Comparative analysis of three recursive real-time river flow forecasting models: Deterministic, stochastic, and coupled deterministic-stochastic (Q1114281)

The authors introduce three methods for river flow forecasting: (i) Deterministic discrete linear cascade model (DLCM); (ii) Stochastic ARMAX model; (iii) DLCM where the forecasting errors are described by an ARMA process. The forecast efficiency is measured by \[ \eta_ k=\{1- [\sigma_{\epsilon}(k)/\sigma_{\Delta}(k)]^ 2\}^{1/2} \] where \(\sigma_{\epsilon}(k)\) is the standard deviation of the k-step-ahead forecasting error and \[ \sigma_{\Delta}(k)=\{t^{- 1}\sum^{t}_{i=1}[\Delta y_ i(k)-\Delta \bar y(k)]^ 2\}^{1/2} \] \[ with\quad \Delta y_ i(k)=y_{i+k}-y_ i,\quad \Delta \bar y(k)=t^{-1}\sum^{t}_{i=1}\Delta y_ i(k). \] For a comparison, daily river flows in 1981 for the Budapest-Dunaföldvár reach of the river Danube were analyzed by each of the three methods. The DLCM alone gave the least efficient forecasts and the combination DLCM-ARMA gave better results than the ARMAX model.