A SIMPLE FORECASTING METHOD APPLIED TO RUNNING WATERS QUALITY TIME-SERIES

Missing values, non-regular sampling censored data and departure from normality are common problems in the analysis of ecological or environ mental time series. For over 20 years, in the scope of the National Ba sin Network in France, numerous water quality time series have been co llected A method is proposed for the regularization of the series, the detection of trend changes and short-term forecasting, on the basis o f the last detected trend, using a non-parametric regression. The meth od is applied to phosphates and nitrogen concentrations measured on th e Sambre river (northern France).