Search

Title: Improving Medium- and Long-Range Hydrological Forecasts with Ensemble Meteorological Forecasts and Climatic Information, Author: Getnet Y. Muluye, Location: Mills The ability to provide reliable and accurate medium- and long-range hydrological forecasts is fundamental for the effective operation and management of water resources systems. The principal objectives of this thesis are (i) to develop a framework for advancing the long-range forecasting skills of hydrological models by coupling pertinent and leading climate information with regional hydro-meteorological variables; and (ii) to develop effective mechanisms for integrating meteorological ensemble systems in a hydrologic prediction system, which would be useful for risk analysis by policy makers for operating both large-scale as well as small-scale water resources systems. This research constitutes three principal components: long-range forecasts, downscaling, and medium-range forecasts. For long-range hydrological forecasting, four data-driven models, including multilayer perceptron (MLP), time-lagged feedforward network (TLFN), Bayesian neural network (BNN) and recurrent multilayer perceptron (RMLP) were designed by incorporating low-frequency climatic indices to forecast seasonal reservoir inflows. The results indicated that the incorporation of modes of climatic indices in a hydrologic forecasting model resulted in a considerable improvement in the seasonal forecast accuracy. Furthermore, the extended Kalman filter approach was used to train the recurrent multilayer perceptron for capturing the complexity associated with the long range streamflow forecasting. Results showed that the proposed methodology was able to provide a robust modeling framework capable of capturing the complex dynamics of the hydrologic system. Different statistical methods were developed and evaluated for downscaling local scale information of precipitation and temperature from the numerical weather prediction model output. Three different methods were considered: (i) hybrids; (ii) neural networks; and (iii) nearest neighbor-based approaches. The findings revealed that the skills in the downscaled temperature forecasts were superior to those in the downscaled precipitation forecasts. In particular, for downscaling daily precipitation, the artificial neural network-logistic regression (ANN-Logst), partial least squares (PLS) regression and recurrent multilayer perceptron trained with the extended Kalman filter (EKF) models yielded greater skill values, and the conditional resampling method (SDSM) and K-nearest neighbor (KNN) based models showed potential for characterizing the variability in daily precipitation. For the case of medium-range hydrological forecasting, the downscaled and the raw numerical model outputs were forced into an HBV hydrologic model in order to generate an ensemble of reservoir inflows. The simulation results indicated that the downscaled-based flows had greater skill values, and yielded more accurate forecasts than the raw-based flows. The potential economic values of flow forecasts were further assessed based on a simple optimal decision-making, cost-loss analysis technique. The principal outcomes emerging from the analyses included: (i) the economic benefits associated with probabilistic flow forecasts were more useful than their deterministic counterparts; and (ii) the downscaled-based flow forecasts offered greater benefits, which are applicable to a much wider range of users, than the raw-based flow forecasts. Thesis Doctor of Philosophy (PhD)