Khan, UsmanSnieder, Everett Joshua2019-11-222019-11-222019-082019-11-22http://hdl.handle.net/10315/36792Floods are the most frequent and costly natural disaster in Canada. Flow forecasting models can be used to provide an advance warning of flood risk and mitigate flood damage. Data-driven models have proven to be suitable for flow forecasting applications, yet there are several outstanding challenges associated with model development. Firstly, this research compares four methods for input variable selection for data-driven models, which are used to minimize model complexity and improve performance. Next, methods for reducing the temporal error for data-driven flood forecasting models are investigated. Two procedures are proposed to minimize timing error: error weighting and least-squares boosting. A class of performance measures called visual measures is used to discriminate between timing and amplitude errors, and hence quantifying the impacts of each correction procedure. These studies showcase methods for improving the performance of flow forecasting models, more reliable flood risk predictions, and better preparedness for flood events.Author owns copyright, except where explicitly noted. Please contact the author directly with licensing requests.Civil engineeringElectronic Thesis or Dissertation2019-11-22flow forecastingfloodingartificial neural networksmachine learninginput variable selectionpartial correlationpartial mutual informationinput omissionneural pathway strength analysistiming erroramplitude errorleast-squares boostingerror weightingvisual performance measurespeak differencehydrograph matchingseries distance