To meet the surging water demand due to rapid population growth and changing climatic conditions around the world, and to reduce the impact of floods and droughts, comprehensive water management and planning is necessary. Climatic variability, hydrologic uncertainty and variability of hydrologic quantities in time and space are inherent to hydrological modeling. Hydrologic modeling using a physically-based model can be very complex and typically requires detailed knowledge of physical

more »

... The availability of data is an important issue to justify the use of these models. Data-driven models are an alternative choice. This is a relatively new and efficient approach to modeling. Data-drive models bridge the gap between the classical regression and physically-based models. By using a data-driven model that relies on the machine learning approach, it is possible to produce reasonable predictions from a limited data set and limited knowledge of underlying physical processes of the system by just relating input and output. This dissertation uses the Multivariate Relevance Vector Machine (MVRVM) and Support Vector Machine (SVM) for predicting a variety of hydrological quantities. These models are used in this dissertation for identifying influential climate indicators, and are used for long-term streamflow prediction for multiple lead times at different locations in Utah. They are also used for prediction of Great Salt Lake (GSL) elevation series. They provide reasonable predictions of hydrological quantities from the available data. The predictions from these models are robust and parsimonious. This research presents the first attempt to identify influential climate indicators and predict long lead-time streamflow in Utah, and to predict lake elevation using machine learning models. The approach presented herein has potential value for water resources planning and management especially for irrigation and flood management.