Date of Award

January 2017

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Civil Engineering

First Advisor

Yeo Howe Lim

Abstract

Streamflow predication is an important task in water management studies. It is needed in the operation and optimization of water resources and flood control projects. The accuracy of these predictions has a great influence on the water resources management and decision making processes. Various models and tool packages have been developed for simulation and prediction of streamflow. Among them, the Soil and Water Assessment Tool (SWAT) is one of the most widely used models, which was originally developed to predict the impacts of land management on water, sediment and agricultural chemical yield in large watershed simulations. Results of the SWAT streamflow simulations have indicated that this tool has deficiencies in simulating the peaks in streamflow generated by snow melting processes in the cold regions. Since global temperature is projected to be increased and the phenomena will change the snow melting characteristics in the snow dominant areas, such as the time of first melt and rate of melting. This trend along with more precipitation will cause more flooding problems in these regions. To improve daily streamflow prediction in these regions, two methods were developed. Firstly, a method was performed by separation of winter and summer seasons simulated streamflow with subsequent validation conducted in two different seasons using Calibration Uncertainty Procedure (SWAT_CUP). It should be noted that sensitivity analysis was performed on each of the seasons separately. The second method was conducted based on coupling Artificial Neural Networks (ANNs ) with calibrated and validated results of SWAT_CUP without any separation of the seasons. The calibrated streamflow, precipitation, maximum temperature, minimum temperature, snow depth, wind speed, and relative humidity were used as inputs to the ANNs model. The results of both methods have indicated significant improvements in the simulated series. In comparison between these two methods, the operation of the second method is considered better than the first method. Although, the first method has shown improvement in the simulated results but there is still a difference between the peak streamflow and the measured streamflow by USGS (United State Geological Survey) stations. However, this difference was found diminished in the simulations using the second method. ANNs method have increased peak streamflow predication in about 70%. With this improvement, the weakness of the SWAT model in simulating sediment accumulation due to improper peak run off simulation was eliminated.

Share

COinS