Summary:

Four methods of forecasting: „no-change", LOESS, local linear regression and Holt-Winters were applied to annual minimum water levels observed at ten cross-sections of two tributaries of the Vistula river. The 1-, 2-, ..., 5-year forecasts were made for each year after some initial year, and four quality measures: bias, root mean square error, mean absolute error and maximum absolute error were calculated for each time series and lead time. The naïve model turned out to be always the worst in it bias and almost always very good, sometimes the best regarding the other measures.