| Summary: | Thesis (M.Eng.)--Memorial University of Newfoundland, 1995. Engineering and Applied Science Bibliography: leaves 88-90. The general purpose of forecasting is to provide the best estimates of what will happen at specified points in time in the future. In hydrology, for example, forecasts of riverflows are often used for operational planning of reservoir and flood control systems. Since, even modest improvements in the operation of a large reservoir system can result in multi-million dollar savings per year, choosing a model which produces reliable and accurate forecasts is therefore essential to the efficient operation of the system. In this study, monthly and quarterly discharge data of Newfoundland rivers were used to forecast future flows using four different statistical approaches: conventional Box and Jenkins's autoregressive integrated moving average (ARIMA), exponential smoothing, periodic autoregressive (PAR), and Harvey's new structural time series (NSM). Each monthly riverflow data was divided into three short term series to study forecasting accuracy. Ten quarterly series were used to predict flows for three forecasting scenarios and thirty monthly series were considered for 3 month, 6 month, 9 month and 12 month ahead forecast horizons. Forecast performance was assessed using the mean absolute percentage error (MAPE) criterion. -- Based on the MAPE criterion, it is concluded that forecasts using the NSM approach for short term monthly riverflow data in general are better than ARIMA, exponential smoothing and PAR approaches. For quarterly data, forecasts using the exponential smoothing approach in general are better than NSM, ARIMA and PAR approaches.
|
|---|