weather prediction should be to aid in the handling of random errors and disturbances. The actual prediction is a problem in dynamics. When the initial conditions are observed with error, there is information in the past forecasts which could increase the accuracy of the numerical predictions. The techniques of control theory provide an optimal method for combining past forecasts with current observations. This paper demonstrates the method on simulated non-linear time series.

Save