Robust recursive estimation of auto-regressive updating model parameters for real-time flood forecasting

In this paper, a new robust recursive method of estimating auto-regressive updating model parameters for real-time flood forecasting using weighted least squares with a forgetting factor is described. The proposed robust recursive least squares (RRLS) method differs from the conventional recursive least squares method by the insertion of a non-linear transformation of the residuals. The RRLS algorithm takes into account the contaminated Gaussian nature of the gross errors for the observed discharge, and assigns less weight to a small portion of large residuals, and gives unity weight to the bulk of moderate residuals generated by the nominal Gaussian distribution. It is the reason why the RRLS method is insensitive to outliers. The proposed method has the potential to give less biased estimates in the presence of outliers. The feasibility of the robust approach is demonstrated with synthetic and real data.