Multiscale error analysis, correction, and predictive uncertainty estimation in a flood forecasting system

摘要: [1] River discharge predictions often show errors that degrade the quality of forecasts. Three different methods error correction are compared, namely, an autoregressive model with and without exogenous input (ARX AR, respectively), a method based on wavelet transforms. For method, Vector-Autoregressive (VARX) is simultaneously fitted for levels decomposition; after predicting next time steps each scale, reconstruction formula applied to transform in domain back original domain. The combined Hydrological Uncertainty Processor (HUP) order estimate predictive conditional distribution. three stations along Danube catchment, using output from European Flood Alert System (EFAS), we demonstrate wavelets outperforms simpler uncorrected respect mean absolute error, Nash-Sutcliffe efficiency coefficient (and its decomposed performance criteria), informativeness score, particular forecast reliability. approach efficiently accounts scale properties unknown source statistical structure.