Adaptive prediction: a sequential approach to forecasting and estimation of nonstationary environmental systems

The interplay between prediction and estimation is crucial in adaptive prediction problems. In its general form, the adaptive prediction problem is a difficult multiperiod optimization problem and thus too complex for practical applications. By minimizing the one‐step‐ahead forecasting error, subject to a constraint on the (weighted) trace of the inverse on the one‐step‐ahead information matrix we arrive at a very simple approach, which should be interesting for more complex situations also. In the context suggesting the problem, a pollution dispersion model with time‐varying parameters is considered, and the target is to obtain a sensible and easily implementable adaptive predictor.