Please use this identifier to cite or link to this item:
|Title:||Downscaling Meteorological Predictions for Short-Term Hydrologic Forecasting|
|Keywords:||Meteorological Predictions;Hydrologic Forecasting;Short-Term;weather predictions|
|Abstract:||<p> This study investigates the use of large scale ensemble weather predictions provided by the National Centers for Environmental Prediction (NCEP) medium range forecast (MRF) modeling system, for short-term hydrologic forecasting. The weather predictors are used to downscale daily precipitation and temperature series at two meteorological stations in the Saguenay watershed in northeastern Canada. Three data-driven methods, namely, statistical downscaling model (SDSM), time lagged feedforward neural network (TLFN), and evolutionary polynomial regression (EPR), are used as downscaling models and their downscaling results are compared. The downscaled results of the best models are used as additional inputs in two hydrological models, Hydrologiska Byrans Vattenbalansavdelning (HBV) and Bayesian neural networks (BNN), for up to 14 day ahead reservoir inflow and river flow forecasting. The performance of the two hydrological forecasting models is compared, the ultimate objective being to improve 7 to 14 day ahead forecasts. </p> <p> The downscaling results show that all the three models have good performance in downscaling temperature time series, the correlation between the observed and downscaled data is more than 0.90, however the downscaling results are less accurate for precipitation, the correlation coefficient is no more than 0.62. TLFN and EPR models have quite close performance in most cases, and they both perform better than SDSM. </p> <p> Therefore the TLFN downscaled meteorological data are used as predictors in the HBV and BNN hydrological models for up to 14 day ahead reservoir inflow and river flow forecasting, and the forecasting results are compared with the case where no downscaled data is included. The results show that for both reservoir inflow and river flow, HBV models have better performance when including downscaled meteorological data, while there is no significant improvement for the BNN models. When comparing the performance of HBV and BNN models through scatter plots, it can be found that BNN models perform better in low flow forecasting than HBV models, while less good in peak flow forecasting. </p>|
|Appears in Collections:||Digitized Open Access Dissertations and Theses|
Files in This Item:
|Liu_Xiaoli_2007Jun_Masters.pdf||8.9 MB||Adobe PDF||View/Open|
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.